INFORMATION TO USERS The most advanced technology has been used to photo­ graph and reproduce this manuscript from the microfilm master. UMI films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter face, while others may be from any type of computer printer. The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send UMI a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright m aterial had to be removed, a note will indicate the deletion. Oversize materials (e.g., maps, drawings, charts) are re­ produced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps. Each original is also photographed in one exposure and is included in reduced form at the back of the book. These are also available as one exposure on a standard 35mm slide or as a 17" x 23" black and w hite photographic p rin t for an additional charge. Photographs included in the original m anuscript have been reproduced xerographically in this copy. H igher quality 6" x 9" black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact UMI directly to order. University Microfilms International A Bell & Howell Information C om p an y 3 0 0 North Z e e b Road, Ann Arbor, Ml 4 8 1 0 6 -1 3 4 6 USA 3 1 3 /7 6 1 -4 7 0 0 8 0 0 /5 2 1 -0 6 0 0 Order N um ber 8912563 V ariation in M ichigan hospital use rates: D o physician and hospital characteristics provide the explanation? Clark, Jane Deane, Ph.D. Michigan State University, 1988 UMI 300 N. Zeeb Rd. Ann Arbor, MI 48106 VARIATION IN MICHIGAN HOSPITAL USE RATES: DO PHYSICIAN AND HOSPITAL CHARACTERISTICS PROVIDE THE EXPLANATION? By Jane Deane Clack A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Geography ABSTRACT VARIATION IN MICHIGAN HOSPITAL USE RATES: DO PHYSICIAN AND HOSPITAL CHARACTERISTICS PROVIDE THE EXPLANATION? By Jane Deane Clark Previous small area analysis studies have shown that hospital admission rates (total, medical and surgical) vary among hospital service areas. Using 1983 Michigan hospital inpatient data from 53 nonMetropolitan Detroit lower peninsula hospital service areas, one physician characteristic and thirteen hospital characteristics (in the categories of resource supply, services offered and organization) were tested for their association with and explanation of fourteen hospital use rates. Registered nurses per bed and the weighted proportion of board certified physicians to total physicians were inversely related to and offered significant contribution to the explanation of the variation in total use rates and in four medical causes for admission rates (cir­ culatory, respiratory, digestive and genito-urinary). Physician and hospital variables provided significant explanation for six of the seven surgical procedure rates tested (appendectomy, hemorrhoidectomy, chole­ cystectomy, inguinal hernia repair, prostatectomy and hysterectomy). Four causative factors derived from the characteristics studied are postulated to influence the hospital use rates. The first factor was the small rural nature of the average high use hospital service area. High use areas had a lower proportion of board certified physicians and fewer RNs per bed, beds per hospital, and house staff per 10,000 population than did low use areas. Another factor was the inequality in the distribution of high technology diagnostic services. High use hospital service areas had fewer diagnostic services than did low use areas. The third factor was the inequality in the rural hospital environment produced by the designation of some hospitals as rural referral centers. The fourth factor was the impact of the definition and size of a hospital service area. Current small area analysis methodology assigns every small area to a hospital service area, no matter what the probability of the population using the hospital(s) within the service area. This research questions that methodology, and suggests the need for hospital service area definitions based upon the specific diagnosis or procedure being studied and postulates that some rural hospital distance decay curves may turn upward at farther distances when the timing of treatment is too critical to allow patients to return to distant residences. Dedicated to Russell G. Clark, for shared moments, minutes, months ... a lifetime. ACKNOWLEDGEMENTS The author wishes to acknowledge the financial support provided by the Michigan Hospital Association and the Department of Geography, Michigan State University. I am grateful for the MHA's permission to use several data bases and for its assistance in gaining access to others. I also wish to thank the MHA for the use of the word processing facilities at the association's headquarters in Lansing. I am indebted to Dr. Philip TedescM for allowing me to use the 1983 hospital service areas defined by the small area analysis research team at the University of Michigan. The author also wishes to thank policy staff members at the MHA who have provided keen observations and valuable Insight Into the hospital industry: Charles Mannlnen, Charles Ellsteln, Dennis Paradis, Marlene Soderstrom, Richard A. Hamilton, and Robert Zorn. Several hospital administrators and consultants have also listened to the hypotheses and commented on the results. Among them I wish to acknowledge and thank Melvin Creeley, Robert Irwin, Phil McCorkle, Jack Weiner, Helen Thompson, Jeff Schilling, and Larry Willis. All these people gave freely of their time and expertise. In addition to these people, I wish to thank my sister, Mary-Anne Deane Rappole, my mother-in-law, Harriett Harper Clark, my "adopted parents," Joseph J. and Laurent1a Irwin, and the many MHA staff members and friends who have offered encouragement throughout this long process. And, in particular, I wish to acknowledge the rooting section v in the Research, Data Policy and Services Department led by my staff who really "make the difference," Julie Lester and Hilda Michaels. I wish my parents could have lived to see the successful completion of this research because they are the ones who provided my educational and emotional foundations and insisted that any task undertaken must be completed. This document has a polished look in large part because of the tremendous skills of Hilda Michaels with the NBI word processing system and Russell G. Clark, Jr. with his innovative use of the graphics capabilities of the IBM AT and his precise editing. My cartographic efforts for Russell's dissertation have been repaid many many times over. My special thanks to Bruce Pigozzi for his advice, counsel, and rigorous editing and to my committee members for their patience. The inspiration forthis work comes from John Wennberg, Phil Caper, Noralou and Leslie Roos, John Griffith, Peter Wilson, and Phil Tedeschi. They are not geographers; one has, in fact, challenged me to tell him what geography 1s. But they have used an inherently geographic method to Influence health policy, and that is Inspiring. Inspiration alone does not get the job done, and I could not have completed this research, nor the requirements for a doctoral degree, without the active support and love of the three dearest people in the world to me: Amy, Ian, and Russell. Without our four-way partnership, I could not have gotten past the household chores and parenting that they willingly accepted as their part of this shared activity. And finally, to the three men in my life who assured me that I could do it and then insisted that I do — Bruce Pigozzi, Richard Hamilton, and Russell Clark. Thank you. vi TABLE OF CONTENTS .Page List of Tables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x List of Figures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii I. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Background. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Conceptual Approach. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Small Area Analysis Methodology. . . . . . . . . . . . . . . . . . . . 6 II. Hospital Utilization Model. . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Andersen Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Clark Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Component 1: The Individual. . . . . . . . . . . . . . . . . . . 12 Component 2: TheCommunity. . . . . . . . . . . . . . . . . . . . 12 Component 3: ThePhysician. . . . . . . . . . . . . . . . . . . . 18 Component 4: TheHospital. . . . . . . . . . . . . . . . . . . . . 21 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 Definition of Hospital Service Area. . . . . . . . . . . . . . . . . 37 Small Area Analysis Definitions. . . . . . . . . . . . . . . . . 42 III. Methods and General Hypotheses. . . . . . . . . . . . . . . . . . . . . . . 48 Focus of the Study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 Study Area and Time Frame. . . . . . . . . . . . . . . . . . . . . . . . 49 Sources of the Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Procedure and Diagnosis Occurrences. . . . . . . . . . . . . . 55 Population. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Physician and Hospital Characteristics. . . . . . . . . . . . 58 Research Design. . . . . . . . . . . . . . . . . . . 58 Goal 1: Document Variation in Use Rates and . i a. « uumpai c rvcdu i id IV. i. A n . . • n i jl - cu r i cv IUU5 rvc^u I C5 from Michigan and Elsewhere. . . . . . . . . . . . . 59 Goal 2: Analyze the Use Rates. . . . . . . . . . . . . . . . . 65 Goal 3: Determine the Relationships Between and Among the Hospital Use Rates and Provider Characterisitcs. . . . . . . . . . . . . . . 66 Goal 4: Use Rates as a Function of Physician and Hospital Characteristics. . . . . . . . . . . . 74 General Hypotheses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 Physician Component. . . . . . . . . . . . . . . . . . . . . . . . . 77 Hospital Component. . . . . . . . . . . . . . . . . . . . . . . . . . 78 Michigan Hospital Use Rates. . . . . . . . . . . . . . . . . . . . . . . . . 85 Documentation of Variation in Hospital Use Rates Found in this Study. . . . . . . . . . . . . . . . . . . . . . . . . . . 85 Comparison of the Variation in Use Rates Reported In this Study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Total Hospital Use Rates. . . . . . . . . . . . . . . . . . . . . . 89 Surgical Procedure Rates. . . . . . . . . . . . . . . . . . . . . . 89 Medical Causes for Admission Rates. . . . . . . . . . . . . . 105 vi i V. VI. VII. Patterns in Michigan Hospital Use Rates. . . . . . . . . . . . ,. . . 109 Four Use Rate Patterns. . . . . . . . . . . . . . . . . . . . . . . . . 109 Pattern 1: Consistency in Procedure Specific Variation Ranking. . . . . . . . . . . . . . . . . 109 Pattern 2: Variation in Medical Causes for Admission Rates Consistently Greater Than Variation in Surgical Procedure Rates. . . . . . . . . . . . . . . . . . 115 Pattern 3: Unique Surgical Procedure Rate Patterns Within a Hospital Service Area. . . . . . . . . . . . . . . . . . . . . . . . . . 116 Pattern 4: Consistency Within Hospital Service Areas of High Use or Low Use Across Several Procedures orDiagnoses. . . . . . . . 122 The Relationships Between and Among the Hospital Use Rates and Provider Characteristics. . . . . . . . . . . . . . . . . . . 149 Correlations Among the Hospital UseRates. . . . . . . . . . . . 149 Correlations Among the Provider Characteristics153 Correlations Between the Hospital Use Rates and the Provider Characteristics. . . . . . . . . . . . . . . . . . . . 155 Total Hospital Use Rates and Provider Characteri st ics. . . . . . . . . . . . . . . . . . . . . . . . . . 155 Surgical Procedure Rates and Provider Characteri sti cs. . . . . . . . . . . . . . . . . . . . . . . . . . 155 Medical Causes for Admission Rates and Provider Characteristics. . . . . . . . . . . . . . . . . . . . 157 Comparison of the General Hypotheses with Results of the Correlations. . . . . . . . . . . . . . . . . . . . . 158 Hospital Use Rates as a Function of Physician and U /*«•»**41 i i v ^ p i w u i v i i u i w c i 4-4 1 X U U Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Regressions of Age and Sex Adjusted Dependent Variables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Total Male, Total Female, and Total Admission Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendectomy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hemorrhoi dectomy. . . . . . . . . . . . . . . . . . . . . . . . . . Cholecystectomy. . . . . . . . . . . . . . . . . . . . . . . . . . . Inguinal Hernia Repair, Hysterectomy, and Cesearean Section. . . . . . . . . . . . . . . . . . . . . . . . . Prostatectomy. . . . . . . . . . . . . . . . . . . . . . . . . . . . Circulatory Admissions. . . . . . . . . . . . . . . . . . . . . . Respiratory Admissions and Digestive Admissions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Genito-Urinary Admissions. . . . . . . . . . . . . . . . . . . . vi ii 160 161 161 168 172 176 180 186 190 194 199 VIII. Summary and Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . 206 Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 Research Goal 1: Documentation of Variation inHospital Use Rates. . . . . . . . . . . 207 Research Goal 2: Analysis of Medical Practice Patterns. . . . . . . . . . . . . . . . . . . 210 Research Goal 3: Determination of the Relation­ ships between and among the Hospital Use Rates and Provider Characteristics. . . . . . . . . . . . . . 214 Research Goal 4: Hospital Use Rates as a Function of Physician and Hospital Characteri st ics. . . . . . . . . . . . . . 216 Conclusions and Future Research. . . . . . . . . . . . . . . . . . . 220 Factor 1: The Small Rural Nature of High Use Hospital Service Areas. . . . . . . . . . . . . . 221 Factor 2: Inequality of the Distribution of High Technology Diagnostic Equipment. . . . . . . . 225 Factor 3: Rural Referral Centers. . . . . . . . . . . . . . . 226 Factor 4: The Impact Upon Hospital Use Rates of the Size and Definition of the Hospital Service Area. . . . . . . . . . . . . . . . . . . . . . 229 IX. Glossary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 X. Appendices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 Appendix A: Hospital Use Rates by Hospital Service Area. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 Appendix B: Physician and Hospital Characteristics by Hospital Service Area. . . . . . . . . . . . . . . . . 240 Appendix C: Comparison of Hospital Use Rate by Hospital Cov*t/* 5r o Av*oa W K jr w VCfVUI IUUI * • V S. I II w u WWI XI. V i How a af i a w ww * I U V I Wi l l * • • • • • • • • « List of References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 ix LIST OF TABLES Page 2.1 2.6 2.7 Correlations Between Community Characteristics and Hospital Utilization. . . . . . . . . . . . . . . . . . . . . . . . . . 14 Percent of Total Variation Explained by Community and Provider Characteristics. . . . . . . . . . . . . . . . . . . . . . . 15 Correlations Between Community Characteristics and Surgery Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Relationship Between Hospital Bed Supply and Utilization...23 Relationship Between Non-Specialist Physician Supply and Utilization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Relationship Between Surgeon Supply and Utilization. . . . . . 27 Relationship Between Physician Supply and Utilization. . . . 29 3.1 Patient Migration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 4.1 4.2 Results of Three Measures of Use Rate Variation.. . . . . . . . 86 Use Rates Ranked within Each of Three Measures of Variation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Total Admission or Discharge Rates. . . . . . . . . . . . . . . . . . 90 Appendectomy Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 Hemorrhoidectomy Rates. . . . . . . . . . . . . . . . . . . . . . . . . . 94 Cholecystectomy Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . 95 Inguinal Hernia Repair Rates. . . . . . . . . . . . . . . . . . . . . . 97 Prostatectomy Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Hysterectomy Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 C-Sectior. Rates. . . . . . . . . . . . iG3 Comparison of Systematic Component of Variation Results...104 Medical Causes for Admission Rates. . . . . . . . . . . . . . . . . 107 2.2 2.3 2.4 2.5 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 4.11 4.12 5.1 5.2 5.3 5.4 6.1 6.2 6.3 6.4 Medical and Surgical Causes of Admissions Ranked in Ascending Order of Variation in Incidence of Hospitalization (1980-1982). . . . . . . . . . . . . . . . . . . . 112 Comparison of Variation Ranking by Systematic Component of Variation. . . . . . . . . . . . . . . . . . . . . . . . 113 High Use Hospital Service Areas in Michigan. . . . . . . . . . 144 Low Use Hospital Service Areas in Michigan. . . . . . . . . . . 146 Spearman Correlations of Use Rates. . . . . . . . . . . . . . . . . Spearman Correlations of Provider Characteristics. . . . . . Spearman Correlation of Use Rates with Provider Characteristics. . . . . . . . . . . . . . . . . . . . . . Comparison of Hypotheses and Significant Correlation Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x 150 154 156 159 7.1 7.19 Multiple Regression Equation Results Separately Run for Total Male, Total Female, and Total Admission Rates. . . . . . . . . . . . . . . . . . . . . . . . . 162 Significant Variables for Total Male, Total Female, and Total Admission Regressions. . . . . . . . . . . . . . . . . 164 Multiple Regression Equation Results for Appendectomy Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . 169 Significant Variables for Appendectomy Regression. . . . . . 170 Multiple Regression Equation REsults for Hemorrhoidectomy Rates. . . . . . . . . . . . . . . . . . . . . . . . 173 Significant Variables for Hemorrhoiedectomy Regression....174 Multiple Regression Equation Results for Cholecystectomy Rates. . . . . . . . . . . . . . . . . . . . . . . . . 177 Significant Variables for Cholecystectomy Regression.... 178 Multiple Regresssion Equation Results Separately Run for Inguinal Hernia Repair, Hysterectomy and Ceserean Section Rates. . . . . . . . . . . . . . . . . . . . . 181 Significant Variables for Inguinal Hernia Repair and Hysterectomy Regressions. . . . . . . . . . . . . . . . . . . . 182 Multiple Regression Equation Results for Prostatectomy Rates. . . . . . . . . . . . . . . . . . . . . . . . . . 187 Significant Variables for Prostatectomy Regression. . . . . . 187 Multiple Regression Equation Results for Clrculatorey Causes for Admission Rates. . . . . . . . . . . . 191 Significant Variables for Circulatory Causes for Admission Regression. . . . . . . . . . . . . . . . . . . . . . 191 Multiple Regression Equation Results Run Separately for Respiratory and Digestive Causes for Admission Rates. . . . . . . . . . . . . . . . . . . . . 195 Significant Variables for Respiratory and Digestive Admission Regressions. . . . . . . . . . . . . . . . . 196 Multiiple Regression Equation Results for Gen1to-Ur1nary Admission Rates. . . . . . . . . . . . . . . . . . 200 Significant Variables for Genito-Urinary Admission Regression. . . . . . . . . . . . . . . . . . . . . . . . . 201 Multiple Regression Results: Beta Weights. . . . . . . . . . . . 204 8.1 Comparisons of High Use and Low Use Areas. . . . . . . . . . . . 222 9.1 9.2 Appendix A: Hospital Use Rates by Hospital Service Area..238 Appendix B: Physician and Hospital Characteristics by Hospital Service Area. . . . . . . . . . . . . . 240 Appendix C: Comparison of Hospital Use Rate by Hospital Service Area by Standard Deviation. . . . . . . 242 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 7.11 7.12 7.13 7.14 7.15 7.16 7.17 7.18 9.3 xi LIST OF FIGURES Page 1.1 2.1 2.2 2.3 2.4 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 C Q 5.10 5.11 5.12 5.13 5.14 5.15 5.16 Anderson Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Clark Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Distance Decay Curve. . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Wennberg's Plurality Definition of a Hospital Service Area..44 Griffith's Relevance Index Definition of a Hospital Service Area. . . . . . . . . . . . . . . . . . . . . . . . . . 46 Michigan Hospital Service Areas Included inStudy Area. . . . . 50 Sole Community Provider Hospital ServiceAreas . . . . . . . . . 57 Measures of Hospital Use. . . . . . . . . . . . . . . . . . . . . . . . . 59 Surgical Procedure Codes. . . . . . . . . . . . . . . . . . . . . . . . . 62 Medical Causes for Admission Codes. . . . . . . . . . . . . . . . . . 63 Physician and Hospital Characteristics Studied. . . . . . . . . 67 Variety of Relationships Reported in theLiterature. . . . . . . 76 Relationships Hypothesized in the Methods Chapter. . . . . . . 84 Surgical Signatures for Five Hospital Service Areas in Maine. . . . . . . . . . . . . . . . . . . . . . . . 117 Surgical Signatures -- High Use Areas. . . . . . . . . . . . . . . 119 Surgical Signatures — Low Use Areas. . . . . . . . . . . . . . . . 121 Michigan Hospital Service Area Regions. . . . . . . . . . . . . . 123 Total Admissions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Appendectomy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 Hemorrhoidectomy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 Cholecystectomy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 T n m i ^I IwI aU lI 4 l l ^ w U n vIIn IAU» M W * I N W ^ U II • Prostatectomy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hysterectomy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cesearean Section. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Circulatory Diagnoses. . . . . . . . . . . . . . . . . . . . . . . . . . Respiratory Diagnoses. . . . . . . . . . . . . . . . . . . . . . . . . . ....................... Digestive Diagnoses Genito-Urinary Diagnoses. . . . . . . . . . . . . . . . . . . . . . . . xii 1^1 133 135 136 138 139 141 142 LIST OF FIGURES (continued) Page 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 7.11 7.12 7.13 7.14 7.15 7.16 7.17 7.18 7.19 7.20 7.21 7.22 8.1 8.2 8.3 Hypotheses for Total Male, Total Female, and Total Admission Rate Regression. . . . . . . . . . . . . . . . . . 162 Male Admissions Multiple Regression Analysis Residuals. . . 165 Female Admissions Multiple Regression Analysis Residuals...166 Total Admissions Multiple Regression Analysis Residuals__ 167 Hypothesis for Appendectomy Use Rate Regression. . . . . . . . 168 Appendectomy Multiple Regression Analysis Residuals. . . . . 171 Hypothesis for Hemorrhoidectomy Use Rate Regression. . . . . 172 Hemorrhoidectomy Multiple Regression Analysis Residuals....175 Hypothesis for Cholecystectomy Use Rate Regression. . . . . . 176 Cholecystectomy Multiple Regression Analysis Residuals. . . 179 Hypotheses for Inguinal Hernia Repair, Hysterectomy and Ceserean Section Use Rate Regressions. . . . . . . . . . . 180 Inguinal Hernia Repair Multiple Regression Analysis Residuals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 Hysterectomy Multiple Regression Analysis Residuals. . . . . 185 Hypothesis for Prostatectomy Use Rate Regression. . . . . . . 186 Prostatectomy Multiple Regression Analysis Residuals. . . . 189 Hypothesis for Circulatory Admission Rate Regression. . . . 190 Circulatory Diagnoses Multiple Regression Analysis Residuals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 Hypotheses for Respiratory and Digestive Admission Rate Regressions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 Respiratory Diagnoses Multiple Regression Analysis Residuals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Digestive Diagnoses Multiple Regression Analysis Residuals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 Hypothesis for Genito-Urinary Admission Rate Regressions...199 Genito-Urinary Diagnoses Multiple Regression Analysis Residuals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 Frequency Histogram for RNs' per Bed. . . . . . . . . . . . . . . 223 Hypothetical Diagnosis-Specific Distance Decay Curve. . . . 234 Total Admissions Distance Decay Curve. . . . . . . . . . . . . . . 234 xiii CHAPTER 1 INTRODUCTION Background Variations in the use of health services by the population of a geographic area have long interested geographers and other health researchers. Perhaps the first attempt to analyze the geographic variation of the per capita utilization of hospital services was reported in 1856 by William A. Guy, a physician at King's College Hospital, London, and one of the Honorary Secretaries of the Statistical Society. Guy noted that the annual per capita rates for hospitalization in King's Hospital varied from 325 per 1,000 population in the parish of St. Mary-1e-Strand to one per 1,000 in the district of Marylebone. Based on these observations Guy hypothesized that charitable medical care was being consumed in parishes such as St. Mary-1e-Strand by an "increasing class of working men, in receipt of good wages, who are 1n the habit of applying to hospitals as a matter of course, even for trifling attacks of illness, to say nothing of those which sometimes follow immediately on expensive acts of self-indulgence" (Barnes, 1982). Guy clearly felt that the explanation for the variation in use rates between the two areas was related to behavioral characteristics of the populations concerned. 1 2 The study of variations in hospital use rates has continued to the present. For example, Wennberg (1982) reported that at the then current rates at which local physicians were performing hysterectomies, in one community in Maine seventy percent of the women would have their uteruses removed by the time they reached seventy-five years of age, while in a similar community less than twenty miles away only twentyfive percent of the women would have had their uteruses removed by the same age. Wennberg felt that since the communities were so similar, the explanation for the variation in hospital use rates would not be found among community variables but elsewhere, and he suggested that the explanation was related to the degree of consensus in medical diagnosis and treatment among physicians. No matter where the final answers to variations in use rates are found, the question that was asked in 1856 is essentially the same question that is asked today: what can explain the large variations in use rates between apparently similar neighboring communities? Are the explanations to be found among community characteristics as suggested by Guy, or among health care provider characteristics as suggested by Wennberg? This current research attempts to answer these and other questions about hospital utilization in Michigan. First, is there variation in hospital utilization among Michigan communities? If there is variation in hospital use rates, how does the variation compare to the results of previous utilization research? Does Michigan exhibit the same range and patterns of variation found in other geographic areas? Are there any spatial patterns to the variations? And finally, what characteristics of the hospitals and their medical staffs are related to and explain 3 variations in hospital utilization rates? Are the explanations to be found among characteristics of the physicians or among hospital characteristics which describe the supply of personnel, the services offered, the organization of the hospitals themselves, or a combination of these factors? Conceptual Approach In order to ask appropriate questions about the explanation of the variation in hospital use rates, it is necessary to define a model including the possible factors which may cause or contribute to the variation in those rates. Researchers have often tried to identify and describe those factors responsible for an individual's seeking medical care and the subsequent care provided. Andersen (1973) provided a model (Figure 1.1) describing the major elements that he believed contributed to variation in the use of health services. The major elements he identified were: the community, the health services system, and the individual. Andersen's model showed community determinants such as socioeconomic variables, the physical environment, and morbidity influencing health services system determinants and the individual. Andersen believed that both the community and the health system determinants influenced the individual and, when added to the individual's own predisposing and enabling determinants, as well as illness level, would influence the individual's utilization of the 4 HEALTH SERVICES SYSTEM DETERMINANTS community DETERMINANTS Poverty Facilities Unemployment Services SES Personnel Physical Environment Morbidity Mortality I INDIVIDUAL DETERMINANTS HEALTH SERVICES UTILIZATION Predisposing Enabling Illness Level Figure 1.1 The Anderson Model (adapted from Anderson, 1973) .5 health services within the community. Andersen did not put his model into a spatial context. My model (hereafter referred to as the Clark Model), which will be described in detail in the next chapter, is more geographic in nature since it defines the community as an area within which are located all of the participants who determine a community's hospital use rate. These participants include the individual who is the potential patient, the total aggregated individuals who make up the population, the physicians, and the hospitals. Another major difference between this model and Andersen's is the distinction made between the physicians (who act as the gatekeepers to the hospital in this model) and the hospital itself. The distinction is necessary because of the influence or control that physicians have in deciding who shall be hospitalized and therefore shall move from being an "individual" who is part of the population to also being a patient who is hospitalized. One of the major goals of this research was to identify those factors within the hospital and its medical staff which may influence the hospital use rates and to determine how much of the variation in hospital use rates can be explained by those provider characteristics. A second major goal was to examine spatially the hospital use rates, use rate patterns, and residual use rates unexplained by the multivariate regressions. 6 Small Area Analysis Methodology While the explanation for the variation in hospital use rates was sought within the four components of the model (individual, physician, hospital, community), the framework to study them comes from small area analysis methodology. Small area analysis is a method used to analyze the way individuals in a community utilize its health care resources. The approach is analogous to that used in epidemiology. In epidemiology the number of disease occurrences during a defined time period in a specific area is divided by the population at risk for contracting the disease in that area during the same time period. The result is an incidence rate for that disease. In small area analysis the use rate is analogous to the incidence rate for a disease. The number of occurrences of a health care event in an area and within a defined time period is divided by that area's population at risk. The result is a use rate which can be standardized to a given population or subpopulation at risk. Small area analysis has gained prominence as a tool for health services researchers and policy makers in part because, by defining a geographic area of observation, a resident population at risk is also defined. Therefore, not only can a use rate be calculated, but the health resources available and the population most likely to use those resources can also be defined. Researchers can then search for explanation of the community's health care use within the many physical, structural, cultural, and behavioral variables that exist for the same area. This current research used small area analysis methodology to determine how much of the variation in hospital use rates can be explained by two components of the Clark Model: the physicians and the hospital. To do this the non-metropolitan Detroit area of the lower peninsula of Michigan was chosen as the study area. Hospital service areas were defined and the characteristics of the hospitals and their medical staffs were determined. Fourteen measures of hospital use were calculated for each hospital service area and then multiple regression analysis was used to determine how much of the variation in hospital use rates could be explained by the characteristics of the hospitals and the physicians that were located within the hospital service area. CHAPTER 2 A HOSPITAL UTILIZATION MODEL To better understand and explain why there is variation in hospital use rates, one must first understand what factors influence an individual's decision to be hospitalized. These factors are best described using a model or framework for the conceptualization, hypothesis development and analysis. The previously discussed Anderson Model provided the initial conceptualization for this research problem but had to be adapted to meet my needs. The Clark Model was developed to meet those needs and to provide the framework for a discussion of the previous small area analysis literature and to serve as the basis for preliminary hypotheses about the relationships between components of the models and hospital use rates. The Clark Model also provides the geographic framework for a discussion of the definition of hospital service areas. Andersen Model In his model Andersen identified three major components which he believed contributed to health services utilization. The first of Andersen's components was the community. 8 The community determinants 9 were identified as socio-economic characteristics such as poverty and unemployment, the physical environment, and the health of the community's population as measured by morbidity and mortality. The second component in Andersen's model was the health services system which was made up of facilities, services and personnel determinants. The third component in Andersen's model was the individual who was influenced by both the community and the health services system. Andersen identified predisposing, enabling and illness level determinants within the individual. Andersen saw demographic, social structure and personal beliefs as influencing the predisposition (predisposing determinants) of any individual to seek medical care. Family and community characteristics, such as income, health insurance and the price of health care affected the individual's ability to seek medical care (enabling determinants), as did the individual's perceived and diagnosed illness level. Andersen's model provides a framework for the discussion of what factors affect an individual's decision to seek medical care. It does not provide the spatial structure which would allow the researcher to calculate use rates or to determine whether any of the determinants could help to explain the variation in use rates from one area to another. Clark Model The Clark model (Figure 2.1) is different in several respects since it divides the health services determinant into hospital and physician determinants, and adds a spatial dimension. It is more geographic in 10 connuNiTY SEltlCE AREA) HOSPITAL ,ea DEATH INDIVIDUAL Figure 2.1 The Clark Model 11 nature because it defines a hospital service area from historical use patterns and then looks at the hospital use within that area. The hospital service area contains all of the participants who determine the hospital use rates: the individual, the population (also called the community), the physicians, and the hospitals. In addition, the Clark Model acknowledges the influence that different individuals or groups have in the process of hospitalization. When individuals recognize that they are ill, they will usually seek advice from a health care provider such as a physician. The provider then determines if the individual is ill enough to be admitted to a hospital. Individuals admitted to the hospital as patients will be exposed to at least one of a series of possible hospital "events" (admission, medical diagnosis with non-surgical treatment, and/or surgical procedure). Upon completion of a series of events, the patient will either be released from the hospital or will be deceased. In either case, an outcome event has resulted from that hospitalization. Hospital events when standardized to the population at risk, become hospital event-rates, or more simply, hospital use rates. It is one of the goals of this paper to explain the variation in hospital use rates from one hospital service area to another. The Clark Model indicates that there are four components (the individual, the community, the physician, and the hospital) that may be examined in an attempt to explain such use rate variation. The Clark Model will be used as a basis from which to make that examination. Previous small area analysis literature describing the relationship between each of these components and hospital use rates will be examined and initial 12 hypotheses regarding the relationships between provider (physician and hospital) variables and hospital use rates will be suggested. These hypotheses will be further discussed and summarized in Chapter 3. Component 1: The Individual The first component in the Clark Model is the individual within the community. All individuals within the community are potential hospital patients. To become a patient, the individual must recognize a state of disease and then approach a provider for counseling and/or treatment prior to becoming a hospital patient. There are many factors such as age, level of education, socio-economic status, occupation, as well as attitudes about physicians and hospitals that can influence the individual's behavior in seeking health care. But the individual's characteristics cannot be tested as explanatory variables for variations in use rates because those characteristics exist only in that individual. However, all the individuals in a community contribute to that community's characteristics and the community's characteristics can be tested for their explanatory power as noted below. Component 2: The Community The second component of the Clark Model is the community, which actually is the population of the hospital service area and therefore includes all of the individuals (patients and non-patients), physicians, and hospital staff within the area. The community has measurable 13 socio-economic and demographic characteristics such as the level of poverty, unemployment rates, population density and average age. As shown in Tables 2.1 and 2.2 previous research using small area analysis methodology has dealt extensively with the relationships between community characteristics and hospital use rates. Small area analysis researchers have also studied the amount of explanation that community characteristics could provide for the variation in hospital and surgical procedure use rates and results are summarized in Table 2.3. Several authors have analyzed the relationship between community characteristics and measures of hospital utilization using correlation analysis. The four measures studied were admission or discharge rate (hereafter referred to as admission rate), length of stay (LOS), patient day rate, and mortality rate. A summary of their findings is shown in Table 2.1. Using multiple regression analyses, previous small area analysis researchers have also tested the power of conmiunity characteristics to explain the variation in total hospital admissions rate, LOS and patient day rate use as well as surgery rates. As shown in Table 2.2 Chiswick (1976) was able to explain 73 percent and Deacon et aT_. (1979) 49 percent of the variation in total admission rates using equations which Included demographic and health resource characteristics among others. On the other hand, Knickman & Foltz (1985), working at a standard metropolitan statistical area (SMSA) level, found very low explanatory power from a combination of socio-economic, demographic and health system characteristics. Wilson et a K (1985) were able to explain 64 percent of white and 55 percent of black patient day rates in Michigan using only community Table 2.1 CORRELATIONS BETWEEN COMMUNITY CHARACTERISTICS AND HOSPITAL UTILIZATION Study Locati on Variable Tested Anderson, 1973 New Mexico % % t % ii II ii II ii II ii II ii It ii II ti It n It Brewer & Freedman, 1982 Vermont ii ti ii it n n ii ii it it Deacon et al., 1979* " * USA Regions Roos, 1984 Man itoba •Partial correlation it labor in agriculture urban Spanish American non-white Net migration Median age Median education % unemployed Per capita income Population density factor Tax base factor Personal income factor Substandard housing Poverty Farm population Enrol lees over 75 Population density Income Total Patient Adm/Disch LOS -.57 .55 -.11 .25 .16 -.11 .19 -.03 .18 -.29 .26 -.40 -.02 .33 -.14 .15 Pt Days -.37 .55 -.25 -.01 .30 .47 .21 -.27 .31 -.37 .49 -.22 .24 .20 .27 .09 -.14 .17 .24 .38 .30 .26 Mortality -0.315 Table 2.2 PERCENT OF TOTAL VARIATION EXPLAINED BY COMMUNITY AND PROVIDER CHARACTERISTICS Total Patient Study Location Chiswick, 1976 USA States & SMSAs 192 Deacon et aj_., 1979 USA Regions Knickman .. Foltz, 1985 SMSAs WiI son et a K , ii 1985 Michigan HSAs •• N Variable Tested Adm/Disch LOS Pt Days Health sector, demographic, income, January temperature variables 73 190 Demographic and health resource variables 49 76 60 203 A3 socioeconomic, ”6 demographic and '6 health system characteristics 02 14 03 23 23 Population size, race, beds, and surgeons Population size, race, beds, and surgeons 60 32 (White) (Black) 16 . measures of mortality, education, unemployment and poverty. The Wilson group was less successful in explaining patient day rates for whites and blacks when they combined the community variables of population size and race with the provider variables of beds and surgeons. All of the studies testing the relationships between community characteristics and surgical procedure rates were done in Canada, and all but one study was done by Roos in Manitoba (Table 2.3). Positive relationships were found between total surgery rates and education, Canadian, U.S. or U.K. ancestry (Roos & Roos, 1982), income (Roos, 1984) and high economic status (Vayda et §2** 1976). Income was found to be inversely related to cataract surgery rates by Roos & Roos (1982), while Roos (1984) found significant positive correlations between community variables and hysterectomy rates only in high use rate hospital service areas. While some researchers have looked at a combination of community and provider variables, no small area analysis researcher has looked at only community variables for the sources of explanation for variation^in surgical procedure rates. Using data from the National Health Interview Surveys for 1969-1976 that had been organized into 349 - 360 primary sampling units, Mitchell & Cromwell (1982) were able to explain only nine percent of the variation in total surgery rates using a combination of twenty-three demographic, socio-economic and provider characteristics. In summary, researchers have been far more successful in explaining total admissions, LOS and patient days using community and provider variables than they have been in explaining surgical procedure rates. Table 2.3 CORRELATIONS BETWEEN COMMUNITY CHARACTERISTICS AND SURGERY RATES Total Surgery Study Location Variable Tested Roos & Roos, 1982 Manitoba Education .43 Canadian, U.S., UK ancestry .40 Income Women with one or more D&C in 1974 Women with five or more physicians visits per year Women with one or more physicians visits for vague psychological diagnoses Women seeing four or more d ifferent phys ic ians 1ncome Mother tongue French, Italian or Polish Mortality of those >25 -.10 Mortality of those >25 .32 S (male) Mortality of those >25 -.19 (female) Mortality of those >75 .42 S Mortality of those >75 .45 S (male) Income .30 S High Economic Status .77,.70 ii ii Roos, N., 1984 ti ii ii ii ii it ti «i it ii ii ii ii Roos, L., 1984 ii II ti II ii II ii II ii II ii Vayda et a 1., 1976 Canadian Provinces NS=Reported as non-significant S=Reported as significant Cataract Pros Hyst -.25 S (high rate areas) NS (all areas) NS (all areas) S (high rate areas) NS (all areas) S (high rate areas) .32 S 18 Although the community is a component in the Clark Model, community characteristics are not included within the "independent variables" in this current research design. Instead, the portion of the variation in use rates not explained by either physician or hospital characteristics is assumed to be made up of unidentified physician and hospital influences, community influences and other unexplained influences. Since the residuals following a regression analysis are specific to a hospital service area, geographic analysis of those residuals may be useful in indicating where further research, particularly among community characteristics such as the urban or rural character of an area, could help to decrease the amount of unexplained variation in use rates. Component 3: The Physician Physicians are the third component within the Clark Model. They usually have a medical practice outside of the hospital setting at the same time they serve on the medical staff within a hospital. Physicians act as the "gatekeepers" for the hospital, since their decisions control, in most instances, who will be admitted to the hospital and when that admission will occur. Like other individuals, physicians are also an integral part of the community. Their characteristics contribute to the community's characteristics and, as discussed later, the community's characteristics influence the physician. But a physician is also a health care provider, and therefore the supply of physicians as well as their characteristics are of importance in understanding the variation in hospital use rates. The supply of 19 physicians (total and by specialty group) within a hospital service area will be addressed as one of the health care resources within the hospital component of the model. In addition to the supply of physicians within a community, the characteristics of the physicians themselves are believed to explain a portion of the variation in hospital use rates. A physician's age, education, residency training, medical specialty, and the length of time in practice within the community all influence patient admitting decisions and practice style and are therefore important provider characteristics. For example, I have reasoned that a physician who did a surgical residency in one institution was taught by mentors not only how to perform surgical procedures, but also the criteria that should be used to decide if and when surgery is necessary. One might therefore assume that those people who did their surgical residency at the same institution under the guidance of the same teachers would have a surgical practice pattern that was more similar to each other's than to those of graduates of other surgical residency programs. This idea of similarity could also be applied to medical specialties, where the physician practice patterns might be expected to have greater similarity within specialties than between specialties. Practice patterns might also be affected by the length of time individual physicians have practiced in the same community. A young physician arriving in a new community might have practice patterns conditioned largely by his residency program. But, after a length of time in the community, the same physician might be expected to have adopted practice patterns more like the other physicians in the community. One might hypothesize that after a longer length of time 20 practicing in the same community, a physician's practice pattern would be even more like the community practice pattern (community standard) and, given enough time, would probably be indistinguishable from it. The small area analysis literature rarely deals with these hypotheses. Originally, this study was to have considered questions concerning how variations in use rates might be explained by the dominance of one residency training program, or one medical specialty, or by the length of time a physician had practiced within a community, but the data necessary to address these questions are highly confidential and were impossible to obtain within the time frame of this study. One previous study, Roos et al. (1977), investigated the explanatory power of several physician characteristics and found that younger physicians trained in the United Kingdom had lower surgical rates and more restrictive criteria for surgery than did older physicians trained in North America. But overall, physician age, specialty, and place of training (United Kingdom or North America) did not account for the observed variation in surgical rates. Only one measure of the physician component could be obtained for this current research, and that is the proportion of board certified physicians (specialists) to total physicians (specialists and non­ specialists) in a hospital service area. One previous study, Connell et al. (1981), had tested the correlation between a specialist to non­ specialist ratio and hospital admission rates. A non-significant positive association was found between the ratio of pediatricians to general practitioners and all but one of the seven hospital use measures tested. Additional research results from previous small area analysis 21 literature that concern the supply of specialists will be discussed within the hospital component of the model since those studies address only the supply of specialist physicians and not the dominance of one specialty or non-specialty group of physicians within the hospital service area. It would appear from previous research that the characteristics of physicians do affect the hospital use rates in their hospital service area. Therefore, I hypothesized that there would be a positive relationship between the proportion of board certified physicians to total physicians and both total admission and total surgery rates. Component 4: The Hospital The fourth component of the Clark Model (and the major focus of this research) is the hospital, which is the setting for the health care event. The hospital has its own Institutional characteristics and, since it provides a place for the medical staff to practice medicine, has characteristics of its medical staff as well. Three types of hospital characteristics are the focus of this study. They are the supply of health care resources (hospital and staff), the services provided in the hospital, and the organization of the institution. These three types of hospital characteristics are hypothesized to influence the hospital use rates for all hospital events. 22 Supply of Health Care Resources: The supply of hospital beds and the supply of physicians (both general practitioners and specialists) have been the most frequently studied health care resources in the small area analysis literature. 1. The Supply of Hospital Beds Starting with Shain and Roemer (1959), there has been a suspicion that an over-abundance of available hospital beds per capita stimulated increased hospital utilization. This positive relationship between hospital beds and utilization has become known as "Roemer's Law". As shown on Table 2.4, the supply of hospital beds has been positively correlated with total admission, patient day, length of stay, total surgical admission, and non-elective surgery rates. The number of hospital beds has provided significant explanation for the variation 1n appendectomy, cholecystectomy, total surgery, total admissions, and patient day rates (white and total) in at least one study. Six studies, shown in Table 2.4, found no significant correlation between hospital beds and use rates and one study reported that the supply of hospital beds provided a non­ significant explanation for black patient day rates. From the results shown in Table 2.4, one can hypothesize that the number of hospital beds per capita will be positively related to total admission, medical causes for admission, and surgical procedure rates. One possible exception to the hypothesized positive relationship between hospital beds and use rates is hysterectomy rates. 23 TABLE 2.4 THE RELATIONSHIP BETWEEN HOSPITAL BED SUPPLY 4 UTILIZATION* Positive Relationship Shain 4 Roemer (1959) beds / patient days Lewis (1969) beds / appendectomy S contributor to explanation beds / cholecystectomy S contributor to explanation .94 correlation ft Anderson (1973) .84 correlation beds / patient days ft beds / admissions It beds / length of stay .83 correlation .86 correlation Chiswick (1976) beds / admissions S Vayda et a k beds / non-elect, surg. rate S (1976) Joffee (1979) Mindel 1 et a k (1982) Mitchell 4 Cromwell (1982) Wilson 4 Tedeschi (1984) ii beds / admissions beds / tot. surg, rate beds / tot. surg. rate + correlation .90 correlation S S correlation in 2 of 5 yrs. contributor to explanation beds / surgical rate very influential to explanation beds / medical admissions very influential to explanation Knickman 4 Foltz (1985) n beds / admissions beds / patient days Wilson et a k beds / white patient day rate (1985) contributor to explanation s s s contributor to explanation contributor to explanation contributor to explanation No or Negative Relationship Vayda 4 Anderson (1975) beds / elective surgery NS correlation Deacon et a k beds / admissions NS correlation Connel1 et a k n (1979) (1981) beds / admissions NS beds / tot. surgery NS .28 correlation .25 correlation Brewer 4 Freedman (1982) available beds / admissions NS .22 correlation Roos, N. (1984) Wi Ison et a k (1985) beds / hysterectomy rate beds / black patient day rate NS NS correlation contributor to explanation S = significant NS = not significant *The results of the various authors' correlation and regression analyses are reported in this table exactly as given in each original article. The Supply of Physicians The relationship between the supply of physicians and hospital utilization has interested many researchers in small area analysis. Table 2.5 shows the published relationships between non-specialist physicians and hospital use rates. Non­ specialist physicians (medical doctors (MDs), general practitioners (GPs) and physicians) were not consistently related positively or negatively to any of the use rate measures studied. Joffee (1979) found a positive association between physicians and total admission rates, while Deacon et al_. (1979) and Brewer and Freedman (1982) found a negative one. Nor was there any consistent relationship between the supply of physicians and total surgery rates. Wennberg and Gittelsohn (1973) found a positive association and Connell et a U (1981) found a significant positive correlation between the supply of direct care physicians or primary care physicians and surgeons with total surgery rates. On the other hand, Detmer and Tyson (1978) found a significant negative relationship between GPs and surgery rate and Mitchell and Cromwell (1982) found that GPs provided a significant (and negative) contribution to the explanation of the variation in total surgery rates. The relationship between specific surgical procedure rates and the supply of non-specialist physicians (Table 2.5) was found to be unique to the procedure and not always consistent within the articles reporting the results of correlation or regression analyses for each procedure. For example, Roos (1984) found no significant correlation between hysterectomy 25 TABLE 2.5 RELATIONSHIP BETWEEN NON-SPECIALIST PHYSICIAN SUPPLY 4 UTILIZATION* Wennberg 4 Gittelsohn (1973) GPs / total surgery .19 correlat ion .42 correlation GPs / T 4 A** GPs / appendectomy GPs / cholecystectomy .14 correlation GPs / varicose veins .31 correlation .38 correlation .24 correlation GPs / D 4 C GPs / hysterectomy -.21 correlation GPs / mastectomy -.20 correlation Physicians not practicing / Detmer 4 Tyson (1976) II surgical rates GPs / T 4 A GPs / appendectomy Phys./ T 4 A -.44 correlation s s + correlation Detmer 4 Tyson (1978) II GPs / total surgery + correlation Not contributor to explanation S correlation GPs / cholecystectomy S Deacon et a k MDs / admissions S Physicians / admissions Dir. care MDs /tot. surg. S Roos, N. et a k (1977) (1979) Joffee (1979) Connel I et a k (1981) II Prim, care MDs /tot. surg. II Dir. care MDs / ENT surg. S S NS -.30 correlation correlation .42 correlation .58 correlation .42 correlation .54 correlation Brewer 4 Freedman (1982) MDs / admissions MitchelI 4 CromwelI (1982) GPs / tot. surgery S('-) contributor to to explanation -.28 correlation Roos, N. (1984) MDs / hysterectomy NS Wilson 4 Tedeschi (1984) Phys./surg. admissions Very influential Phys./med. admissions Low influence correlation in explanation It in explanation Wi Ison et a k (1985) S = significant Phys./black or white patient day rate NS contributor to explanation NS = not significant •The results of the various authors' correlation and regression analyses are reported in this table exactly as given in each original article. **T 4 A = tons i11ectomy and adeno idectomy 26 rates and the supply of MDs, while Wennberg and Gittelsohn (1973) and Detmer and Tyson (1976) found a negative correlation between the supply of MDs and hysterectomy rates. Wennberg and Gittelsohn (1973) and Detmer and Tyson (1976) found positive correlations between tonsillectomy with adenoidectomy and the supply of GPs. According to Roos et al. (1977), the supply of physicians offered no significant explanation in the variation in tonsillectomy with adenoidectomy rates. Cholecystectomy rates were also reported as both positively (Wennberg and Gittelsohn, 1973) and negatively (Detmer and Tyson, 1978) related to the supply of non-specialists. With the exceptions of Detmer and Tyson (1978), who found a negative relationship between the supply of surgeons and appendectomy, tonsillectomy with adenoidectomy (T & A in Tables 2.5 and 2.6), and inguinal hernia repair rates, and Mindell et al_. (1982) who found no correlation between total surgery rates and surgeons, all other researchers (shown in Table 2.6) have found positive correlations between the supply of surgeons and total admission, total surgery and specific surgical procedure rates. The correlations between the supply of surgeons and hospital utilization have been as low as .07 (Wennberg and Gittelsohn, 1973) for varicose vein stripping rates and as high as .70 (Vayda and Morrison, 1976) for total surgical rates. Vayda et al. (1975) found that the supply of surgeons explained sixty-seven percent of the variation in the total elective surgery rate. 27 TABLE 2.6 RELATIONSHIP BETWEEN SURGEON SUPPLY AND UTILIZATION* Wennberg 4 Gittelsohn(1973) II II II It II II II Gen. Surg. / appen. Gen. Surg. / cholecystectomy Gen. Surg. / hysterectomy Gen. Surg. / mastectomy Gen. Surg. / varicose veins Gen. Surg. / D 4 C Vayda & Anderson (1975) Chiswick (1976) Surg./ tot. elective surgery Surg. / admissions Detmer & Tyson (1976) Surg. / total surgery n Surg. / appendectomy Vayda et aj_. (1976) .54 .46 .31 .48 .39 .48 .07 .08 .67 Gen. Surg. / total surgery Gen. Surg. / T 4 A** Surg. / total surgery s s NS S corre 1at ion correlation correlation correlation correlation correlation correlation correlation R2 + correlation + correlat ion correlation .6 7 , .6 9 , .70 correlation for 3 of years Connel I et aj_. (1981) Surg. / total surgery S s s s s s Mlndel I et a k (1982) Wilson & Tedeschi (1984) Surg. / total surgery No Surg. / surgical admissions Detmer & Tyson (1978) it Gen. Surg. / tot. surg. Gen. Surg. / hysterectomy n Gen. Surg. / appendectomy it Gen. Surg. / T 4 A n Gen. Surg. / hernia correlation + correlation - correlation - correlation correlation .50 correlation correlation Very influential in explanation S = significant NS = not significant •The results of the various authors' correlation and regression analyses are reported in this table exactly as given in each original article. K- T a n - i u n s i i ieciu in y anu diien o id ecio m y 28 Small area analysis researchers have also investigated the relationship between use rates and the combined supply of non­ specialty physicians and surgeons (Table 2.7). Lewis (1969) reported that fifty-two percent of the variation in tonsillectomy with adenoidectomy rates and forty-nine percent of the variation in inguinal hernia repair rates were explained by the supply of non-specialty physicians and surgeons. On the other hand, Knickman and Foltz (1985) found no significant contribution to the explanation of the variation in admission rates when a combined total of non-specialty physicians and surgeons was used in the regression. With the exception of a negative relationship between ear, nose, and throat (ENT) specialists and tonsillectomy with adenoidectomy rates found by Detmer and Tyson (1978), all other specialist to use rate relationships were positive. Based on the previous small area analysis literature, I hypothesized that the supply of surgeons would be positively related to surgery rates but that the relationship between the supply of non-specialized physicians and specific specialists to procedure-specific use rates would be idiosyncratic. A measure of the supply of physicians within a hospital service area was not available in the database used, so related hypotheses could not be tested. 29 TABLE 2.7 RELATIONSHIP BETWEEN PHYSICIAN SUPPLY AND UTILIZATION* Non-Specialists and Surgeons Lewis (1969) " Wennberg 4 Gittelsohn (1973) Cageorge et «H. (1981) Knickman 4 Foltz (1985) Phys. 4 Surg. / T 4 A** .52 R2 Phys. 4 Surg. / hernia .49 R2 Phys. 4 Surg. performing surgery / total surgery rate Operating Phys. / cholecystectomy GPs or Surg. / admissions S NS .64 correlation correlation NS contributor to explanation Spec ia 1ist Phys ic ians Detmer 4 Tyson (1976) Detmer 4 Tyson (1978) II ENT$ specialists / T 4 A ENT specialists / T 4 A S strong Internists / total surgery If Connell et a k (1981) ii Knickman 4 Foltz (1985) Internists / hysterectomy S S Pediatricians / total surgery NS Otolaryngologists / total surgery NS Non-surgical specialists / adm. S correlation + correlation correlation + correlation .18 correlation .33 correlation contributor to explanation S = significant NS = not significant * The results of the various authors' correlation and regression analyses are reported in this table exactly as given in each original article. ** T 4 A = tonsillectomy and adenoidectomy t ENT = ear, nose, and throat 3. The Supply of Registered Nurses Only one published small area analysis study has dealt with the role that registered nurses play in explaining use rates. Hairanond (1985) investigated the relationship between home health nurses and Medicare home health visits. He found a positive correlation between home health nurses per beneficiary and visits per beneficiary. Although this was not conclusive evidence for hospital use rates, using the same logic, one could hypothesize that the supply of registered nurses in a hospital would have a significant positive impact on the total admission rates and the medical causes for admission rates if the number of nurses per bed was free to move in response to economic pressures. But Michigan in 1983 did not have an unregulated environment. In the regulated hospital environment of 1983 the total number of licensed beds in the state was restricted. The number of beds could not increase but could decrease if the hospital felt it was necessary to temporarily take beds out of service. In an unregulated environment, an increase in the number of patients would require an increase in the number of nurses. But, in fact, Michigan was experiencing a decrease in the number of patients and patient days in 1983. Hospitals were decreasing the number of beds they staffed with nurses while regulations continued to require a specific number of nurses per bed (particularly for specialty beds such as in intensive care units). As a consequence, the nurse to bed ratio would be expected to increase as the number of patients decreased. Therefore, I hypothesized an inverse relationship between registered nurses per bed and total admission and medical admission rates. Services Provided in the Hospital: No study within the small area analysis literature has attempted to examine the relationship between the services offered in a hospital and its utilization. One study (Hammond, 1985), determined that a measure of the services offered to medicare home health beneficiaries was the single variable which offered the largest part of the explanation (7.2 percent) of the variation in home health benefits. Using Christaller's 31 central place hierarchy as a model for the hospital industry, I hypothesized that a hospital with a larger number of services would provide care to a larger number of patients and, therefore, the number of services in a hospital would be positively related to use rates. Organizational Characteristics of the Hospital: 1. Progressive or Conservative Hospital Philosophy In this time of increased regulatory and reimbursement pressure on hospitals, the role of the hospital administration is increasingly important and can be critical to the stability and health of the institution. Along with the medical staff and the board of trustees, the hospital administration is responsible for defining the admission policies of the institution. Anderson and Lomas (1985, p. 253) found that the "differences in the patient characteristics and the availability of resources appeared less important in explaining these (cesarean section) rate variations than differences in [administrative] policy." Further, they felt that the variability in the cesarean section rates indicated that new criteria for decision-making which were available in the medical literature were being ignored by obstetricians and/or administrators. This suggested to me that the progressive or conservative attitude of the hospital's administration and medical staff may be important predictors of utilization. 32 One difficulty was finding a measure or measures for the progressive nature of administration and staff. My intuitive reaction was to use the number of hospital bed (bed size) as the surrogate measure for progressiveness. This intuitive reaction was strengthened by the results of one small area analysis study that reported high diabetes admission rates in small hospitals. Connell et al^. (1984) had found that the preponderance of hospitals in counties in Washington with high diabetes admission rates were small institutions and that seventy percent of the diabetes admissions in the high-use counties were to small or medium sized hospitals. They also found that hospitals in high-rate counties admitted proportionally more mildly ill patients and made less thorough use of lab tests. If one can generalize from this one study, all of the evidence from Connell et al. indicated that small hospitals were less well run and slower to respond with administrative policy changes to changes in technology or care protocols than were larger hospitals. But, my observation is that the stereotype of small hospitals in backwater communities with out-of-date administrators is not the current state of affairs in Michigan. I decided that a better surrogate measure for progressive or conservative hospital philosophy would be the change in the number of services offered from one time in the past to the present (1981-1983 for this study). A large change, either positive or negative, would indicate that the hospital administration was making a progressive move by either increasing or decreasing the services offered in an attempt to redesign their product to better position the hospital in their current market area. Outpatient visits per capita were used as a second measure of the progressive or conservative philosophy of the hospital administration because it is to the financial advantage of a hospital to move as much of its surgery to an outpatient setting as possible. Therefore, a high outpatient visit per capita use rate ought to indicate a progressive administration and there should be an inverse relationship between outpatient visits per capita and use rates. Brewer and Freedman (1982) reported a significant negative correlation (-.47) between outpatient visits and admissions in Vermont. It was hypothesized that a large change (positive or negative) 1n the number of services offered in a hospital service area would be associated with increasing hospital use rates. It was also hypothesized that as the outpatient visits per capita increased, the total admission (inpatient) use rates would decrease. Since the surgical procedures studied could not be performed in an outpatient setting, the outpatient visits per capita was hypothesized to have no strong relationship to any of the seven surgical procedure rates. Teaching Status Several small area analysis articles have dealt with the impact of teaching hospitals on an area's use rates. Stockwell and Vayda (1979) found that counties with teaching hospitals had 34 consistently lower surgical procedure rates for tonsillectomy & adenoidectomy, hysterectomy, cholecystectomy, appendectomy and colectomy than counties without teaching hospitals. In a later study Vayda et aJL (1984) added prostatectomy and mastectomy to their list of surgical procedure rates that were lower in areas with teaching hospitals. Cesarean section rates were not found to be lower in areas with teaching hospitals. Vayda and his associates also found a positive correlation between health care resources (surgeons and hospital beds) and teaching status. Knickman and Foltz (1985) found that the supply of medical interns and residents was not a significant contributor to the explanation of the variation in admissions. Anecdotal evidence in Michigan supported the notion that areas with teaching institutions, such as Ann Arbor and Grand Rapids, had lower use rates. This was not consistently observed, since Detroit and Flint both have teaching hospitals but also had higher than average use rates. I hypothesized that areas with teaching hospitals would have lower use rates than areas without teaching hospitals and that there would be an inverse relationship between teaching status and use rates. Teaching status was measured by the total number of interns and residents (house staff) per capita. 35 Summary There are four components in the Clark Model and characteristics of each of these components contribute to the explanation of hospital use rate variation. The first, the individual, cannot be tested as a contributor to the explanation of the variation in hospital use rates because characteristics of an individual exist only in that person and cannot be aggregated. The second component, the community, does have testable characteristics. Community characteristics will not be tested for their explanatory powers in this current research, but will be considered when the residuals from the regression equations are examined. Previous research has tested the explanatory strength of several characteristics of the third component, the physician, but found that age, specialty and place of training (United Kingdom or North America) did not account for the variation in surgical rates (Ross et al., 1977). One measure of the physician component, the weighted proportion of board certified physicians, will be tested in this research and is hypothesized to have a positive relationship with hospital use. The fourth component of the Clark Model, the hospital, is the major focus of this research. Previous small area analysis research has tested the explanatory power of several characteristics of hospitals. A positive relationship was found between the supply of hospital beds and hospital use. Therefore, I have hypothesized a positive relationship between the two in this research. Previous small area analysis research also found a positive relationship between the supply of physicians and hospital use. Since the supply of physicians could not be measured for 36 this current research, the explanatory power of this variable could not be tested. The supply of registered nurses in a hospital has not been previously tested, but the supply of home health nurses was positively related to the number of nurses' visits per Medicare beneficiary. Due to the regulatory atmosphere in Michigan, I hypothesized an inverse relationship between RNs per bed and hospital use. Only one previous study tested the relationship between services offered and use rates. Hammond (1985) found that the services offered to home health beneficiaries explained 7.2% of the variation in home health benefits used. I have hypothesized a positive relationship between the availability of hospital services and hospital use. Two organizational characteristics of hospitals have been previously studied. The first, the progressive or conservative philosophy of the hospital administration was found to be of more importance in explaining Cesarean section rates than either differences 1n patient characteristics or the supply of resources (Anderson and Lomas, 1985). Two measures of the conservative or progressive philosophy of the hospital administration were designed for this current research to test their power in explaining hospital use rates. The second organizational characteristic of the hospital component previously tested was teaching status. Several researchers have found a negative relationship between teaching status and many surgical procedure rates. I have hypothesized a negative relationship for all hospital use rates with the exception of Cesarean section rates. The hypotheses used in this current research are further discussed and summarized in Chapter 3. 37 Definition of Hospital Service Area As discussed earlier, one of the additional aspects of the Clark Model is the recognition that all of the influences and activities that produce a hospital admission take place within a defined hospital service area. Medical geographers have long been interested in the definition of hospital service areas, and at least three lines of research support the integrity of the hospital service area. The first line of research is concerned with distance decay. A classic and seminal research paper was published in the Medical and Surgical Journal of Boston (now the New England Journal of Medicine) in 1850 with a following paper 1n the American Journal of Insanity (now the American Journal of Psychiatry) in 1852. Written by a medical doctor, Jarvis, these two papers form the basis in medical geography literature for Jarvis' Law, which states that the use rate for any medical facility will be greater at a point nearer the facility and less at a point farther away from that facility. In more recent literature geographers have explored the question of access using both spatial and temporal measurements from the location of the patient to the medical facility. In a recent paper, Hunter, Shannon and Sambrook (1985) re-evaluated Jarvis' data and found (as he had) that the state mental asylum of the 1850's was, in reality, a local asylum with almost ninety percent of its patients from within sixty miles of the facility. Drosness et cth (1965), Lubin et a k (1965), and Drosness and Lubin (1966) used distance decay curves of either hospital market share or percent of total inpatient population to determine that the largest portion of a hospital's inpatients came from within fifteen minutes - 38 travel time from the facility. The researchers were able to create hospital service areas which were service or disease specific. In every instance the distance decay curve looked very similar to the curve shown in Figure 2.2. Morrill and Earickson (1968; and subsequent papers), DeVise (1966) and Cherniack and Schneider (1967) used patient travel patterns in Chicago and Cincinnati to define hospital spheres of influence or trade areas. The difficulty with spheres of influence is that the defined areas are not unique and therefore have less usefulness for planning and regulation. A unique population at risk is necessary for the denominator in any use rate equation or for age and sex standardization of use rates. P1gozz1 (1969; see also other references cited therein) used a modification of Christaller's central place theory to produce Thiessen polygons which defined hospital service areas. A second line of study used by researchers to establish independent hospital service areas has been to map trips to hospital facilities and delineate the natural catchment areas where there was little, if any, "border crossing" by patients. Using this method, Mountin et a K (1945) not only stressed the need for independent geographic units for comparison and planning, but also delineated a functional hierarchy of 125 regional areas for health care in the United States. Working in western Pennsylvania, Clocco and Altman (1954) established independent hospital service areas which had almost no patient travel across boundaries for either physician or hospital care. In a similar study in Kansas and Missouri, Poland and Lembcke (1962) delineated 130 hospital service areas that differed in size and population, but had distinctive "population divides" that patients seemed not to cross for medical care. Using physician questionnaires, Dickinson et aT. (1949, 1951, 39 Figure Distance QJ a 3 H 0 > 2.2 Decay Curve \ \ . V_ d istance 40 1954, 1954) established 757 geographically defined service areas and assessed their stability over a period of time, making comparisons on such characteristics as size, population, hospital bed to population ratios and physician to population ratios. The third line of study followed by medical geographers has been to delineate hospital service areas using one of the family of gravity models. The use of gravity models has not been as widespread as one might assume among medical geographers. Shannon et al_. (1969) used a modified gravity model to explain the relationship between the distance from a patient's residence to the hospital where that patient received care. Morrill et al. (1970) developed a simulation model in Chicago to account for distance, size of hospital, and several potential intervening opportunities based on racial and religious characteristics of the patients. Morrill and EarTckson (1968) reached the conclusion that a gravity model was more successful in defining the hospital service areas of medium to small hospitals than for larger, urban hospitals. There are two difficulties in using a gravity model to define hospital service areas. The first is that there is overlap in the catchment areas, so that the service areas are not unique. The second problem is that in urban areas where hospitals cluster it is extremely difficult to define individual hospital service areas. Pyle (1979) graphically displayed the shape, size, and extent of each of sixteen hospital service areas in six counties of northeastern Ohio. After constructing an isoline map with overlapping lines surrounding each of the sixteen hospitals, he then drew demand cones for each hospital, showing their overlap and relative size. Pyle's work demonstrated the 41 need to aggregate individual hospital service areas into larger, community-level hospital service areas when doing population-based research such as small area analysis. Designed by geographers interested in market research, the multiplicative competitive interaction (MCI) model is a new addition to the family of gravity models that has recently appeared in the medical geography literature. Folland (1983) used a MCI model to predict hospital market shares for inter-city hospital trade in a predominantly rural area. Distance alone accounted for over half of the variance in the market shares. Cohen and Lee (1985) produced separate models for several socio-economic and age groups as well as for several medical services, because they felt that the use rates in hospital service areas might be different for subgroups of the population. The model allowed them to relate the probability of hospital selection to such hospital and environmental factors as travel time between residence and hospital, hospital and physician characteristics, and patient characteristics. Cohen and Lee were able to explain eighty percent of the variation in use rates. Erickson and Finkler (1985) used the MCI model to focus on the physician characteristics within individual hospital service areas rather than across an aggregated cluster of hospitals. They found that the number of physician affiliations with a hospital had a significant impact on the market share of that hospital. 42 Small Area Analysis Definitions Small area analysis researchers understood that the definition of the geographic area was very important because the area must contain the patients, the community (population), and its health care resources. Previous small area analysis studies used a range of geographic boundaries from those based on pre-existing administrative or political units to those, such as hospital market and service areas, defined by the researcher. States, standard metropolitan statistical areas (SMSAs), counties and minor civil divisions (MCDs) are common choices for pre-existing political boundaries. Counties have the advantage of being relatively small but the disadvantage of being suspect as a basis upon which to match populations and medical care resources. SMSAs and states have fewer boundary problems, allowing the matching of population and resources, but are so large that internal variations, both of population characteristics and of the level and organization of medical care resources, can result. In consequence, the governing assumption — that the population faces a defined set of medical care resources and that the characteristics of both explain the resulting hospital use — is suspect" (Wilson & Tedeschi, 1984, p. 335). In response to the problems identified in utilizing political boundaries, a number of researchers moved to create geographic areas for analysis which were defined by the population's use of its medical resources. By defining a small geographic area the researcher has the best opportunity to examine the relationships between the health resources present and the community's use of those resources. Whereas in previous research the geographic scale of the study was usually a nation, state, province or region, now with the advent of computer data 43 bases and the application of small area analysis methods, the smaller geographic area of observation allows the researcher to better match the population with the health care resources they use. Two definitions of hospital service areas appeared early in the small area analysis literature: the plurality definition of Wennberg and Gittelsohn (1973) and the Relevance Index definition of Griffith (1978). Each was based on the assumption that the population (and therefore the patients) were equally distributed across the "small area" being studied. Wennberg and Gittelsohn (1973) defined a hospital service area using a patient origin plurality measure. To create a hospital service area using this definition, a zip code's or MCD's population and patients were assigned to the hospital service area where the greatest number of patients from that zip code received care historically. Four zip codes are shown in Figure 2.3. Three of them have a hospital located within them and are assumed, for the purposes of this illustration, to have a plurality of the patients within their own zip code receiving care at the hospital within that zip code. It is the allocation of zip code X, which does not have a hospital within it, that is at question. As shown in Figure 2.3, 40 percent of the patients from zip code X received medical care at hospital A, 35 percent at hospital B and 25 percent at hospital C. Using Wennberg's plurality definition of a hospital service area, all of the patients and population from zip code X would be included in hospital A 's service area since that is where the plurality of patients historically received care. Zip code X would then be mapped as lying within hospital A's service area. 44 HOSPITAL B ZIPCODE B — K 35X HOSPITAL A C « * of 25* ^ Discharges ZIPCODE A k 1. + 1 HOSPITAL C ZIPCODE C ZIPCODE X Figure 2.3 Wennberg's Plurality Definition of a Hospital Service Area 45 The aggregation of individual hospitals and their service areas into a larger hospital service area can also be illustrated using Figure 2.3. In the example above, the allocation of zip code X's patients and population was in question. The same criteria for allocation (the plurality of historical patient use) can be used for zip codes B and C. The patients and population of zip code B and/or C will be allocated to hospital A 's service area if a plurality of zip code B's or C's patients received care at hospital A historically, even if there are hospitals present in zip code B and/or C. For example, using Wennberg's plurality measure in South Central Michigan, even though Eaton Rapids, Mason and St. Johns each has a hospital, the zip codes in each community are allocated to Lansing's hospital service area because historically a plurality of patients from each of the three communities has received care in Lansing hospitals. In contrast to the plurality model previously described, Griffith's (1978) Relevance Index used a more complex method of assigning patients and population to hospital service areas. Within each zip code, the proportion of patients who received care at each competing hospital was determined. Each competing hospital service area was then assigned its proportion of the zip code's patients and population. For zip code X shown in Figure 2.4, 40 percent of the population would be assigned to hospital A's service area, 35 percent to hospital B's and 25 percent to hospital C's. Since each of the hospitals in the example was located in a separate community, the zip code would be split into three pieces, with each piece proportional to its patient volume and assigned to a 46 HOSPITAL B ZIPCODE B 1 TT 5 A an at-.38 4 k HOSPITAL A ■ +| HOSPITAL C •t 1 1 k U i « m ZIPCODE A ZIPCODE C ZIPCODE X Figure 2.4 Griffith's Relevance Index Definition of a Hospital Service Area 47 separate hospital service area. The zip codes were never actually split graphically because Griffith never mapped any hospital service area. Only one paper has reported the results of a comparison of the impact that changes in the areal definition of a service area had upon utilization rates. Tedeschi and Martin (1983) tested Wennberg's plurality model and Griffith's Relevance Indices (using both a straight assignment based on proportion and a 12.5 percent market penetration measure) and found the overall use rates from the three definitions of hospital service area (one from Wennberg and two from Griffith) to be highly intercorrelated and that an approximate linear relationship existed between each pair of variables. Shaughnessy (1982) felt that the Relevance Index more accurately described a hospital's service area, while Sigmond et a]. (1981) felt that Griffith's method systematically underestimated use rates of small rural areas. Because of serious technical problems experienced with the Relevance Index, almost all small area analysis studies which have aggregated zip codes or MCDs and were written after 1978 have used the plurality definition based on historical patterns of total admissions. The plurality definition of hospital service area was also used 1n this current research, where the hospital service area defines the geographic limits of the "community" component of the Clark Model. CHAPTER 3 METHODS AND GENERAL HYPOTHESES Focus of the Study Review of the small area analysis literature (in Chapter 2) established that small area analysis research has attempted to do three things: 1) document the variation in utilization rates from one small area to another; 2) determine what patterns were apparent in the variation; and 3) determine what independent variables explained that variation. Although a few previous researchers in small area analysis have tested the explanatory power of a small number of health care resource supply characteristics (physicians, specialists, hospital beds, and empty hospital beds), the major focus of the previous research has been the explanatory power of community characteristics. My research determined the amount of variation found in fourteen measures of hospital utilization among fifty-three non-metropolitan Detroit communities in the southern peninsula of Michigan. The research answered several questions: Was there variation in hospital use rates, and how did the variability in Michigan hospital use rates compare to previous results in Michigan and elsewhere? Were the use rate patterns reported in the small area analysis literature found also in Michigan 48 49 and were there spatial patterns among the use rates? What relationships were found between different hospital use rates and between use rates and the provider characteristics? How much of the variation in hospital use rates was explained by the supply of health resources within a community and how much was explained by the hospital services offered and the organizational characteristics of the hospitals themselves? Study Area and Time Frame The study area for this research included all of the southern peninsula of Michigan with the exception of metropolitan Detroit. Detroit was excluded from this research for two reasons. First, small area analysis is most successful in defining hospital service areas in regions where there is some distance between one cluster of hospitals and another cluster. Second, since approximately one-half of all of Michigan's hospital inpatients are cared for in the Detroit metropolitan area hospitals, the cost of using the data base, if the Detroit metropolitan area had been included, would have been prohibitively expensive. Figure 3.1 shows the hospital service areas included in the research. The hospital inpatient records for all patients from the study area for the calendar year 1983 were analyzed; the data for 1983 was the most recent data available at the time this study was initiated. Consistent with small area analysis methodology and for reasons of comparability, this study used the 1983 Michigan hospital service areas as defined by John Griffith and his associates at the University of Michigan. They used John Wennberg's plurality method for assignment of 50 Cheboygan/ Rogers City Northern Pv. Michigan Otsego Alpena Traverse City Crawford Tawas Ogemaw Cadillac Manistee Mt Pleasant Mason Co. Bad Axe Bay Reed City Midland Oceana_. Tuscola I Sanilac Fremont Gratiot Saginaw Muskegon * 1 Montcalm/Ionia ont Owos- Grand Rapids Ottawa Flint Lapeer ort Huron Pontiac Lansing olla I owell Mt. Clemens Benton Harbor^ St. Joseph Kalamazoo Battle L ALT* C Creek Jackson I Ann Arbor Monroe Figure 3.1 Michigan Hospital Service Areas Included in Study Area. (Abbreviations used: LV = Lake View; AL = Albion) 51 zip codes to hospital service areas. Every zip code in the study area was assigned to the hospital service area where the plurality of its total patients historically had received care. Every zip code in the study area has, therefore, been assigned to one hospital service area. As illustrated in the Clark Model, two assumptions underlie small area analysis methodology. The first assumption deals with the health care resources a community contains while the second assumption has to do with the allocation of patients to a hospital service area. 1) The population of a hospital service area uses a single set of health care resources within the community, no matter how many hospitals exist within that community. This means that for hospital service areas that contain more than one hospital, the individual hospital characteristics, such as the number of beds, are aggregated into a single hospital measure for that hospital service area. Therefore, for the Clark Model and for the purposes of this research, the term "hospital" may refer to a single hospital or a hospital cluster of two or more hospitals, depending on whether one or several hospitals exist within a hospital service area ("community"). 2) The patients residing in a defined hospital service area (the "community" in the Clark Model) go to the hospital(s) within that service area. The utilization rate for a hospital service area, then, is the number of patients in the service area divided by the total population. No matter where hospital care is received by a patient, the occurrence of that health care event is assumed to have taken place in the hospital service area of the patient's residence. Even if a hysterectomy was performed at a referral center a hundred miles from the patient's residence, the hysterectomy was counted as having occurred at 52 a hospital in the same hospital service area where the patient resided. This assumption is problematic because of the number of individuals who could potentially leave any hospital service area to receive care in another hospital service area. Migration is both in and out of any given service area. But because it would not be possible to adjust the population at risk for every patient removed from or added to a hospital service area, small area analysis methodology returns each migrating patient back to his place of residence so that the numerator (patient) is indeed taken from within the denominator (population). The procedure-specific and diagnosis-specific use rates included in this current research were chosen to minimize the migration problem. The extent of the migration problem was checked by determining the migration from "outstate" zip codes into the metropolitan Detroit area. Since Detroit is the largest referral center in the state, it was used as a case study to test if migration was a problem. To do this, the number of patients who were cared for in metropolitan Detroit was determined. As shown in Table 3.1, the total number of patients who went to Southeast Michigan for hospital care was less than eight percent for all but three of the outstate hospital service areas. It was decided that, with the additional safeguards described later in this chapter, this percent was small enough that not being able to adjust for migration should not affect the results of this study. This methodological problem was discussed with a prominent small area analysis researcher (Tedeschi, personal communication, 1987) who agreed 53 Table 3.1 Patient Migration Cluster # 1 2 3 4 5 6 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Hospital Service Area Total Pts. Port Huron Pontiac Howe 11 Ann Arbor Mt. Clemens Monroe Lansing Hillsdale Adr ian Jackson Battle Creek Albion Benton Harbor/St. Joe Ka1amazoo South Haven Comm. Hastings Coldwater Three Rivers/Sturgis S. Berrien/Cass N. Montcalm Freemont Lakeview Reed City Ottawa Kent Co./Grand Rapids Muskegon Montcalm/Ionia Allegan Mason County Hoi land Oceana County Ionia Flint Lapeer Co. Owosso Bay Sag inaw Tuscola Bad Axe Sanilac Midland Mt. Pleasant Tawas Ogemaw Gratiot Petoskey Cheboygan/Rogers City Otsego Crawford Traverse City Cadi 1lac Man istee Alpena 22,012 63,020 4,549 41,310 39,693 13,123 56,709 5,684 55,022 23,842 22,414 3,131 20,041 33,641 3,965 4,626 6,877 8,101 13,766 5,140 3,578 2,088 7,805 4,695 61,594 23,953 5,348 2,763 4,302 12,316 2,552 3,110 77,433 12,193 7,639 24,951 38.197 5^037 7,412 4,857 12,413 11,170 5,047 5,826 9,460 10,034 4,979 3,449 5,071 16,988 7,683 3,680 7,186 Pts. = Patients NA = Data not available Cared for in SE Hosp itaIs Total P+s. % of TotaT NA NA NA NA NA NA NA NA NA NA NA NA 419 702 88 82 356 233 250 87 96 30 205 69 921 414 77 39 95 184 17 43 0 0 323 904 1 .180 '309 604 778 576 493 393 391 190 290 161 221 319 545 233 89 362 NA NA NA NA NA NA NA NA NA NA NA NA 02 02 02 02 05 03 02 02 03 01 03 01 01 02 01 01 02 01 01 01 00 00 04 04 03 06 08 16 05 04 08 07 02 03 03 06 06 03 03 02 05 54 with my decision and recommended that I use the current small area analysis methodology, which returns the patient to the hospital service area of residence. The fourteen hospital use rates (specific diagnoses and procedures) used in the current study were chosen to minimize the number of patients migrating from one hospital service area to another. These fourteen included three types of use rates (see also Figure 3.3, p. 59). The first type included three measures of total admissions (male, female, total), the second included seven specific surgical procedures (appendectomy, hemorrhoidectomy, cholecystectomy, inguinal hernia repair, prostatectomy, hysterectomy, cesarean section), and the third included four broad classes of medical diagnoses (circulatory, respiratory, digestive, genito-urinary). The surgical procedures and medical causes for admissions were chosen in part because they were not technically difficult for the hospital to perform or to care for, nor were the patients who were admitted, or who received the surgical procedure, extremely ill. As a consequence, all the procedures and diagnoses could be handled in any general acute care hospital within each hospital service area as shown in the model. It was felt that the choice of pre-existing hospital service areas and specific diagnoses and procedures maximized the comparability of the results and minimized the problems of patient migration. 55 Sources of the Data Procedure and Diagnosis Occurrences The fourteen hospital use rates were calculated for each hospital service area from data obtained from the Michigan Health Data Corporation (MHDC) through the Michigan Hospital Association's (MHA) Interactive Data System (IDS). The MHDC is a consortium of twelve Michigan institutions that are interested in health care data. One purpose of the MHDC is to produce a data base that has been politically agreed upon by all members and can, therefore, be used as a single data source for all discussions among the membership. The MHDC membership includes, among others: the MHA, the Michigan State Medical Society, the Michigan Department of Public Health, the Blue Cross and Blue Shield Corporation of Michigan and the three largest publicly funded universities (Michigan State University, Wayne State University and the University of Michigan). This data base is called the Michigan Inpatient Data Base (MIDB) and is now available for the calendar years 1980 - 1986, although only 1980 - 1983 data were available at the time this research was initiated. Each annual data base contains almost 1.5 million records and approximately fifty-six variables for every inpatient admission to acute care hospitals in Michigan for that year. Included in the variables are the international codes (ICD-9-CM codes) for each diagnosis and procedure, the zip code of the patient's residence, the sex and the age of each patient. Access to this data base was granted after review by the Research Subcommittee of the MHA's Data Management Committee, the Data Management 56 Committee, the Executive Board of the Hospital Research and Educational Reserve of Michigan, Inc., and the Access and Oversight Committee of the Michigan Health Data Corporation. Since eighteen of the fifty-three hospital service areas included in the study area are "sole community providers" (a hospital service area with a single hospital within it), access to the Michigan Inpatient Data Base was granted with the understanding that the data from the sole community provider hospital service areas could be analyzed, but that the data from these eighteen areas would never be identified by hospital service area in a table or on a map. Figure 3.2 shows the location of the eighteen sole community provider hospital service areas. The regions within which they are located will be used to give their general location in later chapters. The regions are shown in Figure 5.4. Population The 1980 age and sex-specific population figures by zip code were obtained from the MHDC's population data base, POPZIP, which is available on the IDS and was produced from the official population figures of the Michigan State Demographer. The 1980 population figures were chosen to standardize the 1983 hospital data because the 1980 census data were based on actual census counts whereas the 1983 population data were estimates. 57 Figure 3.2 Sole Community Provider Hospital Service Areas (yi-i&3 Sole Community Provider Hospital Service Areas Areas Not Included within the Study Area 58 Physician and Hospital Characteristics The physician and hospital measures were obtained and calculated from variables obtained from the Department of Research, Data Policy & Services at the MHA and from the American Hospital Association's (AHA) Annual Survey of Hospitals for the years 1981 and 1983. This survey of American hospitals is collected annually and contains information relating to each contributing hospital's personnel, facilities, services, and financial status. Permission to use the non-confidential portions of these data bases was given by the Group Vice President of Research, Data Policy & Services at the MHA. Where data were missing, direct contact was made with hospital personnel. Research Design This research had four goals and each had a separate research design. The four goals are discussed below; the results of the research for each goal will be discussed in Chapters 4, 5, 6, and 7, respectively. 59 Goal 1: Document Variation in Use Rates and Compare Results to Previous Results from Michigan and Elsewhere Fourteen measures of hospital use have been calculated for each of the hospital service areas in the study area of Michigan. As shown in Figure 3.3, there were three types of use rates studied. Figure 3.3 Measures of Hospital Use Abbreviation Three Total Hospital Use Rates: Total Male Admission Rate Total Female Admission Rate Total Admission Rate Tot Male Adm Tot Female Adm Tot Adm Seven Surgical Procedure Rates: Appendectomy Rate Hemorrhoidectomy Rate Cholecystectomy Rate Inguinal Hernia Repair Rate Prostatectomy Rate Hysterectomy Rate Cesarean Section Rate Appen Hemorr Chole Hernia Prost Hyster C-sect Four Medical Causes for Admission Rates: uii cu t a i u i A Jt .* s J i n - j .. -. nuini:>a i u n r \ a t e Respiratory Admission Rate Digestive Admission Rate Genito-Urinary Admission Rate Circ Resp Digest Gen-U 60 Each of the fourteen use rates was age and sex adjusted to the 1980 total Michigan population using the direct method of standardization as described by Mausner and Kramer (1985). These age and sex standardized use rates for each hospital service area are shown in Appendix A. An age and sex adjusted use rate was calculated for each of the fourteen hospital use rates for each hospital service area. 6 TA. £ * SP. I TP, i= 1 Standardized Rate = * 1000 (SP,) i= 1 where: i a specific age and sex category SP, TA, State Population for , TP, = Total Admissions for i Total Population for Total Admissions Three measures of total admissions were used for this research: total male admissions; total female admissions; and total admissions. Surgical Procedures The seven surgical procedures were chosen for several reasons. First, the surgical procedures chosen are ones that are well documented in the small area analysis literature. Since previous research has documented procedure-specific use rates, the procedures chosen for this study represent a range from previously determined low variation (inguinal hernia repair) to high variation (hysterectomy). Second, the 61 procedures chosen can be performed at even the smallest general acute care hospital, since the procedures are not technically difficult to perform, nor are they generally performed on patients who are severely ill. Third, the procedures chosen require inpatient, rather than ambulatory surgery. This criterion was used to assure that the number of procedures included in the study was comparable from one hospital service area to another. Michigan does not have an ambulatory surgery data base, so by limiting the research to inpatient procedures, the accuracy of the number of procedures is greater. Outpatients or ambulatory patients can be cared for in non-hospital settings, such as physician-owned "surgi-centers", and data are not collected from those medical facilities. For example, tonsillectomy (with or without adenoidectomy) was excluded from this research, although usually analyzed by small area analysis researchers, because tonsillectomy is now performed far more often on an outpatient rather than an inpatient basis. Fourth, an effort was made to choose procedures where the physician doing the surgery would also be most likely to be the physician who made the diagnosis. This was done to minimize the potential impact of the physician referral patterns within or between communities. Finally, cesarean section was added to the list of surgical procedures to be included in the study because cesarean section met the criteria mentioned above and because cesarean section rates are rising very rapidly all over the United States. Once the seven surgical procedures were chosen, their ICD-9-CM codes were defined as precisely as possible. Once determined for each procedure, the codes were sent to an informal review panel made up of five Directors of Medical Records within the study area. Each of the 62 five Directors returned a critique of the groups of codes. Where there was a discrepancy in the coding the most senior of the Directors was contacted and that person's decision was accepted. At the advice of this review panel, all oncology codes were excluded from the research. This was done because there is a low degree of medical consensus on hospitalization and treatment of cancer patients and because malignancy patients are often hospitalized in referral centers, not in general acute care hospitals. Hospitalization in referral centers would increase the problem of migration, so by removing oncology procedures, migration from one hospital service area to another was minimized. The ICD-9-CM codes for the seven surgical procedures are given in Figure 3.4. Figure 3.4 Surgical Procedure Codes Procedure Appendectomy Hemorrhoidectomy Cholecystectomy Inguinal Hernia Repair Prostatectomy Hysterectomy Cesarean Section ICD-9-CM Procedure Codes 470, 47i 4943 - 4946 5121, 5122 5300 - 5317 602 - 604, 6061, 6062, 6069 683 - 688 740 - 744, 7491, 7499 Medical Causes for Admission The medical causes for admission were chosen quite differently. Rather than to determine very small precise definitions of a hospital event, as was used for the surgical procedures, the medical causes for 63 admission were purposely designed to be very large, inclusive, definitions. The aggregation of ICD-9-CM codes into circulatory, respiratory, digestive, and genito-urinary medical diagnoses accounts for the greatest portion of all medical admissions to general acute care hospitals. These four broad categories were established to reduce the chance of bias due to the unique character of any one hospital. The medical records review panel reviewed and approved the medical cause for admission coding shown in Figure 3.5. Oncology diagnoses were excluded from the study. Figure 3.5 Medical Causes for Admission Medical Cause for Admission Circulatory Causes for Admission Respiratory Causes for Admission Digestive Causes for Admission Genito-llrinary Causes for Admission ICD-9-CM Diagnosis Codes 390 466 530 580 - 438 - 518.8 - 578.9 - 599.9 Analysis for Goal 1: Document and Compare Use Rates Comparisons were made of these fourteen standardized use rates to previously recorded rates for Michigan and elsewhere. First, the ranges were compared. Then the maximum to minimum ratio (the maximum range divided by the minimum range), the coefficient of variation (the standard deviation divided by the mean), and the systematic component of 64 variation (SCV) for each use rate were compared to previously recorded results. The systematic component of variation was used by McPherson et al_. (1981 and the references cited therein) to facilitate comparisons of the variability of procedure or diagnosis-specific use rates. The SCV calculation produces a single number which is not related to the magnitudes of the use rates. Therefore, using the SCV, the variability of an infrequently occurring procedure or diagnosis can more easily be compared to the variability of a more commonly occurring procedure or diagnosis. A separate SCV was calculated for each use rate (total admission, surgical procedure, and medical cause for admission) using the following equation. Systematic Component of Variation SCV = 1000 * where o = |~variance (o/e) - mean (l/e)~| age and sex adjusted observed use rate for each hospital service area e = age and sex adjusted expected use rate for each hospital service area and the variance and mean are calculated over all 53 hospital service areas 65 Goal 2: Analyze the Use Rates Use Rate Pattern Analysis In a review of the small area analysis literature, Paul-Shaheen et al. (1987) identified four use rate patterns among the results of previous researchers. The four use rate patterns were: 1) there is consistency in procedure specific variation ranking; 2) the variation in rates for admission due to medical causes is consistently greater than the variation in surgical procedure rates; 3) hospital service areas have unique surgical procedure rate patterns; and 4) hospital service areas may show consistently high or low use across several measures of use. Using those four use rate patterns as a framework, the results from this research were analyzed and compared to the results reported in earlier research. The variations in the fourteen hospital use rates were mapped and the analysis of their spatial patterns was included as part of the analysis of the fourth use rate pattern, the consistency within hospital service areas of high or low use across several procedures or diagnoses. A classification system based on standard deviation was chosen so that comparisons of hospital service areas could be made across all of the use rates. It was more important to this research to know if a hospital service area had consistently high or low use rates across several measures than to know the actual use rate distribution and range for a specific procedure or diagnosis. Due to the confidentiality constraints placed upon the display of the results of this research, much of the geographic analysis was done on a regional basis. The hospital service areas shown in Figure 3.1 66 were aggregated into six regions (Figure 5.4, p. 123) for the purposes of this current study. These six regions represent slight modifications of the regions which have been defined by the Michigan Hospital Association for their Hospital District Councils. Each regional boundary in this current research was established to assure that at least two sole community provider hospital service areas were included within the region. As shown in Figure 3.2 four areas were not included in any hospital service area used in this research. The non-included areas shown on the southern border of Michigan are aggregated zip codes where the plurality of the population received hospital care in a contiguous state (Indiana and Ohio). The metropolitan Detroit area was not included within this research because of reasons described earlier in this chapter. Goal 3: Determine the Relationships Between Different Hospital Use Rates and Between Use Rates and Provider Characteristics After a review of the small area analysis literature and the 1983 data available, one measure of physician characteristics, four measures of the supply of health service personnel in hospitals, four measures of the services available in hospitals, and five measures of the organization of the hospitals were designed, collected, and aggregated across all hospitals in each hospital service area. The physician and hospital characteristics for each hospital service area are shown 67 in Appendix B. The fourteen measures are shown on Figure 3.6 and described below. Figure 3.6 Physician and Hospital Characteristics Studied I. Physician Component 1. II. Weighted Proportion of Board Certified Physicians in 1983 (Wgt Prop Bed Cert) Hospital Component A. Supply of Health Care Resources in 1983 1. B. Licensed Hospital Beds per 10,000 Population (Beds) 2. Full Time Equivalent Hospital Employees per 10,000 Population 3. 4. Registered Hospital Pharmacists per 10,000 Population (Pharm) Registered Nurses per Licensed Hospital Bed (RNs/Bed) (FTEs) Services Available in the Hospital in 1983 1. Average Number of HospitaI-Based Services per Hospital 2. Average Number of Services per Hospital Provided by Another Facility (Avg Hosp Serv) 3. Service Level per Hospital Service Area (Serv Level per HSA) 4. Total Number of Services (out of 66 possible) Available in the Hospital Service Area (Tot if Serv per HSA) (Avg Other Fac Serv) C. Hospital Characteristics in 1983 1. Change in Number of Services Offered in the Hospital Service Area from 1981 to 1983 (Change in Serv) 2. 3. Outpatient Visits per 10,000 Population (0PV) Proportion of Corporate Owned Beds (Corp Beds) 4. Salaried House Staff per 10,000 Population 5. Proportion of Osteopathic Hospital Beds (House Staff) (Osteo Beds) Physician Component 1. Weighted Proportion of Board Certified Physicians in 1983 (Wgt Prop Bd Cert) The AHA data base counts a physician in every institution where he admits patients. If a physician was counted in every hospital where he had privileges, the physician count would be inflated for those hospital service areas where more than one hospital exists. Therefore, a new measure was developed. The proportion of board certified physicians to all physicians in each hospital was calculated and then weighted by the number of beds in each institution in order to measure that institution's contribution to the specialist to non-specialist ratio within the hospital service area. This total was then divided by the total number of beds in the hospital service area to allow for comparability. The individual hospital contributions were then summed to create a weighted total for the hospital service area. This weighted proportion of board certified physicians for each hospital service area was calculated as follows: 0 We ighted board certified phys in hosp. Proportion of Board (* beds in hospj) Certified Physicians * total 0 phys in hosp. __ i=1 ( total 0 beds in all hospitals in area ) Hospi Ldi Component Supply of Health Care Resources in 1983 1. Licensed Hospital Beds per 10,000 population (Beds) The number of licensed beds per 10,000 population is a measure of the historical size of the hospital and provides an indication of the capital investment made by a community in its hospital. When that measure was not available (six hospitals), the total number of "set up and staffed" beds was substituted. Full Time Equivalent Hospital Employees per 10,000 Population (FTEs) The number of non-professional hospital personnel (FTE's) was determined and standardized to the population of the hospital service area. This measure indicates the level of staffing a community provided for its own health care. Registered Hospital Pharmacists per 10,000 Population (Pharm) The number of full-time equivalent registered pharmacists was determined and standardized to the hospital service area population. This is a measure of the size and technical sophistication of the hospital, particularly in the area of circulatory admissions because, of the four medical causes for admission included within this study, circulatory diagnoses require the greatest amount of drug use and therefore the greatest number of pharmacists on staff. Registered Nurses per Licensed Hospital Bed (RNs/Bed) The number of full-time equivalent registered nurses was determined and standardized to the number of licensed hospital beds. This measure is an indication of the "interpersonal" quality of care provided within a community and was thought to be of more importance in the explanation of the variation in the total admission rates and medical 70 diagnosis rates than in the explanation of the variation in surgical procedure rates. Services Available in the Hospital in 1983 The 1983 AHA Annual Survey of Hospitals tabulated for each hospital sixty-six comparable services that might have been provided during that year. AHA classified the provision of each service into three categories: Code 1 = hospital-based; Code 2 = provided by another hospital or provider; and Code 4 = service was not provided. Using these codes four measures were devised to estimate the provision of services within a hospital service area, and to make it possible to compare the level of services offered in hospital service areas. The first measure calculated was the average number of hospital-based services (code 1) per hospital. The second measure was the average number of services per hospital provided by another facility (code 2). The third measure was the service level per hospital service area for all of the services provided within the service area (codes 1, 2, and 4). In each of these three measures a lower score meant that more services were offered by the hospitals in the service area (coded 1), and that fewer services were offered by contract (coded 2), or were not available (coded 4). The fourth measure was the total number of services (out of 66) which were offered by at least one hospital (code Is) in the hospital service area. A higher score on this measure meant that more services were offered in the hospital service area. Hospital Characteristics 1. Change in Services Offered from 1981 to 1983 (Change in Serv) Because of the change in the hospital environment in recent years, it was important that some measure be devised that would estimate a change in a hospital administration's management philosophy toward a more progressive approach. The change (either positive or negative) in the number of services available from 1981 to 1983 was used as a measure of the change in the administration's management philosophy, since the change would Indicate a purposeful shift in the number of services offered to better meet the demands of the market place. In 1981 the AHA used four codes to classify the provision of services by a hospital rather than the three it used in 1983. In 1981 services were classified as: Code 1 = hospital-based and staffed; Code 2 = hospitalbased, contracted; Code 3 = provided by another hospital or provider; and Code 4 = service not available. For purposes of comparing 1981 and 1983 data, the first two categories were collapsed into one, hospital-based services. The number of services (out of 66) provided by at least one hospital within the service area was counted. The total number of services offered in each hospital service area in 1981 was subtracted from the 1983 total. Outpatient Visits per 10,000 Population (OPV) The total number of outpatient visits (clinic and surgery) was determined and standardized to the hospital service area population. This is another measure of the progressive or conservative management philosophy of the community's hospital administration because, under the current cost containment pressures, more progressive hospitals are shifting an increasing amount of their business to the outpatient setting. Proportion of Corporate Owned Hospital Beds in Each Service Area (Corp Beds) The proportion of corporate owned beds was calculated for each hospital cluster. This was a measure of the structure of the industry in the hospital service area. There are no for-profit hospitals in Michigan at this time, but there is an increasing environment of competition and with it has come corporate and quasi-corporate alliances. The type of organization (corporate vs. independent) could potentially influence the admission practices of the physicians and hence the utilization rates for the hospital service areas. Salaried House Staff per 10,000 Population (House Staff) The number of medical staff who were given a salary or stipend by the hospital was determined and standardized by the population of the hospital service area. This measures 73 the importance of the teaching institutions within a hospital service area. 5. Proportion of Osteopathic Hospital Beds in Each Hospital Service Area (Osteo Beds) The proportion of hospital beds in osteopathic hospitals was calculated for each hospital service area. There is a difference in the medical philosophy between osteopathic and allopathic physicians which could have an influence on the admission practices of the hospitals and therefore on the utilization rates. Michigan isone of the few areas in the nation where this measure couldbe included in small area analysis research, and therefore it was felt to be important that a measure of the osteopathic influence on utilization rates be tested. The hypothesized impact of osteopathic philosophy will be discussed further in the section of this chapter devoted to general hypotheses. Analyses for Goal 3: Determine the Relationships Between Different Hospital Use Rates and Between Use Rates and Provider Characteristics Using the SPSS-X System (Release 2.1), Spearman Rho correlations among the physician and hospital descriptive measures were calculated to determine the relationships between the variables. Correlations were also run to help the researcher develop specific hypotheses and to 74 determine if some of the physician and hospital variables were so closely correlated that they should not be simultaneously entered into the subsequent regression equations. Goal 4: Use Rates as a Function of Physician and Hospital Characteristics In an effort to explain the variation in hospital use rates, each of the total hospital admission, sex-specific admission, surgical procedure and medical cause for admission use rates was entered into a multiple regression as a dependent variable. The independent variables were the physician characteristics and the hospital resource supply, service and organization characteristics described in Figure 3.6. Analysis for Goal 4: Use Rates as a Function of Physician and Hospital Characteristics A stepwise multiple regression technique was used with Version 6.02 of the SAS System (operating under PC DOS). The minimum significance level for entry into the model and for staying in the model was <*- = .10. The Kolmogorov-Smirnov (K-S) test of the residuals was used to determine if they were normally distributed. The residuals from the multivariate regressions were variation unexplained by the provider and community variables in the equation. Since the residuals from each equation were specific to a geographic location, they were mapped and examined for spatial patterns. 75 General Hypotheses General research hypotheses were developed by reevaluating the results of previous small area analysis research in light of my logic and my experience within the hospital industry. Since the current research looks for explanation of hospital use rates among provider characteristics, both the physician and hospital components of the model were hypothesized to contribute to that explanation. A review of the previous literature showed that the relationships between seven provider characteristics and hospital use rates had been reported. Five of the seven variables relate to the supply of physicians or their characteristics and two relate to hospital characteristics (bed supply and teaching status). To facilitate this discussion, Tables 2.4 - 2.7 were consolidated into a single new figure (3.7) that shows the variety of relationships between hospital use rates and provider characteristics reported in the literature. Correlations are reported as "C" with appropriate super­ scripts to indicate positive (+), negative (-), significant (S), and non-significant (NS) associations. Regression results (R) are shown simply as being significant (S) or non-significant (NS). The results of each study are not shown, but rather each different result is shown, no matter how many researchers had similar findings. As shown in Figure 3.7, there is a good deal of conflicting evidence in the literature about the direction (positive or negative) of the relationships and the strength of the provider characteristics in providing explanation for the variation in use rates. 76 Figure 3.7 VARIETY OF RELATIONSHIPS REPORTED IN THE LITERATURE HOSPITAL COMPONENT PHYSICIAN COMPONENT NonSpecial­ ist Supply Col 1 Surgeon Supply Special­ ist Supply Col 2 Col 3 c+ cs Rs NonSpecial + Surg Supply Prop Bd Cert to NonPhy Col 4 Col 5 Bed Supply Teach ing Status Col 6 Col 7 Tot Male Adm Tot Female Adm c V c scns Total Adm Appendectomy c* cs c+cnsc'cs c+c"cs c+ Rns C + cns C+RsCsCns Rns RS # Rs * cns # Hemorrhoi dectomy Cholecyctectomy cns c" cs Inguinal Hernia Rs Prostatectomy c" cns Hysterectomy c c c+ cs C-Sect ion .w.+ T«<* yw. v*+r“>»Sc>S w w •> f „+„s,*ns V V ^s^s^ns w r> u Circ Admissions Resp Admissions Digest Adms Gen-U Admissions Correlation Resu1ts c+ c" cs RnS RS = positive correlation = negative correlation = significant correlation (without sign) Cns = non-significant or no correlation Regress ion Results Rs = significant or very influential contributor to explanation Rns = non-significant or low influence as contributor to explanation "Negative relationship reported as untested conclusion of author Physician Component There is only one measure of the physician component of the Clark Model used in this research. Based on my analysis of the previous literature and my observations and experience in the hospital industry, I have hypothesized that the weighted proportion of board certified physicians to total physicians will have a positive relationship with all types of use rates. I have hypothesized a positive relationship because of the following information. The weighted proportion of board certified physicians is a measure of the dominance of specialists within a hospital service area, and, as such, is related to columns two through five on Figure 3.7. Columns two through five report the relationships between the supply of specialists and use rates. With the exception of inguinal hernia repair rates, all other previous research has shown a positive relationship between hospital use rates and the supply of surgeons, specialists, non-specialists plus surgeons and the proportion of board certified physicians. The strength of the relationship between the supply of specialists and hospital use rates suggests to me that there will also be a positive relationship between the dominance of specialists in a hospital service area and use rates. The relationship is hypothesized to be stronger for surgery than for total admissions because the surgical procedures being studied must be performed in a hospital setting rather than in a physician's office, thereby reducing the possible effect of non-surgeons doing the surgical procedures. 78 Hospital Component Supply of Resources Variables related to the hospital component of the Clark Model were divided into three categories: the supply of resources; the availability of services; and organizational characteristics (see Figure 3.6). Previous small area analysis research has only tested the relationship between use rates and one variable from each of two of these categories: the supply of hospital beds and the effect of teaching status. As shown in Figure 3.7, a clear positive relationship between hospital beds and use rates has been established. Therefore, without any evidence to the contrary, I have hypothesized a positive relationship between each of the fourteen use rates tested and hospital beds per population. I also have hypothesized that FTEs and pharmacists per population will behave the same way as beds (have a positive relationship with use rates), since FTEs and pharmacists are also measures of the supply of health care resources and will increase or decrease with the number of hospital beds. I have hypothesized, with more certainty, the positive relationship between pharmacists and circulatory causes for admission rates than for any of the other use rates, since cardiac care is very medicine-intensive, requiring the services of more pharmacists in those hospitals where there is a high volume of circulatory admissions. The relationship between RNs per bed and hospital use rates is more complex, largely because of the regulatory environment in Michigan. Although the Department of Public Health's Licensing and Registration Bureau does not mandate a minimum number of nurses that a hospital must 79 employ to remain open for patient care, the licensing process does include review of the level of registered nurse staffing in each unit of the hospital. If the number of registered nurses falls below a threshhold level, then hospital beds will have to be withdrawn from service. If there were to be an increase in the number of patients who needed care in that hospital, the economies of scale would go into effect, and the administration would use the same number of nurses to care for many more patients. As a result of this minimum level of staffing I have hypothesized that there would be a negative relationship between registered nurses per bed and total admission rates as well as between registered nurses per bed and medical causes for admission rates. The relationship between surgical procedure rates and registered nurses is less certain, but felt to be negative as well. There may be a threshhold effect with surgical patients, but it would be difficult to hypothesize about each individual procedure use rate. Availability of Services Four measures of the availability of hospital services were used in this research. These four were: average number of services provided by the hospitals within a hospital service area in 1983; the average number of services provided by other facilities within a hospital service area in 1983; the average number of services per hospital in a hospital service area in 1983; and the total number of services (out of available in each hospital service area in 1983. 66 ) I hypothesized that as hospital services were increased in hospital service areas, that hospital use rates would increase. My reasoning was based on the observed increase in hospital use when a new instrument, facility, or 80 procedure is introduced. For example, when ultra-sound or magnetic resonance imagers (new instruments) became available, patients came to a facility to use the diagnostic equipment. The same is true when a new facility is opened or a new procedure introduced, because these new health care resources fulfill a previously unmet need. When heart bypass surgery was introduced, surgical procedure rates rose dramatically in those service areas where the procedure was performed. Assuming that the service was available often enough to meet the patients' needs, it would not matter whether the service was provided by the hospital or some contractual arrangement with another facility as long as the service was available at a hospital within a hospital service area. I have, therefore, hypothesized a positive relationship between all four measures of the provision of services and hospital use rates. I am less certain of the relationship between services provided by a contractual partner and use because those services may be less well known to a community. Hospital Characteristics Five measures of a hospital's character and philosophy were included 1n this study. Two of these, the change in the number of available services from 1981 to 1983 and outpatient visits per population, were used to estimate the progressive or conservative philosophy of a hospital's administration. I hypothesized that the greater the change in absolute value (positive or negative) in the number of available services within a hospital service area, the more progressive was the hospital administration in that area, and therefore the higher the total use rates would be. I am less certain about the 81 strength of the positive relationship between surgical procedure rates and the change in the number of services than I am of the relationship between total use rates and medical causes for admission and the change in availability of services. This is the case because of the potentially ideosyncratic relationship between specific diagnostic and/or therapeutic equipment and individual surgical procedures. As described previously, new instruments encourage increased use, but those instruments are used in the diagnosis or treatment of a very limited number of patients. Without the data to examine specific new instruments, specific hypotheses cannot be generated. I hypothesized an inverse relationship between outpatient visits and all types of use rates. The reasoning was simply that as more of the health care needs of the community are fulfilled in the outpatient setting inpatient hospital use will decrease. The relationship of outpatient visits to specific surgical procedure rates was less clear, and since the surgical procedures chosen for this study all required hospitalization (inpatient) rather than outpatient surgery, outpatient visits and specific surgical procedure rates were hypothesized to be inversely related, but less strongly so than total and medical causes for admission rates. No published small area analysis research has dealt with the possible Impact that corporate ownership has had on hospital utilization rates. As the hospital environment has changed over the past several years, some hospitals have entered into corporate or quasi-corporate alliances with other hospital facilities. To a degree this might be seen as being another measure of progressive or conservative administrative philosophy, but actually it is different because the 82 initiative for these alliances does not always come from the hospital whose status is changed as a result of the legal move. In other words, one hospital may be encouraged or forced into an alliance with another hospital. The initiative for the alliance may have come from the second hospital's administration and yet the first hospital's status would have changed from non-corporate to a corporate status even though the first hospital's administration might still be considered conservative. As a consequence, the measure should be one of corporate dominance in the hospital service area. Corporate dominance was measured as the proportion of corporate beds in each hospital service area. I have hypothesized that as corporate dominance increases, use rates, particularly total admission and medical cause for admission rates, would also increase as the hospital "worked harder" to stay financially healthy. As shown in Figure 3.7, the explanatory role of teaching status has been tested in previous small area analysis research. Knickman and Foltz (1985) found that the supply of house staff (interns and residents) was not a significant contributor to the explanation of the variation in total admission rates. Studies by Stockwell and Vayda (1979) and by Vayda and his associates (1984) have found that areas with teaching hospitals have lower surgical procedure rates for appendectomies, cholecystectomies, prostatectomies, and hysterectomies. Griffith (personal communication 1986, 1987) found that some of Michigan's hospital service areas with graduate medical programs (Lansing, Grand Rapids and Ann Arbor) had lower surgical procedure rates, while others (Flint, Saginaw and Detroit) did not. Therefore, I hypothesized that all of the surgical procedure rates except Cesarean 83 section rates would have a negative relationship with house staff per 10,000 population. Cesarean section rates are known to be increasing in the United States and are particularly high in teaching hospitals. Therefore, I have hypothesized a positive relationship between Cesarean section rates and house staff per 10,000 population. The influence of osteopathic or allopathic philosophy on utilization rates has not been previously discussed in small area analysis literature. Few places offer the opportunity that Michigan does for studying the impact of osteopathic philosophy on hospital use rates. Thirteen Michigan hospital service areas contained osteopathic hospitals in 1983. This is a large enough number to evaluate their impact on use rates. Osteopathic philosophy emphasizes a more holistic approach to medicine than allopathic philosophy. Osteopathic medicine pays special attention to the musculoskeletal system's relationship to health and disease. One can hypothesize therefore that in areas with osteopathic hospitals the admission and surgery rates would be lower than in areas without osteopathic hospitals. The possible exception to this would be the orthopedic surgical procedure rates which would be hypothesized to be higher in areas with osteopathic hospitals. There are no orthopedic surgical procedures included in this study. A summary of the general hypotheses is shown in Figure 3.8. The direction of the relationships (positive or negative) are shown with + and - signs. The strength of my certainty in the hypotheses is also indicated: hypotheses that I feel most certain about have a double positive or negative sign, while those of which I am less certain have a single sign. Figure 3.8 RELATIONSHIPS HYPOTHESIZED IN THE METHODS CHAPTER . Wgt Prop Bd Cert Hosp Beds per 10,000 FTEs per 10,000 Pharm per 10,000 RNs per Hosp Bed Avg Hosp Serv *83 Avg Other Fac Serv '83 Serv Level per HSA '83 4 + + + + 4 4 Total Serv per HSA '83 0 Change in Serv '81-83 Out Pat Visits per 10,000 Corp Beds House Staff per 10,000 Ostec Beds Total Male Adm 4 + 4 4 4 4 — 4 + Total Female Adm 4 4 4 4 4 + - - + 4 4 + + 4 4 4 4 Total Adm 4 4 + + 4 4 - + + + 4 + 4 + 4 4 Appendectomy 4 + + + 4 4 - + + 4 4 4 + + 4 - 4 Hemhorrhoidectomy 4 4 + 4 4 - + 4 4 4 4 4 4 4 - 4 Choiecystectomy 4 4 + + 4 + - + + + + + + 4 4 - 4 Inguinal Hernia 4 4 + + + + - + 4 4 + + 4 4 4 - 4 Prostatectomy 4 4 + + + + - 4 + + 4 4 4 4 4 - 4 Hysterectomy 4 4 + + 4 + - 4 4 ♦ 4 + 4 4 4 - 4 Cesarean Section 4 + 4 + 4 + - + + + + 4 4- 4 4 - 4 4 4 - - Circulatory Adm + + + + 4 + + - - + + + + + 4- 4 4 4 ----- 4 4 -- — Respiratory Adm 4 4 + + + 4 + + + 4 + 4 4 4 4 Digestive Adm 4 + + + 4 4 - - + + 4 4 4 4- 4 4 4 Genito Urinary Adm + + + + + - + 4 4 + 4- 4 4 4 + + and - - + + indicate a more certain hypothesis , + and - - - + indie ate a less certa in hypot nesi s ----- 4 4 ---- — 4 4 - - 4 4 - - - - 4 4 - - 4 4 - - 4 4 - - CHAPTER 4 MICHIGAN HOSPITAL USE RATES Documentation of Variation in Hospital Use Rates Found in this Study Fourteen hospital use rates were calculated for each of the 53 Michigan hospital service areas and then standardized by the direct method using the age and sex of the 1980 Michigan population. Three types of use rates were studied: three total hospital use rates; seven surgical procedure use rates; and four medical causes for admission rates. Each of these use rates was compared to the results from previous small area analyses carried out in Michigan, North America, the United Kingdom, and Norway. A table of all fourteen admission and procedure rates by hospital service area is shown in Appendix A. In addition to reporting the ranges found, three measures of variation were calculated for each use rate: the maximum use rate divided by the minimum; the standard deviation divided by the mean (the coefficient of variation); and the systematic component of variation. The amount of variation from one hospital service area to another has been of utmost importance to small area analysis researchers. The first small area analysis studies used the simple measure of the maximum 85 86 use rate divided by the minimum use rate and discussed "2-fold" or "6 fold" differences from one hospital service area to another. As shown in Table 4.1, using that method for this study, hemorrhoidectomy rates had the greatest variation (18-fold), followed by Cesarean section rates (almost 6 -fold), and respiratory causes for admission rates (4-fold). The difficulty with this measure is that only the extreme use rates have been used to calculate the measure, giving undue importance to the use rates which are at the ends of the distribution and hence least representative of it. Table 4.1 Results of Three Measures of Use Rate Variability Total Male Admissions Total Female Admissions Total Admissions Appendectomy Hemorrhoidectomy Cholecystectomy Inguinal Hernia Prostatectomy Hysterectomy Cesarean Section Circulatory Diag Respiratory Diag Digestive Diag Genito-Urinary Max/Min . S.D./Mean SC V 2 .00 0.15 0.14 0.14 0.26 0.53 0.19 0.23 0.23 0.17 24 23 1.91 1.95 2.98 18.16 2.34 3.18 2 .86 2.15 5.90 2.97 4.03 2.45 2.41 0 .2 0 0.19 0.30 0.21 0.2 0 22 58 199 26 44 34 17 34 38 115 50 41 The coefficient of variation reduces the effect of the outliers. As shown on Table 4.1, using the coefficient of variation, hemorrhoidectomy rates had the highest variation followed by respiratory causes for admission and appendectomy rates. 87 The newest measure of variation to be found in the small area analysis literature is the systematic component of variation (SCV) (McPherson et al., 1981 and other references cited therein) which reduces the random sources of variation within the measurement. As shown in Table 4.1, the highest SCV was noted for hemorrhoidectomy rates followed by respiratory causes for admission and appendectomy rates. As shown in Table 4.2, hemorrhoidectomy rates had the greatest variation no matter which calculation of variation was used. This is in keeping with previous research. Respiratory admission rates ranked second highest in two of the three measures of variation and third when the maximum rate was divided by the minimum rate. This too was expected from previous results. The appearance of the Cesarean section rate as one of the most variable procedure rates was expected, but the high variation in appendectomy rates and inguinal hernia repair rates was unexpected since they have not been considered to have a great deal of medical uncertainty surrounding either their diagnoses or the treatment of choice, surgery. The smallest amount of variation was seen in the total hospital use rate, the sex-specific use rates and the hysterectomy rate. The ranking of the hysterectomy rate as having low variation was unexpected. 88 Table 4.2 Use Rates Ranked Within Each of Three Measures of Variation Max/Min S.D./Mean SCV Hemorrhoidectomy Cesarean Section Respiratory Inguinal Hernia Appendectomy Circulatory Prostatectomy Digestive Genito-Urinary Cholecystectomy Hysterectomy Male Admissions Total Admissions Female Admissions Hemorrhoidectomy Respiratory Appendectomy Inguinal Hernia Prostatectomy Cesarean Section Genito-Urinary Digestive Cholecystectomy Circulatory Hysterectomy Male Admissions Total Admissions Female Admissions Hemorrhoidectomy Respiratory Appendectomy Digestive Inguinal Hernia Genito-Urinary Circulatory Prostatectomy Cesarean Section Cholecystectomy Male Admissions Female Admissions Total Admissions Hysterectomy Comparison of the Variation in Use Rates Reported in this Study To Previous Results From Michigan and Elsewhere Results from previous small area analysis studies have been tabulated by total admissions, surgical procedures, and medical causes for admissions. Comparisons of the results from this current research to earlier work were made. Of particular note were the comparisons made between the results of research by Griffith et al. (1981) calculated from 1978 Michigan data and this current research with its results calculated from 1983 Michigan data. Because so few studies use the SCV, the comparisons using that measure of variation will be discussed separately. 89 Total Hospital Use Rates Total admissions per 10,000 age and sex adjusted population ranged from 1146 to 2235 in this study. As shown on Table 4.3, the figures from this current study were generally higher than ones reported previously. The mean of the previous comparable studies' minimum rates was 1041 and the mean of the maximum rates was 1764. When compared to previous Michigan research, the maximum/minimum ratio of 1.95 reported from this study is the same as reported by Griffith et al_. (1981) but lower than the ratio reported by either of the 0HMA reports (1985). The mean coefficient of variation reported in the literature was .17, only .01 greater than the coefficient of variation found for total admissions in this study. Gender-specific total admission rates were not available from previous small area analysis studies for comparison. Surgical Procedure Rates A large number of small area analysis studies have focused on seven surgical procedures. These seven have not formally been recognized as "index" procedures but, because they appear so frequently in the literature, many researchers have studied them and use rate ranges have been published for many geographic areas of the world and at many different times. The seven procedures most frequently studied are: appendectomy, cholecystectomy, hemorrhoidectomy, hysterectomy, inguinal hernia repair, prostatectomy, and tonsillectomy with or without adenoidectomy. Because tonsillectomy is now frequently performed in an outpatient setting, it was not included as one of the surgical TABLE 4.3 TOTAL HOSPITAL ADMISSION OR DISCHARGE RATES Location Study Clark, 1988 Clark, 1988 Clark, 1988 (males) (females) Brewer & Freedman, 1982 Chiswick, 1976 Connell, Day & Logerfo, 1981 Deacon et a K , 1979 Deacon et a K , 1979 Griffith et al., 1981 Knickman, 1982 Knickman & Foltz, 1984 Knickman & Foltz, 1984 Knickman & Foltz, 1985 Knickman & Foltz, 1985 0HMA, 1985 OHMA, 1985 Vladeck, 1985 Wennberg & Gittelsohn, 1973 Wennberg et al.., 1975 Wennberg et al., 1975 Wennberg et aH., 1977 Ranges in Admissions or Discharges per 10,000 Population Michigan Hosp Serv Areas Michigan Hosp Serv Areas Michigan Hosp Serv Areas 1146 - 2235 945 - 1893 1338 - 2559 Vermont Hosp Service Areas U.S. States & SMSAs Washington Hosp Serv Areas U.S. PSR0 Regions (Medicare) HEW Regions (Medicare) Michigan Hosp Service Areas Los Angeles & NYC Los Angeles & NYC SMSAs Los Angeles & NYC SMSAs Michigan Counties (Med/Surg) Michigan Counties NYC Zip Codes Vermont Hosp Service Areas Maine Hlth Planning Regions Vt & Maine Hosp Serv Areas Vermont Hosp Service Areas 908 830 653 2280 2960 ^Calculated by Clark, 1988 **Per 10,000 person years (Medicaid children) 1000 1051 1080 1148 1050 1130 604 668 897 1220 1504 1270 1270 - 2007 2480 1617** 4460 4060 1900 1176 1113 1195 1100 1180 1556 1661 2532 1970 2035 2350 2200 Max/Min Ratio 1.95 2 .0 0 1.91 2 .2 3.0 2.48 1.96 1.37 1.9 Coefficient of Variation .16* .15* .14* .17 .24 .30* .11* .10* 1.12 1.03 1.04 1.05 1.04 2.58 2.49 2.82 1.61 1.35 2.04 1.73 .07* .16* .15 .22 * .20 * 91 procedures studied in this current research. Instead, Cesarean section rates were included as the seventh surgical procedure for study. Appendectomy Rates Appendectomy rates for the small area analysis studies are shown in Table 4.4. Appendectomy rates ranged from 8 to 22 per 10,000 age and sex adjusted population in this study. The minimum and maximum rates reported for this study are both lower than the mean minimum rate of 12 and the mean maximum rate of 30 calculated from earlier comparable research. The maximum/minimum ratio of 2.75 is slightly higher than the mean (2.45) of previous comparable small area analysis appendectomy research. The only comparable Michigan research (Griffith et al., 1981), shows far more variation using the maximum/minimum ratio than does this study. With the exception of the Lewis (1969) study in Kansas, the coefficient of variation for appendectomy rates lies between .10 and .32 in all previous small area analysis research. The results from this current study (.26) are within that range. TABLE 4.4 APPENDECTOMY RATES Ranges in Adms. or Discharges per 10,000 Population Max/Mi n Ratio Coefficient of Vari ati 8-22 2.75 .26* Study Locat ion Clark, 1988 Michigan Hosp. Service Areas Barnes et ajk, 1985 Detmer T~Tyson, 1976 Gittelsohn & Wennberg, 1976 Gri ff ith et a I., 1981 Lembcke, T952Lewis, 1969 McPherson et al., 1981 II McPherson et aj_., 1982 Stockwell 5~Vayda, 1979 Vayda 4 Anderson, 1975 Massachusetts Minor Civil Div Wisconsin Hlth Planning Regions Vermont Hosp Service Areas Michigan Hosp Service Areas New York Counties Kansas Hlth Planning Regions Canadian Provinces U.S. Regions New England Hosp Service Areas Ontario Counties Canadian Provinces (male) 11 (female) " (1968) " (1972) Ontario Counties (1973) " (1975) " (1977) Vermont Hosp Service Areas Maine Hosp Service Areas Maine, Vermont Hosp Serv Areas Rhode Island Hosp Service Areas Maine Hospital Service Areas Vermont Hospital Service Areas Rl, ME, VT Hospital Service Areas Rhode Island Hosp Service Areas Maine Hospital Service Areas Vermont Hospital Service Areas Maine Hospital Service Areas 8 - 12** 12 - 26 14 - 31 10 - 48 29 - 71 15 - 62 12 - 20 12 - 16 8-19 12 - 57 21 - 31 17 - 29 20 - 30 16 - 30 13 - 54 10 - 40 12 - 34 10 - 32 11-22 11-23 7 - 15** 10 - 20** 11 - 25** Eng and 4 Wales Health Districts Norway Counties W.Mdlands, U.K. Health Districts 13 - 19 10 - 16** 12 - 24** It Vayda et a I., 1976 II Vayda et a I., 1984 it ii Wennberg Wennberg Wennberg Wennberg 4 Gittelsohn, 1973 4 Gittelsohn, 1975 et a I., 1975 eT 5T., 1982 II 11 Wennberg et al., 1982 Wennberg JTGTFte Isohn, 1982 II II Wennberg, et aj_., 1984 McPherson et a I., 1981 McPherson eT aT., 1982 II ‘Calculated by Clark, 1988 “ Range estimated from graphs 7 - 15** 10 - 20** 10 - 18** 1.5 2.15 2.21 4.8 2.45 4.23 1.67 1.33 2.38 4.8 1.5 1 .7 1 .5 1.88 4.15 4.02 2.84 3.20 2.0 2.09 2.14 2.0 2.27 2.0 2.14 2.0 1.8 2.3 1 .50 1.6 2.0 .21* .27* .10* .52* .26 .32* .27* .26* .18* .20 .21 .18 .12* .16 .16 93 Hemorrhoidectomy Rates The variation in hemorrhoidectomy rates has been studied less frequently. As shown in Table 4.5, the rates reported by Lewis (1969) were the highest. Prior to the completion of this current research, which reported a 17-fold difference between the lowest rate and the highest rate, the highest maximum/minimum ratio for hemorrhoidectomy rates reported was Griffith's 1981 study in Michigan, which found a 14fold variation in hemorrhoidectomy rates. The results of this current research showed both maximum and minimum hemorrhoidectomy rates lower than found by Griffith et 66) Ontario Counties Canadian Provinces (male) " (female) " (1968) II Roos 4 Roos, 1981 Stockwel1 4 Vayda, 1979 Vayda 4 Anderson, 1975 II Vayda et al., 1976 It II 11 Vayda et al., 1984 II It Wennberg Wennberg Wennberg Wennberg Ranges in Adms. or Discharges per 10,000 Population 4 Gittelsohn, 1973 4 Gittelsohn, 1975 et al., 1975 eT ST., 1982 II It Wennberg et al., 1982 Wennberg 4 GTFtelsohn, 1982 II II McPherson, et al., 1981 McPherson, eT aT., 1982 II -- ♦Calculated by Clark, 1988 ♦♦Range estimated from graphs " (1972) Ontario Counties (1973) " (1975) " (1977) Vermont Hosp Service Areas Maine Hosp Service Areas Maine, Vermont Hosp Serv Areas Fihode Island Hosp Serv Areas Maine Hosp Service Areas Vermont Hosp Service Areas Vt, Me, R.l. Hosp Serv Areas Fihode Island Hosp Serv Areas Maine Hosp Service Areas Vermont Hosp Service Areas England 4 Wales Hlth Districts Morway Counties VI.Midlands, U.K. Hlth Districts 15 14 25 21 18 14 12 27 19 18 16 20 36 21 4 27 16 18 22 31 27 16 17 27 25 ia. 19 17 - 36 22” 52 29 53 53 42 46 26 35** 40” 28” 99 102 18 63 40 40 45 68 47 39 57 55 55 33” 29** 25** Max/Min Ratio 2.4 1.57 2.1 1.37 2.94 3.7 3.50 1.70 1.37 1.9 2.5 1.4 2.80 4.9 4.1 2.3 2.58 2.26 2.05 2.22 1.76 2.49 3.35 2.04 2.20 1.83 1.53 1.47 1 . 8 18 - 33” 19 - 30” 14 - 26” 6 - 9 6 - 12” 6 - 12” Coefficient of Variation .\9* .09 .11* .29* .37* .18 .24* .19* .17* .17* .14* .17* .22* .24 .18 1.83 1.58 1 . 8 6 1.58 1.5 1.5 . 1 1 * .18 .16 96 procedures per 10,000 population, and a maximum/minimum ratio of 2.4. The Vayda and Anderson study (1975) confirmed previous clinical findings that cholecystectomy was predominantly a procedure done on females. Within the Canadian provinces studied, the cholecystectomy rates were from 4 to 18 per 10,000 males and from 27 to 63 per 10,000 females. The coefficient of variation ranged from .09 in Manitoba to .37 in Kansas. The coefficient of variation for cholecystectomy in this current study was .19, the mean for all the comparable previous research results. Inguinal Hernia Repair In previous studies, inguinal hernia repair rates showed very little variation (Table 4.7) with most studies reporting less than a two-fold difference between the lowest to the highest rate. McPherson found very little variation in inguinal hernia repair rates in New England, Norway, and the West Midland area of the United Kingdom (McPherson et al., 1982), reporting a coefficient of variation of .001 and .002. The results of this current research were dissimilar on two points. First, my minimum Inguinal hernia repair rate was lower than that reported in most North American studies and more closely resembled the rates reported by McPherson et al. (1981, 1982) in England and Wales. The mean minimum rate reported previously was 22 and the mean maximum rate was 36. The second dissimilarity was that the results of this current research showed a greater amount of variability (3-fold) than any previous study. No previous published Michigan results were available for comparison. TABLE 4.7 INGUINAL HERNIA REPAIR RATES Ranges in Adms. or Discharges per 10,000 Population Max/Min Ratio Coefficient of Variation Study Location Clark, 1988 Michigan Hospital Serv Areas 10 - 31 3.1 Barnes et a k , 1985 Detmer 2TTyson, 1976 Gittelsohn & Wennberg, 1976 Lewis, 1969 McPherson, et£l_., 1981 Massachusetts Minor Civil Div Wisconsin Hlth Ping Regions Vermont Hosp Serv Areas (male) Kansas Hlth Planning Regions Canadian Provinces U.S. Regions Vt, Me, R.l. Hosp Serv Areas Manitoba Hosp Service Areas (>66) Canadian Provinces (male) " (female) Canadian Provinces (1968) 18 28 38 18 15 24 22 31 25 3 14 14 13 29 35 34 39 35 21 21 19 21 - 31“ 21 - 32“ 19 - 34“ 1.94 1.36 1.42 2.39 1.73 1.33 1.7 2.7 1.8 1.9 1 .82 1.78 1 .94 1.66 1 .71 1.71 1.69 1.71 1.48 1.52 1.79 1.5 1.48 1.52 1.79 19 - 28“ 1.7 .12 9-15 17 - 21“ 8 - 18“ 1.71 1.3 2.0 .15* .002 .002 McPherson et al., 1982 Roos & Roos, T981 Vayda 4 Anderson, 1975 II Vayda et al., 1976 ii — — it Wennberg Wennberg Wennberg Wennberg II & Gittelsohn, 1973 & Gittelsohn, 1975 et al., 1975 et aT., 1977,1979 - - Wennberg et al., 1982 ii m Wennberg et aj_., 1982 Wennberg 4 Gittelsohn, 1982 n ti Wennberg et a K , 1984 McPherson, et a I., 1981 McPherson, eT aT., 1982 II ‘Calculated by Clark, 1988 “ Range estimated from graphs (1970) ’ (1972) Vermont Hosp Service Areas Maine' Hosp Service Areas Maine, Vermont Hosp Serv Areas Vermont Hosp Service Areas Maine Hosp Service Areas Rhode: Island Hosp Service Areas Maine Hosp Service Areas Vermont Hosp Service Areas Vt, Me, R.l. Hosp Serv Areas Rhode Island Hosp Service Areas Maine Hosp Service Areas Vermont Hosp Service Areas Maine Hosp Service Areas (per 10,000 person years) England 4 Wales Hlth Districts Norway Counties W.MicIands, U.K. Hlth Districts - 35“ - 38 - 54 - 43 - 26 - 32 - 38“ - 85 - 46 -6 - 26 - 24 - 25 - 48 - 60 - 58 - 66 - 60 - 31“ - 32“ - 34“ .23* .11* .12* .26* .001 .16* .15* .16* .14* .16 .12 98 Vayda and Anderson (1975) calculated both male and female inguinal hernia repair rates. The results from that study indicated that inguinal hernia repair was a procedure done predominantly on males, but a later study (Vayda et al.., 1976) did not confirm higher rates for males. Prostatectomy With the exception of one Vermont study (Wennberg and Gittelsohn, 1973), prostatectomy rates (Table 4.8) showed great consistency across all of the Wennberg studies of New England (Gittelsohn and Wennberg, 1976; Wennberg and Gittelsohn, 1982; and Wennberg et al., 1982) with ranges that had a slightly more than two-fold difference between low and high rates. The results of this current study had a slightly higher maximum/minimum ratio than those Wennberg and Gittelsohn reported in New England and had a higher than average minimum (15) and maximum (42) rate. Griffith et al. (1981) found an extremely high variation of over 16-fold between the lowest and highest prostatectomy rates in the 1978 Michigan data. A similar extreme variation result (maximum/minimum ratio of 15) was reported by Vayda et al.. (1984) in the prostatectomy rates of the Ontario counties studied. Comparison of the coefficient of variation results showed that they ranged from .19 to .33 in previous studies, with a mean of .25. The coefficient of variation for this current study was .24. TABLE 4.8 PROSTATECTOMY RATES Clark, 1988 Michigan Hosp Service Areas Barnes et a I., 1985 GittelsoHn-! Wennberg, 1976 Griffith et al., 1981 McPherson, eT“al., 1981 n — McPherson et a K , 1982 MindelI et al., 1982 Massachusetts Minor Civil Div Vermont Hosp Service Areas Michigan Hosp Service Areas Canadian Provinces U.S. Regions Vt, Me, R . t . Hosp Serv Areas Canadian Provinces (1968) " (1977) Manitoba Regions (>66) Canadian Provinces " (1968) " (1970) (1972) Ontario Counties (1973) " (1975) " (1977) Vermont Hosp Service Areas Maine Hosp Service Areas Rhode Island Hosp Serv Areas Maine Hosp Service Areas Vermont Hosp Service Areas Vt, Me, R.l. Hosp Serv Areas Rhode Island Hosp Serv Areas Maine Hosp Service Areas Vermont Hosp Service Areas — ii — Roos A Roos, 1981 Vayda A Anderson, 1975 Vayda et al., 1976 TT“ M Vayda et a I., 1984 T1 ii Wennberg A Gittelsohn, 1973 Wennberg A Gittelsohn, 1975 Wennberg et a I., 1982 II - II Wennberg et al., 1982 Wennberg A GrFtelsohn, 1982 It It McPherson, et al., 1981 McPherson, eT ST., 1982 I t ------------------------------ ------ ‘Calculated by Clark, 1988 “ Range estimated from graphs England A Wales Hlth Districts Norway Counties W.Midlands, U.K. Hlth Districts 15 - 42 29 - 33** 15 - 32 3-50 6-33 21 - 37 18 - 44** 14 - 26** 16 - 25** 125 - 282 14 - 25 14 - 26 13 - 25 14 - 27 3-38 6-37 6-34 11-38 18 - 40 18 - 40** 20 - 41** 12 - 33** - 19 - 39** 16 - 41** 13 - 33** 6-13 16 - 37** 7 - 20** 2.8 1.14 2.13 16.7 5.5 1.76 2.2 1.86 1.56 2.3 1.8 1.81 1.88 1.94 15.08 6.08 5.73 3.45 2.22 2.22 2.05 2.75 2.5 2.05 2.56 2.54 Coefficient of Var iatii m Locat ion Max/Min Ratio CN Study Ranges in Adros. or Discharges per 10,000 Popu1at ion .19* .30 .21* .20* .19* .30* .27* .25* .26* .30 2.28 .20* 2.2 2.1 .33 .24 100 Hysterectomy As shown in Table 4.9, the hysterectomy rates reported in the current research (32 to 69 procedures per 10,000 females) were similar to many of the other reported rates in North America. The mean minimum rate was 32 and the mean maximum rate was 74 for the comparable studies in the literature. The Canadian experience did not appear to be significantly different than that of the United States, although three studies (Dyck et al., 1977; Stockwell and Vayda, 1979; and Vayda et al_., 1984) found extraordinarily high upper ranges in their studies in Canada. The hysterectomy rates in England and Wales were considerably lower, as were the rates reported by Barnes et a K (1985) in Massachusetts. Griffith et al. (1981) found a greater than 6-fold difference between the highest and lowest hysterectomy rates in the 1978 Michigan data, while only a 2-fold difference was found in this current study using the 1983 Michigan data. The highest coefficient of variation (.31) was found in Norway, while the mean was .26. The coefficient of variation for hysterectomy in this current study was .17, far lower than expected. Cesarean Section Cesarean section procedure rates are of increasing interest to researchers because of the rapid increase in the number of Cesarean sections performed in the United States over the last several years. Anderson and Lomas (1985), Barnes et al. (1985), Mindell et al. (1982), Vayda and Anderson (1975), and Vayda et al_. (1984) have all documented TABLE 4.9 HYSTERECTOMY RATES Study Location Clark, 1988 Barnes et a[., 1985 Detmer S~Tyson, 1976 Dyck et a I 1977 Michigan Hospital Service Areas Massachusetts Minor Civil Div Wisconsin Hlth Planning Regions Saskatchewan Cities I P it Gittelsohn A Wennberg, 1976 Gri ffith et al., 1981 McPherson, eT”al., 1981 II McPherson et a K , 1982 MindelI et al., 1982 IP* Roos, N., 1984 StockwelI & Vayda, 1979 Vayda A Anderson 1975 Vayda et al_., 1976 Vayda et a I., 1984 Wennberg Wennberg Wennberg Wennberg A Gittelsohn, 1973 A Gittelsohn, 1975 et al., 1975 et aT., 1977,1979 *TT Wennberg et aj_., 1982 Wennberg et al., 1982 Wennberg A Gittelsohn, 1982 Wennberg et a k , 1984 McPherson, et al., 1981 McPherson, et aT., 1982 •Calculated by Clark, 1988 Vermont Hospital Service Areas Michigan Hospital Service Areas Canadian Provinces U.S. Regions Vt, Mu, R.l. Hosp Service Areas Canadian Provinces (1968) " (1977) Manitoba Regions Ontario Counties Canadian Provinces " (1968) " (1970) " (1972) Ontario Counties (1973) " (1975) " (1977) Vermont Hospital Service Areas Maine Hlth Planning Regions Me, Vt Hospital Service Areas Vermont Hospital Service Areas Maine Hospital Service Areas Rhode Island Hosp Service Areas Maine Hospital Service Areas Vermont Hospital Service Areas Vt, Mo, Rl Hosp Service Areas Rhode Island Hosp Service Areas Maine Hospital Service Areas Vermont Hospital Service Areas Maine Hospital Service Areas (per 10,000 person years) England A Wales Hlth Districts Norway Counties W.Midiands, U.K. Hlth Districts Ranges in Adms. or Discharges per 10,000 Max/Mi n Population Ratio 32 12 20 50 36 30 15 42 58 44 33 33 41 41 32 32 47 50 30 29 22 20 39 40 30 39 42 34 37 - 69 23** 34 126 63 60 95 73 79 96** 58** 68** 123 203 58 58 81 87 109 87 80 60 93 92 61 93 73** 88** 65** Coefficient Of Vari at ion .17* 42 - 74** 34 - 89** 25 - 64** 11-33 2.16 1.92 1.65 2.52 1.75 2.0 6.5 1.74 1.36 2.2 1 .76 2.06 2.8 4.9 1.81 1.81 1.72 1.73 3.65 2.98 3.69 3.0 2.38 2.30 2.03 2.38 1 .74 2.59 1.76 2.0 1.76 2.62 2.5 3.5 18 - 29 5 - 18** 13 - 30** 1.59 3.0 2.1 .12* .31 .20 .15* .22* .22 .18* .16* .18* .28* .25* .24* .25* .26 .23 .23 ••Range estimated from graphs 102 variations in the use of Cesarean sections for deliveries, but Anderson and Lomas (1985) and Vayda et al. (1984) calculated their rates as the number of Cesarean sections performed per 1000 births. As shown on Table 4.10, the minimum Cesarean section rate of 13 per 10,000 females reported by Vayda and Anderson (1975) was most similar to the results of this current research. The difference was that Vayda and Anderson (1975) reported a 2.5-fold variation in Cesarean section rates, while this research found a 6-fold difference. Barnes et a U (1985) found very little variability in Massachusetts. No earlier published results for Cesarean section rates were available from Michigan. Comparison of the Systematic Component of Variation Results Systematic component of variation results from previous research are available for six of the seven surgical procedure use rates studied. Table 4.11 shows comparable results from Wennberg et al. (1984), McPherson et al., (1981, 1982), and McCracken (personal correspondence). These results allowed comparison of the Michigan surgical procedure SCV's to those in New England, England and Wales, Norway, West Midlands of the U. K., Canada, and Iowa. With the exception of Iowa, there was more variation in the procedure rates for appendectomy, hemorrhoidectomy, cholecystectomy and inguinal hernia repair in Michigan than in any of the other areas for which there are SCV data available. Prostatectomy rates in Michigan had the lowest amount of variation, and only England and Wales had lower variation in hysterectomy use rates than Michigan. Michigan's use rate variation in TABLE 4.10 CESEAREAN SECTION RATES Ranges in Adms. or Discharges per 10,000 Population Max/Min Ratio Coefficient of Variation .20 Study Location Clark, 1988 Michigan Hosp Serv Areas 15 - 90 6.0 Barnes et al., 1985 Vayda S""An/ <' M 1 Jh I*'. rrr >/ •10 -20 -r BOO AXE T N. NONT T S A N ILAC 1— T " SCP 1 SCP 2 T SCP 3 -r SCP 4 T TUSCOLA HOSPITAL SERVICE AREA i-APPEN 5-PROST 2-+0CRR B-HYSTER a 3-CHOLE 7-C SECT K 3 4-HERNIA PX = Procedure SURGICAL SIGNATURES— HIGH USE AREAS 120 Low Use Areas Six hospital service areas were identified as low use areas. As shown in Figure 5.3, the surgical signatures for these six areas were dissimilar with one exception. The surgical signatures for Kalamazoo and Lansing were similar in direction (each procedure is higher or lower than the study area average) for six of the seven surgical procedure rates. Only Cesarean section rates were different, with Kalamazoo's rate above the study area's mean and Lansing's rate below. Five of the six remaining surgical procedure rates were very similar in magnitude as well as in direction. Appendectomy rates were lower in Lansing than in Kalamazoo. Grand Rapids had no surgical procedure rates higher than the study area mean, while only the prostatectomy rate was higher than the mean in Ann Arbor. 121 Figure 5.3 P X D I F F F R 0 M S T A T E A V E -to ■ -20 J ANN ARBOR OR RAPIDS KALANAZOO LANSINS SCP 5 SCP 6 HOSPITAL SERVICE AREA 1-APPEN 5-PROST 2-HEMORR 6-HYSTER m 3-CHOLE 7-C SECT E 3 4-+ERNIA PX = Procedure SURGICAL SIGNATURES— LOW USE AREAS 122 Pattern 4: Consistency Within Hospital Service Areas of High Use or Low Use Across Several Procedures or Diagnoses Another pattern which appears to have emerged from the small area analysis literature was that of use rate consistency across high or across low use areas. Griffith and his colleagues (1985), using information on hospital admissions in Michigan, analyzed admissions along a number of clinically descriptive dimensions: organ group categories, selected surgical procedures, and characteristics of length of stay and frequency of diagnosis. They found that hospital service areas with high total admission rates also tended to have high medical causes for admission rates, and, specifically, high respiratory admission rates. To investigate the fourth use rate pattern, the variations in the fourteen hospital use rates were mapped by hospital service area and their distributions described. Their spatial patterns are discussed using the six regions shown on Figure 5.4. A tabulation of the standard deviations for each use rate by hospital service area is shown in Appendix C. The regions are used to give some identification of relative location since the sole community provider areas cannot be described or shown on a map. A standard deviation classification system was chosen so that comparisons of hospital service areas could be made across all of the use rates. It was more important to this research to know if a hospital service area had consistently high or low use rates across several measures than to know the actual use rate distribution 123 NORTH EAST CENTRAL WEST CENTRAL SOUTH- SOUTHEAST WEST SOUTH CENTRAL Figure 5.4 Study Area Regions 124 and range for total admissions, procedure-specific admissions, or medical causes for admission. Total Hospital Use Rates When mapped by standard deviations, the spatial patterns for total male admissions, total female admissions, and total admissions were all similar. The areas with higher admission rates were in the northern Thumb area of the East Central Region, the center of the North Region, immediately northwest of Lansing in the West Central Region and one sole community provider hospital service area of the South Central Region. High use (> 1 S.D.) areas were all north of a line drawn from Detroit to Oceana, while, with the exception of one sole community provider area in the South Central Region, the low use areas (< -1 S.D.) were all south of that line. The areas of lower admission rates were the referral centers of the western part of the state, South Central and Southeast Regions, plus two contiguous sole community provider hospital service areas. Figure 5.5, Total Admissions per 10,000 Population, was representative of all three total hospital use rate maps. Surgical Procedure Rates Each of the seven surgical procedure rates (Figures 5.6 - 5.12) had a unique spatial pattern of high and low use areas. Most shared one or more of the high use areas identified from the maps of the total admission use rates. One hospital service area in the high use portion of the North Region had high use rates for cholecystectomy, hernia 125 Total Admissions Admissions/10,000 Population by Hospital Service Area J Mean = 1638 sd = 228 +2 sd 2094 +1 sd 1866 Mean 1638 Figure 5.5 Values between -1 sd and +1 sd are mapped with the same shade of grey. .Qoa P i m i r a 1 1 far th a h n o n ital panH aa araa natnaa 126 repair, prostatectomy and hysterectomy. One of the high use hospital service areas in the Thumb area of the East Central Region had high rates for appendectomy, hernia repair, prostatectomy and cesarean section, while both of the hospital service areas in the third high use area northwest of Lansing in the West Central Region had high rates for appendectomy and hernia repair procedures. The fourth high use area, a sole community provider hospital service area in the South Central Region, had high use for only one surgical procedure, prostatectomy. Appendectomy Appendectomy rates (Figure 5.6) were very high (>2 standard deviations above the mean) in the Bad Axe and Montcalm/Ionia hospital service areas and in one of the sole community provider areas in the North Region. One sole community provider service area in each of three regions (South Central, North and West Central), and the Holland hospital service area had appendectomy rates between 1 and 2 standard deviations above the mean. A sole community provider service area in the North Region plus one in the South Central Region, one in the West Central Region as well as the Midland, Oceana and Allegan hospital service areas had appendectomy rates between one and two standard deviations below the mean. Hemorrhoidectomy As shown in Figure 5.7, only one hospital service area (a North Region sole community provider) had a hemorrhoidectomy rate above 2 standard deviations above the mean. Eight hospital service areas (in a line from Port Huron southwest to Battle Creek) had hemorrhoidectomy 127 Appendectomy Admissions/10,000 Population by Hospital Service Area Mean = 13.2 sd = 3.4 +2 sd 20.0 +1 sd 16.6 Mean 13.2 -1 sd -2 sd Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 5.6 Values between -1 sd and +1 sd are mapped with the same shade of grey. See Figure 3.1 for the hospital servioe area names. 128 Hemo rrhoidectomy Admissions/10,000 Population by Hospital Service Area Mean = 4.32 sd = 2.28 +2 sd 8.88 +1 sd 6.60 4.32 -1 sd 2.04 -2 sd 0.00 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 5.7 Values between -1 sd and +1 sd are mapped with the same shade of grey. See Figure 3*1 for the hospital service area names* 129 rates that were between 1 and 2 standard deviations above the mean. Those hospital service areas between -1 and -2 standard deviations below the mean were Petoskey, one of the North Region sole community providers, and a distinct cluster of seven hospital service areas in the West Central Region that included Grand Rapids. Cholecystectomy Three hospital service areas (Figure 5.8) had a cholecystectomy rate greater than two standard deviations above the mean. One was a sole community provider in the North region, one was a sole community provider in the West Central Region, and the third was Allegan. Areas of high (1 S.D. - 2 S.D.) cholecystectomy use rates included one hospital service area in the North Region's cluster of sole community providers, one Southwest sole community provider area, plus the Saginaw, Tuscola, and Jackson hospital service areas. Hospital service areas with low cholecystectomy rates (-1 to -2 S.D.) included Ann Arbor, Northern Montcalm, Berrien and Cass counties, two sole community provider areas in the West Central Region of the state, and three contiguous hospital service areas in the East Central Region (Reed City, Mt. Pleasant, and Midland). Inguinal Hernia Repair Two hospital service areas, Northern Montcalm and Manistee, had very high (>2 S.D.) inguinal hernia repair rates (Figure 5.9). Two of the six sole community provider areas in the North Region, one sole community provider area in each of the West Central and Southwest Regions, plus the Bad Axe, Sanilac, Montcalm/Ionia and Holland service 130 Cholecystectomy Admissions/I0,000 Population by Hospital Service Area Mean = 21.6 sd = 4.0 +2 sd 2 9. 6 25.6 21.6 -1 sd 17.6 -2 sd 13.6 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 5.8 Values between -1 sd and +1 sd are mapped with the same shade of grey. See Figure 2.1 for the hcopitol service eree nemos. 131 Inguinal Hernia Repair Admissions/l0,000 Population by Hospital Service Area A Mean = 20.2 sd = 4.6 Mean Data Source: 20.2 1983 Michigan Inpatient Data Base Clark '88 Figure 5.9 Values between -1 sd and +1 sd are mapped with the same shade of grey. See Figure 3.1 for the hospital service area names. 132 areas had inguinal hernia repair rates that were between 1 and 2 standard deviations above the mean. Areas with low inguinal hernia repair rates (-1 to -2 S.D.) included Freemont, Oceana, Bay, Ann Arbor and one of the sole community providers in the Southeast Region. Only one hospital service area, a sole community provider in the Southwest Region, had a use rate for inguinal hernia repair which was lower than -2 S.D. Prostatectomy One hospital service area, Ann Arbor, had a very high (>2 S.D.) prostatectomy rate (Figure 5.10). Eleven hospital service areas, scat­ tered around the state, had rates that were between 1 and 2 standard deviations above the mean. One sole community provider area in the Southeast Region of the state had very low prostatectomy use rates (<-2 S.D.) while eleven hospital service areas had rates between -1 and -2 standard deviations from the mean. These eleven hospital service areas with low prostatectomy rates included Traverse City, Reed City, Gratiot, and Jackson as well as five others along the southern border (Three Rivers/Sturgis, Berrien County, Benton Harbor/St. Joseph and two sole community providers) plus one sole community provider hospital service area in each of the North and Southeast Regions. Generally, prostatectomy rates were higher in the eastern part of the state and lower in the western part. 133 Prostatectomy Admissions/10,000 Population by Hospital Service Area Mean = 28.0 sd = 6.5 +2 sd 41.0 +1 sd 34.5 Mean 28.0 -1 sd 21.5 -2 sd 15.0 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 5.10 Values between -1 sd and +1 sd are mapped with the same shade of grey. Soo Fi^jyro 3.1 for fho p a t ha s . 134 Hysterectomy Two sole community provider areas in the North Region, as well as the Northern Montcalm hospital service area, had very high (>2 S.D.) hysterectomy use rates (Figure 5.11). hysterectomy rates between 1 and 2 Four hospital service areas had standard deviations from the mean, including two sole community providers in the North Region, as well as Montcalm/Ionia and Port Huron. Eight hospital service areas had hysterectomy rates between -1 and -2 standard deviations. These included Petoskey, Oceana, Mt. Pleasant, Allegan, Benton Harbor/St. Joseph, Ann Arbor, and two sole community providers: one in the West Central Region and one in the North Region. Cesarean Section One hospital service area, a sole community provider in the Southwest Region, had Cesarean section use rates (Figure 5.12) greater than 2 standard deviations above the mean. Manistee, Bad Axe, Mt. Clemens, Montcalm/Ionia, Northern Montcalm and one West Central sole community provider hospital service area had Cesarean section rates between 1 and 2 standard deviations above the mean. In addition to the Tuscola hospital service area (-1 to -2 S.D.), lower than average use rates were found in a cluster of three hospital service areas including Mt. Pleasant, Reed City, and one sole community provider in the North Region. Adrian and Oceana were the only hospital service areas with Cesarean section rates which were lower than 2 standard deviations below the mean. 135 Hysterectomy Admissions/10,000 Population by Hospital Service Area Mean = 47.2 sd = 8.0 +2 sd 63.2 +1 sd 55.2 Mean 47.2 -1 sd 39.2 -2 sd 31.2 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 5.11 Values between -1 sd and +1 sd are mapped with the same shade of grey. Qaa r Hmi r a "5 1 fnr +*Ha V >r\ Hf a l aa a »•« a v>ama a 136 Cesarean Section Admissions/I0,000 Population by Hospital Service Area Mean = 57.6 sd = 11.7 +2 sd 81.0 +1 sd 69.3 Mean 57.6 -1 sd 45.9 -2 sd 34.2 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 5.12 Values between -1 sd and +1 sd are mapped with the same shade of grey. $a a FirriirA 3.1 for hospital area namaa 137 Medical Causes for Admission Rates Circulatory Diagnoses Only one sole community provider hospital service area in the North Region had a circulatory admission rate greater than 2 standard deviations above the mean (Figure 5.13). Hospital service areas with circulatory rates between 1 and 2 standard deviations above the mean were Cheboygan/Rogers City, Bay, Bad Axe, Tuscola, Sanilac, Mt. Clemens, Montcalm/Ionia and one sole community provider in each of the South Central and North Regions. Freemont, Holland, Kalamazoo, Lansing and one South Central sole community provider hospital service area had circulatory use rates between -1 and -2 standard deviations below the mean, while the Grand Rapids service area's circulatory admission rate was more than 2 standard deviations below the mean. With the exception of the sole community provider hospital service area in the South Central Region, all of the high use areas for circulatory admission use rates were north of an imaginary line running from Detroit to Oceana and all of the low use hospital service areas were south of that imaginary line. Respiratory Diagnoses One sole community provider hospital service area plus the Montcalm/Ionia hospital service area in the West Central Region had respiratory admission rates greater than 2 S.D. above the mean (Figure 5.14). High use ( 1 - 2 S.D.) areas were clustered around 138 Circulatory Diagnoses Admissions/10, 000 Population by Hospital Service Area Mean = 1946 sd = 387 +2 sd 2700 +1 sd 2323 Mean 1946 -1 sd 1569 -2 sd ................ 1182 Areas not studied and sole provider hospital service areas are shown in white. Data Source; Clark '88 1983 Michigan Inpatient Data Base Figure 5.13 Values between -1 sd and +1 sd are mapped with the same shade of grey. i ^ ^ a a iiv x iw e a4> «a n a iiie ^ • 139 Respiratory Diagnoses Admissions/10,000 Population by Hospital Service Area Mean = 1116 sd = 347 +2 sd 1790 +1 sd 1453 Mean 1116 -1 s d -2 sd SSI 779 432 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 5.14 Values between -1 sd and +1 sd are mapped with the same shade of grey. See Figure 3.1 for the hospital service area names. 140 Montcalm/Ionia, the northern Thumb area of the Eastern Region as well as two sole community provider hospital service areas in the South Central Region. Lower than average respiratory causes for admission rates were found in two sole community provider areas in each of the Southeast and West Central Regions as well as Grand Rapids, Kalamazoo and Sturgis/Three Rivers hospital service areas. Digestive and Genito-urinary Diagnoses The spatial patterns for the two other medical causes for admission use rates (Figures 5.15 - 5.16) were quite similar to the circulatory admission rate pattern and to the total hospital use rate map discussed above. Areas that were consistent (>1 S.D.) across these two use rate measures were the Thumb in the East Central Region and the Montcalm/Ionia hospital service areas. One sole community provider area in the North Region had higher than average digestive and genito-urinary use rates. The Kalamazoo, Grand Rapids and Lansing hospital service areas had medical causes for admission rates that were consistently lower than most other hospital service areas, while the Ann Arbor hospital service area only had low digestive use rates. With the exception of one sole community provider area, high (> 1 S.D.) digestive cause for admission rates were found north of a line from Detroit to Oceana and low (< -1 S.D.) digestive cause for admission rates were found south of the line. The area with the highest genito-urinary cause 141 Digestive Diagnoses Admissions/10/000 Population by Hospital Service Area Mean = 1926 sd = 407 +2 sd 2720 +1 sd 2323 Mean 1926 -1 sd 1529 -2 sd 1122 Areas not studied and sole provider hospital service areas are shown In white. Data Source: 1963 Michigan Inpatient Data Base Clark '88 Figure 5.15 Values between -1 sd and +1 sd are mapped with the same shade of grey. See Figure 3.1 for the hospital service area names. 142 Genito-Urinary Diagnose Admissions/10, 000 Population by Hospital Service Area Mean = 47.6 sd = 9.7 +2 sd 67.0 +1 sd 57.3 Mean 47.6 -1 sd 37.9 -2 sd 28.2 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 5.16 Values between -1 sd and +1 sd are mapped with the same shade of grey. SCC FljUrS 3.2. for the hospital ssrvics araa nawa a 143 for admission rates was in the East Central Region, particularly around the Saginaw Bay. Discussion of Use Rate Pattern #4 Previous small area analysis studies have indicated that for many hospital service areas (or other geographical units) there has been a consistently high or low use pattern across total admissions and several medical causes for admission; particularly respiratory causes for admission. This was the fourth use rate pattern identified by PaulShaheen et a K (1987) and was first discussed by Griffith et al. (1985). A tabulation of the standard deviations for each use rate by hospital service area is shown in Appendix C. Four high use areas were identified on the map of total admission use rates (Figure 5.5) and confirmed by inspection of the total male and total female admission rates. Some or all of these same four general areas (two hospital service areas in the North Region, three hospital service areas at the tip of the Thumb in the East Central Region, two hospital service areas northwest of Lansing in the West Central Region and one sole community provider hospital service area in the South Central Region) appeared in many of the maps of medical causes for admission rates (Figures 5.13 5.16) with higher than average use rates. Many of the surgical procedures had higher than average use rates within these high use areas as well (Figures 5.6 - 5.12). When at least one hospital service area in a high use area had a use rate of more than one standard deviation above the mean, the area was noted on Table 5.3. As shown in Table 5.3, high appendectomy use rates were present in three of the areas, high 144 Table 5.3 HIGH USE HOSPITAL SERVICE AREAS IN MICHIGAN NORTH REGION EAST CENTRAL REGION REGION (2 hospital USE RATE service areas) WEST CENTRAL REGION (3 hospital (2 hospital service areas) Total Male Admissions X X X Total Female Admissions X X X Total Admissions X X X Appendectomy X X Cholecystectomy X X X Inguinal Hernia Repair X X X Prostatectomy X Hysterectomy X service area) X X X X X Cesearean Section X X REGION (1 hospital X X Hemorrhoidectomy Circulatory Admissions service areas) SOUTH CENTRAL X Respiratory Admissions X X X X X Digestive Admissions X X X Genito-Urinary Admissions X X X X = at least one HSA in region has use rate > I SD above mean X 145 hemorrhoidectomy rates were present in one, high cholecystectomy and inguinal hernia repair rates were present in three of the four, high prostatectomy rates were present in three, high hysterectomy rates were present in two, and high Cesarean section rates were present in two. One of the four medical causes for admission (circulatory diagnoses) had high rates in all four high use areas. Respiratory, digestive and genito-urinary causes for admission rates were high in three of the four high use areas. From the results of this study, there appeared to be areas of consistently high use for total admissions, several surgical procedures and most medical causes for admission. Respiratory causes for admission rates did not seem to have more consistency in high use than did the other three medical causes for admission rates. When the low use pattern was analyzed, six specific hospital serv­ ice areas (rather than more generalized clusters of hospital service areas) had low total male, total female and total hospital admission rates. Four of the low use areas (Ann Arbor, Lansing, Grand Rapids, and Kalamazoo) were urban with tertiary care hospitals located within them while the fifth and sixth were sole community provider areas in the Southeastern and West Central Regions of the state. Each of the low use sole community provider areas was contiguous to one of the low use tertiary care hospital service areas. As shown on Table 5.4, all six low use hospital service areas were consistently low in their useage across all three measures of total admission plus digestive causes for medical admissions. They showed slightly less consistency across the 146 Table 5.4 LOW USE HOSPITAL SERVICE AREAS IN MICHIGAN USE RATE Total Male Admissions Total Female Admissions Total Admissions Appendectomy Hemorrhoidectomy Cholecystectomy Inguinal Hernia Repair Prostatectomy Hysterectomy Cesearean Section Circulatory Admissions Respiratory Admissions Digestive Admissions Genito-Urinary Admissions Sole Community Providers Grand KalaWest South Rapids mazoo Lansing Ann Arbor Central East X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X = at least one HSA has rate < -1 SD below the mean X X X X X X 147 remaining three medical causes for admission and no consistency within the surgical procedures. In several instances use rates would appear to cluster, taking in more than a single hospital service area. There are a cluster of areas above 1 standard deviation from the mean for hemorrhoidectomies (Figure 5.7). The cluster trends southwest from Sanilac and Port Huron through Flint and Lansing to Battle Creek plus one sole community provider area on the southern border. There is also a cluster of below average hemorrhoidectomy use rate areas in the west central region centered north of Grand Rapids. A cluster of hospital service areas with lower than -1 standard deviation below the mean was found for cholecystectomy use rates. It included three hospital service areas (Reed City, Mt. Pleasant, and Midland) in central Michigan. Two of those three hospital service areas plus one sole community provider in the North Region and Oceana also were in a low use cluster of hospital service areas for Cesarean section rates. It is not surprising that the patterns of use for these two procedures would be similar since cholecystectomy is more often performed on females than males and cesarean sections are only performed on females. In summary, high use areas appeared to be less localized geographically than low use areas and have more consistently high use across more measures, including surgical procedures. Low use appeared to be most consistent among the three total admission rates and the medical causes for admission rates and less consistent among the surgical procedure rates. This result supported Wennberg's hypothesis of surgical signatures, where specific procedures were either high or low within a hospital service area, and that the "surgical signature" of 148 that hospital service area would remain relatively constant over time. It was. not possible to fully test the surgical signature hypothesis without data from several years, but the uniqueness of the surgical use rate pattern in low use areas supported his theory. Further research into the relationships between the dominance of a physician graduate medical education program and/or physician specialty dominance within a hospital service area and the use rates in that area is necessary. CHAPTER 6 THE RELATIONSHIPS BETWEEN AND AMONG THE HOSPITAL USE RATES AND PROVIDER CHARACTERISTICS The third research goal was to test the relationships between and among the hospital use rates and the provider characteristics. Three sets of correlations were run using the data from 53 hospital service areas: among the hospital use rates; among the provider (both physician and hospital) characteristics; and between the hospital use rates and the provider characteristics. The correlations were run to test the relationships within the use rates, to test for multicollinearity among the physician and hospital characteristics and to help develop the research hypotheses to be used in the multiple regressions. The correlations were run using the SPSS-X System, Release 2.1 for IBM VM/MTS. Correlations Among the Hospital Use Rates As shown in Table 6.1, the Spearman correlations among hospital use rates were all positive with the exception of the relationships between prostatectomy procedure rates and both hemorrhoidectomy and hysterectomy 149 Table 6.1 SPEARMAN CORRELATIONS OF USE RATES Total Male Adm Total Female Adm Total Adm Appendectomy Cho1ecystectomy Inguinal Hernia Prostatectomy Hysterectomy Cesarean Section Circulatory Adm Respiratory Adm Digestive Adm Gen ito Ur inary Adm * = p <.001 $ = p <.01 Female Adm Total Adm Appen 1.000 .852* .946* .262 .360$ .463* .297 .971* .256 .285 .507* .245 .323 1.000 .150 1.000 1.000 Hemorr 1.000 Resp Adm Digest Adm Gen itoAdm C-Sect .127 .197 .007 .699* .836* .793* .691* .321 .087 .273 .212 .589* .799* .857* .716* .486* .321 .119 .248 .126 .649* .848* .861* .720* .322 .504* .068 .234 .185 .022 .305 .343 .278 .358$ .269 -.022 .275 .227 .208 .198 .275 .179 .218 .240 .220 .304 .347 .589* .312 .251 .291 .258 .114 .256 .465* .297 1.000 -.044 .154 .284 .005 .196 .279 1.000 .237 .167 .101 .255 .048 1.000 .118 .033 .134 .093 .513* .570* .550* .788* .610* 1.000 Hernia .430$ 1.000 Prost Ci re Adm Hyster Chole 1 .000 1.000 1.000 .762* 1.000 150 Hemorrhoi dectomy Male Adm 151 rates. In each instance, the correlation was weakly negative and not significantly related. As expected the three measures of total admissions and the four measures of medical causes for admission were strongly associated. The three measures of total admission rates were significantly (p<.001 ) associated with correlations of greater than .850. The medical causes for admission rates were all significantly (p<.001 ) associated, with correlations greater than .500. All three of the measures of total admission rates were significantly (p<.001 ) correlated with each of the four medical causes for admission rates. When surgical procedures rate correlations were examined only appendectomy with inguinal hernia repair rates and cholecystectomy with digestive admission rates had significant (p<.001 ) correlations. Cholecystectomy rates were also significantly (p<.001) correlated with all three measures of total admissions. The greatest association was between female total admission rates and cholecystectomy rates (.507) which is understandable since cholecystectomy is performed on females more frequently than on males. As a result of the correlation analysis of use rates, I hypothesized that the total admission rates and medical cause for admission rates would behave the same way when explanations were sought among the provider characteristics and when spatial analyses were done. Indeed, results described in Chapters 4 and 5 showed that areas of high use for total admissions also displayed higher than average use rates for many of the medical causes for admission. The relationships were even more pronounced for areas with both low total hospitalization use rates and areas with low medical causes for admission use rates. 152 Only two surgical procedure rates had significant (p<.001) correlations with medical causes for admission rates. Digestive admission rates were significantly correlated with cholecystectomy rates (.589) and with inguinal hernia repair rates (.465). Therefore, I hypothesized that there would be similarities in the explanations provided by the independent variables and in the spatial pattern of use among these three measures of use. The results of the regressions will be discussed in the next chapter. A comparison of the maps in Chapter 5 showed that the use patterns for these three measures of use (digestive admission, cholecystectomy, and inguinal repair rates) had very little similarity. Only two hospital service areas, a sole community provider service area in each of the North and West Central high use areas, were consistently between 1 and 2 standard deviations above the mean in all three use measures while Ann Arbor was the only hospital service area which had consistently low use (-1 to -2 S.D.) across all three measures of use. A significant association (p<.001) was also found between appendectomy and inguinal hernia procedure rates. A comparison of the two maps showed considerable similarity among the high use areas. The appendectomy use rate map (Figure 5.6) showed only three hospital service areas with higher than average use. All three of these hospital service areas also had higher than average inguinal hernia repair rates (Figure 5.9). Only one hospital service area had both lower than average appendectomy and inguinal hernia repair rates. In summary, the spatial similarities between total admission rates and medical causes for admission rates that have been described earlier in this paper were reconfirmed by the correlation results. With the 153 exception of appendectomy and inguinal hernia repair rates, spatial similarities were not found among surgical procedure rates that were significantly correlated. Correlations Among the Provider Characteristics A Spearman correlation was run among the fourteen provider characteristics. As reported on Table 6.2, the results showed that although there were significant associations among some provider characteristics, not one of the provider relationships was strong enough to be considered to show multicollinearity (.800), and therefore, all fourteen provider characteristics were entered into the multiple regressions. One measure of the physician component of the model, weighted proportion of board certified physicians to total physicians, was positively and significantly related to three measures of the hospital component. The significance level of the relationship between weighted proportion of board certified physicians and RNs per bed, total number of services available (out of 66 ) and corporate beds was p<.0 1 . Table 6.2 SPE/tRMAN CORRELATIONS OF PROVIDER CHARACTERISTICS Wgt Prop Bd Cert Phys Hosp Beds/ 10,000 FTEs/ 10,000 Pharm/ 10,000 1.000 .090 .204 .367* 1.000 Beds 1.000 FTEs Pharm Other Fac Serv '83 Serv Level/Hosp '83 f Serv out of 66 Serv *81-'83 OPD/Pop Corp Beds/Pop House Staff/Pop Ostseo Beds/Pop * = p <.001 Change in Serv ’81-83 .387$ -.031 .088 .066 .054 RNs/ Bed .274 .428$ .292 .068 -.326 .097 -.402$ -.130 -.052 .157 -.104 -.232 .500* OPD Visits per 10,000 House Staff per 10,000 Osteo Beds .367$ .191 .136 .286 .085 .257 .103 .184 .083 .268 .602* .312 -.056 -.063 .112 .287 .226 Corp Beds .185 .362$ .457$ .347 .136 -.445$ .320 .512* .001 -.440$ .568* .089 -.059 .362$ .367$ .121 -.141 -.765* .779* .391$ -.226 .299 .457$ .283 1.000 -.474* -.195 -.346 .056 -.170 -.090 .048 1.000 -.523* -.069 .120 -.178 -.319 -.293 1.000 Hosp Serv '83 Total Serv per HSA '83 Hosp Serv '83 .563* Serv Level/ HSA '83 0 Other Fac Serv '83 1.000 RN/Bed Tot Wgt Prop Bd Cert Phys 1.000 1.000 .410$ 1.000 -.185 .455$ .584* -.176 .250 .163 -.049 1.000 -.129 .068 -.235 1.000 .196 .189 1.000 .443$ .387$ 1.000 155 Correlations Between the Hospital Use Rates and the Provider Characteristics Total Hospital Use Rates and Provider Characteristics As shown in Table 6.3, there was a significant (p<.001) negative relationship between registered nurses per bed and both total female and total admission rates, and at the p<.01 level for total male admission rates as well. Significant negative relationships at the p<.01 level were found between the weighted proportion of board certified physicians and all three measures of total admission rates. At the p<.01 level of significance, negative associations were found between both total female admissions and total admission rates and both the services provided by the hospitals and total number of services available (out of 66 ) in the hospital service area. Surgical Procedure Rates and Provider Characteristics The correlations between the seven surgical procedure rates and the provider characteristics showed no significant relationships. Unlike the consistency in direction shown between the provider characteristics and the total admission rates, the surgical procedure rates had no directional consistency. No provider characteristic was either consistently positively or negatively related to all of the procedure specific use rates, nor did any surgical procedure rate have a consistent pattern of positive or negative relationship to the provider characteristics. This observation reinforces my hypothesis that Table 6.3 SPEARMAN CORRELATIONS OF USE RATES WITH PROVIDER CHARACTERISTICS Hosp Beds/ 10,000 FTEs/ 10,000 Total Male Adm -.356$ .231 .053 Total Female Adm -.382$ .314 Total Adm -.402$ Appendectomy Serv Leve1/ HSA '83 Total # Serv per HSA '83 Change in Serv '81-83 0PD Visits per 10,000 Corp Beds .105 -.226 RNs/ Bed Hosp Serv •83 Other Fac Serv '83 -.171 -.453$ -.268 .074 .221 -.277 .055 -.206 -.590* -.439$ .153 .293 -.428$ -.033 .281 .049 -.217 -.548* -.377$ .113 .279 -.373$ -.114 .093 .027 .013 -.301 -.278 -.099 .274 Hemorrhoi dectomy -.257 -.069 .123 -.003 .051 .066 -.277 Cho1ecystectomy -.256 .294 .217 .046 -.352 -.265 Inguinal Hernia -.330 .062 .088 .032 -.193 .104 .166 .271 .179 Hysterectomy -.331 -.112 .043 Cesarean Section -.131 .038 Circulatory Adm -.200 Respiratory Adm House Staff per 10,000 Osteo Beds -.145 -.181 .007 -.029 -.187 -.237 .012 .029 -.104 -.180 -.230 .002 -.292 .135 -.143 .014 -.076 -.098 .088 .098 .038 -.240 .153 .047 .156 .091 .145 -.294 -.145 -.086 -.209 -.071 -.010 -.103 -.084 .112 -.125 .137 -.054 -.093 -.008 .040 .078 -.055 .087 -.086 -.045 -.131 .081 -.143 .053 .009 -.052 -.153 .080 .057 -.110 -.013 .064 -.258 .081 .008 .325 .092 .125 -.042 -.252 -.044 .167 -.186 -.026 .180 -.163 -.076 .035 .294 .210 .017 -.293 -.087 -.004 .058 -.087 .257 -.138 -. 188 -.069 .054 -.417$ .172 -.025 -.223 -.458$ -.424$ .107 .331 -.366$ .054 -.043 -.097 -.223 -.041 Digestive Adm -.359$ .336 .113 -.139 -.514* -.472* -.004 .427$ -.391$ -.079 -.077 -.240 -.184 .025 Genito Urinary Adm -.115 .225 -.046 -.034 -.385$ -.436$ .184 .259 -.403$ -.114 -.044 -.260 -.256 .018 Prostatectomy * = p <.001 $ = p <.01 Pharm/ 10,000 156 Wgt Prop Bd Cert Phys 157 surgical procedure rates are ideosyncratic, responding to physician characteristics (largely untested in this research) rather than to hospital characteristics. Medical Causes for Admission Rates and Provider Characteristics The digestive admission rate was significantly (p<.001) negatively associated with both RNs per bed and services available in the hospital. The other significant associations between medical causes for admission rates and provider characteristics were at the p<.01 level and indicated that three of the medical causes for admission (respiratory, digestive and genito-urinary) might exhibit more similar behavior than the fourth, circulatory causes for admission. The only consistent positive relationship (although not significant) was between all four of the medical admission rates and the beds per population. This is in agreement with all previous research. In summary, the relationships found between the provider characteristics and the surgical procedure rates showed no strong associations and no consistent positive or negative pattern. Therefore, I hypothesized that very little similarity would be found in their explanations. The three total admission rates almost always had the same positive or negative direction in their correlation to each of the provider characteristics, although very few were significant. Therefore, I hypothesized similarities in the variables that would provide explanation for the three total admission rates and for all three of the four medical causes for admission rates. 158 Comparison of the General Hypotheses with Results of the Correlations The general hypotheses generated from the review of the small area analysis literature and from personal observation and experience were stated at the end of the Methods Chapter (Figure 3.8). These hypotheses were reviewed following analysis of the correlation results. The directional hypotheses I had generated are shown as the top line in each cell in Table 6.4. The lower symbol, shown in parentheses, is the direction of the correlation for that cell. The correlation results are only shown for significant relationships. In several cells the directional symbols do not agree. For example, I had hypothesized that the relationship between the proportion of board certified physicians to total physicians in each hospital service area and the hospital use rates would be positive. The significant correlations between the proportion of board certified physicians and five measures of use were all negative. Other areas of disagreement between my hypotheses and the correlation results include the relationship between both the mean number of services offered by the hospitals in 1983 and the total number of services (out of 66 possible) available in the hospital service area in 1983 and hospital use. Since the correlations test only a bivariate relationship, I decided, after review of the results of the correlation, to continue with the hypotheses as stated in the multivariate regressions. Table 6.4 COMPARISON OF HYPOTHESES AND SIGNIFICANT CORRELATION RESULTS Note: top line = hypo­ theses derived from literature and observation bottom line ( ) = signifleant cor­ relation results Total Male Adm Total Female Adm Total Adm Wgt Prop Bd Cert + (-) + (-) + (-) Hosp Beds per 10v000 FTEs per 10,000 + Pharm per 10,000 RNs per Hosp Bed + + + + + ■f + Avg Hosp Serv •83 Avg Other Fac Serv •83 Serv Level per HSA '83 Total Serv per HSA '83 § Change in Serv '81-83 Out Pat Visits per 10,000 + + + + + (-) (-) (-) + (-) + <-) + + + - + + + (-) + (-) + - + + + - + + Corp Beds House Staff per 10,000 Osteo Beds + - - - - + - - - + - - + Appendectomy + + + ♦ - + + Hemhorrho idectomy + + * + - + + Cholecystectomy + + + - + + + + + - + - - + - + ♦ + + + - + - - + - + + + + + - + - - + - + + + + + - + - - + - + + + + + - + + - ♦ - + + + + + - + - - - + (-) + (-) + (-) + + + - + - - + + + (-) + (-) + (-) + - + - - + - + - - Ui VO Inguinal Hernia Prostatectomy + + + Hysterectomy Cesarean Section + + Circulatory Adm + + Respiratory Adm + (-) + (-) + + Digestive Adm Genito Urinary Adm + +■ + + + ♦ * + + (-) (-) (-) + CHAPTER 7 HOSPITAL USE RATES AS A FUNCTION OF PHYSICIAN AND HOSPITAL CHARACTERISTICS Introduction The fourth goal of this research was to determine the amount of explanation of the variation in hospital use rates that could be provided by the physician and hospital components of the model. To do this the fourteen hospital use rates were used as dependent variables in fourteen multiple regressions. The independent variables used were the one available measure of the physician component of the model, plus the thirteen measures of the hospital component: four variables measuring hospital resources; four measuring services available within the hospi­ tal service area; and five measuring hospital organization and philos­ ophy characteristics. A stepwise multiple regression was run for each use rate using all of the independent variables. The statistical package used was Version 6.02 of the SAS System. The minimum significance level for variable entry and retention in the model was o*- = .10 . The results of each of the regressions are shown below. In each instance the discussion and analysis of the results are presented in four sections (Hypothesis, Equation Results, Variables Entering the 160 161 Equation, and Residuals). The hypothesis for the multiple regression is shown first as a matrix. The columns are the use rates being tested and the rows are the independent variables from the model that are being entered into the regression. Each cell in the matrix shows the positive (+) or negative (-) direction hypothesized for the independent variable if it should enter significantly into the equation. The relative certainty of the hypothesis is shown. Double directional signs (++ or — ) have more certainty than single directional signs (+ or -). The second section, Equation Results, deals with the explanatory value of the equation and the third section discusses the significant independent variables that entered the equation. The final section, Residuals, describes the results of the spatial analysis of the resid­ uals from the equation. Since community variables were not among the independent variables tested, any community contribution to the variation in use rates should appear in the residuals. A K-S test of the residuals was used to determine if they were normally distributed. Regressions of Age and Sex Adjusted Dependent Variables Total Male, Total Female and Total Admission Rates Hypothesis Fourteen measures of the physician and hospital components of the Clark Model were tested for their power in explaining the variation in total male, total female, and total admission rates. The hypothesized positive or negative entry of each independent variable into the regression equation is shown in Figure 7.1. 162 Figure 7.1 Hypotheses for Total Male, Total Female, and Total Admission Rate Regressions Total Male Admissions Physician Wgt Prop Bd Cert Hosp Beds per 10,000 Hospital Supply of Resources FTEs per 10,000 Pharm per 10,000 RNs per Hosp Bed Services Available Avg Hosp Serv '83 Avg Other Fac Serv '83 Avg Serv per Hosp '83 Total # Serv per HSA '83 Organization Change in Serv '81-'83 Out Pat Visits per 10,000 Corp Beds House Staff per 10,000 Osteo Beds H0 H.j : : b1 = b.j t Total Female Admissions Total Admissions + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ + + + + for 1 for at least one n- 0 0 Equation Results Table 7.1 Multiple Regression Equation Results Separately Run for Total Male, Total Female, and Total Admission Rates Dependent Var r£ Adj R2 F Value Tot Male Adm Tot Female Adm Total Adm 0.3892 0.5077 0.4827 0.3601 0.4842 0.4581 13.383*** 21.656*** 19.598*** ***P<.01 + ++ 163 For each of the first three hypotheses shown above, the critical F (2,42) for the equation was 2.44; therefore, in each instance (total male admissions, total female admissions, and total admissions), the null hypothesis was rejected and the alternative was accepted. The adjusted R2 in Table 7.1 shows that 36% of the variation in the total male admission rate, 48% of the variation in the total female admission rate, and 46% of the variation in total admission rate was explained by the independent variables entered into the regression equation. Each equation was significant at the p<.01 level. Variables Entering the Equations There was a negative relationship between hospital admissions rates (total male, total female, and total) and the number of registered nurses per capita, and the proportion of board certified physicians. As the number of registered nurses per bed and the proportion of board certified physicians decreased, the total use rate and gender-specific use rates increased. The proportion of board certified physicians had been hypothesized to enter the equation with a positive sign. As Table 7.2 shows, each of the terms in each of the three equa­ tions was significant at the p<.10 level. The beta weights in each equation showed that the contribution to the explanation of variation made by the registered nurses per bed variable was greater than that of the weighted proportion of board certified physicians variable. The tolerance levels of the two independent variables were above .84 in each of the three equations indicating their mutual independence. A correl­ ation analysis done previously showed that the association between RNs per Bed and Weighted Proportion of Board Certified Physicians was .428. 164 Table 7.2 Significant Variables for Total Male, Total Female and Total Admission Rate Regressions Intercept & Ind Var. Intercept RNs/Bed Wgt Prop Bd Cert Coefficient 10.47 14.47 Total Male Admissions 189.6 -50.5 18.10*** -3.49*** -0.46 17.45 -37.1 -2.13** -0.28 21.71*** -5.12*** -0.61 -1.77* - 11.69 16.16 Intercept RNs/Bed Wgt Prop Bd Cert Intercept RNs/Bed Wgt Prop Bd Cert *p<.10 Standard Error of B P< ,05 Total Female Admissions 253.8 -82.7 19.48 -34.5 10.49 14.50 Total Admissions 222.5 -67.0 17.48 -35.7 t statistic Beta Weights 0 .2 1 2 1 .22 *** -4.63*** -0.56 -2.05** -0.25 ***p<.01 Residuals The residuals for all three total admission multiple regression equations were analyzed for any spatial pattern (Figures 7.2, 7.3 and 7.4). The three equations all over-predicted the male, female and total admission rates for Port Huron, Reed City, South Berrien/Cass County plus two sole community provider hospital service areas: one in the South Central Region; and one in the West Central Region. The three equations under-predicted the total admission rates for Grand Rapids, Tuscola, Cheboygan/Rogers City, Ann Arbor, N. Montcalm, and Saginaw hospital service areas. One sole community provider service area in the 165 Male Admissions Multiple Regression Analysis Studentized Residuals [ +2.0 +1.0 - 2.0 Areas not studied and sole provider hospital service areas are shown In white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.2 Values between -1 sd and +1 sd are mapped with the same shade of grey. PH AaiieeA 9 1 f a** V U a U ah ssH + e l n a a a n r% m A e* 166 Female Admissions Multiple Regression Analysis Studentized Residuals +2.0 + 1.0 - 2.0 Areas not studied and aole provider hospital service areaa are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.3 Values between -1 sd and +1 sd are mapped with the same shade of grey. o _ 94 T uoo c r t 1 < 4•a r - - iiWii ■»_ v . _ uviaipAWCidb • 9eo iidiinsae 167 Total Admissions Multiple Regression Analysis Studentized Residuals +2.0 + 1.0 - 2/0 Areaa not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.4 Values between -1 sd and +1 sd are mapped with the same shade of grey. C>AA O m. 1 -wa_ waaw aa^w^-WM^-.1 MM MM M*M m*“ M' M- ■ a . Mil — v a w ww-.w-. _w w -w w » «* 168 North Region was under-predicted. There appeared to be no urban-rural or regional component to the residuals from the multiple regressions using the three total admission rates as the dependent variables. A K-S test of the residuals verified their normal distribution. Appendectomy Hypothesis Fourteen measures of the physician and hospital components of the Clark Model were tested for their power in explaining the variation in appendectomy use rates. The hypothesized positive or negative entry into the regression equation are shown in Figure 7.5. Figure 7.5 Hypothesis for Appendectomy Use Rate Regression Appendectomy Physician Wgt Prop Bd Cert Hosp Beds per 10,000 Hospital Supply of Resources FTEs per 10,000 Pharm per 10,000 RNs per Hosp Bed Services Available Avg Hosp Serv '83 Avg Other Fac Serv '83 Avg Serv per Hosp '83 Total # Serv per HSA '83 Organization Change in Serv '81-'83 Out Pat Visits per 10,000 Corp Beds House Staff per 10,000 Osteo Beds + + + + + + + + + + + + + + 169 H0 : : bi = bn- f 0 0 for i for at least one n- Equation Results The critical F (3,41) for the equation was 2.23. The calculated value of F was 5.71. Therefore, the null hypothesis was rejected and the alternative was accepted. As shown in Table 7.3, the adjusted R2 indicated that 24% of the variation in appendectomy rates could be explained by the independent variables that entered into the regression equation. The equation was significant at the p<.01 level. Table 7.3 Multiple Regression Equation Results for Appendectomy Rates R2 . Adjusted R^ F Value 0.2947 = 0.2431 = 5.71 p<.01 Variables Entering the Equation As Table 7.4 shows, all of the terms of the equation were signifi­ cant at the p<.10 level. There was a positive relationship between the appendectomy rate and the FTE's per capita and a negative relationship between the appendectomy rate and the outpatient visits per capita and the number of services (out of a possible 66 ) available within the hospital service area. As the number of FTE's per capita increased and the number of outpatient visits and available services decreased, the appendectomy rate increased. The number of available services had been hypothesized to enter the equation with a positive sign. The beta weights showed that the outpatient visits per capita contributed most to 170 the equation, followed by the total number of services available in the hospital service area and the number of FTE's per capita. The tolerance levels were all above .58, indicating their mutual independence. A correlation analysis done previously showed that the highest correlation among the independent variables in this equation was between the number of FTE's per population and the number of services available in the hospital service area. The two independent variables were positively correlated (.500). Table 7.4 Significant Variables for Appendectomy Rate Regression Intercept & Inde Var. Coeffi­ cient Intercept 1.73 FTEs/Capita 0.003 Tot # Serv per HSA -0.00003 0.01 OPD/Capita ★** p<.01 V.IO - Standard Error of B t statistic Beta Weights 0.001 11.03*** 1.75* 0.30 0.58 0.000008 0.004 -3.17*** -3.41*** -0.46 -0.57 0.83 0.61 0.16 Toler­ ance Residuals The residuals were mapped (Figure 7.6) and showed that the equation under-predicted the appendectomy rates in Midland, Saginaw, Grand Rapids, Flint, Berrien/Cass County, Allegan and two sole community providers (one in the Southeast Region and one in the South Central region). The equation over-predicted the appendectomy rates in Bad Axe, Lapeer, Jackson, Benton Harbor/St. Joseph, Northern Michigan and in five sole community providers; one in the North Region, one in the Southeast 171 Appendectomy Multiple Regression Analysis Studentized Residuals +2.0 +1.0 - 2.0 Areaa not studied and sole provider hospital service areas are shown In white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.6 Values between -1 sd and +1 sd are mapped with the same shade of grey. See Figure 3.1 for the hospital service area names. 172 Region, two in the West Central Region and one in the Southwest Region. There does not appear to be any urban/rural or regional component to the residual. The K-S test of the residuals verified their normal distribution. Hemorrhoidectomy Hypothesis Fourteen measures of the physician and hospital components of the Clark Model were tested for their power in explaining the variation in hemorrhoidectomy use rates. The hypothesized positive or negative entry into the regression equation are shown in Figure 7.7. Figure 7.7 Hypothesis for Hemorrhoidectomy Use Rate Regression Hemorrhoidectomy Physician Wgt Prop Bd Cert Hosp Beds per 10,000 Hospital Supply of Resources FTEs per 10,000 Pharm per 10,000 RNs per Hosp Bed Services Available Avg Hosp Serv '83 Avg Other Fac Serv '83 Avg Serv per Hosp '83 Total # Serv per HSA '83 Organization Change in Serv '81-'83 Out Pat Visits per 10,000 Corp Beds House Staff per 10,000 Osteo Beds + + + + + + + + + + + 173 H0 : b1 =0 ^ 0 for i for at least one 1- Equation Results The critical F (3,41) for the equation was 2.23. The calculated value of F was 3.4. Therefore, the null hypothesis was rejected and the alternative was accepted. As shown in Table 7.5, the adjusted R2 indicated that 14% of the hemorrhoidectomy rate could be explained by the independent variables in the equation. The equation was significant at a p<.05 level. Table 7.5 Multiple Regression Equation Results for Hemorrhoidectomy Rates R2 _ Adjusted R F Value 0.2001 = 0.1416 = 3.419 p<.05 Variables Entering the Equation As Table 7.6 shows, each of the terms of the equation was significant at the p<.10 level. There was a negative relationship between the hemorrhoidectomy rate and three independent variables. As the number of outpatient visits per capita, the proportion of board certified physicians and the number of services provided by a contractual party decreased, the hemorrhoidectomy rate increased. Both the outpatient visits per capita and average number of services provided by another facility had been hypothesized to enter the equation with a positive sign. The beta weights showed that the three independent variables contributed almost equally to the explanation. The tolerance 174 levels for all the independent variables were very high, confirming their orthogonality. A correlation analysis had previously indicated their independence. Table 7.6 Significant Variables for Hemorrhoidectomy Regression Intercept & Inde Var. Coeffi­ cient Intercept OPD/Capita Wgt Prop Bd Cert Avg Other Fac serv *p<.10 Standard Error of B t statistic 0.84 -0.00001 0.15 0.000006 5.78*** -1.70* -0.24 0.98 -0.37 0.22 -1.70* -0.24 0.98 -0.004 0.002 -1.71* -0.24 0.99 Beta Tolei Weights ance ***p<.01 Residuals The residuals were mapped (Figure 7.8) and showed that the equation under-predicted the hemorrhoidectomy rates for Northern Michigan, Fusco la, Ann Arbor, Mt. Clemens, Adrian, Midland, Manistee, Oceana, and one of the sole community provider areas in each of the South Central and West Central Regions. The equation over-predicted the hemorrhoidectomy rates in Port Huron, Mt. Pleasant, Allegan, Three Rivers/Sturgis, Fremont, and Reed City. There was one cluster of three hospital service areas in the West Central Region that were over­ predicted by the equation. The K-S test of the residuals verified their normal distribution. 175 Hemor rhoidect omy Multiple Regression Analysis Studentized Residuals +2.0 +1.0 - 2.0 Areaa not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.8 Values between -1 sd and +1 sd are mapped with the same shade of grey. See Figure 3.1 fcr the hospital 5017/102 area names. 176 Cholecystectomy Hypothesis Fourteen measures of the physician and hospital components of the Clark Model were tested for their power in explaining the variation in cholecystectomy use rates. The hypothesized positive or negative entry into the regression equation are shown in Figure 7.9. Figure 7.9 Hypothesis for Cholecystectomy Use Rate Regression Cholecystectomy Physician Wgt Prop Bd Cert Hosp Beds per 10,000 Hospital Supply of Resources FTEs per 10,000 Pharm per 10,000 RNs per Hosp Bed Services Available Avg Hosp Serv '83 Avg Other Fac Serv '83 Avg Serv per Hosp '83 lotai # Serv per H5A '83 Organization Change in Serv '81-'83 Out Pat Visits per 10,000 Corp Beds House Staff per 10,000 Osteo Beds H0 Hn- : : bi b.j = 0 f 0 + + + + + + + + + + + for 1 for at least one ^ 177 Equation Results The critical F (2,42) for the equation was 2.44. The calculated value of F was 4.9. Therefore, the null hypothesis was rejected and the alternative was accepted. As shown in Table 7.7, the adjusted R2 indicated that 15% of the variation in cholecystectomy rates could be explained by the two independent variables entered into the regression equation. The equation was significant at a p<.01 level. Table 7.7 Multiple Regression Equation Results for Cholecystectomy Rates r 2 Adjusted R2 F Value 0.1893 0.1507 4.904 p<.0l Variables Entering the Equation As Table 7.8 shows, each of the terms of the equation was significant at the p<.10 level. There was a positive relationship between the cholecystectomy rate and the number of beds per capita, while there was a negative relationship between the cholecystectomy rate and the weighted proportion of board certified physicians. As the number of beds per capita increased and the proportion of board certified physicians decreased, the cholecystectomy rates increased. The weighted proportion of board certified physicians had been hypothesized to enter the equation with a positive sign. The beta weights showed that the proportion of board certified physicians contributed more to the equation than did the variable measuring beds 178 per capita. The tolerance level for each independent variable was extremely high, showing their mutual independence which was previously reported from a correlation analysis (0.090). Table 7.8 Significant Variables for Cholecystectomy Regression Intercept & Inde Var. Intercept Beds/Capita Wgt Prop Bd Cert *p<.10 Coeffi­ cient Standard Error of B t statistic Beta Weights Toler ance 2.33 0.31 0.01 0.01 7.53*** 1.83* 0.26 0.98 -1.08 0.39 -2.75*** -0.38 0.98 ***p<.01 Residuals When the residuals were mapped (Figure 7.10), the cholecystectomy use rates for Northern Michigan, Midland, Saginaw, North Montcalm, Muskegon, Allegan, Kalamazoo plus one sole community provider in each of the Southeast and South Central Regions were under predicted by the equation. The cholecystectomy rates in Three Rivers/Sturgis, Fremont, Bad Axe, Sanilac, Port Huron, Benton Harbor/St. Joseph, and Pontiac hospital service areas as well as one Southwest Region sole community provider area were over-predicted. There appeared to be no urban/ rural component to the residuals, but three hospital service areas in the eastern edge of the Thumb were all over-predicted by the equation. The K-S test of the residuals verified their normal distribution. 179 Cholecystectomy Multiple Regression Analysis Studentized Residuals +2.0 +1.0 - 2.0 Areas not studied and sole provider hospital service areas are shown in white. Data Source; 1963 Michigan Inpatient Data Base Clark '88 Figure 7.10 Values between -1 sd and +1 sd are mapped with the same shade of grey. Sea Figure 3.1 for the hospital service area names. Inguinal Hernia Repair, Hysterectomy and Cesarean Section Hypotheses Fourteen measures of the physician and hospital components of the Clark Model were tested for their power in explaining the variation in inguinal hernia repair, hysterectomy, and Cesarean section use rates. The hypothesized positive or negative entry into the regression equation are shown in Figure 7.11. Figure 7.11 Hypotheses for Inguinal Hernia Repair, Hysterectomy, and Cesarean Section Use Rate Regressions Physician Wgt Prop Bd Cert Hosp Beds per 10,000 Hospital Supply of Resources FTEs per 10,000 Pharm per 10,000 RNs per Hosp Bed Services Available Avg Hosp Serv *83 Avg Other Fac Serv '83 Avg Serv per Hosp '83 Total # Serv per HSA '83 Organization Change in Serv '81-'83 Out Pat Visits per 10,000 Corp Beds House Staff per 10,000 Osteo Beds H0 H.j : : bn* = bi ? 0 0 Ing Hernia Hysterectomy C-Section + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for ^ for at least one ^ 181 Equation Results Table 7.9 Multiple Regression Equation Results Separately Run For Inguinal Hernia Repair, Hysterectomy and Cesarean Section Rates Dependent Var. R^ Adjusted R2 F Value Inguinal Hernia Hysterectomy Cesarean Sect 0.1198 0.2005 0.0993 0.1819 0 0 5.853** 10.784*** 2.250 **p<.05 ***p<.01 The critical F (1,43) for each equation was 2.84. Therefore, the null hypothesis was rejected and the alternative hypothesis was accepted for the inguinal hernia repair regression equation and for the hysterectomy equation. The adjusted R2 indicated that 9% of the variation in inguinal hernia procedure rates and 18% of the variation in hysterectomy rates were explained by the independent variable entering the equation. These two equations were significant at at least a p<.05 level. In the multiple regression for the Cesarean section rates, the critical F (1,43) for the equation was 2.84, while the F value was 2.2520; therefore, the null hypothesis was accepted. There was no relationship between the independent variables and the Cesarean section use rate. Variables Entering in the Equation As shown in Table 7.10, the weighted proportion of board certified physicians was the only independent variable which was significant in explaining inguinal hernia repair rates and hysterectomy rates. There 182 was a negative relationship between the weighted proportion of board certified physicians and the inguinal hernia procedure rate and also it and the hysterectomy rate. As the proportion of board certified physicians decreased in a hospital service area, the inguinal hernia repair rate and the hysterectomy rate increased. The weighted proportion of board certified physicians had been hypothesized to enter the equations with a positive sign. No physician or hospital characteristic provided significant explanation to the Cesarean section use rate. Equation Results Table 7.10 Significant Variables for Inguinal Hernia Repair Rates and Hysterectomy Use Rate Regressions Intercept & Inde Var. Intercept Wgt Prop Bd Cert Intercept Wgt Prop Bd Cert *★1p<.05 Coeffi­ cient Standard Error of B t statistic Inquinal Hernia Repair 2.54 0.24 10.50*** -0.99 0.41 -2.42** 6.01 Hysterectomy 0.43 14.07*** -2.38 0.73 -3.28*** *** p<.01 Residuals The residuals from the inguinal hernia use rate equation were mapped (Figure 7.12). The regression equation for inguinal hernia repair under-predicted the use rates for nine hospital service areas. These service areas included Tuscola, Saginaw, and Mt. Pleasant in the 183 Inguinal Hernia Repair Multiple Regression Analysis Studentized Residuals +2.0 +1.0 - 1.0 - 2.0 Areas not studied and sole provider hospital service areas are shown In white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.12 Values between -1 sd and +1 sd are mapped with the same shade of grey. SCC Fi^UJTe 3.1 servic® Aroa narno-q 184 East Central Region, Grand Rapids, Montcalm/Ionia, and Allegan in the West Central Region, and Kalamazoo and Three Rivers/Sturgis in the Southwest Region. The equation under-predicted the Traverse City hospital service area in the North Region. The regression equation over-predicted the inguinal hernia repair rates for Port Huron, Benton Harbor/St. Joseph, Cheboygan/Rogers City, Pontiac, Lapeer and one sole community provider hospital service area in the West Central Region. There appeared to be no urban/rural or regional pattern to the residuals. The K-S test of the residuals verified their normal distribution. The residuals from the hysterectomy multiple regression equation were mapped (Figure 7.13) and showed under-prediction by the regression equation for nine hospital service areas. These nine included Northern Michigan, Manistee and two sole community provider areas in the North Region, as well as the Grand Rapids, Kalamazoo, Three Rivers/Sturgis and Mt. Pleasant hospital service areas, plus one sole community provider area in the Southeast Region. Over-prediction of the hysterectomy rates occurred in Muskegon, Sanilac, Bay, Lansing, and Benton Harbor/St. Joseph hospital service areas. There appeared to be no urban/rural or regional component to the residuals. The K-S test of the residuals verified their normal distribution. 185 Hysterectomy Multiple Regression Analysis Studentized Residuals +2.0 + 1.0 - 2.0 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.13 Values between -1 sd and +1 sd are mapped with the same shade of grey. caa 3 .i fAt* +ka hcsp^ ssrvics a nsntss 186 Prostatectomy Hypothesis Fourteen measures of the physician and hospital components of the Clark Model were tested for their power in explaining the variation in prostatectomy use rates. The hypothesized positive or negative entry into the regression equation are shown in Figure 7.14. Figure 7.14 Hypothesis for Prostatectomy Use Rate Regression Prostatectomy Physician Wgt Prop Bd Cert Hosp Beds per 10,000 Hospital Supply of Resources FTEs per 10,000 Pharm per 10,000 RNs per Hosp Bed Services Available Avg Hosp Serv '83 Avg Other Fac Serv '83 Avg Serv per Hosp '83 Total # Serv per HSA '83 Organization Change in Serv '81-'83 Out Pat Visits per 10,000 Corp Beds House Staff per 10,000 Osteo Beds H_ H^ : : b.b.j = 0 f 0 + + + + + + + + + + + + + + + for .• for at least one ^ 187 Equation Results Table 7.11 Multiple Regression Equation Results for Prostatectomy Rates R2 _ Adjusted R F Value = = = 0.1005 0.0796 4.805 p<.05 The critical F (1,43) for the equation was 2.84. The calculated value of F was 4.8. Therefore, the null hypothesis was rejected and the alternative hypothesis was accepted. As shown on Table 7.11, the adjusted R^ indicated that slightly less that 8 % of the variation in prostatectomy rates could be explained by the independent variable. The equation was significant at the p<.05 level. Variables Entering the Equation As shown on Table 7.12, one term in the equation was significant at the p<.05 level. There was a positive relationship between the prostatectomy procedure rate and the number of full time equivalent personnel per population. As the number of FTEs per capita increased, the prostatectomy rate increased. Table 7.12 Significant Variables for Prostatectomy Regression Intercept & Inde Var. Coefficient Standard Error of B t statistic Intercept FTEs/Capita 2.08 0.34 0.01 0.00 6.13*** 2.19** **p<.05 ***p<.01 188 Residuals When the residuals were mapped (Figure 7.15), the regression equation over-predicted the prostatectomy rates for eleven hospital service areas. These eleven hospital service areas included two in the North Region, two in the West Central Region, four in the Southwest Region, one in the South Central Region and two in the Southeast Region. The prostatectomy use rate equation under-predicted the twelve hospital service areas shown in Figure 7.15, plus one sole community provider hospital service area in the Southeast Region. The pattern did not appear to be regional or urban/rural in nature. The K-S test of the residuals verified their normal distribution. 189 Prostatectomy Multiple Regression Analysis Studentized Residuals +2.0 - 2.0 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.15 Values between -1 sd and +1 sd are mapped with the same shade of grey. Sa a TPS miro .1 fnr +•>■>** hnflnHhal p o t -tH a roa nam® p 190 Circulatory Admissions Hypothesis Fourteen measures of the physician and hospital components of the Clark Model were tested for their power in explaining the variation in circulatory causes for admission use rates. The hypothesized positive or negative entry into the regression equation are shown in Figure 7.16. Figure 7.16 Hypothesis for Circulatory Admission Rate Regression Circulatory Physician Wgt Prop Bd Cert Hosp Beds per 10,000 Hospital Supply of Resources FTEs per 10,000 Pharm per 10,000 RNs per Hosp Bed Services Available Avg Hosp Serv '83 Avg Other Fac Serv '83 Avg Serv per Hosp '83 iota! # Serv per HSA '83 Organization Change in Serv '81-'83 Out Pat Visits per 10,000 Corp Beds House Staff per 10,000 Osteo Beds + + + + + + + + + + + + + + + + + + for ^ for at least one .• 191 Equation Results Table 7.13 Multiple Regression Equation Results for Circulatory Causes for Admission Rates R2 _ Adjusted R F Value = = 0.3231 0.2735 6.523 p<.01 The critical F (3,41) for the equation was 2.23. The calculated value of F was 6.5. Therefore, the null hypothesis was rejected and the alternative wasaccepted. As shown on Table 7.13, the adjusted R2 indicated that21% of the variation in circulatory admissionrates could be explained by the three independent variables entered into the regression equation. The equation was significant at a p<.01 level. Variables Entering the Equation Table 7.14 Significant Variables for Circulatory Causes for Admission Regression intercept & Inde Var. Coeffi­ cient Intercept Pharm/Capita RNs/Bed Change in Serv 22.72 2.84 -11.73 0.29 **p<.05 Standard Error of B 1.69 1.22 3.11 0.11 t statistic 13.46*** 2.34** -3.77*** 2 .66*** Beta Weights Toler ance 0.35 -0.56 0.35 0.74 0.75 0.98 ***p<.01 As Table 7.14 shows, each of the terms in the equation was significant at a p<.05 level. There was a positive relationship between the circulatory admission rate and two of the three independent variables and a negative relationship between the circulatory admission 192 rate and one of the independent variables. As the supply of pharmacists per population increased, and the change in the number of services offered from 1981 to 1983 increased, the circulatory admission rate increased. As the registered nurses per bed decreased, the circulatory admission rate increased. The beta weights showed that the contribution made by the registered nurses per bed to the equation was slightly larger than that of the pharmacists per population and the change in the services (from 1981 to 1983) available in the hospital service area. The tolerance levels of the three independent variables were all above .73 and confirmed their mutual independence. A correlation analysis previously run had indicated this finding. Residuals When the residuals from the circulatory admission equation were mapped (Figure 7.17) the circulatory admission rates for eight hospital service areas were under-predicted. These hospital service areas included Jackson, Mt. Pleasant, Lapeer, Bay, Gratiot, Cheboygan/Rogers City, and two sole community provider areas; one in the West Central Region and one in the Southeast Region. With the exception of Port Huron and one Southeast Region sole community provider area, over­ prediction of the circulatory admission rates occurred in the west from Reed City south through Muskegon, Allegan, Berrien/Cass County, Three Rivers/Sturgis and one sole community provider area in the Southwest Region. There did not appear to be any urban/rural component to the residuals. The K-S test of the residuals verified their normal distribution. 193 Circulatory Diagnoses Multiple Regression Analysis Studentized Residuals +2.0 +1.0 - 1.0 - 2.0 Areas not studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.17 Values between -1 sd and +1 sd are mapped with the same shade of grey. Sow rly ulo 3.1 fCx uuo hwspxual boiviCw ctxud ucuuea • 194 Respiratory Admissions and Digestive Admissions Hypothesis Fourteen measures of the physician and hospital components of the Clark Model were tested for their power in explaining the variation in respiratory and digestive causes for admission rates. The hypothesized positive or negative entries for each independent variable into the regression equation are shown in Figure 7.18. Figure 7.18 Hypotheses for Respiratory and Digestive Causes for Admission Rate Regressions Physician Wgt Prop Bd Cert Hosp Beds per 10,000 Hospital Supply of Resources FTEs per 10,000 Pharm per 10,000 RNs per Hosp Bed Services Available Avg Hosp Serv '83 Avg Other Fac Serv '83 Avg Serv per Hosp '83 Total # Serv per HSA '83 Organization Change in Serv '81-'83 Out Pat Visits per 10,000 Corp Beds House Staff per 10,000 Osteo Beds H0 H^ : : bi b.j = 0 f 0 Respiratory Digestive + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for 1 for at least one n- + + 195 Equation Results Table 7.15 Multiple Regression Equation Results Run Separately For Respiratory and Digestive Causes for Admission Rates Dependent Var. r£ Respiratory Digestive 0.3768 0.4046 Adjusted R2 F Value 0.3471 0.3762 12.695*** 14.269*** ***p<.01 The critical F (2,42) for each equation was 2.44; therefore, the null hypothesis was rejected and the alternative was accepted in each case. The adjusted R^ (shown in Table 7.15) indicated that slightly less than 35% of the variation in respiratory admission rates and 37% of the variation in digestive admission rates could be explained by the two independent variables entered into the equation. Variables Entering the Equation As shown on Table 7.16, each term in the equation was significant at the p<.10 level. There was a negative relationship between each of the two causes for admission rates and two independent variables (RNs per bed and weighted proportion of board certified physicians). As the supply of registered nurses per hospital bed and the weighted proportion of board certified physicians decreased, the respiratory and digestive admission rates increased. The weighted proportion of board certified physicians had been hypothesized to enter the equations with a positive sign. The beta weights showed that the contribution made by the two independent variables was almost identical in the respiratory regression equation. The registered nurses per bed contributed more than the 196 weighted proportion of board certified physicians did to the digestive admission equation. High tolerance levels (>.84) indicated the mutual independence of the two independent variables. Previously, a correlation of - 0.428 between RNs per bed and weighted proportion of board certified physicians was reported. Table 7.16 Significant Variables for Respiratory and Digestive Admission Rate Regressions Intercept & Ind Var. Coefficient Standard Error of B t statistic Beta Weights Respiratory Intercept RNs/Bed Wgt Prop Bd Cert 19.17 -6.30 1.67 2.31 11.46*** -2.72*** -0.36 -7.85 2.79 -2.82*** -0.37 Digestive Intercept RNs/Bed Wgt Prop Bd Cert *p<.10 27.67 -9.50 1.80 2.48 15.41*** -3.83*** -0.50 -5.72 2.99 -1.91* -0.25 ***p<.01 Residuals The residuals were mapped (Figures 7.19 and 7.20) and showed some similarity in their patterns for under-prediction. Both equations under-predicted respiratory and digestive admission rates for the Bay, Saginaw, North Montcalm and Holland hospital service areas. Six other hospital service areas were under-predicted for respiratory use rates and three were under-predicted for digestive use rates. Over-prediction of the respiratory and digestive admission rates occurred in Port Huron, Muskegon, Reed City, and one sole community provider area in the West 197 Respiratory Diagnoses Multiple Regression Analysis Studentized Residuals +2.0 + 1.0 - 1.0 - 2.0 Areas noC studied and sole provider hospital service areas are shown in white. Data Source: 1983 Michigan Inpatient Data Base Clark '88 Figure 7.19 Values between -1 sd and +1 sd are mapped with the same shade of grey. Stiti Fi^uitt 3*1 Z1 S.D) than average. The urban or rural character of a hospital service area is a community variable and therefore was not studied in this research. The rural nature of many of the high use areas indicates that it is an important variable that should be included in future work. Twelve of the fourteen hospitals within the high use areas, and therefore six of the eight hospital service areas, have fewer than 100 beds. As shown in Table 8.1, the average number of licensed hospital beds in the high use hospital service areas was 74, while in the low use areas it was 271. This is not solely a question of the size of the hospitals, but rather of the whole health care delivery system and the stress that the current industry environment places on small rural hospitals. Recent research I directed at the MHA (Hamilton, Clark and Lester, 1988) showed that the smaller hospitals in Michigan are currently under a great deal of hospital industry environmental stress. That would have been true, although to a lesser degree, in 1983 as well as today. The hospital industry environmental stress on smaller hospitals has exhibited itself in several ways. First, the small rural hospitals have had difficulty recruiting registered nurses and physicians. As a result of recruitment problems and lay-offs, the number of registered nurses 222 per bed has decreased. This is true of all hospitals, but is particularly critical in smaller hospitals where the margin for change is reduced because of the smaller starting point. The potential impact on hospital use rates of the reduced number of RN's per bed is supported by the comparison shown on Table 8.1. The high use areas had an average of .44 registered nurses per bed, while the low use areas had an average of .85 registered nurses per bed. Table 8.1 Comparisons of High Use and Low Use Areas identified in Chapter 5 High Use Area Mean (8 HSAs) Licensed Beds per Hospital Weight Proportion of Board Certified Phys. Registered Nurses per Bed House Staff per 10,000 Pop Diagnostic Services (Out of 7 Possible) 74 Study Area Mean (53 HSAs) 172 Low Use Area Mean (6 HSAs) 271 .42 .44 1.32 .58 .58 1.71 .61 .85 6.80 3.00 4.10 5.50 The negative relationship between registered nurses per bed and use rates can be better understood if one considers the histogram (Figure 8 .1) of the distribution of registered nurses per bed by hospital service area. There is virtually no tail to the left of this distribution. The sharp rise at .4 RNs per bed indicates a threshhold effect. Clearly one cannot have a hospital bed in operation without nurses on staff to administer care to the patient in that bed. But as hospital industry environmental pressures such as recruitment problems, 223 Figure 8.1 Frequency Histogram for RNs per Bed (shown with normal curve) 10 V +1 1 C O3 * L CA * 8 6 4 2 0 0 0.3 0.6 0.9 BNs per Bed 1.2 1.5 224 declining revenues, and occupancy rates make an impact on a hospital, the nurses are asked to provide care to more and more patients in the occupied beds. The number of patients an RN is able to care for can be increased to the threshhold level that the state and the Health Care Financing Administration (Medicare) require. Also, because of the difficulty in recruiting physicians, the proportion of board certified physicians to total physicians has decreased in the small rural hospitals. As shown on Table 8.1, the high use areas had a weighted average of 42% board certified physicians, while the low use areas had a weighted average of 61% board certified physicians. Anecdotal information provided by the hospital administrators suggests that the physicians in many smaller hospitals are aging and that physicians who are currently initiating practices in smaller hospitals are less apt to be board certified specialists, less apt to be graduates of United States medical schools, and more apt to be graduates of foreign medical schools who are board eligible but not board certified. I think that one way that the decrease in the proportion of board certified physicians affects use is by the following mechanism: as physicians increase their training and expertise in their area of specialty, they are more likely to use more diagnostic tests and to use alternative health care settings outside the hospital. Both the increased use of diagnostic tests and the use of alternative health care settings would cause a reduction in inpatient hospital use. 225 Factor 2: Inequality of the Distribution of High Technology Diagnostic Equipment The second factor in the explanation of hospital use rates also is within the hospital component of the model. Because of financial and regulatory constraints, smaller hospitals have had a great deal of difficulty in obtaining the new technology that has become available to larger, urban hospitals. To test the effect that the lack of access to new technology may have upon use, the hospital services available in 1983 were re-examined, and those that were diagnostic services were noted. Seven types of diagnostic services or equipment were included within the list of available services in 1983. These seven diagnostic technologies included cardiac catheterization, CT scanner, diagnostic radioisotope, electrocardiogram, histopathology, ultrasound and nuclearmagnetic resonator equipment. As shown on Table 8.1, the high use areas had an average number of 3.0 diagnostic services per hospital service area while the low use areas had an average number of 5.5 diagnostic services available. I think that the lack of access to new diagnostic technology increases the hospital use rates because physicians cannot be as certain of their diagnoses, nor can the diagnoses be made as rapidly when high technology diagnostic equipment is not present. In these times of defensive medicine due to the difficulties associated with and expense of malpractice insurance, I think physicians in hospital service areas lacking high technology diagnostic equipment are more apt to use the inpatient facilities of the hospital and less apt to delay or avoid hospitalization for their patients. 226 Factor 3: Rural Referral Centers The third factor in the explanation of higher use in small rural hospital service areas also lies within the hospital component of the model. Specifically, there is an inequality in the rural hospital environment produced by the designation of some hospitals as rural referral centers. With the advent in 1983 of DRGs and the prospective payment system for Medicare patients, a higher level of reimbursement was given to rural hospitals who met the criteria for being designated a rural referral center. These criteria included being an acute care hospital in a rural area that had either: 1) more than 500 beds or 2) had both 50% of its Medicare patients referred from another hospital or by physicians who did not have admitting privileges at the hospital and had at least 60% of its Medicare patients and services provided to people who lived more than 25 miles from the hospital. None of the fourteen hospitals in the high use hospital service areas has been designated as a rural referral center. The two high use sole community provider hospital service areas in the North Region are surrounded by four rural referral center hospital service areas. As a consequence of the designation as a rural referral center, those hospitals that are not rural referral centers have lower patient revenues, and less ability to maintain a medical education program or to purchase high technology equipment. As a consequence the rural hospitals that are not referral centers tend to attract fewer board certified physicians and have fewer house staff (interns, residents and salaried physicians). As a consequence, the weighted proportion of Ill board certified physicians in the high use areas was 4.2 and in the low use areas was 6.1. The average number of house staff per 10,000 population in the high use areas was 1.32, while in the low use areas it was 6.80. The importance of these hypothesized causative factors has yet to be tested, but the differences in the measures between high use and low use areas are striking. Further research will need to be conducted to test the power of other physician and hospital characteristics in explaining hospital use rates. Two hospital characteristics that should be tested are immediately apparent from the previous discussion. They are the number of beds that are currently being used (set up and staffed) per hospital and the number of high technology diagnostic services per hospital service area. An additional hospital characteristic that may have an impact on the hospital use rates is the occupancy rate. By testing the explanatory power of occupancy rate, one would be isolating the unused and available hospital resources from a measure of the total historical resources (beds per population) in the hospital service area. Further research is necessary in the area of physician character­ istics as well. Few physician characteristics have been studied previously, and those studies yielded inconclusive results. As described earlier (Chapter 2), I have hypothesized that physician training, specialty and length of time in a community will be influential in explaining hospital use rates. The results of this research have strengthened this hypothesis for several reasons. the percent of explanation found among the physician and hospital characteristics tested in this study was low. There is further First, 228 explanation of hospital use rates to be determined. Second, results from this current research indicated that Wennberg's classification of procedures and medical causes for admission into low, medium, high, and very high variation groups did not match the classification in Michigan. One possible way to explain this difference is that Wennberg's classification system is not stable. The differences found between Michigan and New England bring into question Wennberg's hypothesis that procedures that have high levels of medical consensus have low variation in use rates and that procedures that have low levels of medical consensus have high variation in use rates. Procedures that were thought to have high medical consensus did not necessarily have lower use rates. This suggests to me that surgical training may be of importance in creating medical consensus and, therefore, in influencing surgical procedure rates. Further work needs to be done in Michigan to determine if the procedure-specific use rates are stable over time and if Michigan perhaps has its own (and different) classification of low, medium, high and very high variation procedures due possibly to a difference in the medical education of its physicians. The influence of medical training, particularly surgical medical education should be studied further. One word of caution should be introduced here. This research used 1983 data and the hospital industry is not the same today as it was in 1983. The beginning of some very important changes in the hospital industry within Michigan and elsewhere within the United States were initiated in the fourth quarter of 1983 when the DR6 system was introduced into some of the hospitals in the United States. The DRG system was phased in at all United States hospitals following 1983. The 229 change to a prospective payment system for Medicare and then Medicaid patients has had an immense impact upon Michigan's hospitals and health care. It has been observed that dramatic decreases in admissions and length of stay had begun prior to the institution of the DRG system and have continued to the present. Although the influence of this revolutionary change in incentives has not been addressed in this paper, the changes that have occurred in the hospital industry since the end of 1983 are continuations of trends that started in 1983 or before. Therefore, this research gives an indication of changes that were to become more extreme over time and longitudinal studies should be undertaken to study the impact of the DRG system and to monitor the use rate patterns over time. Factor 4: The Impact Upon Hospital Use Rates of the Size and Definition of the Hospital Service Area Small area analysis methodology is based upon the notion of a hospital service area defined by the historical patterns of hospital use by the population within the area. As explained earlier, the definition of the hospital service area is of great importance because it delineates a boundary and within that boundary are found the hospital resources and the population assumed to use those resources. Although the delineation of the hospital service area boundary is important, it has been very casually derived by small area analysis researchers who are non-geographers and therefore not as conscious as they might be of its potential importance. Small area analysis methodology has been accepted by many within the health services 230 research and policy community with very little validation. Several questions about the validity of the currently used hospital service area boundary definition methodology (Wennberg's plurality method) have been raised by this current research. To my knowledge, only one previous paper has reported the results of a comparison of the impact that changes in the assignment of small areas (zip codes or minor civil divisions) to a hospital service area have had upon use rates. Tedeschi and Martin (1983) tested Wennberg's plurality model and Griffith's relevance indices (using both a plurality and a 12.5 percent market penetration measure) and found the overall use rates from the three definitions of a hospital service area to be highly intercorrelated, and that an approximate linear relationship existed between each pair of variables. This satisfied the non-geographers, but does not satisfy me. Preliminary work done by this author indicated that there may be no significant difference between use rates when the hospital service areas are defined using either Wennberg's plurality model or Griffith's majority model. There probably would not be any significant difference because they both assign all small areas to a hospital service area, leaving no small areas unassigned. On the other hand, there may be a very great difference between use rates when a narrowly defined hospital service area is compared to a larger, more inclusive hospital service area. Using a probabilistic gravity model (Taylor, 1977) as the basis for further work, I would hypothesize that use rates and the explanation for those use rates from a hospital service area where 75 - 90 percent of the population used the same hospital would be significantly different than use rates and the explanation for those use rates calculated in a hospital service area where only 20 - 30 percent of the 231 population used the same hospital. I suggest that only those small areas that have a great probability of all or a large percentage of the patients using the same hospital should be assigned to the hospital service area. Small areas that do not meet the high level of probability required would not be assigned to any hospital service area. In this way, the impact of the physician and hospital components of the model would not be "diluted" by patients who had a low probability of using those physicians or that hospital. This notion of a more narrowly defined hospital service area would be particularly important if hospital service areas were defined by historical procedure-specific or medical diagnosis-specific use rates rather than by historical total admission rates. Defining the hospital service area more narrowly and by diagnosis or procedure-specific use would be of particular importance in any attempt to determine the true impact of physician training or specialty. One of the family of gravity models could be used to define a hospital service area as I have suggested. The criteria could be established so that 75, 80, or any percent of the population had a high probability of using the hospital located within it. Once the new hospital service areas were defined, then testing the explanatory power of a dominant physician residency program or specialty or any physician or hospital characteristic would have more meaningful results. If the results were significant, then re-educating the health services research and policy community to understand the importance of the areal units would have to be undertaken. 232 The delineation of the boundary of a hospital service area is also of particular importance to this research in yet another way. Most of the high use hospital service areas found in this current research were rural in character and a few were very large in size. I think that this is very important because I think that hospital use rates increase in areas where there is a long distance between the hospital and the hospital service area boundary. This hypothesis is based on earlier work done by Girt (1973) who studied patient visits to physicians' offices in Newfoundland. Girt found that when graphed, distance decay curves were very different for some diseases or complaints. He found that distance had both a positive and negative effect on a patient's decision to initiate contact with a physician. Girt found that a patient increased his sensitivity to the development of a disease as the distance to medical help increased. At the same time, Girt discovered that the increase in distance reduced the likelihood of the patient actually consulting a physician. If you move this notion from a physician's office in a rural setting to a hospital in a rural setting, there may be pressures not only for the patient to increase his desire for hospitalization but also for a physician to increase his desire to order a hospital stay rather than to observe the patient in the patient's home or to treat the patient in a non-hospital setting. For example, let us look at the hypothetical case of a child with an acute respiratory infection or an elderly person with chest pain. In each instance there is a need for observation of the patient and there is also the need to have immediate availability of care in a medical crisis. Both of these pressures will encourage the physician to order a hospital admission if the distance 233 from the patient's home is considered to be too far to provide the immediacy of treatment necessary. Both of these pressures will also encourage the patient to seek hospitalization. In these two examples the diagnosis-specific distance decay curve could conceivably look like Figure 8.2, with an upturn in admissions at longer distances, rather than looking like Figure 8.3, the normal distance decay curve for total admissions. The potential importance of the definition of the hospital service area to both of these discussions re-emphasizes the need for a rigorous re-examination of the impact that the definition of the hospital service area can have on the hospital use rates and on the explanation of those use rates. 234 F i g u r e 8.2 Hypothetical Diagnosis-Specific Distance Decay Curve CD m & i— i o > distance F i g u r e 8.3 Hypothetical Total Admissions Distance Decay Curve CD m & r— I O > distance GLOSSARY 235 GLOSSARY appendectomy surgical removal of the appendix board eligible physicians physicians who are qualified to take the examinations for specialty certification board certified physicians physicians who have passed specific requirements for specialty certification cesarean section surgical delivery of baby through the abdominal wall cholecystectomy surgical removal of gall bladder ORG Diagnosis Related Group; Medicare's prospective payment coding system ENT physician who specializes in diseases of the ear, nose and throat epidemiology the study of the incidence, distribution, and control of a disease within a population events (hospital, outcome) HCFA Health Care Finance Authority; federal administrators of the Medicare system hemorrhoidectomy surgical removal of hemorrhoids home health health care delivered in a home hospital service area geographic area served by the hospital/s within it hysterectomy surgical procedure to remove uterus ICD-9-CM international codification of diseases and procedures IDS Interactive Data System; interactive computer system produced by the MHA and containing patient level clinical data as well as other data bases incidence rate occurrence of a disease per population inguinal hernia repair surgical procedure to repair protrusion of tissue through the abdominal wall inpatient patient who is admitted to a hospital, i.e., spends at least one night in a hospital bed 236 licensed beds hospital beds licensed by the state market share (hospital) percent of population within a hospital service area which receives care from the hospital/s within it MHA Michigan Hospital Association MHDC Michigan Health Data Corporation MIDB The Michigan Inpatient Data Base; an annual compilation of all of Michigan's hospital inpatient clinical abstracts migration patient movement from hospital service area of residence to hospital service area of the hospital where care was provided morbidity occurrence mortality death outpatient patient who receives hospital care in an outpatient setting, i.e., without being admitted to a hospital patient origin geographic area where patient resides procedure surgical procedure prostatectomy surgical procedure to remove or repair prostate gland relevance index Griffith's method of allocating population to a hospital service area referral center tertiary care facility providing more highly sophisticated technology and care than other hospitals rural referral center hospital that meets the criteria or a larger DRG reimbursement for each Medicare patient treated set up and staff beds hospital beds currently being used small area analysis method used to study the variation in health care use rates SMSA Standard Metropolitan Statistical Area sole community provider a single hospital within the hospital service area 237 surgical signature unique pattern of procedure-specific use rates seen when each hospital service area's surgical rates are graphed systematic component of variation (SCV) - a measure of variation T & A tonsillectomy and adenoidectomy use rate hospital use per population APPENDICES APPENDIX A Table 9.1 HOSPITAL USE RATES BY HOSPITAL SERVICE AREA (Standardized by Age and Sex per 10,000 Population) PORT HURON PONTIAC ANN ARBOR MT CLEMENS LANSING ADRIAN JACKSON BATTLE CREEK B HARBOR/ST JOE KALAMAZOO STURGIS/3 RIVERS S BERRIEN/CASS N MONTCALM FREMONT REED CITY GRAND RAPIDS MUSKEGON MONTCALM/1 ON IA ALLEGAN HOLLAND OCEANA CO flint LAPEER CO BAY SAGINAW TUSCOLA BAD AXE SANILAC MIDLAND MT. PLEASANT GRATIOT N.MICHIGAN CHEBOYGAN/ROGERS TRAVERSE CITY MANISTEE Male Adm Female Adm Total Adm 144.48 130.10 113.43 146.41 116.39 147.48 130.30 130.67 135.84 105.44 118.64 149.96 143.24 119.85 133.65 94.49 126.42 189.31 124.72 115.84 129.92 140.64 150.08 160.00 151.77 169.59 171.42 166.84 120.95 148.81 150.90 128.10 153.18 127.51 123.42 181.89 180.31 141.57 201.33 158.26 184.58 180.93 185.82 170.32 136.64 175.30 205.65 221.80 166.81 165.99 133.78 166.24 255.92 180.02 165.50 209.78 181.87 192.89 191.49 188.31 230.43 224.06 218.33 166.64 174.20 201.77 172.42 212.83 169.10 190.14 163.65 155.84 127.85 174.56 137.85 166.5C 156.25 158.94 153.51 121.43 147.68 178.50 183.51 143.92 150.23 114.63 146.83 223.45 153.06 141.29 170.85 161.77 172.02 176.14 170.50 200.78 198.40 193.23 144.37 161.82 176.97 150.82 183.76 148.83 157.62 Appen 1.12 1.31 1.11 1.15 0.98 1.33 1.18 1.12 1.07 1.26 1.15 0.98 1.50 1.63 1.43 1.02 1.23 2.24 0.93 1.73 0.75 1.15 1.14 1.11 0.99 1.58 2.22 1.31 0.87 1.17 1.53 1.38 1.55 1.24 1.34 Hemrr Chole Hern Prost 0.67 0.39 0.31 0.67 0.67 0.38 0.38 0.72 0.23 0.61 0.29 0.29 0.16 0.08 0.17 0.14 0.45 0.57 0.33 0.13 0.07 0.67 0.84 0.38 0.51 0.60 0.37 0.87 0.22 0.49 0.58 0.20 0.35 0.32 0.66 2.06 2.13 1.53 2.00 1.86 2.11 2.57 2.03 2.02 1.83 1.83 1.72 1.70 2.31 1.56 1.94 2.31 2.45 2.97 2.38 2.07 2.15 2.33 2.05 2.63 2.76 2.49 2.32 1.63 1.72 2.34 1.92 2.28 1.86 2.33 2.17 1.89 1.33 2.10 1.81 1.79 1.77 1.85 2.02 1.86 1.62 1.78 3.03 1.42 1.91 1.85 2.16 2.86 2.19 2.71 1.26 1.86 1.83 1.48 2.08 2.32 2.77 2.63 1.86 1.97 1.90 1.70 2.08 1.70 3.11 2.43 2.94 4.24 3.53 2.50 3.55 1.94 2.64 2.01 2.58 1.78 1.96 2.69 2.68 1.58 2.70 2.47 3.11 3.00 3.18 2.30 3.10 2.88 3.55 3.15 3.06 3.81 2.68 3.21 3.54 2.13 3.03 3.71 2.00 2.80 Hyster 5.76 5.11 3.91 5.52 5.05 3.93 5.07 4.36 3.91 4.95 4.74 4.12 6.65 4.49 4.19 4.10 5.02 6.31 3.18 4.45 3.87 5.06 5.35 4.75 4.12 4.27 4.82 4.15 4.90 3.31 5.41 3.73 4.50 4.92 4.65 C-Sect 5.93 5.94 4.94 7.24 5.47 3.08 4.70 5.56 6.08 6.34 6.26 4.62 7.59 5.68 3.77 5.65 4.87 7.27 6.30 5.98 1.52 5.50 6.37 5.60 5.33 4.41 6.94 6.29 5.96 4.31 6.65 6.02 6.18 5.97 7.63 Ci re Adm 21.92 22.77 17.08 26.04 15.49 17.03 17.83 15.77 20.00 13.83 17.27 16.69 16.79 13.45 19.05 10.10 17.44 23.46 17.92 13.26 18.99 20.82 21.91 24.24 22.20 24.73 24.57 24.21 18.75 20.81 20.18 18.54 24.35 18.87 19.11 Resp Adm 9.54 10.21 8.82 11.09 8.49 12.55 10.48 10.44 12.55 6.74 7.12 10.23 17.68 9.11 14.95 5.03 8.08 20.28 9.08 8.14 11.61 9.93 10.62 10.55 8.94 17.18 17.05 14.36 9.05 12.86 1 1 .86 9.59 12.33 8.55 9.51 Digest Adm 17.90 17.99 13.45 18.71 14.81 19.51 17.52 18.74 17.98 13.43 16.71 17.76 23.09 15.55 19.22 12.66 17.18 30.97 16.75 17.88 24.06 18.28 20.72 19.44 21 .44 28.55 30.14 23.27 15.24 21 .01 21.53 17.10 21.12 16.09 21 .64 Gen ito-U Adm 4.57 4.22 4.18 4.74 3.73 4.37 4.46 4.65 3.90 3.14 3.82 4.34 5.18 3.91 4.58 2.97 3.60 7.14 3.92 4.40 6.10 4.57 5.19 6.85 6.09 6.09 6.84 6.46 3.82 5.48 5.24 4.31 4.36 3.99 4.96 N> U) 00 APPENDIX A (Continued) Table 9.1 HOSPITAL USE RATES BY HOSPITAL SERVICE AREA (Standardized by Age and Sex per 10,000 Population) SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP Male Adm Female Adm Total Adm 126.81 148.95 127.12 151.05 111.47 176.43 130.37 136.03 178.87 120.59 149.42 130.89 188.91 114.34 135.72 137.53 163.07 157.32 157.03 196.56 189.40 195.53 146.95 200.41 196.58 179.32 243.16 166.73 205.94 172.98 228.25 145.48 187.69 177.39 216.35 213.46 142.30 173.30 159.04 173.85 129.60 188.72 164.31 158.22 211.82 144.24 178.35' 152.40 209.00 130.30 162.30 157.90 190.30 186.10 Appen 0.91 0.92 1.54 1.88 1.24 0.90 1.05 1.60 1.75 1.63 1.12 2.10 1.29 0.95 1.18 1.58 1.56 1.71 SCP = Sole Community Provider Hospital Service Area Hemrr Chole Hern Prost 0.23 0.52 0.46 0.74 0.32 0.56 0.25 0.54 0.55 0.34 0.21 0.55 0.13 0.19 1.23 0.32 0.60 0.39 2.02 1.94 2.65 2.03 1.81 2.44 1.56 2.20 3.01 1.84 2.24 2.26 2.69 1.61 3.58 1.81 2.45 2.20 1.56 1.67 2.72 2.46 1.28 1.74 0.98 2.13 2.58 2.07 1.64 1.98 2.90 2.16 2.55 2.30 1.72 2.15 2.92 1.48 3.62 1.78 2.07 3.69 2.73 1.82 2.24 3.70 2.71 3.12 3.76 3.03 3.57 2.90 2.11 2.93 Hyster 4.74 3.98 5.50 4.74 5.24 4.10 4.10 4.13 4.79 3.60 4.79 3.79 6.86 4.62 6.41 4.92 5.80 5.53 C-Sect 5.29 5.58 8.99 5.11 6.21 6.31 7.20 5.50 6.83 6.18 6.25 6.27 4.82 4.97 6.78 5.29 5.47 4.54 Ci rc Adm 19.60 18.66 16.51 14.75 17.92 26.53 21.58 16.05 21.26 17.77 18.96 18.41 29.95 16.02 15.85 19.84 24.17 19.10 Resp Adm 8.45 11.96 9.90 14.69 6.28 14.92 10.22 10.91 19.38 6.90 14.77 9.94 14.44 5.96 9.86 7.64 14.20 14.30 Digest Adm 14.12 17.57 20.83 18.61 13.94 21.07 15.47 16.72 26.59 17.34 22.31 16.32 23.71 13.28 21.25 17.68 21.86 21.39 4.01 4.75 4.71 4.90 3.64 4.65 4.75 3.87 4.88 5.72 6.48 4.73 5.88 3.56 4.24 5.58 3.94 5.73 APPENDIX B Table 9.2 PHYSICIAN AND HOSPITAL CHARACTERISTICS BY HOSPITAL SERVICE AREA Wgt Prop Bd Cert Phys PORT HURON PONTIAC ANN ARBOR MT CLEMENS LANSING ADRIAN JACKSON BATTLE CREEK BHARBOR/ST JOE KALAMAZOO STURGIS/3 RIVERS S BERRIEN/CASS N MONTCALM FREMONT REED CITY GRAND RAPIDS MUSKEGON MONTCALM/IONIA ALLEGAN HOLLAND OCEANA CO FLINT LAPEER CO BAY SAGINAW TUSCOLA BAD AXE SANILAC MIDLAND MT. PLEASANT GRATIOT N.MICHIGAN CHEBOYGAN TRAVERSE CITY MANISTEE 0.64 0.76 0.80 0.55 0.68 0.43 0.50 0.71 0.65 0.56 0.68 0.59 0.32 0.75 0.49 0.71 0.66 0.33 0.46 0.70 0.66 0.66 0.38 0.61 0.74 0.31 0.73 0.58 0.54 0.50 0.46 0.76 0.68 0.74 0.44 Hosp Beds/ FTEs/ Pharm/ 10,000 10,000 10,000 36.44 33.78 52.15 32.61 33.76 42.90 36.36 45.25 38.83 37.58 31.32 38.94 42.10 49.14 32.00 30.58 48.49 65.84 69.93 36.14 49.39 41.54 32.88 41.48 49.96 46.97 43.36 45.57 33.74 30.49 35.54 61.36 69.92 45.12 57.05 104.97 120.10 244.00 118.92 124.28 109.75 108.56 146.37 129.76 176.68 81.50 102.46 93.35 96.12 92.72 110.59 130.58 166.70 223.99 93.15 109.91 142.75 78.50 132.65 146.80 104.14 131.53 123.99 125.32 86.21 105.10 204.87 142.66 128.27 121.36 0.76 1.18 2.03 0.93 1.15 0.56 0.78 1.39 1.17 1.28 0.75 0.67 0.73 0.43 0.96 1.11 1.30 0.84 2.27 0.80 0.00 1.29 0.52 1.57 1.47 0.41 1.46 0.41 0.78 0.83 1.13 1.45 0.00 1.19 1.45 RNs/ Bed 0.68 0.83 1.04 0.62 0.89 0.40 0.39 0.70 0.69 1.12 0.46 0.52 0.35 0.47 0.63 0.87 0.49 0.40 0.58 0.67 0.37 0.77 0.56 0.58 0.65 0.40 0.47 0.48 0.86 0.66 0.56 0.85 0.32 0.52 0.39 Hosp Serv '83 26 41 38 33 26 21 27 26 27 34 19 24 19 16 22 34 33 17 30 21 15 33 20 26 30 19 24 20 28 25 22 22 26 26 14 Other Fac Serv '83 26 14 10 25 13 28 14 22 22 18 39 21 17 38 32 18 13 22 17 19 18 14 8 8 12 22 8 36 20 41 25 20 21 18 50 Serv Leve1/ HSA '83 172 130 132 157 159 184 158 168 166 148 199 170 177 205 186 148 145 185 156 176 187 146 167 154 149 182 158 194 163 190 180 175 168 164 220 Total / Serv per HSA •83 45 57 61 62 52 39 39 43 48 59 22 35 24 22 29 56 48 21 37 30 20 60 24 41 45 21 34 36 44 32 31 48 31 52 23 Change in Serv •81-83 , -1 2 -4 -1 1 . 5 1 -4 1 -2 3 2 1 # 0 1 -3 6 0 -3 -2 -1 12 5 6 2 8 5 5 . OPD Visits per 10,000 9,210 6,679 16,393 8,036 12,222 11,640 9,169 8,947 10,781 17,503 11,188 8,399 15,005 23,042 11,402 6,790 12,896 11,177 27,045 7,062 30,459 7,381 5,845 4,493 7,057 11,481 11,437 15,325 12,431 4,993 8,113 10,137 9,954 9,217 10,416 Corp Beds 0.53 0.56 0.38 0.55 0.36 0.00 0.00 0.37 0.74 0.49 0.00 0.00 1.00 0.00 0.00 0.30 0.00 0.00 0.00 0.00 0.43 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.78 0.00 0.00 0.00 0.26 House Staff per 10,000 0.53 3.57 21.59 0.66 4.11 1.45 0.52 1.25 0.70 6.23 0.00 0,21 0.00 0.34 0.16 0.00 0.13 0.17 0.00 0.00 0.00 0.00 0.27 0.00 0.73 0.49 0.43 0.00 0.39 0.00 7.76 0.15 1.12 0.18 2.95 1.00 1.71 0.00 0.00 0.00 1.39 0.00 7.37 0.19 0.78 0.00 0.21 0.17 2.78 0.18 0.00 0.00 0.88 0.00 2.05 0.00 2.02 0.00 0.14 0.00 0.38 0.00 2.42 0.00 0.40 0.00 2.02 0.16 0.00 0.00 Appendix B (continued) Table 9.2 PHYSICIAN AND HOSPITAL CHARACTERISTICS BY HOSPITAL SERVICE AREA Wgt Prop Bd Cert Phys 0.64 0.74 0.63 0.42 0.40 0.52 0.49 0.48 0.29 0.89 0.69 0.68 0.26 0.51 0.38 0.66 0.33 0.74 FTEs/ Pharm/ 10,000 10,000 10,000 34.44 30.94 32.46 27.57 36.21 51.74 35.24 32.86 55.61 35.72 39.75 23.25 37.05 30.80 26.39 47.49 38.52 44.81 94.91 86.48 129.13 84.32 105.16 112.78 80.80 81.91 149.72 112.79 89.31 94.45 116.80 93.23 84.14 102.03 150.22 128.55 0.82 1.04 1.41 0.64 1.06 0.58 0.86 0.95 1.07 1.50 1.03 1.43 2.01 0.82 0.50 0.60 1.28 1.02 SCP - Sole Community Provider Hospital Service Area RNs/ Bed 0.50 0.63 0.72 0.49 0.47 0.39 0.49 0.47 0.35 0.59 0.57 0.80 0.40 0.72 0.51 0.41 0.60 0.45 Hosp Serv '83 30 36 26 23 29 18 18 25 26 21 16 20 19 31 20 35 27 27 Other Fac Serv '83 0 0 0 35 22 42 0 1 38 42 0 0 0 1 44 31 36 0 Serv Leve1/ HSA '83 Total # Serv per HSA '83 138 126 146 187 162 204 162 149 184 198 166 158 160 137 30 36 26 23 29 18 18 25 26 202 20 159 180 144 35 27 27 21 16 20 19 31 Change in Serv '81-83 5 -1 3 5 -1 0 -2 5 -4 -6 0 2 4 -3 9 2 1 OPD Visits per 10,000 Corp Beds 5,464 14,985 13,914 4,113 7,713 19,151 14,251 6,576 8,492 12,502 15,080 12,984 9,645 14,804 9,634 13,470 9,025 4,361 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 House Staff per 10,000 Osteo Beds 1.23 0.00 0.00 0.00 0.71 0.00 0.32 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.19 0.00 2.14 0.00 0.75 0.00 0.00 0.00 0.36 0.00 0.40 0.00 1.09 0.00 0.00 0.00 1.61 0.00 2.14 0.00 0.00 0.00 241 SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP SCP Hosp Beds/ Appendix C Table 9.3 COMPARISON OF HOSPITAL USE RATES BY HOSPITAL SERVICE AREA BY STANDARD DEVIATIONS Ma le Adm Appen Hemorr Chole Hernia Prost 1 South Central Region Lans ing Jackson Adrian SCP SCP -1 -1 -1 -1 -1 -1 -1 -1 -1 1 Southwest Region Battle CreeR Kalamazoo Sturgis/3 Rivers B Harbor/St. Joe S Berrien/Cass SCP SCP SCP West Central Region Reed City Fremont Oceana Total Adm -1 -1 1 -1 1 1 1 -1 -1 1 -1 1 SCP = Sole Community Provider 2 1 -1 -1 -2 1 -1 1 C-Sect 1 -2 1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -2 1 -1 -1 Ci rc Adm Resp Adm Digest Adm Gen itoAdm -1 1 -1 -1 -1 1 1 -1 1 -1 -1 1 - -1 Hyster -1 -1 -1 -1 1 1 242 Southeast Region Port Huron Pont iac Ann Arbor Mt. Clemens Lapeer F 1int SCP SCP SCP Female Adm -1 1 1 -1 1 1 -1 -1 -1 -1 -t 1 1 2 -1 -1 -1 -3 -1 1 Appendix C (Continued) Table 9.3 COMPARISON OF HOSPITAL USE RATES BY HOSPITAL SERVICE AREA BY STANDARD DEVIATIONS Muskegon Grand Rapids N. Montcalm Montcalm/Ionia Hoi 1and Allegan SCP SCP SCP SCP North Region Man istee Traverse City N. Michigan Cheboygan/Rgrs City SCP SCP SCP SCP SCP SCP Fema1e Adm Total Adm -2 2 -1 -2 1 2 -2 2 -1 1 -1 2 -1 2 Appen 2 1 -1 -1 1 Hemorr -1 -1 -1 -1 1 1 1 1 1 1 -1 2 -1 2 -1 -1 -1 1 1 1 Chole J 2 1 1 Hern ia 2 1 1 1 -1 1 1 2 -1 2 1 SCP = Sole Community Provider 1 -1 1 1 1 1 3 -1 1 2 -1 3 1 Prost 1 1 Hyster 2 1 -1 1 -1 1 -1 1 1 -1 -1 1 1 1 -1 C-Sect 1 1 Circ Adm -2 1 -1 Resp Adm Digest Adm Gen itoAdm -1 1 2 -1 2 -1 -1 2 1 -1 2 -1 -1 1 1 -1 243 East Central Region Mt. Pleasant Grat iot Midi and Bay Saginaw Bad Axe Tuscola San i1ac Ma le Adm -1 1 -1 1 1 1 1 1 1 2 2 1 2 1 2 1 1 1 -1 2 2 1 1 -1 1 2 -1 1 1 -1 1 1 LIST OF REFERENCES 244 LIST OF REFERENCES Anderson, James G. (1973) "Demographic Factors Affecting Health Services Utilization: A Casual Model," Medical Care, Vol. XI, No. 2, (April), pp. 104-120. Anderson, Geoffrey M. and Jonathan Lomas (1985) "Explaining variations in Cesarean Section Rates: Patients, Facilities or Policies", Canadian Medical Association Journal, Vol. 133, February 1, pp. 253-259. Barnes, B.A. (1982) "Geographic and Temporal Patterns of Care in the United States," Regional Variation in Hospital Use, D.L. Rothberg (ed.), Lexington MA: Lexington Books, Chapt. 7. , B.A., O'Brien, E., Comstock, C., De'Arpa, D.G., Donahue, C.L. (1985) "Report on Variation in Rates of Utilization of Surgical Services in the Commonwealth of Massachusetts," JAMA Vol. 354, No. 3, pp. 371-375. Brewer, W. Ross and Mary Anne Freedman (1982) "Causes and Implications of Variation in Hospital Utilization," Journal of Public Health Policy, No. 3, pp. 445-454. Cageorge, Sandra M., Roos, Leslie L. Jr., and Danzinger, Rudy (1981) "Gallbladder Operations: A Population-based Analysis," Medical Care, Vol. 19, No. 5, pp. 510-525. Cherniack, H.D. and J.B. Schneider (1967) A New Approach to the Delineation of Hospital Service Regions, Discussion Paper No. 16, Regional Science Institute, University of Pennslyvania, Philadelphia. Chiswick, Barry R. (1976) "Hospital Utilization: An Analysis of SMHS Differences in Occupancy Rates, Admission, Rates, and Bed Rates," Exploration in Economic Research, Vol. 3 (Summer). Ciocco, Antonio and Isidore Altman (1954) "Medical Service Areas and Distances Traveled for Physician Care in Western Pennslyvania, Part I: Medical Services Areas as Indicated by Intercounty Movement of Patients, Public Health Monograph, pp. 3-20. 245 Cohen, M.A. and H.L. Lee (1985) "The Determinants of Spatial Distribution of Hospital Utilization in a Region," Medical Care, Vol. 23, No. 1, pp.27-38. Connell, F. A., Day, R. W., LoGerfo, J P. (1981) "Hospitalization of Medicaid Children: Analysis of Small Area Variations in Admission Rates," American Journal of Public Health, Vol. 71, No. 6 , (June), pp. 606-613. , F. A., Blide, L.A., Hanken (1984) "Clinical Correlates of Small Area Variations in Population-Based Admission Rates for Diabetes," Medical Care, Vol. 22, No. 10 (October), pp. 939949. deVise', P. (1966) "Hospital Study Districts for Metropolitan Chicago," Technical Report No. 2. Chicago: Hospital Planning Council for Metropolitan Chicago. Deacon, R., Lubitz, J., Cornish, M. and Newton, M. (1979) "Analysis of Variations in Hospital Use by Medicare Patients in PSRO Areas, Health Care Financing Review, 1:79-109. Detmer, Don E. and Timothy J. Tyson (1976) "Delivery of Surgical Care: Inferences Based on Hospital Discharge Abstract Data," Surgical Forum. _ _ _ _ _ , Don E. and Timothy J. Tyson (1978) "Regional Differences in Surgical Care Based Upon Uniform Physician and Hospital Discharge Abstract Data," Annals of Surgery. Vol. 187, No. 1, pp. 166-169. Dickinson, F. (1949) "A Medical Service Area Map of the United States, A Progress Report," Journal of the American Medical Association, Vol. 139, pp. 1021-1028. _ _ _ _ _ _ and C. Bradley (1951) "Medical Service Aeas, Population, Square Miles, and Primary Centers," American Medical Association, Bureau of Medical Economic Research, Bulletin No. 80, Chicago. _ _ _ _ _ _ (1954) "Ranks of Medical Service Areas, 1-759," American Medical Association, Bureau of Medical Economic Research, Miscellaneous Publication M-85, Chicago. _ _ _ _ _ _ (1954) "How Bad is the Distribution of Physicians?" American Medical Association, Bureau of Medical Economic Research, Bulletin No. 94B, Chicago. Drosness, D., Reed, I.M., and Lubin, J.W. (1965) "The Application of Computer Graphics to Patient Origin Study Techniques," Public Health Reports, Vol. 80, No. 1, pp. 33-40. and J.W. Lubin (1966) "Planning Can Be Based on Patient Travel," The Modern Hospital, Vol. 106, pp. 92-94. 246 Dyck,F.G., Murphy, F.A., et.al. (1977), "Effects of Surveillance on the Number of Hysterectomies in the Province of Saskatchewan," New England Journal of Medicine, (June), pp. 1326-1328. Erickson, G.M. and S.A. Finkler (1985) "Determinants of Market Share for a Hospital's Services," Medical Care, Vol. 23, No. 8, pp. 1003-1018. Folland, S.T. (1983) "Predicting Hospital Market Shares," Inquiry. Vol. 20, pp. 34-44. Girt, John L. (1973) "Distance to General Medical Practice and Its Effect on Revealed Ill-Health in a Rural Environment," Canadian Geographer, Vol. XVII, No. 2, pp. 154-166. Gittelsohn, A.M. and J.E. Wennberg (1976) "On the Incidence of Tonsillectomy and Other Common Surgical Procedures," Cost Risks and Benefits of Surgery, Ed. John P. Bunker, Benjamin A. Barnes, Frederick Mosteller, (N.Y.: Oxford University Press), pp. 91-106. Griffith, J.R. (1978) Measuring Hospital Performance, Chicago: Books. Inquiry _ _ _ _ _ , J.R., Wilson, P.A., Wolfe, R.A. and Bischak, D.P., (1985) "Clinical Profiles of Hospital Discharge Rates in Local Communities," Health Services Research Vol. 20, No. 2. pp. 131-151. _ _ _ _ _ , J.R., Restuccia, J.D., Tedeschi, P.J., Wilson, P.A. and Zuckerman, H.S. (1981), "Measuring Community Hospital Services in Michigan," Health Services Research, pp. 135-173. Hamilton, R.A., Clark, J.D., and Lester, J.L. (1988) "The Status of Michigan's Smaller Hospitals," Michigan Hospitals, Vol. 24, No. 4, pp. 13-20. Hammond, John (1985) "Analysis of County-Level Data Concerning the Use of Medicare Home Health Benefits," Public Health Reports, Vol. 100, No. 1, pp. 48-55. Hunter, J.W., Shannon, G.W., and Sambrook, S.L. (1985) "Rings of Madness: Service Areas of 19th Century Asylums in North America," Paper presented in the Institute of British Georgraphers and the Association of American Geographers, Joint Symposium in Medical Geography, Nottingham, England, July 15-19, 1985. Jarvis, (1850) Medical and Surqical Journal of Boston, Vol. 42, pp. 209222. , (1852) "On the Supposed Increase of Insanity," American Journal of Insanity, Vol. 8, pp. 333-364. 247 Joffee, Jerome (1979) "Mobility Adjustments for Small Area Utilization Studies," Inquiry, Vol. 16, pp. 350-355. Knickman, J.R. and A.M. Foltz (1982) "Hospital Utilization: Reasons for Regional Differences Between East and West: Final Report," Washington, DC: Health Care Financing Administration. _ _ _ _ _ , J.R., (1982) "Variations in Hospital Use Across Cities: A Comparison of Utilization Rates in New York and Los Angeles," Regional Variations in Hospital Use, D.L. Rothberg (ed.), Lexington Books, pp. 23-64. _ _ _ _ _ , J.R. and A.M. Foltz (1984) "Regional Differences in Hospital Utilization," Medical Care, Vol. 22, No. 11, pp. 971-986. _ _ _ _ _ ,J.R. and A.M. Foltz (1985) "A Statistical Analysis of Reasons for East-West Differences in Hospital Use," Inquiry, Vol. 22, pp. 45-58. Lembcke, P.A. (1952), "Measuring the Quality of Medical Care Through Vital Statistics Based on Hospital Services Areas: Comparative Study of Appendectomy Rates," American Journal of Public Health, Vol. 42, (March), pp. 276-286. Lewis, Charles E. (1969) "Variations in the Incidence of Surgery" New England Journal of Medicine. Vol. 281, No. 16, (October), pp. 880-884. Lubin, J.W., Drosness, D.L., and Wylie, L.G. (1965) "Highway Network Minimum Path Selection Applied to Health Facility Planning," Public Health Reports, Vol. 80, No. 9, pp. 771-778. Mausner, J.S. and S. Kramer (1985) Epidemiology — An Introductory Text. Philadelphia: W.B. Saunders Co. McCracken, Gene (1980) "Patterns of Hospital Use," unpublished paper. McPherson, Klim, Strong, P.M., Epstein, A., Jones, L. (1981) "Regional Variations in the Use of Common Surgical Procedures: Within and Between England and Wales, Canada and the United States of America, Social Science and Medicine, Vol. 15A, pp. 273-288. _ _ _ _ _ _ , K., Wennberg, J.E., Hovind, O.B., Clifford, P., (1982) "Small-Area Variations in the Use of Common Surgical Procedures: An International Comparison of New England, England and Norway, " The New England Journal of Medicine, Vol. 307, No. 21, (November 18), pp. 1310-1314. Mindell, W.R., Vayda, E., Cardillo, B. (1982) "Ten-Year Trends in Canada for Selected Operations," Canadian Medical Association Journal, Vol. 127, pp. 23-27. Mitchell, J.B. and Cromwell, J. (1982) "Variations in Surgical Rates and the Supply of Surgeons," Regional Variations in Hospital Use, 248 ed. D.L. Rothberg. 129. Lexington MA: Lexington Books, pp. 103- Morrill, Richard L. and Robert Earickson (1968) "Variations in the Character and Use of Chicago Area Hospitals," Health Services Research, No. 3, pp. 224-238. _ _ _ _ _ , Richard L. and Robert Earickson (1968) "Hospital Variation and Patient Travel Distances," Inquiry, Vol. 5, No. 4, pp. 26-34. , R.L. Earickson, R.J., Rees, P. (1970) "Factors Influencing Distances Traveled to Hospitals," Economic Geoqraphy, Vol. 46, No. 2, pp. 161-171. Mountin, J.W., Pennell, E.H., and Hoge, V.M. (1945) Health Service areas: Requirements for General Hospital and Health Centers, United States Public Health Service, Public Health Bulletin No. 292, Washington, D.C.: United States Government Printing Office. OHMA (1985) "Regional Variation in Hospital Use in Michigan: A Symptom of Unnecessary Hospital Utilization," Office of Health and Medical Affairs, Department of Management and Budget, State of Michigan, No. 5 (January). OHMA (1985) "Michigan Health Care Costs Review: Variation in Hospital Expenditure Levels by County," Office of Health & Medical Affairs, Department of Management & Budget, State of Michigan, No. 5 (August). Paul-Shaheen, P., Clark, J.D., Williams, D. (1987) "Small Area Analysis: A Review and Analysis of the North American Literature," Journal of Health Politics, Policy and Law, Vol. 12, No. 4, pp. 741-809. rigozzi, B.W. (1969) "Optimal utilization of Ohio’s Miami valley Hospitals: Medical Service Areas Based on Minimal Distance Criteria," M.A. Thesis, Miami University, Oxford, Ohio. Poland, E. and P.A. Lembcke (1962) Delineation of Hospital Service Districts: A Fundamental Requirement of Hospital Planning. Kansas City, Mo: Community Studies, Inc. Pyle, Gerald F. (1979) "Applied Medical Geography," Scripta Series in Geography, New York, John Wiley & Sons, 282 p., ISBN 0-47026643-0. Roos, N.P., Roos, L.L. and Henteleff, P.D. (1977) "Elective Surgical Rates — Do High Rates Mean Lower Standards? Tonsillectomy and Adinoidectomy in Manitoba," The New England Journal of Medicine, (August 18) Vol. 297, No. 7, pp. 360-365. , N.P. and L.L. Roos (1982) "Surgical Rate Variations: Do They Reflect the Health or Socioeconomic Characteristics of the 249 Population?" Medical Care, Vol. XX, No. 9, (September) pp. 945-958. , Roos, Noralou (1984) "Hysterectomy: Variations in Rates Across Small Areas and Across Physician Practices," American Journal of Public Health, Vol. 74, No. 4, (April), pp. 327-335. , Leslie L. (1984) "Surgical Rates and Mortality, A Correlational Analysis," Medical Care, Vol. 22, No. 6, ppg. 586-588. , Noralou P., Shapiro, Evelyn, and Roos, Leslie L. (1984) "Aging and the Demand for Health Services: Which Aged and Whose Demand?" The Gerontologist, Vol. 24, No. 1, pp. 31-36. Shain, M. and M.I. Roemer (1959) "Hospital Costs Relate to the Supply of Beds," The Modern Hospital, Vol. 92, No. 4. Shannon, G.W., Bashur, R.L., and Metzner, C.A. (1969) "The Concept of Distance as a Factor in Accessibility and Utilization of Health Care," Medical Care Review, Vol. 26, No. 2, pp.149-150. Shaughnessy, Peter W. (1982) "Methodological Issues in Per Capita Measurement in Health Care," Health Services Research, Vol. 17, No. 1, pp. 61-81. Sigmond, Robert M. (1981) "Commentaries on Measuring Community Hospital Services in Michigan," Health Services Research, Vol. 16, No. 2 (summer), pp. 161-173. Stockwell, H., Vayda, E. (1979) "Variations in Surgery in Ontario," Medical Care, Vol. XVII, No. 4, (April), pp. 390-396. Taylor, Peter J. (1977) Quantitative Methods in Geography. Boston: Houghton Mifflin Company. Tedeschi, r.u. ana J.5. Martin (1983) "Inpatient Use Rate — A Comparison of Geographic and Product Moment Approaches," Michigan Community Hospital Service Measures Project, Analytic Paper #12, Ann Arbor: Program and Bureau of Hospital Administration. Vayda, E. and G.D. Anderson (1975) "Comparison of Provincial Surgical Rates in 1968," The Canadian Journal of Surgery, Vol. 18, pp. 18-26. , E., Morison, M., and Anderson, G.D. (1976) "Surgical Rates in the Canadian Provinces 1968 to 1972," The Canadian Journal of Surgery, Vol. 19, pp. 235-242. , E., Barnsley, J.M., Mindell, W.R., Cardillo, B. (1984) "Five-Year Study of Surgical Rates in Ontario's Counties," Canadian Medical Association, Vol. 131, pp. 111-115. 250 Vladeck, Bruce C. (1985) "The Poor Use More: Hospitals and Communities in New York City," United Hospital Fund of New York, President's Letter. Wennberg, John, and Alan Gittelsohn (1973) "Small Area Variations in Health Care Delivery," Science Vol. 182, (December), d d . 11021108. _ _ _ _ _ , and Gittelsohn, Alan (1975) "Health Care Delivery in Maine I: Patterns of Use of Common Surgical Procedures," The Journal of the Maine Medical Association, Vol. 66, No. 5, (May), d d . 123149. _ _ _ _ _ , Gittelsohn, Alan and Soule, David, (1975) "Health Care Delivery in Maine II: Conditions Explaining Hospital Admission," The Journal of the Maine Medical Association. Vol. 66, No. 10, (October), pp. 255-269. _ _ _ _ _ , Gittelsohn, Alan and Shapiro, Nancy (1975) "Health Care Delivery in Maine III: Evaluating the Level of Hospital Performance," The Journal of the Maine Medical Association. Vol 66, (November), pp. 298-306. _ _ _ _ _ , (1977) "Physician Uncertainty, Specialty Ideology and a Second Opinion Prior to Tonsillectomy," Pediatrics, Vol. 59, No. 6 (June) pp. 952. _ _ _ _ _ , Blowers, L., Parker, R., Gittelsohn, A. M., (1977) "Changes in Tonsillectomy Rates Associated With Feedback and Review," Pediatrics, Vol. 59, No. 6, (June), pp. 821-826. _ _ _ _ _ , (1979) "Factors Governing Utilization of Hospital Services," Hospital Practice, (September), pp. 115-127. _ _ _ _ _ , Barnes, B. A., Zubkoff, M., (1982) "Professional Uncertainty and the Problem of Supplier-Induced Demand," Social Science and Medicine, Vol. 16, pp. 811-824. _ _ _ _ _ , Gittelsohn, Alan, (1982) "Variations in Medical Care Among Small Areas," Scientific American, Vol. 246, (April), pp. 120134. _ _ _ _ _ , (1982) "Should the Cost of Insurance Reflect the Cost of Use in Local Hospital Markets?," The New England Journal of Medicine, Vol. 307, No. 22, (November), pp. 1374-1381. _ _ _ _ _ , (1984) "Dealing with Medical Practice Variations: A Proposal for Action," Health Affairs, (summer), pp. 6-32. _ _ _ _ _ , McPherson, Kilm, Caper, Philip (1984) "Will Payment Based on Diagnosis-Related Groups Control Hospital Costs?," The New Enqland Journal of Medicine, Vol. 311, No. 5 (Auqust), d p . 295-300. 251 , (1985) "On Patient Need, Equity, Supplier-Induced Demand, and the Need to Assess the Outcome of Common Medical Practices," Medical Care, Vol. 23, No. 5, (May), pp. 517-520. Wilson, P.A., Griffith, J.R., and Tedeschi, P.J. (1985) "Does Race Affect Hospital Use?" American Journal of Public Health, Vol. 75, No. 3 (March) pp. 263-269. , Peter and Philip Tedeschi (1984) "Community Correlates of Hospital Use," Health Services Research, Vol. 19, No. 3, (August) pp. 333-355.