1 5...": r: . . .211... I. 4 hficxsuhuf 5.. ..a.....v.vl>r 9“ a :4.) I... p‘ , :5... r. f ; d. 29...”, .33..”5... my. a». . . ._ . . .f 5. . 2.....- 3%"? up .1 25...: . .. $13. . "f . 1...: , ‘ . . . V .2... .J :3. any...» he}. . , II. A - .. 313er ‘1!- .2 ,e... .532". .2 . fixwzafiafifldmfimfi i. .. .3. In ‘32) , .; : 3 is... wkumwi. 1.. lu....-li:v..> 3.: i153!" .ufl»..f..t.v\l. 3.... 9.... ... 3:15.. I .) ..:.x)\l..n.. pa 'Ivblbi’lt‘ 11.1.9952 .Di‘liflw 3.11.1333. 1.: i “ kit! ‘1 . .L 206“/ This is to certify that the thesis entitled Occurrence and Recurrence of Physical Abuse in Young Children: A Secondary Analysis of NCANDS presented by Vincent John Palusci, M.D. has been accepted towards fulfillment of the requirements for M.S. degree in EpidemiOIOgy Date 37/917 07" 0-7639 W Major professor MS U is an Affirmatiw Action/Equal Opportunity Institution LIBRARY Michigan State University PLACE IN RETURN Box to remove this checkout from your record. To AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DATE DUE DATE DUE DATE DUE own 9 (Mm 6/01 c:lCiRC/DateDue.p65-p. 15 OCCURRENCE AND RECURRENCE OF PHYSICAL ABUSE IN YOUNG CHILDREN; A SECONDARY ANALYSIS OF NCANDS By Vincent John Palusci, MD. A THESIS Submitted to Michigan State University in partial fitlfillment of the requirements for the degree of MASTER OF SCIENCE Department of Epidemiology 2002 ABSTRACT OCCURRENCE AND RECURRENCE OF PHYSICAL ABUSE IN YOUNG CHILDREN: A SECONDARY ANALYSIS OF NCANDS By Vincent John Palusci, M.D. Child maltreatment report information for 1995-1999 from the Detailed Case Data Component of the National Child Abuse and Neglect Data System (NCANDS-DCDC) was used to study: (1) the first occurrence of physical abuse as compared to other maltreatment in children under three years, and (2) the recurrence of physical abuse in a cohort of children first physically abused during infancy. Among 577,655 reports concerning children under three years of age, there were 27,367 first confirmed physical abuse reports in 16 states during 1995-1999 in NCANDS- DCDC. The mean annual incidence of first confirmed reports for physical abuse was 2.3 cases per 1,000. Physical abuse recurred before age three years in 5.8% of children at a rate of 0.024 cases per person-year. While caretaker problems with drugs (risk ratio, RR= 2.4; 95% confidence interval, 95%CI=[1.2, 4.8]) and emotional disturbance (11.7; [5.4, 25]) were positively associated with physical abuse recurrence in a cohort of children first abused during infancy, medical report source (RR=7.0), male gender (10.8), child medical problems (23.4), foster care services (0.0007) and time to post-investigation services (1 .004/day) were associated with recurrence in multivariate proportional hazard models. These results should be interpreted with caution due to the limitations of these analyses and the substantial lack of information in certain fields in NCANDS-DCDC. COPyright by Vincent John Palusci, MD. 2002 This work is dedicated to the memory of Dr. Ray Helfer, to those who strive to improve the lives of children and families, and to those who support their work. iv ACKNOWLEDGEMENTS I wish to thank Nigel Paneth, M.D., M.P.H., Wenjieng Fu, Ph.D. and Nannette Bowler, J .D. who provided invaluable direction for this work. Dr. Paneth and the Faculty of the Department of Epidemiology at Michigan State University have nurtured my epidemiologic skills to understand how and why children are maltreated. My TRECOS fellowship at Michigan State University and administrative support from DeVos Children’s Hospital have also been greatly appreciated. Elliott Smith, Ph.D. and Cornell’s Summer Research Institute allowed me to focus my work. I must also thank my wife, Roz, and my children, John and Katie, who helped me to get my homework done and who suffered through my many late nights and busy weekends. The data utilized in this publication were made available by the National Data Archive on Child Abuse and Neglect, Cornell University, Ithaca, NY, and have been used by permission. Data from the National Child Abuse and Neglect Data System were supplied by state child protective services agencies to the Children's Bureau, the Administration on Children, Youth and Families, US. Department of Health and Human Services. Funding for NCANDS was provided by the Children's Bureau, US. Department of Health and Human Services. Technical support on NCANDS is provided under contract to the Children's Bureau by Walter R. MacDonald & Associates, Inc. The authors are solely responsible for its content and analyses; neither the participating state agencies, Walter R. MacDonald & Associates, Inc., the Children's Bureau, the Archive, Cornell University, or its agents or employees bear any responsibility for the analyses, opinions, or interpretations presented here. TABLE OF CONTENTS LIST of TABLES ................................................................................... vii LIST of FIGURES ................................................................................ viii INTRODUCTION .................................................................................... 1 History and definitions 2 Impact of physical abuse 3 Risk factors 5' Recurrence of physical abuse 7 National Child Abuse and Neglect Data System 9 OBJECTIVES and HYPOTHESES ............................................................. ll METHODS ......................................................................................... 13 Study design 13 Study subjects 18 Data management 18 Protection of human subjects 20 RESULTS ............................................................................................ 21 Assessment of the dataset 21 First occurrence of physical abuse 24 Population-based incidence rates of first physical abuse 27 Recurrence of physical abuse 29 Cohort analysis 32 Analysis of time to physical abuse recurrence 33 DISCUSSION ....................................................................................... 36 Physical abuse occurrence 36 Hypotheses regarding physical abuse recurrence 40 Limitations 45 CONCLUSIONS .................................................................................... 47 APPENDICES ....................................................................................... 50 REFERENCES ...................................................................................... 67 vi LIST OF TABLES Table l NCANDS Child Abuse and Neglect Reports, by State and Year, Children Under 3 Years ........................................................................................ 22 Table 2 NCANDS Child Abuse and Neglect Reports, by Report Disposition and Year ............ 23 Table 3 Comparison of Child Factors in Confirmed Reports, Physical Abuse (PA) vs. Non- Physical Abuse (Non-PA) (N=206,274) ........................................................ 25 Table 4 Comparison of Family and Caretaker Factors in Confirmed Reports, Physical Abuse (PA) vs. Non-Physical Abuse (Non-PA) (N=206,274) ........................................ 26 Table 5 Comparison of Report and System Response Factors in Confirmed Reports, Physical Abuse (PA) vs. Non-Physical Abuse (Non-PA) (N=206,274) ................................ 27 Table 6 First Confirmed Reports of Physical Abuse, by State and age, rates per 1,000 for age- specific populations ................................................................................. 28 Table 7 Variable Responses for Index Cases in Physical Abuse Recurrence Cohort (N=2,670)... 31 Table 8 Initial reports in Cohort and Physical Abuse Recurrence Reports, by State and Year ..... 32 Table 9 Bivariate Association of Factors with Physical Abuse Recurrence (N=2,670). . . 33 Table 10 Statistical Power Calculations ....................................................................... 34 Table 11 Final Cox regression Model Comparing Time to Second Confirmed Physical Abuse Reports with Child, Family and System Response Factors in Recurrence Cohort ......... 35 Table 12 Child Maltreatment Fatality Reports, by State and Year ........................................ 40 LIST OF FIGURES Figure 1 Conceptual model ................................................................................ 11 Figure 2. NCANDS Variables ............................................................................... 14 Figure 3. Calculation of risk ratios and 95% confidence intervals .................... I .................. 15 Figure 4. Calculation of statistical power ................................................................... 16 Figure 5 Cox proportional hazard models ................................................................. 17 Figure 6 Population subsets ................................................................................. 21 Figure 7 Proportional hazards equation for fitll recurrence model ..................................... 35 Figure 8. Effects of missing or ‘unknown’ data ........................................................... 44 viii INTRODUCTION The physical abuse of young children caused injury and/or death in 166,626 US. children in 1999 (Administration on Children, Youth and Families, 2001). Despite the creation of reporting laws and governmental agencies for child abuse investigation, many surviving children are then re-abused (Alexander, Crabbe, Sato, Smith, Bennett, 1990; Benedict, Zuravin, Brandt, Abbey, 1994; DePanfilis, Zuravin, 1998). Young children and infants are at particular risk for severe physical injury and death given their small size, with 50% of fatal child abuse occurring during infancy and 75% occurring from birth to age three years (Hicks & Gaughan, 1995; Ellis, 1997). Shaken Baby Syndrome and other inflicted neurologic injuries during the first three years of life result in significant long-term disability and handicap in infants (Hermes, Kini and Palusci, 2001). While much has been learned about the patterns of abusive injury in young children, little has been reported regarding the incidence and risk factors for abuse and re- abuse in large populations (DePanfilis & Zuravin, 1998). Understanding factors associated with re-abuse has the potential to save lives by improving our assessment of a family’s risk for re-abuse, enhancing our measurement of the outcome of investigation and treatment programs, and efficiently targeting limited resources (Brayden, Altemeier, Dietrich, et al. 1993; Oates & Bross, 1995; Kasim, Cheah & Shafie, 1995). This study assesses the frequency, incidence and risk factors associated with the occurrence and re-occurrence of physical abuse in infants and children under three years of age by secondary analysis of child maltreatment reports during 1995-1999 in the Detailed Case Data Component of the National Child Abuse and Neglect Data System. History and definitions In 1946, John Caffey described fractures and intracranial injuries in children that were caused by their parents, which Kempe and others later described as “the battered child syndrome” (Cafl‘ey, 1946; Kempe, Silverrnan, Steele, Droegmueller and Silver, 1962). With the adoption of the US Child Abuse Prevention and Treatment Act in 1974, medical and other professionals were mandated to report suspected child abuse and neglect to governmental agencies. State child protective systems were created to collect and investigate such reports, and the National Child Abuse and Neglect Data System (N CANDS) began to compile data concerning reports fiom all US states and territories as required by law (CAPTA, 1988). Child Protective Services agencies (CPS) were created in each U.S. county or other jurisdiction to receive child maltreatment reports, decide if investigation is necessary, investigate, then determine if investigated reports are substantiated or indicated (have credible evidence) or unfounded (are without credible evidence). CPS agencies may provide family preservation or family support services, or CPS may remove children from the home and/or petition courts to terminate parental rights based on state statutes (Administration on Children, Youth and Families, 2001). The first NCANDS report was published in 1992 with summary information from 1990. In an effort to assess risk factors and service outcomes, state CPS agencies were later given the opportunity to voluntarily submit expanded information about child, family and services provided to the Detailed Case Data Component (DCDC) ofNCANDS (Gaudiosi et al., 2001; Administration on Children, Youth and Families, 2001). Although precise definitions vary from state to state, child abuse and neglect reports in NCANDS [\J are generally defined based on federal guidelines to include the following (CAPTA, 1988; Sedlak, 1996; National Data Archive on Abuse Child Abuse and Neglect, 2000): Child physitgtl abuse has been defined as non-accidental acts by their parents or caretakers that cause or could cause physical injury to children less than 18 years of age. This is operationalized to mean that physical abuse is characterized by physical injury (for example, bruises and fiactures) resulting from punching, beating, kicking, biting, burning, or otherwise harming a child. Although the injury is not an accident, the parent or caretaker may not have intended to hurt the child. The injury may have resulted fi‘om over-discipline or physical punishment that is inappropriate to the child's age or condition. Child sexual abuse is the involvement of children under the age of 18 in sexual or incestuous acts they do not fully comprehend, cannot give legal consent to, and that violate criminal sexual conduct laws or social taboos. CAPTA defines sexual abuse as "the use, persuasion, or coercion of any child to engage in any sexually explicit conduct (or any simulation of such conduct) for the purpose of producing any visual depiction of such conduct, or rape, molestation, prostitution, or incest with children." Emotional Maltreatment / chhological Abuse includes acts or omissions by the parents or other persons responsible for the child's care that have caused, or could cause, serious behavioral, cognitive, emotional, or mental disorders. In some cases of emotional/psychological abuse, the parental acts alone, without any harm evident in the child's behavior or condition, are sufficient to warrant CPS intervention; for example, the parents/caretakers use extreme or bizarre forms of punishment, such as torture or confinement of a child in a dark closet. For less severe acts, such as habitual scapegoating, belittling, or rejecting treatment, demonstrable _ harm to the child is often required for CPS (the public agency providing services to abused and neglected children and their families) to intervene. Child neglect is the physical neglect, child abandonment and expulsion, medical neglect, inadequate supervision, emotional neglect and educational neglect by parents, parent substitutes, and other adult caretakers of children. Impact of physical abuse There were 166,626 known victims of physical abuse (PA) in the US. in 1999 with 1,082 fatalities (Administration on Children, Youth and Families, 2001). Physical abuse consistently accounts for approximately one-fifth of confirmed reports to Child Protective Services, and while other forms of maltreatment appear to be declining, the annual incidence of physical abuse has declined only 29% since its peak in 1995. This compares to the contemporaneous decline of 39% in sexual abuse substantiated reports during 1995-1999 (Jones, Finkelhor, Kopiec, 2001). Additional injuries are increasingly labeled as physical abuse as patterns of abusive injuries have become clinically apparent beyond those initially described in a ‘battered child’ (Gillenwater, Quan & F eldman, 1996; Goldstein, Kelly Bruton and Cox, 1993; Kasirn, Cheah & Shafie, 1995; Oates & Bross, 1995; Rivara, 1985; Wolfner & Gelles, 1993). Despite these declines, Reece & Sege noted that abuse accounted for a large proportion of serious head trauma, subdural hematoma, subarachnoid and retinal hemorrhage in children under six years of age. Higher mortality and longer hospitalizations have been noted among children with physical abuse as compared to children with accidental injuries (Reece & Sege, 2000; Thomas, Leventhal & F riedlaender, 2001). Infants are at particular risk; ‘serious’ forms of physical abuse including head trauma, bone fracture and other “grievous harm” accounted for nearly one-halfof reported PA in Wales infants (Sibert et al., 2002). Marshall noted that while child abuse hospitalizations comprise less than 1% of pediatric hospital admissions and while only 0.2% of children reported to Child Protective Services require hospitalization, more than halfof those hospitalized were infants and 15% died (Marshall, 1997). Children identified as ‘high risk’ for abuse by clinicians on the postpartum record in New Haven were hospitalized more often and for longer periods of time during the first four years of life than were children matched on socioeconomic status and other factors drawn from the same population (Leventhal, Pew, Berg & Garber, 1996). Among 5,446 children in three cities, children were more likely to be reported to CPS for head trauma than for neglect, and poverty, young age of the child and single parent family were independently associated with increased frequencies of concern and reporting to CPS (Thyen, Leventhal, Yazdgerdi & Perrin, 1997). Studies of physical abuse have been limited by small sample sizes. Large databases such as NCANDS and national registries ofi’er opportunities to improve our understanding (Wissow & Wilson, 1988). For example, DiScala noted that child abuse accounted for 10.6% of all blunt trauma in children under age five years in the National Pediatric Trauma Registry over a 10 year period and that abused children were mainly injured by battering (53%) and by shaking (10.3%), with head injuries accounting for more than 60% of those affected, followed by extremities, thorax, abdomen, face and skin (DiScala, Sege, Li & Reece, 2000). Bertolli (1995) has suggested that analyses using several designs with large populations offer the most promise for understanding child maltreatment. Risk factors Several risk factors have been identified in families who maltreat children. Low maternal education (odds ratio, OR=2.6), low religious attendance (OR=2.2), young maternal age (OR=3.5), single parent (OR=2.3) and poverty (OR=3.7) were all statistically associated with increased risk for abuse in New York City in 1992 (Brown, Cohen, Johnson & Salzinger, 1998). Other family characteristics such as early parental separation, dissatisfaction, illness, pregnancy complications, child disability and low parental involvement were also significantly associated with abuse (Brown et al., 1998). In one of the few case-control studies of disability and maltreatment, Crosse noted that ‘disabled’ children were 1.7 times more likely to be maltreated than were non-disabled children, with neglect thought to be more common than physical abuse, and disabled children being older and having higher levels of emotional abuse (Crosse, 1993). Physical abuse has been also associated with young maternal age in several studies (Connelly & Straus, 1992). Among children assessed in the neonatal period, maternal depression, poor education, disability, alcohol consumption, poverty, and maternal separation from their own mothers were significant predictors of child maltreatment reports through the fourth year of life (Kotch, Browne, Dufort & Winsor, 1992). Furthermore, parental education, psychiatric illness, sexual abuse as a child, and absence of the father has also been associated with abuse in the UK. (Sidebotham, Golding, ALSPAC, 2001). The presence of violence between the parents has been shown to limit the positive impact of home visitation programs designed to prevent child maltreatment (Eckenrode, et al., 2000). The Third National Incidence Study of Child Abuse and Neglect (NIS-3) surveyed ‘sentinel’ child welfare professionals in 83 US. counties in 1992-1993 to identify all cases of child abuse and neglect in child protective services, criminal justice, healthcare and social service systems. Using a weighted, randomized county selection process designed to be nationally representative, NIS-3 identified poverty and family size but not race/ethnicity as risks for all forms of maltreatment, although there is controversy regarding sample selection bias and potential racial differences in child abuse reporting (Ards and Harrel, 1993; Ards, Chung and Myers, 2001; Cappelleri, Eckenrode and Powers, 1993; Sedlak and Broadhurst, 1996; Sedlak, Bruce & Schultz, 2001). Rates of physical abuse were found to be 21.3 per 1,000 in N18 studies, compared to 35.5 per 1,000 in disabled children. More physical abuse was likely to be reported to Child Protective Services (CPS) agencies in NIS-3 among younger children as compared to other forms of abuse, but race, gender and income were not predictive of physical abuse (Sedlak & Broadhurst, 1996; Administration on Children, Youth and Families, 2001). The relationship of abuse to poverty remains incompletely understood, and physical abuse in infancy and young childhood was not specifically addressed in NIS-3 (Ards et al., 1993; Drake et al., 1996). Recurrence of physical abuse Despite CPS reporting and investigation, physical abuse recurs. Small studies of abusers have reported re-abuse rates ranging from 16-67%. Alexander noted that 71% of children with Shaken Baby Syndrome had evidence of prior abuse or neglect, with one third having prior shaking injury (Alexander, Crabbe, Sata, Smith & Bennett, 1990). Ferleger noted that physical abuse can recur in the form of excessive corporal punishment, or other maltreatment such as emotional abuse can continue or escalate despite a decrease in physical abuse after investigation and treatment termination by Child Protective Services (Ferleger, Glenwick, Gaines & Green, 1988). Levy noted a 16.8% frequency of re-abuse among 304 children discharged from an inpatient assessment and treatment program over 5-6 years (Levy, 1995). It is difficult to predict which families will reabuse their children. In a review of 53 studies of abuse recurrence in cases reported to CPS, more recurrence was noted with increasing ‘risk’ as determined by CPS, ranging from 1-2% for ‘low risk’ families to over 50% for ‘high risk’ families over five years (DePanfilis & Zuravin, 1998). ‘Risk’ is variably identified during case investigation by CPS agencies and is often used to determine the level of services provided to ‘reunify’ children with their families (Chaflin, Bonner, Hill, 2001). Marital status and poverty were associated with recurrence in some studies, and there was a tendency for re-abuse to occur with more frequency among children not reported to CPS. A carefiil review of risk assessment data from several CPS state agencies noted that White children had greater risk for abuse recurrence than did Black children, who were more likely to be neglected (Baird, 1999). Child maltreatment can recur during investigation or after removal of the child from the parents (DePanfillis, 1999). Delays in removal from the caretakers or petition for juvenile court jurisdiction may also be associated with repeat abuse. In a small sample, Ferleger noted that 18 of 45 parents during treatment were cited for re-abuse, only 2/3 of whom were reported to CPS. Physical abuse has been shown to recur after a child is placed into foster care, albeit at low rates (Benedict, Zuravin, Brandt & Abbey, 1994): Lower rates of recurrence are noted after unsubstantiated as compared to substantiated physical abuse reports (Thompson et al., 2001). Given the serious morbidity and mortality of physical abuse in infants, recurrence of physical abuse is of great concern (Brown, et al., 1998; Leventhal, et al., 1996). Repeat physical abuse may be inflicted on a child with additional physical and mental disabilities as a result of the first physical abuse. Placement of a child into foster care improves a child’s function over a 12-month period (Horwitz, 2001), and clinicians are concerned about properly identifying physically abused infants in order to take steps to prevent such recurrence during young childhood when the risk is greatest for serious neurological and developmental impairment (Alexander et al., 1990). National Child Abuse and Neglect Data System The National Child Abuse and Neglect Data System (NCANDS) collects data from all US. States as mandated by the Child Abuse Prevention and Treatment Act (CAPTA, 1988; Administration on Children, Youth and Families, 2001). NCANDS collects regarding cases in which local child protective service agencies investigate suspected maltreatment. When CPS agencies find credible evidence that abuse or neglect has occurred, the report is labeled ‘substantiated’ or ‘indicated’; if not, the report is labeled ‘unsubstantiated’. Aggregate information about all reports from all states is collected by NCANDS and reported in the annual NCANDS Summary Data Component (SDC). Substantiated and indicated reports are combined as ‘confirmed’ cases for analyses. Additional case-level information is provided voluntarily by some States to the NCANDS Detailed Case Data Component (DCDC). The DCDC is archived and maintained in the National Data Archive in Child Abuse and Neglect (N DACAN ) at the Family Life Development Center in the College of Human Ecology of Cornell University under a cooperative agreement with the Children’s Bureau, Administration on Children, Youth and Families in the US. Department of Health and Human Services. It is important to note that NCANDS-DCDC contains only ‘numerator’ data, i.e. fiequencies of reported child maltreatment cases, although comparisons have been made to State populations in published NCANDS analyses. Most States have not provided five sequential years of data for NCANDS-DCDC during 1995-1999. During 1995-1999, NCANDS-DCDC included over 2.8 million total reports of suspected child abuse and neglect, with 936,999 confirmed reports. NCANDS-DCDC contained information from 16 states with over 30% of the US. child population during 1995-1999. There were 577,655 total reports concerning children under age three years, and of these, over 30,000 were confirmed with child physical abuse. DCDC case-level data includes fields with information concerning child age, gender, race, ethnicity, CPS report date, source and disposition, services provided, type(s) of abuse or neglect, child factors (such as special medical needs, developmental delays, drug exposure, impairments, disabilities, behavior and medical problems) and caretaker characteristics (such as drug or alcohol problem, mental or physical problem, mental retardation, emotional disturbance, disability, medical problems, violence between caretakers, receipt of financial assistance, or problems with finances or housing) (Appendix). For unsubstantiated reports, only state of residence, CPS report date and disposition, report source, child age, gender, race, and ethnicity are included. Coded child identifiers are used by NDACAN to protect privacy while allowing assessment of maltreatment recurrence (Gaudiosi, et al, 2001). Records with fatality have been excluded, although other records prior to death for a particular child remain in the dataset (Elliott Smith, personal communication, 2002). 10 OBJECTIVES and HYPOTHESES The principle objectives of this study are to describe the incidence of first physical abuse occurrence and risk factors associated with physical abuse recurrence in a large population of young children and families reported to state Child Protective Services agencies (CPS) for suspected maltreatment. After describing child, family and system response factors noted among children with physical abuse less than 3 years of age, several hypotheses are tested regarding the relationship of risk factors to occurrence and recurrence in an overall conceptual model (Figure 1): Figure 1 Conceptual model Physical Abuse Recurrence: Prior Physical Abuse before age 1 year + Family Factors + Child Factors + System Responsefitctors '9 Recurrence Substance Abuse Age Post-Investigation Services before age Alcohol Gender Time to services 3 years Drugs Race Family Support Services Mental Illness Ethnicity Family Preservation Services Retardation Substance Exposure Foster Care Emotional Alcohol Time to Removal Learning Disability Drugs Juvenile Court Petition Physical Illness Mental Illness Time to Court Petition Sensory Impair Retardation Physical Disability Emotional Medical Illness Learning Disability Violence Behavior Problem Betw Caretakers Physical Illness Poverty Sensory Impairment Inadequate Housing Physical Disability Financial Problems Medical Illness Receipt of Assist ll The following hypotheses will be tested in this study: I. II. III. Child factors A. Less than 5% of physically abused infants will have recurrence; B. Infancy, non-White race, male gender and disability place children at increased risk for recurrence of physical abuse; Family factors A. Poverty will not be associated with increased risk of recurrence of physical abuse; B. Geographic differences in recurrence rates will not be noted among states with higher proportions of their children living in urban settings; System response factors A. Removal to foster care and juvenile court petition will be associated with lower rates of physical abuse recurrence but increasing time to removal or petition will be associated with higher rates; B. Medical source of report and post-investigation, family support, and family preservation services will be associated with higher rates of physical abuse recurrence as compared to other forms of maltreatment. 12 METHODS Study design Several study designs were used to assess the first occurrence and recurrence of physical abuse (PA) in infants and young children. For occurrence, a case series was constructed consisting of all confirmed physical abuse reports for children less than three years of age in NCANDS-DCDC, and the frequency of confirmed physical abuse was compared with other confirmed maltreatment types. Frequencies of child, family and system response factors (Figure 2) were calculated comparing PA to other maltreatment types. Effect of urban population was assessed by creating a new variable representing an estimate of the percent of the state’s child p0pulation living in urban metropolitan areas as noted in US. Population Estimates (U .8. Census Bureau, 2000). In a second study design, report dispositions were stratified by State, year and abuse type, and confirmed first physical abuse reports were linked state-by-state and year- by-year with census data on the age-specific population of children at risk. Physical abuse rates per 1,000 children were calculated by state and child age over the five-year study period in this population-based incidence rate series. For recurrence, a cohort was constructed from records of children aged less than 1 year with confirmed physical abuse in those states and years with data fiom three or more consecutive years in NCANDS-DCDC. Within this cohort, records of children with repeat confirmed PA reports were compared to records of children with no additional confirmed physical abuse reports in a cohort analysis using risk ratio calculations and proportional hazards models. An event or case of physical abuse recurrence was defined 13 Figure 2. NCANDS Variables Child Factors Farnil and aretaker Facto Report & System Response Factors Age in years Gender Race Hispanic Ethnicity Problem / exposure to Alcohol Problem / exposure to Drugs Mentally Retarded Emotionally Disturbed Visual or Hearing Impaired Learning Disability Physically Disabled Behavior Problem Other Medical Problem Caretaker Problem with Alcohol Caretaker Problem with Drugs Caretaker Mentally Retarded Caretaker Emotionally Disturbed Caretaker Visual or Hearing Problem Caretaker Learning Disability Caretaker Physically Disabled Caretaker Other Medical Problem Violence Between Caretakers Inadequate Housing Family Financial Problems Family Receives Public Assistance Report ID Child ID State Report Date Disposition Date Report Disposition Report Source Post Investigation Services and Date Family Support Services Provided Family Preservation Services Provided Foster Care Services and Removal Date Juvenile Court Petition and Date Physical Abuse Disposition Neglect Disposition Medical Neglect Disposition Sexual Abuse Disposition Psychological Abuse Disposition Other Abuse Disposition Unknown Maltreatment Disposition, 14 as a second confirmed report for physical abuse occurring in the same state for the same child before age three years. Risk ratios with 95% confidence intervals were calculated for the child, farme and system response factors listed in Figure 2 to assess whether factors in first reports were associated with physical abuse recurrence. Chi square analysis was used to assess the statistical significance of the association of the risk factor with physical abuse recurrence in these analyses (Figure 3). Figure 3. Calculation of risk ratios and 95% confidence intervals In a two by two contingency table of risk factor exposure by outcome: Disease Totals Exposure + - + A B A + B = nl - C D C + D = II; Total A+C B+D A+B+C+D=n,+n2=N Where: A= Number exposed with the outcome of interest B= Number exposed without the outcome C= Number not exposed with the outcome D= Number not exposed without the outcome p, = probability of developing disease for exposed individual = A/n, p2 = probability of developing disease for non-exposed individual = C/n2 Estimate of Exposure Risk Ratio = R = pl / p2 = [A/(A + B)]/[C/(C+ D)]= (A/n,)/ (C/nz) 95% Confidence Interval (95%CI) for Exposure Risk Ratio (c., cl): cl is the lower bounds of the 95%CI = em“) ' Z ””2 W“ + ”“2” c2 is the upper bounds of the 95%CI = enm’ + Z “”2 Wm " ““2” Where Z is the inverse normal function, or is the estimate of type I error (set at 0.05 to obtain 95%CI). Statistical association calculated using chi square or Fishers exact test. Valid with normal distribution, [n,*p,*(1-p,)] a 5 and n2*p2*(1-p2)] a 5. Adapted from Rosner (1995), pp. 362-364. Statistical power (l-B) was calculated to assess the adequacy of the sample size of this cohort (Figure 4). Time to recurrence of physical abuse was calculated for cases, and 15 time elapsed to provision of post-investigation services, juvenile court petition and removal to foster care were also calculated using the date of the confirmed PA second report. Controls in the cohort analyses consisted of all records with complete information in children in the cohort who did not have PA recurrence. Figure 4. Calculation of statistical power Statistical power (1 —- [3) comparing two proportions in a cohort can be' calculated from 213 = [(N (d"')2 R)/((R+1)(p)(1-P))]°'5 - Za/Z Where: R = ratio of unexposed / exposed records for the selected exposure R is the estimate of the Risk Ratio, based on the cohort NT= total number in cohort = N1 / (RR) Nl = number of cases in the cohort with the selected outcome = (NT)RR p = weighted average of the proportion of the entire cohort with the selected outcome (p0) and the proportion of those exposed with the selected outcome (p.) such that p = [p, + R(po)] / [l + R] (1* = difference of proportions between exposed and overall cohort = p , — p2 Z“,2 = Z0025 = 1.96 for a= 0.05 Power (1 - ,8) can be read from Z in Table 12-14, in Kelsey et a1. (1996), p.331. Adapted fi'om Kelsey, et al. (1996), pp. 331-335. Cox regression models (Allison, 1995; Kelsey, 1996) were used to calculate proportional hazard rate-ratios for risk factors and their association with time to a second episode of confirmed physical abuse in a right-censored model, with recurrence date up three years after first occurrence among children in the cohort (Figure 5). These models are useful when exact person-years at risk are unavailable in the data or inappropriate to use (Kelsey, p. 168). Cox regressions model exposure hazard rates as varying fimctions of time which is considered a continuous variable, and they do not assume that rates are constant within exposure categories. All risk factors and elapsed times for services were entered into the model, and variables with the highest variance were removed one at a 16 time and the model re-calculated until all remaining variables had P<0.05. Changes in model P were monitored to assure model convergence. The null hypothesis that the risk ratio of the factor was equal to 1.0 was rejected if the probability (P) of the association was found to be less than 0.05 in these analyses (a=0.05). Risk ratios with 95% confidence intervals were calculated for the variables remaining in the adjusted, final model. Figure 5. Cox proportional hazard models Cox regression models use estimates of maximizing partial likelihood in a proportional hazards model: 111.“) / hj(t) _—_- e(131(in-le)+...+ Bum-gr) in which the type of function, or(t), does not have to be chosen apriori, the estimates are consistent and asymptotically normal, and depend on ranks of event times. 95% CI for factori == [ e"'""“""‘1 “a“ “'7 ' w'““”‘ “"1”" "'6“ ”m, eflnflnzard ratio, Bi) + ZaIZ‘stnndard error (hazard ratio. [30] ] Adapted from Allison (1995), pp. 113-115. The Investigator obtained Dataset #75 and its updates from the National Data Archive on Child Abuse and Neglect (NDACAN). NDACAN datasets are available to researchers for $75 for non-commercial statistical research and teaching and have personal identifying information removed to protect individual confidentiality. Costs associated with dataset acquisition were paid for by research funds available to the Investigator. 17 Study subjects Records in NDACAN Dataset #75 represent de-identified reports to Child Protective Services during 1995-1999 which were collected by the participating states and entered into the National Child Abuse and Neglect Data System (NCANDS). Data management NCANDS-DCDC report files contain administrative data entered by CPS caseworkers about the child and family, with other case and service characteristics as determined by investigation (Appendix). NDACAN Dataset #75 (1995-1997) and its updates (1998-1999) contain 14 variables in the ‘J ’ files pertaining to all reports and 62 variables in the ‘K’ files pertaining to those reports in ‘J’ that were ‘confirmed’ (National Data Archive on Child Abuse and Neglect, 2000). Several variables such as child or family ‘problems’ with alcohol or drugs reflect the worker’s determination as to the presence or absence of a factor identified by the state CPS agency as needed case information. The ‘J’ and ‘K’ files can be matched and combined using the fields common to both files, with children matched using a unique child identification code. The Investigator used SAS statistical software package, version 8.1 for data management and analysis (SAS Institute Inc., Cary, NC). Individual files provided by NDACAN are provided for each state and year in SAS transport file format. Two files for each state are provided by NDACAN containing information about all reports (unsubstantiated and substantiated) and substantiated reports only (J and K files, respectively). Transport files for each year and state were imported using SAS code provided by NDACAN and were combined for the years 1995 through 1999. Duplicative 18 records and reports occurring on the same day were deleted using a ‘roll up’ procedure provided with the dataset, and records for children ages three years and older were removed. The resulting dataset contained records that pertained to unique child ID-report date pairs for children less than three years of age, and substantiated and indicated reports have data additional fields available after CPS investigation. Records were sorted by state, disposition and year in the dataset and compared with published information to assess overall dataset integrity (Gaudiosi, 2001). Variable ranges were reviewed to assure consistency with permitted dataset entries (NDACAN, 2000). Records were then sorted by state, child ID and type of maltreatment, and the number of unsubstantiated, substantiated and indicated reports were calculated for each State and form of maltreatment. Substantiated and indicated reports were combined for further analyses as ‘confirmed’ reports. A variable was created indicating report year, and flag variables were created to indicate type of abuse and records of children with confirmed physical abuse. Counter variables were created to calculate the number of confirmed reports for physical abuse, sexual abuse, neglect, medical neglect, psychological abuse and other abuse and the total number of all reports for each unique state-child ID pair. Values in all variable fields were assessed to determine whether they were categorical or continuous in nature. The distribution of continuous variables was assessed. The frequency of ‘unknown’ or missing responses was counted for each variable, and steps were taken to assess patterns of missing data to determine if methods should be used to exclude cases or fields or impute data using standard methods (Frongillo & Rowe, 1999). Once the dataset was finalized, the fiequencies of responses 19 for categorical variables, and the means and standard deviations for continuous variables were assessed. Protection of human subjects The dataset provided by NDACAN has been protected with safeguards such as encrypted case identifiers and removal of fatality information. The User’s Agreement required several standards for dataset use, including maintaining data security, limiting others’ access to data, preserving confidentiality, and corresponding with NDACAN and the Children’s Bureau before publication. The investigator complied with these standards and kept the dataset and analyses in password-protected files on a computer separate from the institutional network. No patient names were used, and all reports of study outcomes and conclusions use aggregate information only. The study protocol (#02-046) was reviewed and approved 1/23/2002 by the Michigan State University Committee on Research Involving Human Subjects (UCRIHS) and the Spectrum Health Research Committee. 20 RESULTS Assessment of the dataset The Detailed Case Data Component of the National Child Abuse and Neglect Data System (NCANDS-DCDC) contains 2,834,336 records fi'om 1995-1999, including 577,655 reports of suspected or confirmed abuse among children less than three years of age (Figure 6). Confirmed PA records made up 14% of all confirmed reports among children ages less than 3 years, and more than 90% of these records were first confirmed reports. Values in all variable fields were categorical in distribution, except for child age in years, date fields and calculated fields counting the number of reports. Variable responses conformed to published dataset parameters (NDACAN, 2000) and published reports (Administration on Children, Youth and Families, 1997-2001). Figure 6. Population subsets NCANDS-DCDC, 1995-1999 Reports, all children 2,834,336 Reports, children ages under three years 577Y655 Confirmed reports, all types, age < 3y 216268 Confirmed physical abuse reports. age < 3yr 30:56 First confirmed physical abuse reports, age < 3yr 27:67 Index physical abuse reports in cohort 2,670 Follow-up physical abuse reports in cohort 155 All NCANDS files provided by NDACAN were combined and duplicate records removed, resulting in a single file with reports from 16 US. states, with 8-14 states 21 reporting for each of the included years (Table 1). Additionally, 7 of 16 states did not report data for three or more consecutive years. Colorado, Florida, Kentucky, Massachusetts, Oklahoma, Rhode Island, Utah, Vermont, and Wyoming (whose state populations amounted to 53% of the total number of NCANDS-DCDC reports) reported data for at least one consecutive three—year period during 1995-1999. Table 1 NCANDS Child Abuse and Neglect Reports,1=92; (0) Child Problem with Alcohol Yes=2; No=1.050 (61) Child Problem with Drugs Yes=5; No=49 ‘ (61) Child Mentally Retarded Yes=0; No=986 (63) Child Emotionally Disturbed Yes=0; No=986 (63) Child Visually or Hearing Impaired Yes=1 1; No=975 (63) Child Learning Disability Yes=2; No=984 (63) Child Physically Disabled Yes=9; No=977 (63) Child Behavior Problem Yes=2; No=947 (64) Child Other Medical Problem Yes=9; No=375 (86) Caretaker Problem with Alcohol Yes=56; No=926 (63) Caretaker Problem with Drugs Yes=71; No=878 (64) Caretaker Mentally Retarded Yes=1; No=882 (67) Caretaker Emotionally Disturbed Yes=5; No=878 (67) Caretaker Visually/Hearing Impaired Yes=2; No=881 (67) Caretaker Learning Disability Yes=1; No=8 82 (67) Caretaker Physically Disabled Yes=0; No=8 83 (67) Caretaker Other Medical Problem Yes=0; No=281 (89) Violence Between Caretakers Yes=28; No=3 85 (85) Inadequate Housing Yes=8; No=339 (87) F arnily Financial Problems Yes=15; No=333 (87) Family Receives Public Assistance Yes=121; No=869 (63) Post-Investigation Services Yes= 1,625; No=1,042 (<1) Family Support Services Yes= 62; No=3 83 (83) Family Preservation Services Yes= 86; No=974 (60) Foster Care Services Yes= 605; No=856 (45) Juvenile Court Petition Yes= 399; No=950 (49) *All variables categorical except child age in years; total records=661 "Missing or labeled ‘Unknown’; NA=not applicable, all labeled as ‘yes’ or ‘no’ From among these 2,670 children, there were 155 (5.8%) cases of second confirmed physical abuse before three years of age (Table 8). While most states had consistent decreasing recurrence in follow-up years, some states had few recurrences 31 recorded (Colorado, Rhode Island), no recurrences at all (Vermont) or no index cases (Utah). Using rates for individual states fi'om Table 8, the 95% confidence interval for recurrence can be calculated as [0.0%, 16.4%] (data not shown). Table 8 Initial reports in Cohort and Physical Abuse Recurrence Reports, by State and Year State Year: 1995 1996 1997 1998 ~ 1999 Total Colorado: *Index Reports 54 49 - - - 103 +Recurrence 0 4 0 0 - 4 Florida: Index Reports 453 427 432 - - 1,312 Recurrence 1 1 19 19 1 1 3 63 Kentucky: Index Reports - - 271 - - 271 Recurrence - - 9 S 2 16 Massachusetts: Index Reports 221 - - - - 221 Recurrence 32 3 3 - - 38 Oklahoma: Index Reports 275 327 - - - 602 Recurrence 0 10 1 3 2 - 25 Rhode Island: .Index Reports - - 132 - - 132 Recurrence — - 1 l 0 2 Utah: Index Reports 0 0 0 - - O Recurrence 0 0 0 O 0 0 Vermont: Index Reports 9 5 5 - - 19 Recurrence 0 0 0 0 0 0 Wyoming: Index Reports - 10 - - - 10 Recurrence - 0 1 O - 1 TOTALS Index Reports 737 756 1,177 O 0 2,670 Recurrence 43 32 43 30 7 155 *First confirmed physical abuse before age 1 year in index States and years - State and year not selected +Second confirmed physical abuse before age 3 years in follow-up States and years Cohort analysis Relative risk ratios were calculated to compare the child, family and system response factors among children with and without physical abuse recurrence in the cohort (Table 9). In these bivariate analyses, variable responses of interest were compared to 32 those without the response while excluding missing or ‘unknown’ responses. Only two associations were statistically significant at P < 0.05. Significant factor associations with second confirmed physical abuse reports were noted in children in families with caretaker drug problems (risk ratio, RR=2.4) and caretaker emotional disturbance (RR=11.7): Table 9 Bivariate Association of Factors with Physical Abuse Recurrence (N=2,670) Significant factor associations N* RR“ 95% Cl*** P Caretaker Problem with Drugs 949 2.4 1.2, 4.8 0.02 Caretaker Emotionally Disturbed 883 11.7 5.4, 25 <0.001 *Remaining factors with P> 0.05, N=cases and controls with complete data; "Risk ratio ***95% Confidence Interval Statistical power calculations (Table 10) indicate that the entire cohort has a power (1 — B) greater than 0.80 to detect a risk ratio (R) of 2.0 for factors with greater than 6% (R=14) exposed cases (report source, gender, race, percent urban population, prior unsubstantiated reports and whether post-investigation services were provided). When only half(50%) of the data is present, statistical power is greatly diminished and declined to less than 0.3 for the many child and family factors with less than 10% exposure. Analysis of time to physical abuse recurrence Mean time to physical abuse recurrence was 250 days (range, r= 1-1084 days; 95% confidence interval, 95%CI= [210, 291]). For recurrence cases, mean time from first report to provision of foster care services was 41 days (r= 0-868 days [16, 67]), time to post-investigation services was 153 days (r= 0-2,019 days [112, 194]), and time to juvenile court petition was 11 days (r= 0-804 days [0, 25]). The crude hazard rate for 33 Table 10 Statistical power calculations M911. W Percent Exposed to factor (R)* Z Power” Z Power 50 (1) 2.94 >0.99 0.55 0.71 33 (2) 2.92 >0.99 0.53 0.70 25 (3) 2.66 >099 0.35 0.64 20 (4) 2.39 >0.99 0.16 0.56 17 (5) 2.15 0.98 -0.02 0.49 14 (6) 1.93 0.97 -0.17 0.43 13 (7) 1.75 0.96 -O.30 0.38 11 (8) 1.58 0.94 -0:41 0.34 10 (9) 1.44 0.93 -0.52 0.30 6 (14) 0.91 0.82 -0.89 0.19 5 (19) 0.56 0.71 -1.13 0.13 4 (24) 0.32 0.63 -1.31 0.10 3 (32) 0.04 0.52 -1.50 <0.08 2 (49) - —0.32 0.38 -1.76 <0.08 1 (99) —0.79 0.22 -2.09 <0.08 *Full cohort NT=2,670, target RR=2.0, R = [l/percentage exposed - 1] "Power (1 - ,B ) can be read using Zpin Table 12-14, in Kelsey et al.(1996), p.331. physical abuse recurrence can be calculated by dividing the total number of cases (155) by the sum of the person-years at risk for the cases and controls. Assuming that cases were at risk for 250 days (0.68y) and controls were at risk during one-half of the year of first occurrence (6 months) plus the two years of follow-up, the hazard rate (HR) is given by: (the number of cases) / (person-years-cases + person—years-controls) = 15 5 / [(15 5 *0.68) + (2,515*2.5)] = 155 / 6,393 = 0.024 cases/person-yr (Kelsey, p. 315). This indicates that there are approximately 2.4 cases of recurrent physical abuse per year for each 100 children physically abused during infancy. Cox regression models were used to calculate proportional hazard rate-ratios for child, family and system response factors. All factors plus times elapsed between first and second confirmed physical abuse reports for post-investigation services, removal to foster care and court petition were entered into an initial model of the form (Figure 7): 34 Figure 7. Proportional hazards equation for full recurrence model 11.“) / hi0) : C [ Bl(Medical Report Source- Non-medical source) + B2 (Percent urban population) + [32 (Male - female gender) r B3 (White — non-White race) + B4 (Hispanic - non-Hispanic ethnicity) + B5 (Number of prior unconfimied reports) ~'- B6 (Child problem with alcohol — No problem) + B7 (Child problem with drugs - No problem) + B8 (Child visual/hearing impairment -— No impaimient) + B9 (Child learning disability — No disability) + B 10 (Child physical disability - No disability) + B1 1 (Child Behavior problem — No problem) + B12 (Child other medical problem — No problem) + B13 (Caretaker alcohol problem — No problem) + B l 4 (Caretaker drug problem — No problem) + B15 (Caretaker mentally retarded — Not retarded) r B16 (Caretaker emotionally disturbed - Not disturbed) + B17 (Caretaker visually impaired — Not impaired) + B18 (Caretaker Ieaming disability — No disability) + B19 (Violence between caretakers - No violence) + B20 (Inadequate housing - Not inadequate) + B21 (Financial problem - No problems) + B22 (Public assistance - No assistance) + B23 (Pm-investigation services - No services) + B24 (Time to P1 services) + B25 (Family support services - No services) + B26 (Family preservation services - No services) + B27 (Foster care services )+ B28 (Time to FC services) + B29 (Juvenile court petition - No petition) + B30 (Time to petition)] Variables with P values > 0.05 were removed one at a time beginning with the highest P, resulting in a final, adjusted model (Table 11). Medical report source, male gender, child medical problems and time in days to post-investigation services were significantly associated with an increased hazard rate of recurrence while use of foster care services was associated with decreased recurrence in a model which had only 152 records because of missing data. Among those records in the model, 16 were recurrence cases and 136 were censored because of non-recurrence. Table 11 Final Cox regression model comparing time to second confirmed physical abuse reports with child, family and system response factors in recurrence cohort Significant Factor Associations* B“ HR 95% Cl P Medical Report Source 1.945 7.0 1.0, 50 0.05 Male gender 2.379 10.8 1.3, 90 0.03 Child Medical Problems 3.151 23.4 1.3, 927 0.03 Removal to Foster Care -7224 0.0007 2x106, 0.3 0.02 Time to Post-Investigation Services 0.0043 1.004 / day 1.000, 1.007 0.05 *Model Statistics: 16 recurrence cases, 136 records censored (no recurrence); -2 Log L=59.256; Likelihood ratio P=0.008 **B= Parameter estimate; HRR=Hazard Rate Ratio 35 DISCUSSION Physical abuse occurrence This study has identified that child factors of infancy, male gender, Black race, Hispanic ethnicity and drug exposure are noted in a higher proportion of children with a first confirmed record of physical abuse in NCANDS-DCDC during 1995-1999. Medical reporting source and family factors of alcohol and drug problems and violence between caretakers are also associated with increased first confirmed physical abuse. Children with first confirmed physical abuse have fewer prior unconfirmed reports or have received fewer family support services compared to children with other types of maltreatment. They were more likely to have used foster care services, received family preservation services or juvenile court petition. Differences in rates were noted by state, although no differences were noted based on percent urban state population or among other child or family factors. Differences in maltreatment reporting rates related to race or ethnicity have been found in some national incidence studies (Ards, 2001). In Michigan and California, African-American families have been found to have decreased rates of CPS reports for physical abuse as compared to white families, whereas white families have decreased rates of CPS reports for neglect. The overall effect is thought to ‘cancel out’ any effect of race on CPS reporting, report confirmation or confirmation of additional reports by CPS when overall maltreatment statistics are evaluated (Baird, 1999), which may explain these results. 36 There were state-to-state differences in mean annual incidence rates of physical abuse, varying from 0.5 cases per 1,000 children in North Carolina to 4.8 cases per 1,000 in Oklahoma. The overall rate of 2.5 cases per 1,000 is much lower than the rates noted in NIS. Given its use of ‘sentinel’ sources, NIS is thought to include more cases than those only known to CPS (Sedlak, 1996). However, this rate is double the rate of physical abuse in infants in a recently reported population—based study in Wales, suggesting marked international differences in PA occurrence (Sibert et al., 2002). ' Children living the US. in the present study in states with higher rates of urban population actually had less physical abuse than other forms of maltreatment (Table 3). The reasons for this are unclear but may relate to how state urban populations were characterized (the percent of children living in urban settings was assumed to be the same as that of the total state population) or to how child maltreatment reports are handled in more urban states (CPS may confirm fewer numbers of physical abuse in face of more neglect in urban settings). This may also reflect decreasing national rates of urban violent crime overall, dropping over 30% to 33 violent crimes per 1,000 persons age 12 or older in 1999 (Bureau of Justice Statistics, 2000). Much still remains to be understood regarding the state-to-state variation in child maltreatment reporting and intervention as was suggested by Murray Straus in 1990 (Straus & Moore, 1990). Young children with first confirmed reports of physical abuse have few or no prior unconfirmed reports in this study. This may reflect the CPS response to reports of physical abuse in young children (which have been noted to be more likely to be confirmed than reports of other maltreatment) and which can be influenced by the harm or potential harm to the child (in whom physical abuse causes physical injuries which are 37 thought to be more apparent and may be perceived as more harmful than neglect or other forms of maltreatment). This has resulted in reports that consistently show that younger children with abuse are more likely reported to CPS than are older children (Ards, 1993). A medical source of confirmed physical abuse reports has been associated with increased first confirmed physical abuse reports in this study. Given the young age of the child and the injuries associated with these reports, it appears logical that these would be more likely to come from medical sources such as physicians or nurses rather than from school or legal professionals. However, it is particularly interesting to note that the null hypothesis could not be rejected in that no statistically significant diflerences could be found in other system response factors such as juvenile court petition, or other post- investigation, family support, and family preservation services. One might expect the system to have a greater response to serious harm in young children, yet these differences could not be confirmed in the models. This study has noted an increased rate of PA recurrence in families with caretakers with ‘drug problems.’ Others have noted caretaker problems with drugs or other medical problems and violence between the caretakers have an association with increased risk of physical abuse. In a longitudinal study of 14,138 children over 6 years in the UK, 162 were abused and the factors associated with this abuse were maternal age less than 20 years, lower educational attainment, history of maternal sexual abuse, abuse of the father, psychiatric illness, paternal age less than 20 years and prior parental foster care (Sidebotherrr, 2001). Connolly adds that a crucial factor is the age of the mother at birth and not at the time of maltreatment which predicts physical abuse (Connolly, 1992). 38 This suggests that NCANDS-DCDC variables do not get to the heart of the mechanisms by which maltreatment affects the child. Brown found that that attachment, maladaptive personalities, parental history of personal abuse, single (parenting, sociopathy, marital quality and prenatal and perinatal factors have higher associations with physical abuse before age 4 years (Brown, 1998). Kotch identified infants at birth who were at increased risk for abuse based on maternal education, maternal depression, presence of the mother’s mother during childhood and total life events and social well-being (Kotch, 1999). In a study of US. infant mortality using linked birth and death certificates over 8 years, Brenner noted that the highest rate of homicide was noted in infants born less than 2500 grams and to mothers with little or no prenatal care, young age, or who were single, Black or Native American (Brenner et al., 1999). Other than race and ethnicity, NCANDS does not provide this type of information and is therefore not an ideal source of data for these analyses. Calculation of physical abuse occurrence is also hampered by removal of child fatality information. While records other than the fatal event for an individual child have been left in NCANDS-DCDC, the actual entire record for the confirmed maltreatment report leading to death has been removed (Elliott Smith, NDACAN, personal communication, 2002). Using published reports of child maltreatment fatality (Table 12), one can estimate the revised rates of first confirmed physical abuse, assuming: (1) 50% of maltreatment fatalities occur because of physical abuse; (2) 50% of maltreatment fatalities occur during the first year of life; (3) an additional 25% of child maltreatment fatalities occur during the second and third year of life; and (4) more than 95% of children will have had no prior reports or confirmation of physical abuse. This calculation results in 39 fewer than 5% additional reports in NCANDS States, resulting in small changes in the overall crude rates in Table 6. This is not inconsistent with the estimated 1.5 per 100,000 children who die annually as the result of maltreatment (Herman-Giddens et al., 1999). Table 12 Child Maltreatment Fatality Reports, by State and Year State 1 995 l 996 l 997 1998 1999 ' Total Mean Arkansas 16 7 12 10 8 53 10.6 Colorado 23 20 24 28 32 127 25.4 Delaware 6 10 * 3 3 22 5.5 Florida 68 49 78 54 57 306 61.2 Kentucky 24 22 * 18 5 69 17.3 Louisiana 16 22 19 27 21 105 21.0 Massachusetts 9 3 1 ll * 24 6.0 Missouri 43 36 * 28 36 143 35.8 N. Carolina 17 43 * 23 21 104 26.0 Oklahoma 34 29 42 45 47 197 39.4 Rhode Island 4 4 3 2 4 17 3.4 Texas 96 110 * 176 143 525 131.3 Utah 14 9 7 12 7 49 9.8 Vermont 0 1 3 0 4 8 1.6 Washington 9 * 9 8 6 32 8.0 Wyoming 3 2 4 3 1 13 2.6 Total 382 367 202 448 395 1,794 25 .3 *not reported to NCANDS-DCDC Source: ACYF Child Maltreatment Reports, 1995-1999 Hypotheses regarding physical abuse recurrence Among the 2,670 children in the cohort, 155 (5.8%) had a second confirmed report of physical abuse before age three years (95CI= [0%, 16.4%]). While this was more than the hypothesized rate of <5%, the null hypothesis cannot be disproved given the wide 95% confidence interval. Caretaker problems with drugs and emotional disturbance were significantly associated with recurrence. Financial problems, receipt of public 40 assistance, and inadequate housing were not significantly associated with recurrence. A null hypothesis of no difference in recurrence based on percentage urban population could be rejected with greater than 0.9 power. Given that the present study looked at physical abuse in young children, these findings are understandably different from published analyses of NCANDS which have reported that 7.5% of children of all ages with substantiated abuse or neglect in the first six months of 1999 were noted to have additional confirmed reports of any type of maltreatment during the second half of that year (Administration on Children, Youth and Families, 2001). Children with substantiated reports early in 1999 were almost three times more likely to experience recurrence during the six months following their first victirrrization than children without a prior history of victimization, and neglected children were 44% more likely than physically abused children to experience recurrence of any form of maltreatment. There was more recurrence of neglect than for physical abuse during 1999, but recurrence outcomes over longer periods were not addressed (Administration on Children, Youth and Families, 2001). The relatively low recurrence rate of maltreatment in this population of children under three years of age is reassuring given the high rates of recurrence or recidivism noted in other studies where reported re-abuse rates varied from 16% to 66.8%. The true incidence of re-abuse is not known because there are thought to be discrepancies between official reporting and actual occurrence. More recurrence is thought to be associated with more than one type of initial maltreatment and with neglect as compared to physical or sexual abuse, although this was not measured in the present study (Ferleger, 1988). Alexander notes a recidivism rate for all abuse of 33% in Iowa and found that 71% of 41 infants with shaken baby syndrome (a serious form of physical abuse) had evidence of prior abuse, neglect or both (Alexander, 1990). In the National Child Abuse Evaluation Program, 30% of 1700 children were re-abused during treatment and 58% were felt to be no less likely to abuse their children at the end of treatment. Higher rates of recurrence were noted after physical abuse in children under age 3 years, 69% of whom were hospitalized and 75% of whom were removed from their parents care (Rivara, 1985). Time to recurrence in proportional hazard models was significantly associated with medical report source, child male gender, medical problems, removal to foster care and time to post-investigation services. Power calculations for other variables in proportional hazards models are beyond the scope of this study. In the present study, time to recurrence averaged over 8 months. NCANDS data in 1999 noted that the highest risk for re-abuse is within 6 months after the initial event, but again, that analysis only included recurrence during the same year (Administration on Children, Youth and Families, 2001). This differs from results in srrrall clinical samples where the risk for recurrence is thought to be increased for up to 5 or six years after treatment (Levy, 1995). Factors associated with physical abuse in prior NCANDS studies were child age less than three years (highest rates among all age groups), White race, and reports from non-medical professionals. Weak but statistically significant associations with maltreatment recurrence were noted in NCANDS in 1999 for children requiring petition to juvenile court or placed in foster care (OR=1.17) and families receiving post-investigation services such as family support and family preservation counseling, parenting classes or other services (OR=1.16) (Administration on Children, Youth and Families, 2001). While this may indicate the inefficacy of such programs, these low-level associations may also 42 indicate the selective provision of these services to high-risk families or other confounding factors (Heneghan, Horwitz, Leventhal, 1996). Differences were also noted in recurrence rates by region, but remarkable similarities in recurrence have been noted across states (Fluke et al., 1999; Gaudiosi et al., 2001). In this study, use of foster care services was associated with a greater than 99.9% decrease in recurrence risk and increased time to post-investigation services was associated with small but statistically increased risk of subsequent physical abuse. No differences in recurrence risk in proportional hazards models were found depending on the source of the report or whether family preservation, or juvenile court services were provided at the time of the first confirmed physical abuse report. This suggests that a ‘delay’ in providing foster care or post-investigation services could cause recurrence; however, the causal relationship (if any) remains to be proven. While Fluke assumed that recurrence would be low in foster care and controlled for foster care placement, the present study’s findings highlight the powerful effect of foster care services in preventing physical abuse recurrence (Fluke et al., 1999). It is not surprising that provision of services to families after physical abuse is not linked to a decrease in recurrence. Not only are services selectively provided to families at high risk for recurrence, several studies have suggested some services are not effective in preventing future child maltreatment (Chafiin, 2001). Family support services are community-based prevention activities designed to alleviate stress, promote parental competence, and improve nurturing to enable families to better use existing resources and social supports, and DePanfilis has suggested that it is the number of sessions attended rather than service characteristics which is most predictive of maltreatment recurrence 43 (Depanfilis & Zuravin, 2002). Family preservation services are short-term, home-based services provided after CPS referral to help stabilize families and reduce the risk of removal to out-of-home placement. Evaluation of these services is thought to be relatively difficult given surveillance effects, and the few studies done have not shown reductions in the removal of children at risk for maltreatment (Henegiarr, 1996). Eckenrode noted that some family preservation services such as home visiting have been associated with decreased abuse, yet these effects are moderated by domestic violence which reduces or eliminates their effectiveness (Eckenrode, 2000). Any proposed relationship between the provision of foster care or post- investigation services with PA recurrence is provisional and should be interpreted with caution. Missing data can potentially bias these results in several ways (Figure 8). For Figure 8. Effects of missing or ‘unknown’ data Removal to Foster Care Recurrence Non-recurrence TOTALS Data Present 96 1365 1461 Data MissinLUnknown 59 1 150 1209 TOTALS 1 55 251 5 2670 Percentage of records with data missing/unknown = 45% OR for recurrence for data present! missinflnknown = 1.37 95% Confidence Interval = 0.98, 1.91 chi square (Yates) p = 0.0756 Child Medical Problems Recurrence Non-recurrence TOTALS Data Present 18 366 384 Data Missigfllnknown 137 2149 2286 TOTALS 1 55 251 5 2670 Percentage of records with data missigmiknown = 86% OR for recurrence for data present/ missing-unknown = 0.77 95% Confidence Interval = 0.27, 1.27 chi square (Yates) p = 0.377 44 example, data concerning removal to foster care was absent in 45% of the records in the cohort, and ‘unknown’ or missing data was noted in 86% of records concerning child ‘other medical problems’. Some States reported little or no data regarding foster care placement and child factors, potentially biasing any association with abuse recurrence. While neither association of absent or missing data is statistically significant in Figure 8, it is interesting to note that child records with missing data regarding foster care were more likely to have recurrence (odds ratio, OR= 1.3 7) and less likely to have recurrence regarding other child medical problems (OR=0.77). It is unclear whether data could be reasonably imputed for missing fields given the high rates of ‘unknown’ or missing data, and future analyses could be directed toward assessing the impact on these results from using various imputation strategies. Limitations This study using NCANDS-DCDC offers a unique opportunity to evaluate child maltreatment reports from large populations of young children in several states but has some important limitations. NCANDS contains information obtained from state CPS agency administrative data systems, and any secondary analysis of NCANDS therefore suffers from the fact that the data were not collected prospectively for the analysis. CPS practices regarding assessing and recording variables in NCANDS has not been standardized, and CPS acceptance and investigation of reports varies across states and may vary over time, leading to several potential differential and non-differential biases. Cases of suspected or confirmed abuse reported to CPS do not necessarily represent the entire population of maltreated children because a significant proportion of maltreatment is 45 never reported to CPS (Sedlak, 1996; Spano & Eckenrode, 1998). Thus, the NCANDS- DCDC sample is not referable to the entire population of physical abuse cases, some of which are not known to CPS. Given the significant health eflects of physical abuse (which are more likely to result in CPS report), this may be less important in physical abuse in young children than in other forms of maltreatment. While large, the DCDC sample is not designed to be nationally representative and analyses cannot be extrapolated to the entire US. population of children reported for suspected maltreatment. Age-specific child populations in states and years for which data were reported to NCANDS amounted to only 21% of the matched age-specific state populations because most states did not report data for all years during the study period. Family and child factors were often not reported by States to NCANDS-DCDC, and many caretaker and child-specific variables were missing or ‘unknown’ in the data. Such ‘missing’ information decreased statistical power and may have lead to differential misclassification of effect. Only nine states contributed data for at least three consecutive years to the DCDC during 1995-1999, and substantial data is missing. The subtype of physical abuse is also not specified in the DCDC. Different types of injuries resulting from physical abuse are postulated to have differing rates of occurrence and recurrence, and the impact of these specific injuries cannot be assessed. Less important is the fact that DCDC data specifically excludes records concerning child fatalities, an important outcome of physical abuse. The absence of fatality data affects our ability to study the most severe forms of abuse but probably has little effect on overall comparisons since fatalities represent less than 5% of physical abuse cases in NCANDS-DCDC and less than 2% of physical abuse overall (Ellis, 1997). 46 CONCLUSIONS Several conclusions can be made about the occurrence and recurrence of physical abuse among infants and young children included in the National Child Abuse and Neglect Data System Detailed Case Data Component (NCANDS-DCDC) for 1995-1999: 1. Overall mean rate of physical abuse was 2.3 per 1,000 children under age three years. Rates of first confirmed physical abuse occurrence vary widely state to state, from 0.5 to 4.8 cases per 1,000 children. This underestimates the true incidence given the fact that NCANDS-DCDC does not include child fatality records. However, fatality incidence is low and only marginally increases the calculated rate. 2. NCANDS-DCDC contains no information about the type or severity of the confirmed physical abuse in maltreatment reports. This prevents further analysis of specific forms of physical abuse such as Shaken Baby Syndrome, although many of the reports concerning infants are presumed to include SBS and other serious head injuries; 3. While most infants and young children with a first confirmed report of physical abuse receive post-investigation services, most do not receive family support services, family preservation services, foster care services or petition to juvenile court; 4. Factors associated with increased first confirmed physical abuse as compared to other forms of confirmed maltreatment include: medical reporting sources, 47 infancy, male gender, Black race, Hispanic ethnicity, child exposure to alcohol or drugs, caretaker problem with alcohol or drugs, and family violence; . Factors associated with decreased first confirmed physical abuse as compared to other forms of confirmed maltreatment include: inadequate housing, receiving public assistance, and urban location; . Overall, the rate of confirmed physical abuse recurrence in the cohort is low (5.8%) before age three years among children first physically abused in infancy, with 0.024 cases per person-year. However, less than 10% of children who were physically abused in infancy could be followed until age three years in the dataset because most states have not consistently provided information to NCANDS-DCDC. Steps should be taken to improve reporting by the States to NCANDS-DCDC to benefit future analyses of recurrence; . Analysis of confirmed physical abuse recurrence in infants and young children in NCANDS-DCDC is problematic because of missing variables that were not reported by the States. In the limited information available, caretaker problems with drugs and emotional disturbance were associated with increased risk of confirmed physical abuse recurrence before age three years. Medical report source, male gender, child medical problems and time to post-investigation services were associated with decreasing time to recurrence in proportional hazard models, and removal to foster care was associated with less risk of recurrence. However, these results should be interpreted with caution given the limitations of this study and the high frequency of rrrissing data in NCANDS-DC DC ; 48 8. Limitations in the data collection and reporting by the States suggest that NC AN DS-DCDC is not representative of all children reported for suspected maltreatment and is not referable to the whole US. population of reported children. Furthermore, given the idiosyncrasies of child maltreatment reporting to CPS, it is suspected that the results of analyses of data in NCANDS-DCDC are not referable to the whole US. population of maltreated children. Further studies should be directed at evaluating the referability of NCANDS-DCDC reports to the entire population of maltreated children, regardless of whether reported to CPS; 9. It would be ideal to replicate these analyses in a large, nationally-representative dataset with additional maternal child variables linked to birth and death certificates, and firrther research can be directed toward comparing the results of this study with: a. The occurrence and recurrence of physical abuse in older children; b. The occurrence and recurrence of other forms of maltreatment in infants and young children; c. The recurrence of other forms of maltreatment after infants and young children are first physically abused; d. The effects of multiple forms of maltreatment co-occurring with physical abuse in infants and young children. 49 APPENDICES Appendix 1 Data Use Agreement Form Appendix 2 List of NCANDS variables Appendix SAS Code 50 NCANDS Terms of Use Agreement This agreement outlines the specific terms under which data obtained from the National Data Archive on Child Abuse and Neglect (hereafter referred to as 'the Archive") may be used. A hard copy of this agreement must be signed and returned to the Archive before data can be released. Data and explanatory documentation provided by the Archive for the National Child Abuse and Neglect Data System, 1995-1999 are hereafter referred to as 'the materials'. I intend to use the materials for the following research purposes: I hereby agree: (1) Purpose.- To use the materiaTs only for the noncommercial statfstical research purposes listed above and teaching. (2) Confidentiality: To act at all times so as to preserve the confidentiality of individuals and institutions recorded in the materials. In particular I agree not to use or attempt to use the materials alone or in combination with any other data to derive information relating specifically to an identified individual or institution nor to claim to have done so, nor to match any datum contained in the materials with any other data or data sets. Additionally, no listings of information from individual records will be published or otherwise released. (3) Acknowledgment: To acknowledge in any publications, whether printed, electronic, or broadcast, based wholly or in part on such materials, the U. S. Department of Health and Human Services, Administration on Children, Youth and Families, Children's Bureau, and the Archive. To declare in any such work that those who carried out the original collection and analysis of the data bear no responsibility for their further analysis or interpretation. To use the following citation in all published works: The data utilized in this publication were made available by the National Data Archive on Child Abuse and Neglect, Cornell University, Ithaca NY; and have been used by permission. Data from the National Child Abuse and Neglect Data System were supplied by state child protective service agencies to the Children's Bureau or the Administration of Children, Youth and Families, US. Department of Health and Human Services Funding for NCANDS was provided by the Children's Bureau, US Department of Health and Human Services. The National Child Abuse and Neglect Data System was implemented by Walter R. McDonald & Associates, Inc. Neither the participating state agencies; Walter R. McDonald & Associates, Inc.; the Children's Bureau, Administration on Children, Youth and Families, US. Department of Health and Human Services, the Archive, Cornell University, or its agents or employees bear any responsibility for the analyses, opinions, or interpretations presented here. (4) Publications: To supply a draft of any manuscript or report based wholly or in part on these data to John Gaudiosi, Children's Bureau, Administration on Children, Youth and Families, 330 C Street SW, Washington, D. C. 20013. I will not submit the manuscript for publication until 60 days following submission to the Children’s Bureau. I authorize the Children's Bureau to submit copies of the manuscript to the state representatives at any time. Finally, I will deposit with the Archive two copies of any published work or report based wholly or in part on these data. 51 (5) Copyright: Not to make any claim to copyright ownership of the materials provided, not to distribute copies of the materials to others, nor to make copies, except as necessary to carry out non-commercial statistical research and teaching. (6) Access to others: To store the data securely, and to restrict access to the data to registered users who have received written permission from the Archive, for the specified purposes. In particular, I will not store the data set on a file server or in any other computational domain where it could be accessed by others. in the case of datasets for which my purpose of use is teaching to give access only to students who have signed the student Terms of Use Agreement and to provide the Archive with copies of the Students' Agreements. (7) Errors: To notify the Archive of any errors discovered in the materials. (8) Liability: To accept that the Archive, the Children's Bureau and its agents, Walter R. McDonald & Associates, Inc., and the states participating in the 1995-1999 NCANDS bear no legal responsibility for the accuracy or comprehensiveness of the materials. To indemnify and hold harmless the NCANDS participating state agencies; Cornell University; the National Data Archive on Child Abuse and Neglect; Walter R. McDonald & Associates, Inc.; the US. Department of Health and Human Services; and any employees or agents of the same against any and all claims for damages, demands, and all other actions and all expenses and costs arising from such demands whatsoever arising from the release of the materials. (9) Completion: To inform the Archive of the completion of the project and, if requested by the Archive, to destroy or return the materials and the copies to the Archive. I understand that violation of any of the above-mentioned conditions will be a breach of this Terms of Use Agreement, will constitute unethical professional practices, and may subject me to legal action under applicable statutes and regulations, including but not limited to injunctive relief. Signature: Date Signed: Name (Print): Organizationzz Position: Only a signed original hard copy of this document can be accepted, not faxes or photocopies. Please mail this completed agreement to: National Data Archive on Child Abuse and Neglect Attention: Data Orders Family Life Development Center Cornell University, Ithaca, NY 14853-4401 52 Appendix 2 NCANDS Variables The ‘J’ and ‘K’ files can be matched and combined using the following twelve fields (with their variable names and ‘YIN’ response type) in both files: Report ID (Y SRPTID): Unique identification label assigned to each report of maltreatment in each state; Child [D (Y SCHID): Unique identification label assigned to a child in each state; State (STATERR): US. State in which report was received and where child lives; Report Date (RPTDT): Date maltreatment was reported; Dieposition Date (RPDISPDT): Date a decision was reached concerning the disposition of alleged maltreatrnent(s); Report Disposition (RPTDISP): Conclusion reached by responsible agency concerning report of alleged maltreatment; Report Sergce (RPTSRC): Role of person who reported alleged maltreatment; thld’s Age (CHAGE): Child’s age in years in date of report; Cm'ld’s Sex (CHSEX): Child’s gender recorded at time of report; Child’s Raee (CI-IRACE): Primary racial group which the child or parent identifies the child as a member; Child’s Hiepanic Ethnicig (CHISP): A Mexican, Puerto Rican, Cuban, Central or South American, or other Spanish ancesu'y, regardless of race (YIN); Pp’eg Victimization Status (CHPRIOR): Existence of previous substantiated or indicated incident of maltreatment (YIN). Two additional fields are included in the ‘1’ files: Notifications (N OTIFS): Mandated or courtesy contacting of other agencies concerning report of child maltreatment; Substentiated or Indicated Maltreagent (F LVICTIM): A computed variable flag of substantiated or indicated maltreatment of child in this report (YIN). The fifty additional fields in the ‘K’ files contain a variety of variables with information about the child, caretaker and services provided after a substantiated or indicated report: Post Investigation Serviees Provided (POSTSERV): CPS or child welfare services provided as a result of needs identified during the investigation (YIN); Date of Best Investigative Services (SERVDATE): Date post investigation services were provided, generally the date of the report disposition; Family Suppprt Servicee Previded (F AMSUP): Family support services were open or panned for the family (YIN); Femily Presematien Services Egovided (F AMPRES): Family preservation services were open or planned for the family (YIN); 53 LesseLCare Servirfis Provided (F OSTERCR): Services or activities associated with 24 hour substitute care for all children placed away from their parents (YIN); WWATE): The month, day and year the child was removed fi'om the care and supervision of his or her parents; Juvenile Court Petition (JUVPET): Legal document filed with the court requesting that the court take action regarding the child’s status (YIN); Court Petition Date (PETDATE): The month, day and year the juvenile court petition was filed; Maltreatment Count (C_MAL_CN): Computed variable representing number of substantiated or indicated maltreatments; Child Vietim of Physical Abuse (FL_CVPHY): Computed flag indicating whether child was substantiated or indicated victim of physical abuse (YIN); Physical Abuse Disposition (C_PHYDIS): Computed variable representing the disposition of maltreatment associated with report of physical abuse; Child Victim of Neglect (F L_CVNEG): Computed flag indicating whether child was substantiated or indicated victim of neglect (YIN); Neglect Disposition (C_NEGDIS): Computed variable representing the disposition of maltreatment associated with report of neglect; Child Victim of Medical Neglect (F L_CVMED): Computed flag indicating whether child was substantiated or indicated victim of medical neglect (Y /N); Medical Neglect Disposition (C_MEDDIS): Computed variable representing the disposition of maltreatment associated with report of medical neglect; Child Victim of Sexual Abuse (F L_CVSEX): Computed flag indicating whether child was substantiated or indicated victim of sexual abuse (YIN); Sexual Abuse Disposition (C_SEXDIS): Computed variable representing the disposition of maltreatment associated with report of sexual abuse; Child Victim of Psychological Abuse (FL_CVPSY): Computed flag indicating whether child was substantiated or indicated victim of psychological abuse (Y /N); Psychological Abuse Dispesition (C_PSYDIS): Computed variable representing the disposition of maltreatment associated with report of psychological abuse; Chile Victim of Other Abuse (FL_CVOTH): Computed flag indicating whether child was substantiated or indicated victim of other abuse (YIN); cher Abuse Diwositiop (C_OTHDIS): Computed variable representing the disposition of maltreatment associated with report of other abuse; Child Victim of Unknom Malgeatment (F L_CVUNK): Computed flag indicating whether child was substantiated or indicated victim of an unknown type of maltreatment (Y /N); annewn Maltreatment Disposition (C_PHYUNK): Computed variable representing the disposition of maltreatment associated with report of an unknown type of maltreatment; 54 Child Problem Recorded (CDPROBFL): A computed variable flag that the child was reported to have a special problem such as drug abuse or a physical handicap (YIN); . Child Problem, Drugs or Alcohol (CDALCDRU): A computed variable flag that the child was reported to have a problem with either drugs or alcohol (Y /N); Child Problem, Mental or Physical (CDPHYSME): A computed variable flag that the child was reported to have a physical or mental problem (YIN); Child fleblem with Alcohol (CDALC): The child was reported to compulsively use or need alcohol (YIN). For young children this generally indicates neonatal alcohol exposure; Child Problem with Drugs (CDDRUG): The child was reported to compulsively use or need narcotics (YIN). For young children this generally indicates neonatal drug exposure; Child Meptally Retarded (CDRTRD): The child was clinically diagnosed to be significantly sub average in general cognitive ability (YIN); Child Emotionally Disturbed (CDEMOTNL): The child was clinically diagnosed to be emotionally disturbed (YIN); Child Visual or Hearing Impaired (CDVISUAL): The child was clinically diagnosed to be visually, hearing or speech impaired (YIN); Child Learning Disabiligy (CDLEARN): The child was reported to have a learning disability (YIN). For young children this generally indicates developmental disability; Child Physieally Disabled (CDPHYS): The child was reported to be physically disabled (YIN); Child Behavior Preblem (CDBEHAV): The child was reported to have a behavior problem in school or in the community (YIN); i1 edical Pr ble (CDMEDICL): The child was reported to have another medical problem not previously identified (YIN); Caretaker Preblem Recorded (F CPROBF L): A variable flag for the caretaker was reported to have a special problem such as drug abuse or a physical handicap (Y /N); Caretaker floblem, Drugs or Alcohol (F CALCDRU): A variable flag for caretaker was reported to have a problem with drugs or alcohol (YIN); Caretaker Mental or Physical Problem (F CPHY SME): A variable flag for caretaker was reported to have a physical or mental problem (YIN); Caretaker Problem with Alcohol (F CALC): Caretaker was reported to compulsively use or need alcohol (YIN); Caretaker Problem m’th Drugs (F CDRUG): Caretaker was reported to compulsively use or need narcotics (YIN); Caretake; Mentally Retarded (F CRTRD): Caretaker was clinically diagnosed to be significantly sub average in general cognitive ability (YIN); Caretakg Emotionally Disturbed (F CEMOTNL): Caretaker was clinically diagnosed to be emotionally disturbed (YIN); Caretake; Visual or Hearing Problem (F CVISUAL): Caretaker was clinically diagnosed to be visually, hearing or speech impaired (Y /N); 55 Caretaker Learning Disability (F CLEARN): Caretaker was reported to have a learning disability (YIN); Caretaker Physically Disabled (F CPHY S): Caretaker was reported to be physically disabled (YIN); Caretake; Cther Medical Problem (F CMEDICL): Caretaker was reported to have another medical problem not previously identified (YIN); Violence Between Caretakers (F CVIOL): Domestic violence between caretakers reported (YIN); Inadequate Housing (F CHOUSE): Family reported to have inadequate housing (Y IN); Family Financial Problems (F CMONEY): Family finances do not meet minimal needs (YIN); ' Emily Receives Public Assistance (F CPUBLIC): Family receives public assistance fi'om AFDC, general Assistance, Medicaid, SSI, Food Stamps, etc. (YIN). 56 Appendix 3 SAS Code /* NCANDS 1995-1999 Analyses */ /* creates library 'indata' and includes ncands formats*/ libname indata 'C:\ncands\big kahuna files'; %include "C:\ncands\ncands formats.sas"; /* combine J and K files and deletes substantiated reports from jkahuna file to create summary file (bigkahun) with add'l fields for confirmed rpts*/ data templ; set indata.jkahuna; if flvictim = 1 then delete; run; data temp; set indata.kkahuna; run; /*appends modified jkahuna file to kkahuna file*/ proc APPEND BASE=temp data=temp1 force; run; /*creates new child id 'stchid' which gets rid of rollups and the first 4 numbers in 'yschid' child id and creates a year field i / data indata.bigkahun; set temp; stchid = substr(yschid,5); year = substr(yschid,1,4); run; /* Creates file (LTTHREE) for all reports for ages LT three, recreates victim flag, sorts, removes rollups, creates and advances report counter * / data indata.LTTHREE; set indata.bigkahun; where chage LT 3; if ((fl_cvphy = 1) or (fl_cvsex = 1) or (fl_cvneg = 1) or (fl_cvmed = 1) or . (fl_cvpsy = 1) or (fl_cvoth = 1) or (fl_cvunk = 1)) then flvictim = 1; else flvictim = 0; run; /*sorts by child-report unique pairs*/ proc sort data=indata.LTTHREE; by stchid rptdt; run; /*gets rid of rollup (duplicate unsub reports in one day)*/ data indata.LTTHREE; set indata.LTTHREE; if ((stchid = lag(stchid)) and (rptdt = lag(rptdt)) and (flvictim = 0)) then delete; 57 run; /*sorts by child-report unique pairs, descending*/ proc sort data=indata.LTTHREE; by stchid descending rptdt; run; /*gets rid of rollup (duplicate reports in one day)*/ data indata.LTTHREE; set indata.LTTHREE; if ((stchid = run; /*re-sorts by child-report unique pairs*/ proc sort data=indata.LTTHREE; by stchid rptdt; run; data indata.LTTHREE; set indata.LTTHREE; by stchid rptdt; retain numsub numunsub rptposit; if (first.stchid and flvictim if (first.stchid and flvictim if ((not first.stchid) and if first.stchid then rptposit . 1; lag(stchid)) and (rptdt = lag(rptdt))) then delete; 1) then numsub =1; 0) then numsub =0; else rptposit = rptposit + 1; numunsub = rptposit - numsub; if last.stchid then lastrpt = run; 1; (flvictim = 1)) then numsub = numsub + /*identifies individual child id’s and counts substantiated reports by type of abuse */ data indata.LTTHREE; set indata.LTTHREE; by stchid rptdt; retain numpa numsex numneg nummed numpsy numoth numunk; if (first.stchid and f1_cvphy = 1) then numpa =1; if (first.stchid and f1_cvsex = 1) then numsex =1; if (first.stchid and f1_cvneg = 1) then numneg =1; if (first.stchid and f1_cvmed = 1) then nummed =1; if (first.stchid and f1_cvpsy = 1) then numpsy =1; if (first.stchid and f1_cvoth = 1) then numoth =1; if (first.stchid and f1_cvunk = 1) then numunk =1; if (first.stchid and f1_cvphy = 0) then numpa =0; if (first.stchid and f1_cvsex = 0) then numsex =0; if (first.stchid and f1_cvneg = 0) then numneg =0; if (first.stchid and f1_cvmed = 0) then nummed =0; if (first.stchid and f1_cvpsy = 0) then numpsy =0; if (first.stchid and f1_cvoth = 0) then numoth =0; if (first.stchid and f1_cvunk = 0) then numunk =0; if (not first.stchid and f1_cvphy = 1) then numpa = numpa + 1; if (not first.stchid and f1_cvsex = 1) then numsex = numsex + 1; if (not first.stchid and f1_cvneg = 1) then numneg = numneg + 1; if (not first.stchid and f1_cvmed = 1) then nummed = nummed + 1; if (not first.stchid and f1_cvpsy = 1) then numpsy = numpsy + 1; if (not first.stchid and f1_cvoth = 1) then numoth = numoth + 1; 58 if (not first.stchid and f1_cvunk = 1) then numunk = numunk + 1; run; /* Adds percent urban as 'pcturban', labels quartiles 'qrturban' top=4, bottom=1 */ data indata.ltthree; set indata.ltthree; if staterr = 'AR' then pcturban = 53.5; if staterr = 'AR' then qrturban = 1; if staterr = 'C0' then pcturban = 82.4; if staterr = 'C0' then qrturban 3; if staterr = 'DE' then pcturban = 73; if staterr = 'DE' then qrturban = 3; if staterr = 'FL' then pcturban = 84.8; if staterr = 'FL' then qrturban 4; if staterr = 'KY' then pcturban = 51.8; if staterr = 'KY' then qrturban = 1; ' if staterr = 'LA' then pcturban = 68.1; if staterr a 'LA' then qrturban a 2; if staterr = 'MA' then pcturban = 84.3; if staterr a 'MA' then qrturban = 4; if staterr = 'M0' then pcturban = 68.7, if staterr = 'M0' then qrturban = 2; if staterr = 'NC' then pcturban = 50.4, if staterr = 'NC' then qrturban = 1; if staterr = 'OK' then pcturban = 67.7, if staterr = 'OK' then qrturban 2; if staterr = 'RI' then pcturban = 86; if staterr = 'RI' then qrturban = 4; if staterr = 'TX' then pcturban = 80.3; if staterr - 'TX' then qrturban = 3; if staterr - 'UT' then pcturban = 87; if staterr - 'UT' then qrturban = 4; if staterr . 'VT' then pcturban - 32.3, if staterr = 'VT' then qrturban 1; if staterr = 'WA' then pcturban 76.4, if staterr = 'WA' then qrturban 3; if staterr - 'WY' then pcturban — 65; if staterr = 'WY' then qrturban - 2; run; /*creates dummy label variableso (O or 1) for all fields by setting all known values >zero to 0 and then resetting variableo = 1 for value of choice */ data indata.LTTHREE; set indata.LTTHREE; if (rptsrc GT 0 and rptsrc LT 99) then rptsrco - 1; if (chsex GT 0 and chsex LT 9) then chsexo - 0; if chsex = 1 then chsexo - 1; if chage = 0 then chageo = 0; if chage > 0 then chageo = 1; if (chrace GT 0 and chrace LT 99) then chraceo = 0; if chrace = 1 then chraceo = 1; if (chisp GT 0 and chisp LT 9) then chispO = 0; if chisp = 1 then chispo a 1; if (postserv GT 0 and postserv LT 9) then postservo = 0; if postserv - 1 then postservo = 1; then rptsrco = 0; if rptsrc = 2 59 if (famsup GT 0 and famsup LT 9) then famsupo = 0; if famsup — 1 then famsupo = 1; . if (fampres GT 0 and fampres LT 9) then fampreso = 0; if fampres = 1 then fampreso = 1; . if (fostercr GT 0 and fostercr LT 9) then fostercro = 0; 1f fostercr = 1 then fostercro = 1; . if (juvpet GT 0 and juvpet LT 9) then juvpeto = 0; if juvpet = 1 then juvpeto = 1; if (cdalc GT 0 and cdalc LT 9) then cdalco = 0; if cdalc = 1 then cdalcO = 1; if (cddrug GT 0 and cddrug LT 9) then cddrugo = 0; if cddrug = 1 then cddrugo a 1; if (cdrtrd GT 0 and cdrtrd LT 9) then cdrtrdo . 0; if cdrtrd = 1 then cdrtrd = 1; if (cdemotnl GT 0 and cdemotnl LT 9) then cdemotnlo = 0; if cdemotnl = 1 then cdemotnlo = 1; if (cdvisual GT 0 and cdvisual LT 9) then cdvisualo = 0; if cdvisual = 1 then cdvisualo = 1; if (cdlearn GT 0 and cdlearn LT 9) then cdlearnO = 0; if cdlearn 1 then cdlearno = 1; if (cdphys GT 0 and cdphys LT 9) then cdphyso - 0; if cdphys = 1 then cdphyso = 1; if (cdbehav GT 0 and cdbehav LT 9) then cdbehavo = 0; if cdbehav 1 then cdbehavo = 1; if (cdmedicl GT 0 and cdmedicl LT 9) then cdmediclo = 0; if cdmedicl = 1 then cdmediclo = 1; if (fcalc GT 0 and fcalc LT 9) then fcalco = 0; if fcalc = 1 then fcalco = 1; if (fcdrug GT 0 and fcdrug LT 9) then fcdrugo = 0; if fcdrug = 1 then fcdrugo = 1; if (fcrtrd GT 0 and fcrtrd LT 9) then fcrtrdo = 0; if fcrtrd = 1 then fcrtrdo a 1; if (fcemotnl GT 0 and fcemotnl LT 9) then fcemotnlo = 0; if fcemotnl= 1 then fcemotnlo . 1; if (fcvisual GT 0 and fcvisual LT 9) then fcvisualo = 0; if fcvisual = 1 then fcvisualo = 1; if (fclearn GT 0 and fclearn LT 9) then fclearno = 0; if fclearn 1 then fclearno - 1; if (fcphys GT 0 and fcphys LT 9) then fcphyso = 0; if fcphys = 1 then fcphyso = 1; if (fcmedicl GT 0 and fcmedicl LT 9) then fcmediclo = 0; if fcmedicl = 1 then fcmediclo = 1; , if (fcviol GT 0 and fcviol LT 9) then fcviolo = 0; if fcviol = 1 then fcviolo = 1; if (fchouse GT 0 and fchouse LT 9) then fchouseo 1 then fchouseo = 1; if (fcmoney GT 0 and fcmoney LT 9) then fcmoneyo 1 then fcmoneyo = 1; if (fcpublic GT 0 and fcpublic LT 9) then fcpublico = 0; if fcpublic = 1 then fcpublico = 1; if qrturban = 1 then qrturbano = 0; if qrturban = 4 then qrturbano = 1; run; 0; if fchouse 0; if fcmoney /* DESCRIPTIVE EPI of children less than 3 years */ proc contents varnum data=indata.LTTHREE; 60 run; proc freq data=indata.LTTHREE; titlel 'Table 1: Crosstabs of state by year, children < 3y, all reports'; TABLES staterr * year; run; proc freq data=indata.LTTHREE; titlel 'State quartiles for percentage urban population, 1990 Census data'; tables staterr * qrturban; run; proc freq data=indata.LTTHREE; titlel 'Table 2: Crosstabs of report dispositions, by year, by type of abuse'; TABLES (C_phydis C_negdis C_meddis C_sexdis C_psydis C_othdis C_unkdis) * year; run; proc freq data=indata.LTTHREE; titlel 'Table 3a: Demographic characteristics, individual children age < 3y'; where lastrpt = 1; tables chage chsex chrace chisp qrturban numsub numunsub rptposit; run; proe means data-indata.ltthree; titlel 'Table 3b: Mean total number reports by type, individual children, age < 3y'; where lastrpt = 1; var numpa numsex numneg nummed numpsy numoth numunk numsub numunsub rptposit ; run; proc univariate data=indata.LTTHREE; where lastrpt BO 1; titlel 'Table 3c: Quartiles for percent urban location, individual children, age < 3y' ; var pcturban; run; proc freq data=indata.LTTHREE; titlel 'Table 4: Crosstabs of variables, PA and total, all confirmed records'; where flvictim = 1; tables (year rptsrc rptsrco chage chageo chsex chsexo chrace chraceo chisp chispo qrturban qrturbano cdalco cddrugo cdrtrdo cdemotnlo cdvisualo cdlearno cdphyso cdbehavo cdmediclo fcalco fcdrugo fcrtrdo fcemotnlo fcvisualo fclearno fcphyso fcmediclo fcviolo fchouseo fcmoneyo fcpublicO postservo famsupo fampreso fostercro juvpeto numsub numunsub rptposit) * f1_cvphy; run; proc freq data=indata.LTTHREE; titlel 'Table 5a: First confirmed PA reports, by state and year, by age'; where (numsub = 1 and f1_cvphy = 1); tables year* staterr * chage ; 61 run; proc freq data=indata.LTTHREE; titlel 'Table 5b: First confirmed PA reports, by state and age'; where (numsub = 1 and f1_cvphy = 1); tables staterr * chage; run; /* COHORT STUDY of recurrence Cohort definition: children less than 3y with first confirmed PA before age 1y, in states and years with 2 or more consecutive years of followup Cohort Analysis: CONTROL: No additional confirmed PA: CASE: 2nd PA in child: sub2pa=1; sub2pa=1 Labels cases and controls in COHORT study and creates new file (PAcohort) */ data indata.PAcohort; set indata.LTTHREE; if ((numpa=1 and f1_cvphy=1 and chage=0)and ((staterr = 'C0' and year = 1995) or (staterr = 'C0' and year = 1996) or (staterr = 'FL' and year = 1995) or (staterr = 'FL' and year = 1996) or (staterr = 'FL' and year = 1997) or (staterr = 'KY' and year = 1997) or (staterr = 'MA' and year = 1995) or (staterr = 'OK' and year = 1996) or (staterr = 'OK' and year = 1997) or (staterr = 'RI' and year = 1997) or (staterr = 'UT' and year = 1995) or (staterr = 'UT' and year = 1996) or (staterr = 'UT' and year = 1997) or (staterr = 'VT' and year = 1995) or (staterr = 'VT' and year = 1996) or (staterr - 'VT' and year = 1997) or (staterr = 'WY' and year = 1997))) then sub2pa=0; if (numpa=2 and fl_cvphy=1) then sub2pa=1; run; /*Removes records not pertaining for first or second confirmed PA reports */ data indata.PAcohort; set indata.PAcohort; where (sub2pa=0 or sub2pa=1); run; proc freq data=indata.PAcohort; titlel 'Table 6a: Cohort status by state and year (sub2pa=0 => PA)‘: where (sub2pa=0); tables staterr * year; run; first /*sorts by child=report date unique pairs*/ proc sort data=indata.PAcohort; 62 by stchid rptdt; run; /*deletes additional records with no first PA before 1 year*/ data indata.PAcohort; set indata.PAcohort; if ((stchid NE lag(stchid)) and (sub2pa =1)) then delete; run; proc freq data=indata.PAcohort; titlel 'Table 6b: Cohort status by state and year (second PA)‘; where (sub2pa=1); tables staterr * year; run; /*dataset descriptives for COHORT study */ proc contents varnum data=indata.PAcohort; titlel 'COHORT dataset descriptives'; run; proc freq data=indata.PAcohort; titlel 'Table 8: Age of child by recurrence status (sub2pa=1 -> repeat PA)‘; tables chage*sub2pa; run; /*sorts by child~report date unique pairs*/ proc sort data=indata.PAcohort; by stchid rptdt; run; /* copies fields from first report to second record 'f' variables*/ data indata.PAcohort; set indata.PAcohort; fpostservo = lag(postservO); ffamsupo = lag(famsupO); ffampreso = lag(fampresO); ffostercro = lag(fostercrO); fjuvpeto = lag(juvpeto); frptdt = lag(rptdt): fservdate = lag(servdate): frmvdate = lag(rmvdate); fpetdate = lag(petdate); fchage = lag(chage); fyear = lag(year); run; /* calculates time intervals in weeks between first and second reports for report, service, removals and court petition dates */ data indata.PAcohort; set indata.PAcohort; time2 = (rptdt - frptdt); timeserv = (rptdt - fservdate); timermv = (rptdt - frmvdate); timepet - (rptdt - fpetdate); run; data indata.PAcohort; set indata.PAcohort; if timeserv < 0 then timeserv = 0; 63 I C if timermv < 0 then timermv - if timepet < 0 then timepet — run; . proc freq data=indata.PAcohort; titlel 'Crosstab of first year of PA (fyear) by year of recurrence of PA'; where sub2pa =1; tables fyear*year; run; I O /*Transfers info from second report to first in children with multiple reports */ /*sorts by child-ID and position, descending*/ proc sort data=indata.PAcohort out=indata.PAcohort; by stchid descending rptdt descending sub2pa ; run; /* copies fields from second report to first record '5' variables*/ data indata.PAcohort; set indata.PAcohort; spostservo = lag(postservO); sfamsupo = lag(famsupO); sfampreso = lag(fampresO); sfostercro = lag(fostercrO); sjuvpeto = lag(juvpetO); srptdt = lag(rptdt); sservdate = lag(servdate); srmvdate = lag(rmvdate); spetdate = lag(petdate); schage = lag(chage); syear = lag(year); stime2 = lag(timeZ); stimeserv . lag(timeserv); stimermv = lag(timermv); stimepet = lag(timepet); run; /* creates variable labeling PA recurrence cases (PArecur=l) vs cases with no recurrence (PArecur=0) and back-labels first record */ data indata.PAcohort; set indata.PAcohort; retain PArecur; if sub2pa=1 then PArecur=1; else PArecur = 0; if ((stchid = lag(stchid)) and (lag(PArecur)=1)) then PArecur=1; run; /* Creates new dataset (PArecur) which selects first records only */ data indata.PArecur; set indata.PAcohort; where sub2pa=0; run; /*dataset descriptives for recurrence records */ proc contents varnum data=indata.PArecur; titlel 'PArecur dataset descriptives'; run; 64 proc freq data=indata.PArecur; titlel 'Age of child by recurrence status (PArecur=1 : repeat PA)‘; tables chage*PArecur; run; /* Comparison of recurrence vs non-PA recurrence cases */ proc freq data=indata.PArecur; titlel 'Table 7: Crosstabs of variables by PA recurrence (PArecur=1) in COHORT'; tables (qrturban year rptdisp rptsrc chage chsex chrace chisp chprior cdalc cddrug cdrtrd cdemotnl cdvisual cdlearn cdphys cdbehav cdmedicl fcalc fcdrug fcrtrd fcemotnl fcvisual fclearn fcphys fcmedicl fcviol fchouse fcmoney fcpublic C_MAL_CN FL_CVPHY C_PHYDIS FL_CVNEG C_NEGDIS FL_CVMED C_MEDDIS FL_CVSEX c_saxnrs FL_CVPSY C_PSYDIS FL_CVOTH C_oranrs FL_CVUNK C_UNKDIS CDPROBFL CDALCDRU CDPHYSME FCPROBFL FCALCDRU FCPHYSME postserv famsup fampres fostercr juvpet numunsub) * PArecur; run; proc freq data=indata.PArecur; titlel 'Table 9a: 2x2 tables and stats, variables for recurrence comparisons in COHORT'; tables (rptsrco qrturbano chageo chsexo chraceo chispo cdalcO cddrugo cdrtrdo cdemotnlo cdvisualo cdlearno cdphyso cdbehavo cdmediclo fcalcO fcdrugo fcrtrdo fcemotnlo fcvisualo fclearno fcphyso fcmediclo fcviolo fchouseo fcmoneyo fcpublico postservo famsupo fampreso fostercro juvpetO) * PArecur /cmh; run; /* Analysis of time variables */ Proc ttest data=indata.PArecur; titlel 'Table 9b: Comparison of variables by recurrence (PArecur)'; var numunsub stime2 stimeserv stimermv stimepet; class PArecur; run; Proc means data=indata.PArecur; titlel 'Table 9c: Comparison of variables by recurrence (PArecur)'; var numunsub stime2 stimeserv stimermv stimepet; class PArecur; run; proc phreg data=indata.PArecur; titlel 'Table 11a: Cox proportional hazards model of recurrence, initial model'; model stime2*PArecur(0)= rptsrco pcturban chsexo chraceo cdvisualo cdlearnO cdbehavo cdmediclo 65 fcalco fcviolo fchouseo fcpublico postservo fampreso fostercro numunsub stimeserv stimermv; run; proc phreg data=indata.PArecur; titlel 'Table 11b: Cox proportional hazards model final adjusted model'; model stime2*PArecur(0)= rptsrco chsexo chraceo cdmediclo fostercro stimeserv; run; 66 of recurrence, REFERENCES Administration on Children, Youth and Families, US. Department of Health and Human Services. (1997). Child Maltreatment 1295: Reports from the States to the National Child Abuse a_ngi Neglect Data Sysiem. Washington, DC: US. Government Printing Office. Administration on Children, Youth and Families, US. Department of Health and Human Services. (1998). Child Maltreatment 1996: Reports from the States to the National Child Abuse and Neglect Data System. Washington, DC: US Government Printing Office. Administration on Children, Youth and Families, US. Department of Health and Human Services. (1999). Child Maltreatment 1997: Reports fi'om the Slates tn the National Cbild Abme and Neglect Data System. Washington, DC: US. Government Printing Office. Administration on Children, Youth and Families, US. Department of Health and Human Services. (2000). hild Maltreatment 1998: th tatest e National Abuse end Neglect Data System. Washington, DC: US. Government Printing Office. Administration on Children, Youth and Families, US. Department of Health and Human Services. (2001). hild Ma en : orts from the tates to the Nat'onal Abuse and Neglect Data System. Washington, DC: US. Government Printing Office. Alexander R., Crabbe L., Sato Y., Smith, W. and Bennett T. (1990). Serial abuse in children who are shaken. Men'can Journal of Diseases of Childhood, 144,58-60. Allison, PD. (1995). Survival Analysis Csing the SQ System: A Bmctical Cuide. Cary, NC: SAS Institute. Ards, S. and Harrell, A. (1993). Reporting of child maltreatment: A secondary analysis of the national incidence surveys. Child Abuse and Neglect, 17, 337-344. Ards, S., Chung, C. and Myers S.L. (2001). Sample selection bias and racial differences in child abuse reporting: Once again (letter). Child Abuse mg Negieei, 25, 7-12. Baird, C., Ereth, J., Wagner, D. (1999). Research-Based Risk Assessment Adding Eguiiy 19 CBS Deeision Making. Madison, WI: Children’s Research Center, 15-18. Benedict, M.I., Zuravin, S., Brandt, D.and Abbey, H. (1994). Types and frequency of child maltreatment by family foster care providers in an urban population. Cm'ld Abuse and Neglect, 7, 577-85. Bertolli, J ., Morgenstem H. and Sorenson, SB. (1995). Estimating the occurrence of child maltreatment and risk factor effects: Benefits of a mixed design strategy in epidemiologic research. Child Abuse and Neglmi, 19, 1007-1016. 67 Brayden, R.M., Altemeier, W.A., Dietrich, M.S., Tucker, D.D., Christensen, M.J., . McLaughlin, F.J. and Sherrod, KB. (1993). A prospective study of secondary prevention of child maltreatment. ieumal of Pegiatrics, 122, 511-6. Brenner, R.A., Overpeck, M.D., Trumble, A.C., DesSirnonian, R., Berendes, H. (1999). Deaths attributable to injuries in infants, 1983-1991. Eediatrics 103, 968-974. Brown, J ., Cohen, R, Johnson, J.G. and Salzinger, S. (1998). A longitudinal analysis of risk factors for child maltreatment: Findings of a 17-year prospective study of officially recorded and self-reported child abuse and neglect. Cbild Abuse and Neglect, 22, 1065- 1078. Bureau of Justice Statistics, US. Department of Justice. (2000). Criminal Victimization, 1299: Changes 1998-29 with Tgends 1993-99. Washington, DC: NCJ 182734, 1-3. Cappelleri, J .C., Eckenrode, J. and Powers, J .L. (1993). The epidemiology of child abuse: findings fi'om the Second National Incidence and Prevalence Study of Child Abuse and Neglect. American Journal bf Public Health, 83, 1622-4. Caffey, J. (19446). Multiple fractures of the long bones of infants suffering from chronic subdural hematoma. American Jenmal ef Roentgenology, 56, 163-173. Chafl'rn, M., Bonner, BL. and Hill, RF . (2001). Family preservation and family support programs: child maltreatment outcomes across client risk levels and program types. Child Abuse and Neglect, 25, 1269-1289. Child Abuse Prevention and Treatment Act (CAPTA) of 1974. US. Public Law 100—294 as amended, 1988. Connelly, CD. and Straus, MA. (1992). Mother's age and risk for physical abuse. Child Abuse and Neglect, 16, 709-18. Crosse, S.B., Kaye, E. & Ratnofsky, AC. (1993). A Report on the Maltreatmen; 91 Chi ingen Wiin Disabilin'es, Washington, DC: National Center of Child Abuse and Neglect, Department of Human Services. DePanfilis, D. & Zuravin, SJ. (1998). Rates, patterns and frequency of child maltreatment recurrences among families known to CPS. Cbiid Maltreannent, 3, 27-42. DePanfilis D. & Zuravin, SJ. (1999). Predicting child maltreatment recurrences during treatment. Cbild Abnse and Neglect, 23, 729-43. DePanfilis D. & Zuravin, SJ. (2001). The effect of services on the recurrence of child maltreatment. Child Abuse and Neglect, 26, 187-205. DiScala, C., Sege, R., Li, G. and Reece, RM. (2000). Child abuse and unintentional 68 injuries: a lO-year retrospective. Archives of Pediatric and Adolescent Medicine, 154, 16- 22. Drake, B. and Pandey, S. (1996). Understanding the relationship between neighborhood poverty and specific types of child maltreatment. Cm'ld Abuse and Negleet, 20, 1003-18. Eckenrode, J ., Ganzel, B., Henderson, C.R., Smith, E., Olds, D.L., Powers, J ., Cole, R. and Kitzrnan, H. (2000). Preventing child abuse and neglect with a program of nurse home visitation. Jonmal oi the American Medical Association, 284, 1385-1391. Ellis, PS. (1997). The pathology of fatal child abuse. Ratholbgy, 29(2), 113-21. F erleger, N., Glenwick, D.S., Gaines, R.R.W. and Green, AH. (1988). Identifying correlates of reabuse in maltreating parents. Child Abuse and Neglecl, 12, 41-49. Fluke, J .D., Yuan, Y.Y.T. and Edwards, M. (1999). Recurrence of maltreatment: An application of the national child abuse and neglect data system (N CANDS). Cnilgl Abuse mg Neglect, 23, 633-650. Frongillo, EA. and Rowe, EM. (1999). Challenges and solutions in using and analyzing longitudinal growth data. In Johnston, F.E., Zemel, B. and Eveleth, P.B. Humsn Qewtb in Context. London, UK: Smith-Gordon. Gaudiosi, J ., Yuan, Y.Y., Fluke, J. and Johnson, J. (2001). Reports, victims, maltreatrnents: What is the difference? An introduction to understanding child maltreatment statistics. Presented at the 13th National Conference on Child Abuse and Neglect. April 27, 2001. Gillenwater, J .M., Quan, L. and F eldman, K.W. (1996). Inflicted submersion in childhood. Arcbives ol Pedietric and Adolescen; Medicine, 150, 298-303. Goldstein, B., Kelly, M.M., Bruton, D. and Cox, C. (1993). Inflicted versus accidental head injury in critically injured children. Critical Care Medieine, 21, 1328-32. Heneghan AM, Horwitz SM, Leventhal JM. Evaluating intensive family preservation programs: A methodological review. Pediatrics. 1996, 97, 535-542. Hermes, H., Kini, N. and Palusci, V.J. (2001). The epidemiology, clinical characteristics and public health implications of Shaken Baby Syndrome. lowal oi Aggrfision, Maltreatment and Trauma, 5(1), 19-40. Herrnan-Giddens, M.E., Brown, G., Verbiest, S., et a1. (1999). Underascertainment of child abuse mortality in the United States. oum f the erican ical ocia 'on, 282, 463-467. 69 Hicks, RA. and Gaughan, DC. (1995). Understanding fatal child abuse. Child Abuse and Neglect, 19, 855-63. Horwitz, S.M., Balestracci, K.M.B., Simms, MD. (2001). Foster Care Placement Improves Functioning. ficbives of Rediatric and Adolescent Medicine, 155, 1255-1260. Jones, L.M., Finkelhor, D. and Kopiec, K. (2001). Why is sexual abuse declining: A survey of state child protection administrators. Child Abuse and Neglect, 25 , 1139-1158. Kasirn, M.S., Cheah, I. and Shafie, HM. (1995). Childhood deaths from physical abuse. Child Abuse and Neglect, 19(7), 847-54. Kelsey, J .L., Whittemore, A.S., Evans, A.S. & Thompson, W.D. (1996). Methods in Cbservational Epidemiology, 2"" Ed. New York, N.Y.: Oxford University Press, 330335. Kempe, C.H., Silverrnan, F.N., Steele, B.F., Droegemueller, W. and Silver, H.K.(l962). The battered child. loumal of the Ameriean Medieal Association, 181, 17-24. Kleinbaum, DC. (1994). Logistic Regiession: A Self-Leaming Text. New York, NY: Springer-Verlag. Kotch, J .B., Browne, D.C., Dufort, V., Winsor, J. and Catellier, D. (1999). Predicting child maltreatment in the first 4 years of life fi'om characteristics assessed in the neonatal Period- Wig. 23. 22-32- Leventhal, J .M, Pew, M.C., Berg, AT. and Garber, RB. (1996). Use of services by children who were identified during the postpartum period as being at high risk of child abuse or neglect. Pediatrics 97, 331-335. Levy, H.B., Markovic, J., Chaudhry, U., Ahart, S. and Torres, H. (1995). Reabuse rates in a sample of children followed for five years after discharge fi'om a child abuse inpatient assessment program. Child Abuse and Neglect, 19, 1363-1377. Marshall, W.N. (1997). Hospitalization of abused and neglected children. Archives of Pediatric and Adeleseenl Mfiieine, 151, 273-275. National Data Archive on Child Abuse and Neglect. (2000). National Child Abuse md Ne lect Data S stem ANDS : Detailed ase ata om one 995- 997. NDACAN Dataset #75. Ithaca, NY: Cornell University Family Life Development Center. Oates, R.K. and Bross, DC. (1995). What have we learned about treating child physical abuse? A literature review of the last decade. Child Abuse and Neglecl, 19, 463-475. Reece, RM. and Sege, R. (2000). Childhood head injuries: Accidental or inflicted? Archives of Pediatrie md Adelesceni Medicine, 154, 11-15. 70 Rivara, F .P. (1985). Physical abuse in children under two: A study of therapeutic outcomes. Child Abuse and Neglecl, 9(1), 81-7. Rosner, B. (1995). Fundamentals of Biostatistics 4‘” Ed. Behnont, CA: Wadsworth Publishing Co., 365-370, 469-483. Sedlak, A.J and Broadhurst, DD. (1996). The Third National Incidence Study of Child Abuse mid Neglect (NlS-3 ). Washington, DC: US Dept of Health and Human Services. Sedlak, A..,J Bruce, C. and Schultz, D..J (2001). Sample selection bias. in NIS- 3 (letter). Child Abuse snd Neglect t,25 l -.5 Sibert, J .R., Payne, E.H., Kemp, A.M., et al. (2002). The incidence of severe physical abuse in Wales. Cbild Abuse and Neglect, 26, 267-276. Sidebothan, P., Golding, J ., and the ALSPAC Study Team. (2001). Child maltreatment in the “Children of the Nineties:” A longitudinal study of parental risk factors. Cbild Abuse and Neglect, 25, 1177-1200. Spano, S. and Eckenrode, J. (1998). Administrative data in child welfare: Research strategies and prospects for the future. Presented at the National Data Archive on Child Abuse and Neglect, Cornell University. June 26-27, 1998. Strauss, M.A. & Moore, D.W. “Differences among States in Child Abuse Rates and Programs” in Besharov, D.J., Ed. (1990). Fmiily Violence: Researcb and Public Eolicy Issues. Washington, D.D.: The AEI Press, pp. 150-163. Thomas, D.E., Leventhal, J .M. and F riedlaender, E. (2001). Referrals to a hospital-based child abuse committee: a comparison of the 19603 and 1990s. Cbild Abnse and Neglect, 25, 203-13. Thompson, K.M., Wonderlich, S.A., Crosby, R.D., Ammerrnan, F .F ., Brownfield D. (2001). An assessment of the recidivism rates of substantiated and unsubstantiated maltreatment cases. Child Abuse end Neglect, 25, 1207-1218. Thyen, U., Leventhal, J .M., Yazdgerdi, SR. and Perrin, J .M. (1997). Concerns about child maltreatment in hospitalized children. Child Abdse and Neglect, 21(2), 187-98. Wissow, LS. and Wilson, M.H. (1988). The use of consumer injury registry data to evaluate physical abuse. Child Abuse end Neglect, 12, 25-31. Wolfner, G.D. and Gelles, R.J. (1993). A profile of violence toward children: a national StudY- Winch 17(2). 197-212- 71 US. Census Bureau. (2000). 1990 to 1999 Annual Time Series of State Population Estimates B in 1e Year f A e and ex. Http://www.census.gov/popluation/www/estimates/st-99-10.htrnl. Accessed 12/27/2001. US Census Bureau. (2000). Statistical Abstract of the llnited States, Table 37 ‘Urban and Rural Population, 1960-1990, and by State, 1990’. Washington, DC; US. Government Printing Office. 72 IIIIIIIIIIIIIIIIIIIII inu)[trillirjllilijijul