A COMPARISON OF OUTCOMES FOR ACUTE STROKE PATIENTS HOSPITALIZED IN MICHIGAN, USA AND ONTARIO, CANADA USING HOSPITAL DISCHARGE DATA (2010 - 2012) By Joshua Orlando David Cerasuolo A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Epidemiology Master of Science 2015 ABSTRACT A COMPARISON OF OUTCOMES FOR ACUTE STROKE PATIENTS HOSPITALIZED IN MICHIGAN, USA AND ONTARIO, CANADA USING HOSPITAL DISCHARGE DATA (2010 - 2012) By Joshua Orlando David Cerasuolo P revious studies have compared cardiovascular disease outcomes between Canad a and the United States; however, there are limited data for stroke. This thesis compares hospital discharge data to compare mortality and readmission rate s after stroke between Michigan and Ontario . Eligible acute stroke patients (both ischemic and hemorrhagic) were hospitalized between January 1, 2010 and December 31, 2012. A total of 47,364 and 35,648 patients were included in Michigan and Ontario, respectively. To ensure comparabi lity of patient risk profiles between Michigan and Ontario, we applied a Michigan risk - adjustment model to Ontario patients to generate directly standardized outcome rates for Ontario. Results indicate that Ontario stroke patient population was older (mean age: 72.4 vs. 6 9.5 years) , had longer hospital length of stay ( mean length of stay : 12.5 vs . 5.4 days ), and experienced higher frequencies of acute ischemic heart disease and cancer, whereas the Michig an stroke patient population exhibited higher frequencies of chronic ischemic heart disease, diabetes, heart failure, and renal failure. Ontario had a higher risk - standardized in - hospital mortality rate (13.3%) compared to Michigan ( 7.6% ) ; h owever, risk - standardized 30 - day readmission rates we re similar ( 5.3% vs. 4.5%) . Other performance metrics , such as 30 - day mortality , are required to make valid comparisons regarding mortality , but was not possible with the datasets used in this study. iii To my family and friends for their love and support. iv ACKNOWLEDGMENTS First and foremost, I would like to thank my thesis committee chair, Dr. Mat Reeves , for being patient with me throughout the thesis process. I am greatly appreciative of his significant contribution to this work, including the endless hours he spent meeting with me, as well as reviewing this document . I would like to thank the entiret y of my thesis committee (Drs. Reeves, Luo, and Barondess) for their valued contribution and guidance throughout this process. T his work would not be possible without the assistance of Dr s. Moira Kapral, Ruth Hall, and Jiming Fang, as well as Limei Zhou, Sean Leonard, Mike Thompson and Adrienne Nickles. These individuals have assist ed me in va rious capacities throughout my thesis work, including data acquisition, statistical guidance, workspace , and supervision. Their time spent on this project is and will always be greatly appreciated. Lastly, my family and friends have provided support during the many hours I spent compl eting this thesis, and they deserve every bit of gratitude. v TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ......................... vi LIST OF FIGURES ................................ ................................ ................................ ...................... vii CHAPTER 1: BACKGROUND ................................ ................................ ................................ ...... 1 Significance ................................ ................................ ................................ .......................... 1 American and Canadian Health Care Systems ................................ ................................ .... 1 Stroke Burden and Delivery of Care ................................ ................................ .................... 4 Canada/US Comparison: Cardiovascular - Related Mortality and Risk Factors .................. 7 Canada/US Comparison: Cardiovascular Outcomes and Processes of Care ...................... 9 Study Aims ................................ ................................ ................................ ........................ 11 CHAPTER 2: METHODS ................................ ................................ ................................ ............. 12 Hospital Discharge Databases ................................ ................................ ............................ 12 Case Ascertainment ................................ ................................ ................................ ........... 13 Study Exclusions ................................ ................................ ................................ ................ 1 5 Outcome Definitions ................................ ................................ ................................ .......... 17 Statistical Analys e s: Descriptive ................................ ................................ ........................ 18 Rationale for Risk Adjustment ................................ ................................ ........................... 18 Statistical Analys e s: Risk - Adjustment Model s ................................ ................................ .. 19 Aim 1: Hospital - Level Risk Profiling Comparison ................................ ........................... 20 Aim 2: Patient - Level Risk Standardization ................................ ................................ ....... 22 Ethics Approval ................................ ................................ ................................ ................. 24 CHAPTER 3: RESULTS ................................ ................................ ................................ ............... 25 Descriptive Characteristics ................................ ................................ ................................ 25 Length of Stay ................................ ................................ ................................ .................... 2 6 Aim 1: Hospital - Level Risk Profiling Comparison ................................ ........................... 26 Aim 2: Patient - Level Risk Standardization ................................ ................................ ....... 30 CHAPTER 4: DISCUSSION ................................ ................................ ................................ ......... 33 APPENDICES ................................ ................................ ................................ ............................... 42 Appendix A: Tables ................................ ................................ ................................ ........... 43 Appendix B: Figures ................................ ................................ ................................ .......... 55 Appendix C: Supplemental Tables ................................ ................................ .................... 64 Appendix D: IRB Approval Letter ................................ ................................ .................... 68 REFERENCES ................................ ................................ ................................ .............................. 70 vi LIST OF TABLES Table 1 : Comparison of demographic, geographic, and health care characteristics between Michigan, USA and Ontario, Canada (2000 - 2013) ................................ ................................ ....... 4 4 Table 2 : Regional comparison of patient - and hospital - level characteristics in the Michigan and Ontario study samples. ................................ ................................ ................................ .................. 45 Table 3 : Regional comparison of adjusted odds ratios from the hierarchical logistic regression model used to profile hospital performance for in - hospital mortality . ................................ .......... 46 Table 4 : Regional comparison of adjusted odds ratios from the hierarchical logist ic regression model used to profile hospital performance for 30 - day readmission . ................................ ........... 47 Table 5 : Fit statistics for all models performed in aims 1 and 2. ................................ ................. 48 Table 6 : List of predicted over expected ratios (in ascending sequence) used to quantify hospital performance for in - hospital mortality. ................................ ................................ .......................... 49 Table 7 : List of predicted over expected ratios (in ascending sequence) used to quantify hospital performance for 30 - day readmission. ................................ ................................ ........................... 50 Table 8 : Comparison of adjusted odds ratios , for the Michigan sample, generated from the hierarchical logistic regression (HLM) and generalized estimating equations (GEE) modelling methods for in - hospital mortality. ................................ ................................ ................................ .. 51 Table 9 : Comparison of adjusted odds ratios , for the Michigan sample, generated from the hierarchical logistic regression (HLM) and generalized estimating equations (GEE) modelling methods for 30 - day readmission . ................................ ................................ ................................ ... 52 Table 1 0 : Direct standardization procedure to produce comparable in - hospital mortal ity rates between Michigan, USA and Ontario, Canada ................................ ................................ .............. 53 Table 1 1 : Direct standardization procedure to produce comparable 30 - day readmission rates between Michigan, USA and Ontario, Canada ................................ ................................ .............. 54 Table S 1 : Description of codes used in Michigan (ICD - 9) and Ontario (ICD - 10) to identify strokes ................................ ................................ ................................ ................................ ............ 65 Table S2 : List of diagnosis codes used to identify past medical history in the respective discharge databases in Michigan and Ontario ................................ ................................ ............... 66 Table S3 : Description of ICD - 9 procedure codes used to identify elective readmissions in the MIDB ................................ ................................ ................................ ................................ ............. 67 vii LIST OF FIGURES Figure 1 : Exclusions applied to the Michigan and Ontario study samples (in the order shown) . 56 Figure 2 : Outline of the procedure used to account for differences in risk profile and distribution between the Michigan and Ontario patient samples . ................................ ................................ ..... 57 Figure 3 : Length of stay distributions for the Michigan and Ontario patient samples ................. 58 Figure 4 : Discharge patterns for hospitalized stroke patients during the first 7 days following hospital admission in Michigan, USA and Ontario, Canada. . . ................................ ...................... 59 Figure 5 : Distribution of predicted over expected ratios of in - hospital mortality performance for Michigan and Ontario hospitals, stratified by stroke center certification . ................................ ..... 60 Figure 6 : Distribution of predicted over expected ratios of 30 - day readmission performance for Michigan and Ontario hospitals, stratified by stroke center certification . ................................ ..... 61 Figure 7 : Rank correlation between observed and risk - adjusted in - hospital mortality rates for Michigan and O ntario hospitals (2010 - 2012) . ................................ ................................ ............... 62 Figure 8 : Rank correlation between observed and risk - adjusted 30 - day readmission rates for Michigan and Ontario hospitals (2010 - 2012) . ................................ ................................ ............... 63 1 CHAPTER 1: BACKGROUND Significance E very country seeks to provide efficient, effective, and equit able health care , yet health care systems vary across countries to reflect the unique ness of its population, economic status, and disease burden. The substantial variation in health systems between countries provides the opportunity to compare health outcomes across systems that differ in terms of organization, quality of care, and economics. Specific ally , in stroke research, comparing two contrastin g health systems can aid in understanding how differences in organization and delivery of stroke care may be reflected in differences in stroke outcomes. Disparities in stroke outcome s across different heal th systems may also aid in the identification of strengths and weaknesses of specific systems that influen ce quality of care and outcomes. Canada and the United States are bordering developed nations with significant cultural, infrastructural, and economic similarities , but their fundamentally different health care systems offer an intriguing opportunity to conduct cross boarder comparison s of health outcomes . American and Canadian Health Care Systems There are notable differences between the American and Canadian health care systems , particularly related to health care organization, specialty medical care allocation, and payer source. The American health care system consists of public and p rivate stakeholders that market privatized health insurance to gove rnment al and corporate entities , as well as directly to individuals, with minimal government al oversight . 1 In 1965, the US government legislated the creation of the Medicare and Medicaid programs, both being government - provided (public) health care coverage for the elderly and citizens on social assistance , respectively. 2 Since their 2 implementation in the mid - 1960s, both Medicare and Medicaid have undergone changes to become more inclusive programs. Medicare was initially for the elderly population ( at least 65 years of age), but as of 1973, additionally included persons with certain disabilities. Likewise for Medicaid, its initial purpose was to provide coverage for persons on welfare assistance, but now also includes adults with qualifying disabilit ies, pregnant women and children of families in poverty. Since state governments regulate their own Medicaid programs, there is statewide variation in Medicaid eligibility. In 2012, Medicaid covered 35% of all children, 41% of pregnant women, 40% of parent s below the federal poverty line (FPL) , and 45% of adults younger than 65 years below the FPL . 3 Collectively in 2013, the Medicare and Medicaid programs accounted for 35 % of all US national health expenditures (NHE); $585.7 billion USD (20% of NHE) and $44 9.4 billion USD (15% of NHE), respecitvely. 4 The largest health care reform since the establishment of Medicare and Medicaid was the Patient Protection and Affordable Care Act (ACA) , which was implemented in the United States in 2011. 1 The ACA requires all US residents to have some form of health coverage. Before the ACA was legislated, a major difference between the American and Canadian health systems was the proportion of uninsured or under - insured residents, because Canada possesses a universal healt h care system that guarantees health care to all its citizens 5 . In the US prior to the implementation of key components of the ACA, 18% of US residents under the age of 65 lacked any form of health insurance. 1 More recent data from the US Department of Hea lth & Human Services show that the ACA now include s 11. 4 million enrollees as of early 2015 6 , and the proportion of uninsured US residents younger than 65 year of age has fallen to 1 5 . 3 % 7 . Specifically for Medicaid, 2015 enrollment in the US has seen a 20% increase since before the implementation of the ACA. 8 3 In contrast , health care in Canada is governmentally mandated to provide universal access to essential hospital and physician services for all eligible Canadian residents. 5 If a service is deemed medically necessary, it is considered an insured service under the Canada Health Act . 9 Seventy percent of the C anadian health system is funded by tax revenue from the federal, provincial, and territorial governments. 5 As federally mandated in the Canada Health Act, the individual provincial governments are responsible for passing legislation that ensures universal health coverage for necessary medical serv ices . Furthermore, coverage must be public , inclusive of the broad scope of a ll essential medical services, and be universally accessible to all eligible residents. In addition to the mandate set forth for the provinces to follow , t he federal government oversees health surveillance, public health initiatives, and safety, while also administering federal revenue to provinces for use on health care - related expenses. The provincial governments also set physician remuneration rates . Prior research has shown that t he substantial organizational differences between Canada and the United States , result in substantial differences in services utilization between these two systems. 10 Authors of a 2013 study of nationally representative surveys in the US and Canada showed that Canadians are more likely to visit a specialist, have a medical doc tor, and stay overnight in hospital. Specifically among the poor and less educated, Canadians have demonstrated the higher likelihood to utilize health care services , such as specialist and general physician visits, than Americans of the same class . 10 Amon g the elderly in both countries, s , but less procedures than Americans (i.e. major orthopedic procedures) ; the authors attributed the more liberal approach to evaluatio n to the lower fee for service in Canada , but the budgetary restraints of a universal health system cause lower procedure utilization in comparison to the multi - payer 4 system of the United States . 11 Diagnostic testing use was 32% higher in the United States compared to Canada, particularly in the use of CT and MRI imaging. 12 Contrary to the above health care utilization differences; emergency department use was found to be very similar between Canada and the US. 1 3 T he US spends more on health care than any other developed nation . 1 I n 20 11 , the US spent $2.7 trillion USD on health care that accounts for 17.9% of US gross domestic product (GDP) 14 . By 2019, this figure is projected to reach $4.5 trillion USD representing 19.3% of the US GDP. 15 In 201 2 , Canada allocated $ 205.4 billion CDN or 11.3% of their GDP to health care spending. 16 On a per capita basis, Canada spends substantially less, $4,445 USD compared to $8,233 USD the US spends. 1 Specifically re lated to stroke, annual direct and indirect costs totaled $33.6 billion in the US 17 , and $3.6 billion in Canada 18 . Stroke Burden and Delivery of Care Even though the organization of health care is different between the two countries, the burden of stroke is very similar. Stroke is the 3 rd leading cause of death in Canada, accounting for 5.5% of all Canadian deaths in 2011. 19 Similarly in the United States, stroke is the 4 th leading cause of death a ccounting for 5.1% of all deaths in 2011 . 20 In 2011, the age - adjusted stroke mortality rates for Canada and the United States were 24.8 and 37.9, respectively per 100,000 population. 17,21 Stroke is the leading cause of disability in both Canada 2 2 and the United States 2 3 . Stroke burden was discovered to be different i n t erms of stroke hospitalizations; t he rate in the US was 31.8 per 10,000 population in 2009 2 4 , compared to only 12 per 10,000 population in Canada in 2005 2 5 . To counter the burden of stroke in Canada and the US, both countries have implemented their own organized stroke health systems to improve the delivery of stroke care. 2 6 5 Stroke care in the United States and Canada has undergone substantial changes in recent decades as new systems of stroke care have been developed in response to the availability of new acute treatments for stroke . In the US s ince 2000 , the Brain Attack Coalition of the American Stroke Association (ASA) has laid the groundwork for necessary improvements to stroke c are that are required to lessen the stroke burden on mortality and morbidity. 27 , 2 8 Since these recommendations were released , the ASA has implement ed stroke quality improvement initiatives 29 , 30 , and promoting the delivery of specialized stroke care service s 3 1 . Regional stroke systems of care in the United States have been shown to increase access to stroke - specific care and services. 26 , 2 7 In conjunction with The Joint Commission, the A SA created a disease - specific primary stroke center (PSC) certification program 2 8 , to recognize centers that have more intensive stroke - specific procedural capabilities, dedicated stroke units, and actively participate in stroke research 3 2 . The Brain Attack Coalition made recommendations on a two - tier system of stroke care in the United States: primary stroke centers (PSC) and comprehensive strokes centers (CSC) . 28 , 3 3 Primary stroke centers are designed to provide acute treatments , such as tissue plasminogen activator ( tPA ), and stroke unit care to all acute stroke patients as necessary 2 8 , whereas a CSC is designed to provide care to the most severe and complex patients, who may require highly specialized endo vascular procedures 3 3 . Unlike trauma centers in the United States, there is no central organization for the placement of comprehensive and primary stroke centers , which has resulted in geographic disparities in access to primary stroke centers in the US. 3 4 Michigan, USA currently has 30 PSC and 3 CSC . 3 5 In addition to the comprehensive and primary stroke center certification s, the Joint Commission is implementing a third level of certification in July 2015 called Acute Stroke - Ready hospitals; these centers will have the capability of administering thrombolysis and have stroke specialists on standby via telephone. 36 6 Substantial changes in the organization of acute stroke care have also occurred in Canada in recent years. In 1998, the Ontario Ministry of Health and Long - Term Care answered the demand of the Canadian Stroke Systems Coalition, which was to implement a sy stems - based approach to stroke care in Canada. 3 7 As in the United States, Canada implemented a similar stroke care d elivery system led by the Heart and Stroke Foundation. 3 8 This new system focused on province - wide organized systems of car e and was intended to decrease the burden of stroke nationwide by focusing on all facets in the continuum of stroke care . Hospitals in Ontario were designated into 3 categories: regional stroke centers (RSC) , district stroke centers (DSC) , and non - designated community hospitals. Regional stroke centers provide care to all stroke patients, regardless of severity and requirement for surgery; district stroke centers can admit stroke patients and administer thrombolytic therapy, but do not have the infrastructure for advanc ed surgical procedures. Regarding the hierarchy of stroke care delivery in Canada and the United States, regional and district stroke centers in Ontario can broadly be regarded as the equivalen ce of CSC and PSC in the US , respectively. Non - designated commu nity hospitals in Ontario accept stroke patients who are not in requirement of advanced surgical procedures or thrombolytic therapy , but also receive patients that are transferred from more advanced centers following initial intervention (i.e. thrombolytic therapy). In contrast to the US stroke care system, the Canadian system is more centrally organized , and thus the placement of stroke centers depends upon the d istance to other stroke centers, as well as the population size and hospital resources in the region it is serving. 3 8 Ontario currently possesses 11 regional stroke centers, and 18 district stroke centers. 3 9 Improved patient outcomes have been shown to be associated with the implementation of an organized system of stroke care delivery in Ontario. 40 In addition to specialized inpatient stroke care, the Ontario 7 government established 45 stroke prevention clinics 3 9 , which includes post - stroke outpatient care , focusing on secondary prevention , for those who were eith er admitted to hospital or sought emergency department care for a transient ischemic attack or minor stroke. 4 1 R eferral to an outpatient stroke prevention clinic reduced 1 - year all - cause mortality among ischemic stroke patients . Comparative analyses of stroke outcomes between Canada and the US could determine which system produces better patient outcomes , and lead to further studies that help identify the drivers of outcome di fferences . Michigan, USA and Ontario, Canada would be a ppropriate regions to compare since they are similar in population size and distribution 42 , 4 3 , number of hospitals 44 , 4 5 , and a regional stroke care delivery system 39 , 4 6 . Even though stroke hospitalization rates have been steadily decreasing in both regions, Ontario still has substantially lower rates than Michigan. 47 , 4 8 Table 1 shows a breakdown of demographic, geographic, and hospital characteristics between Michigan and Ontario. 39 , 42 - 5 5 Canada/US Comparison: Cardiovascular - Related Mortality and Risk Factor s In 2011, Canada and the United States shared the same top 5 leading causes of death, but in different order. 20 , 2 1 disease, stroke, chronic lower respiratory disorders, and accidental death. 20 Similarly in the US in descending order were: heart disease, cancer, chronic lower respiratory disorders, stroke, and accidental death. 2 1 R eported common causes of death due to vascular - related disease or complications were kidney disease , and heart disease . Diabetes contributed to slightly more Canadian deaths (3.0% vs. 2.9%) in 2011, whereas the opposite was , kidney disease (1.8% vs. 1.4%) , and heart disease (23.7% vs. 19.7%) . 20,21 8 Aside from mortality data , Canada and the United States both possess nationally representative surveys that provide useful information on cardiovascular risk factor s in the respective countries : the US National Health and Nutrition Examination Survey (NHANES) and Canadian Health Measures Survey (CHMS). 56 , 5 7 Hypertension is one of the most important risk fa ctors for stroke 5 8 ; recent data shows that hypertension is more prevalent in the United States compared to Canada (31% vs. 23%). 59 , 60 T he n ational databases (NHANES and CHMS) were used to compare hypertension prevalence and control be tween Canada and the US, and assess the impact of other cardiovascular risk factors on hypertension . 61 McAlister and colleagues found that the US NHANES sample had higher prevalence rates of hypertension (40.2% vs. 27.1%) compared to the CHMS sample. Additionally, uncontrolled hypertension (i.e. average blood pressure greater than 140/90 mm Hg) was more prevalent among the NHANES sample (57.6% vs. 41.4%) , and thus saw higher prevalent rates of controlled or treated hypertension among the Canadian sample (58.6% vs. 42.4%) as a r esult thereof . Furthermore, several other studies that have compar ed Canada and US cardiovascular outcomes have found higher prevalence rates of hypertension in the US . 62 - 6 7 This may be indicative of a poorer cardiovascular health state in the US . Other comparisons of cardiovascular risk factors between Canada and the United States include cholesterol and smoking. 68 - 72 National data shows that high levels of low - density lipoproteins (LDL) are more prevalent in the US (37.8% vs. 23%), but high total cholesterol is more prevalent in Canada (39% vs. 30%). 68 - 70 However, t he age range at which this cholesterol data is based complicates the i nterpretation of these differences; the American data comes from only adults 69 , 70 , whereas the Canadian data consists of a n age range 6 - 79 years old 6 8 . Apart from cholesterol, r ecent 2013 data for smoking shows that the prevalence of current regular smok ers is 9 slightly higher in Canada ( 19.3%) compared to the US (17.8 %). 71 , 72 The Canadian data is more inclusive of age, as it includes regular smokers aged 12 and older, whereas the US data includes regular smoking adults only; these inclusion cri teria may be the reason for the national difference s in smoking prevalence. Canada/US Comparison: Cardiovascular Outcomes and Processes of Care S everal previous studies have compare d outcomes, quality of care, and service utilization between Canada and the United States for cardiovascular diseases, including acute myocardial infarction , heart disease , and stroke . 62 - 6 6 . Studies that compare d heart failure (HF) outcomes identified more favorable short - term outcomes in the United States , compared to Canada, including lower unadjusted in - hospital mortality 62 (3.4% vs. 11.1% ) and 30 - day mortality 6 3 (8.9% v s. 10.7%) , however, the mean length of stay among the American samples of both studies was significantly lower than the Canadian samples which makes the direct comparison of in - hospital mortality rates invalid 62 , 6 3 . However , the difference s in short - term outcomes between Canada and the US was not reflected in long er - term outcome s of HF patients , such as 1 - year mortality (32.2% vs. 32.3%). 6 3 Ko and colleagues 6 3 discussed that t he difference s between short - term and long - term outcomes may be because of differences in the allocation of services and resources between the two countries, with more intensive in - hospital care provided in the US, but better post - discharge care in Canada . More intens ive in - hospital care in the US was also common among studies comparing acute myocardial infarction (AMI) patients. 6 4 - 6 6 The authors found that US patients had undergo ne a more intensive hospital stay, which includ ed higher rates of cardiac procedures . Tu et al. found that a difference s in services utilization ( i.e. coronary angiography, coronary artery bypass surgery, etc.) w ere reflect ed in slightly more favorable short - term mortality in the US 10 elderly (at least 65 years of age) compared to the Canadian elderly (30 - day mortality: 21.4% vs. 22.3%) , but d id not result in better long - term outcome s (1 - year mortality: 34.3% vs. 34.4%) . 6 6 In summary , from previous comparative analyses of patients with cardiac - related diseases , US cohorts generally have better s hort - term outcomes, while longer - term outcomes are comparable between Canada and the US. 62 - 64 ,6 6 Two trials (one each for HF and AMI, respectively) compared outcomes between Canada and the US patients 62 ,6 5 and found contradictory results compared to prior comparative analyses 63,64 ,6 6 . A 2004 study of the GUSTO - I trial ( which compared the effectiveness of four thrombolytic treatments in patients with AMI 7 3 ) comparing long - term mortality in Canada and US AMI patients show ed better long - term outcomes in the U S (5 - year mortality: 19.6% vs. 21.4%) . 6 5 The authors speculated that the more intensive regimen in the US, which entailed 3 - fold higher rates of revascularization procedures, yield ed better mortality in the US. R elat ed to heart failure, a 2013 comparison of American and Can adian patients enrolled in the Acute Study of Clinical Effectiveness of Nesiritide in Decompensated Heart Failure (ASCEND - HF) trial showed that despite a similarity in unadjusted 30 - day mortality rate s between the American and Canadian samples (3.7% vs. 2.2%, p - value = 0.09), Canadians had lower odds (OR= 0.46 ) of 30 - day mortality after adjustment for baseline factors. 62 Additionally, Canadians had a more improved functional status 30 - days post - dischar ge as measured by a trial - related index score on quality of life, even after adjustment for age, sex, and baseline quality of life index score . Specifically related to stroke, a 1999 comparison of aneurysmal subarachnoid hemorrhage patients between Canada and the United States in a trial testing the benefit of tirilazad mesylate ( a medication designed to improve cerebral blood flow 7 4 ) , did not find any significant differences in survival 90 days after hospitalization. 6 6 A previous comparative 11 analys es of stroke care delivery conducted on North Carolina, USA and Ontario, Canada found that patients in North Carolina had a more intensive hospital stay (i.e. higher rates of tPA); however, outcomes were not compared (mortality or readmission) between the two regions. 7 5 To the best of our knowledge, previous studies have not compared stroke outcomes between Canada and the United States using population - based data sources , especially in regions that have implemented regional systems of stroke care delivery. Study Aims To compare acute stroke outcomes between Michigan and Ontario, this study has two primary aims: Aim 1 : To a ssess the impact of applying an established administrative data - based risk adjustment model for in - hospital stroke mortality and 30 - day r eadmission by comparing the distributions of crude (unadjusted) rates and risk - adjusted rates in both Michigan and Ontario hospitals. Aim 2 : In order to produce comparable outcome rates in Michigan and Ontario, we generated directly standardized outcome ra tes for Ontario by applying the Michigan administrative data - based risk - adjustment model to the Ontario sample. This allowed us to determine whether the outcome (mortality and readmission) was different between Michigan and Ontario, having accounted for th e differences in risk profile between the two health care systems. 12 CHAPTER 2: METHODS Hospital Discharge Databases Hospital discharge databases were used to identify acute stroke patient s in both Michigan and Ontario . These databases include information on patient demographics, including age, sex, primary and secondary diagnoses, comorbid conditions, length of stay, and discharge destination. 76 , 7 7 Hospital - level information, including stroke center designation, is publically available and was linked to all hospitals in the two databases . 37,41 D ischarge data for the Michigan hospitals was accessed from the Michigan Department of Community Health (MDCH). Michigan hospitals submit all discharge abstracts to the Michigan Health & Hospital A ssociation (MHA), who facilitates the compilation of all discharge data in the Mich igan Inpatient Database (MIDB). 7 8 The data is then released to other organizations, such as the MDCH and the Healthcare Cost and Utilization Project (HCUP) which is organize d by the Agency for Healthcare Research and Quality (AHRQ) . I n the US , HCUP maintains all state - level inpatient databases, like the MIDB in Michigan. 7 6 The State Inpatient Databases (SID) includes 47 of the US states, and 97% of all public hospitals across the US. The SID collects discharge abstract databases from participating health institutions, and formats patient - level data to be accessible for researchers. Regardless of payer source, the SID includes information on patient demographics, diagnose s, procedures, length of stay, admission and discharge characteristics, and medical costs. The Michigan version of the SID, MIDB, contains data on 150 Michigan hospitals, with 146 of these hospitals reporting information on all hospital stays of Michigan resi dents. 7 9 Currently to our knowledge , there are no previous reports regarding data quality of the MIDB. 13 The Canadian Institute for Health Information Discharge Abstract Database (CIHI - DAD) represents the entire Ontario hospital patient population since al l hospital discharges are included in this database. 80 By law, each acute inpatient facility is required to submit discharge information to the CIHI - DAD. 81 Because the data contains every stroke admission from every hospital in the province it can be regar ded as population - based. Similar to their counterpart in Michigan, the CIHI - DAD includes information on patient demographics, admission and discharge information, diagnoses, past medical history, length of stay, and medical costs. 80 We accessed the CIHI - DAD from the Institute for Clinical Evaluative Sciences in Toronto , Ontario . By re - abstracting admissions records of 18 Ontario hospitals, a 2006 validation study of the CIHI - DAD found an agreement over 97% for non - medical information (demographics, a dmission/discharge information), and kappa scores of 0.81, 0.74, and 0.79 for primary diagnoses of cerebral infarction, non - specified stroke, and intracerebral hemorrhage, respectively. 82 One notable difference between these databases in Michigan and Ont ario that warrant ed special consideration before case ascertainment is the presence of unique patient identifiers . In the MIDB, a unique patient can only be identified within each institution using the medical record number (MRN) , thus a single patient wil l have a different MRN for each hospital that she/he was admitted to . On the contrary, a unique patient in Ontario can be traced across all institutions using the ICES key number (IKN) , which is a unique patient identifier centrally assigned by the Institute for Clinical Evaluative Sciences (ICES), not the hospitals themselves. Case Ascertainment The study sample consisted of hospitalized patients in Michigan or Ontario, during the 3 - year period between January 1, 2010 and December 31, 2012 (inclusive), with a principal 14 diagnosis of acute stroke (ischemic or hemorrhagic) at time of discharge . If an eligible subject was admitted more than once in the 3 - year period with a principal diagnosis of stro ke , we excluded subsequent stroke - related hospitalizations and only utilized their first admission for analysis in this study . T he CIHI - DAD uses a more recent revision of the International Classification of Disease (ICD) compared to the MIDB. In the MIDB , acute stroke discharges were identified using ICD - 9 (International Classification of Disease, 9 th Revision) codes 430, 431, 432, 433, 434, and 436 , whereas in Ontario , the study sample was identified using ICD - 10 (International Classification of Disease, 10 th Revision) codes I60 (excluding I60.8), I61, I62, I63 (excluding I63.8), and I64. Descriptions of the codes used to identify stroke s are listed in T able S1 . The Centers for Disease Control and Prevention (CDC) publishes comparability ratios to show the impact on the number of events after implementation of a new ICD revision. 8 3 For any specific cause of death, a comparability ratio is defined as the number of deaths reported by the ICD - 9 code divided by the number of deaths reported by the ICD - 10 code. The comparability ratio published by the CDC for cerebrovascular diseases was 1.06, which indicates a 6% increase in the attribution of cerebrovascular disease as the underlying cause of death after ICD - 10 implementation. Statistics Canada also reported 1.06 as the comparability ratio for cerebrovascular diseases . 8 4 Patients under the age of 18, or admitted to hospital for a transient ischemic attack were not included in this study. To ensure completeness of patient comorbidity information in both Michig an and Ontario databases , patient discharge records were retrospectively reviewed for three years prior to their index stroke admission date to search for comorbidity diagnosis codes. Since the unique patient identifier in Michigan can only be tracked with in the same hospital, comorbidity data for Michigan stroke patients was only searched within 15 the same admitting hospital up to 3 years from the index stroke admission . ICD codes used to identify past medical history (patient comorbidity data) are seen in T able S2 . Study Exclusions Prior to applying the study exclusions, we started with 7 0,259 Michigan and 40,976 Ontario stroke patients ( Figure 1 ) . Exclusions for this study are presented in Figure 1 , along with the number of patients excluded following each criterion . The values in Figure 1 are mutually exclusive, since each exclusion criteria was applied in a stepwise manner, and not simultaneously. Only non - elective admissions are included in th e analysis , which is standard practice of the Canadian Hospital Reporting Project , when reporting risk - adjusted 30 - day in - hospital stroke mortality rates and 28 - day stroke readmission rates for stroke 8 5 . In both the varia ble that identified an admission as being elective . The total number of elective admissions that were subsequently excluded totaled 12,370 ( 17.6 % of starting sample) Michigan patients and 622 (1.5%) Ontario patients. The substantial discrepancy in elective admissions between Michigan and Ontario is not easily explainable, but obviously is more liberally utilized in the MIDB. This discrepancy in elective admissions is further discussed in Chapter 4. This study also exclude d in - hospital strokes ( i.e. stroke events that occur after being admitted for another reason ) (Figure 1) . Both datasets (MIDB and CIHI - DAD) have specific variables that identify all diagnoses as either pre - admit or post - admit, and was thus used to identify strokes occurring post - admi ssion (i.e. in - hospital strokes) ; this excluded 3,804 ( 5.4 % of starting sample) Michigan and 743 (1.8%) Ontario patients. As this study aims to rank and compare hospitals based on stroke outcome measures, excluding these patients would reduce the variation due to non - stroke diagnoses that occur before an in - hospital stroke event. 8 5 Since 16 hospice care focuses on end - of - life care, we excluded 2,957 ( 4.2 %) Michigan and 275 (0.7%) Ontario patients discharged to palliative care since we were unable to track deaths once a patient was discharged from hospital. 8 7 CMS excludes patients enrolled in the Medicare Hospice program from their stroke mortality measures. 8 8 To ascertain an inception str oke cohort (patients with first strokes only), we employed a washout method previously undertaken in prior stroke research. 8 9 All patients who met the study inclusion criteria, but were admitted to hospital for stroke in the 3 year period prior to the star t of the cohort , i.e. between January 1, 2007 and December 31, 2009 (inclusive), were excluded from the study. Therefore, a single patient could only be counted once in the denominator population for both primary outcomes - in - hospital mortality and 30 - day readmission. Since previous strokes affect subsequent stroke outcomes, this methodology creates a study sample with a more similar cerebrovascular health state and baseline risk . The washout period applied to 150 (0.2%) Michigan patients and 1,491 (3.6%) Ontario patients; this difference in patients identified in Michigan and Ontario using the washout procedure 89 is likely due to the fact that the IKN in Ontario was able to track a unique patient over time and across multiple hospitals , whereas the MRN in Michigan was not . Since readmission measures are used to gauge quality of care, planned readmissions ( usually indicative of elective procedures ) were not included (Figure 1) . A specific variable exists in the CIHI - DAD that identifies a ll elective readmissions, and thus allowed us to perform the exclusion to 142 (0.3% of starting sample) Ontario patients using this variable ; we sought alternatives to identify planned/elective readmissions in Michigan since a variable of this nature does not exist in the MIDB . As in a previous publication using the Michigan Stroke Registry 90 , we used the following procedure codes (ICD - 9) to help identify 2,357 ( 3.4 %) Michigan patients 17 with elective readmissions: 0061, 0063, 380.2, 381.2, 382.2, 383, 384.1, and 384.2. Descriptions of these procedure codes are included in T able S3 . Twelve (0.01%) Michigan patients with missing hospital - level information (ex. stroke center certification, teaching status, or acute bed size) were excluded from this study. We also excluded hospitals with a stroke case load less than 75 patients over the three year period of case ascertainment (2010 - 2012) , and as such eliminated 1,275 ( 1.8 %) Michigan and 2,043 (5.0%) Ontario pat ients. As a measure of data quality, 12 (0.03%) Ontario patients were excluded for having an index admission listed chronologically after death; this procedure was conducted at the Institute for Clinical Evaluative Sciences before the data was given to inv estigators. Outcome Definitions There were two primary outcomes used in this study: in - hospital mortality and 30 - day readmission following discharge. In - hospital mortality was defined as death due to any cause during the hospital stay after being admitted for stroke; 30 - day readmission was a non - elective readmission to hospital, for any reason (not just stroke) , within 30 days of discharge. The denominator population for the in - hospital mortality measure includes all stroke patients (satisfying study eligibility criteria listed in Figure 1 ) admitted to hospital from 2010 to 2012, whereas the denominator population for 30 - day readmission includes stroke patients who were discharged alive , excluding those patients discharged to hospice /palliative care . Since unique patient tracking post - discharge is not possible in Michigan, we were unable to use 30 - day mortality as an outcome in this study, and so instead relied on in - hospital mortality. Readmission in Michigan cannot be tracked across multiple health institutions , so o nly readmissions to the same facility as the index admission were included in Michigan. Although readmission s across different facilities can be tracked in O ntario, we also restricted Ontario readmissions to those at 18 the same facility as the index event to promote comparability between the two data sources . A 2014 population - based cohort study by Staples and colleagues found that approximately 82% of patients are readmitted to the original hospital. 91 Although this stu dy was not disease - specific, it can be assumed that our capture of Ontario stroke readmissions was underestimated by approximately 18% . Statistical Analyses: Descriptive Due to privacy restrictions in Michigan and Ontario, the respective datasets remaine d in their own regions, and the datasets analyzed separately. Summary estimates of p atient and hospital characteristics were compared between the two regions using a t - test for continuous variables, and chi - square test s for categorical variables. Rationale for Risk Adjustment When comparing health outcomes , like mortality and readmission , across health institutions or jurisdictions (i.e. hospitals or provinces ) , it is crucial to account for the differences in patient risk (i.e. presence of risk factors that pre - dispose patient to a particular health outcome) . 92 - 9 4 R isk adjustment models are used to account for differences in patient characteristics that result in one patient having a higher likelihood of experiencing an outcome compared to another . If risk adjustment models account for all differences in patient risk, then any outcome variation after risk - adjustment can be assumed to be due to differences in hospital performance (i.e. quality) and not patient case mix 92 . In terms of aggregate - level data w hen comparing stroke outcomes across hospitals, comparability is essential since some hospitals, particularly referral centers , may treat patients that are more severe, or of higher risk profile, and would therefore have more adverse outcomes as a re sult. 9 5 It would have been unfair to compare 19 a hospital that does not treat the same patient risk profile as another hospital; risk - adjustment allows for a fair comparison across hospitals. Statistical Analys e s: Risk - Adjustment Models CIHI implemented the Canadian Hospital Reporting Project (CHRP) to create outcome measures that are generalizable to all regions across Canada, and so that health officials can use these measures to compare their hospital, health jurisdiction, or province against others; these measures include both mortality and readmission. 8 5 Outcome measures reported by CIHI are risk - adjusted for age and sex, as well as comorbidities related to the specific disease; these outcome measures include various diseases (or conditions) in addition t o stroke (i.e. AMI , obstetrics, surgery ) . In this study, w e employed an established administrative data - based risk - adjustment model developed by CIHI for 30 - day in - hospital stroke mortality and 28 - day readmission after stroke . 96 Although these CIHI models are stroke - specific, it is important to note its similarities to the CIHI acute myocardial infarction (AMI) models with the same outcomes (30 - day in - hospital mortality and 28 - day readmission) . Likenesses among the 30 - day in - hospital mortality models for A MI and stroke include: age, gender, shock, renal failure, heart failure, cancer, and pulmonary edema. Similarities for the 28 - day readmission models include: age, gender, diabetes, and renal failure. This model was selected because it was developed using the CIHI - DAD 8 5 , which is our data source for Ontario stroke patients . Furthermore, the model can be applied to the MIDB (Michigan data source) , since the variables required for the model can also be searched for in the MIDB . Currently to our knowledge, thi s model has not been validated outside of Canada . Variables included in the in - hospital mortality model are shown in Table 3 and include age, gender, stroke type, and past medical history of cancer, shock, heart failure, pulmonary edema, ischemic heart disease (acute and chronic), renal failure, liver disease, and 20 hospital stroke certification. Variables included in the 30 - day readmission model included ( Table 4 ) include age, gender, stroke type, diabetes, acute ischemic heart disease, renal failure, and hospital stroke certification. For this study, a stroke - certified center in Michigan included all comprehensive and primary stroke centers, and likewise in Ontario included all regional and district stroke centers. The ICD - 10 codes used to identify past m edical history in the O ntario data are listed in T able S2 . These same risk - adjustment models wer e applied to the MIDB; however, past medical history/comorbidity data was identified using ICD - 9 codes which are again shown in T able S2 . Since the mortality and readmission outcomes in this study were binary (i.e. both outcomes are dichotomous), we used multivariable logistic regression models , which describe the rela tionship of the binary outcome with this set of covariates . 9 7 Aim 1: Hospital - Level Risk Profiling Comparison Hierarchical logistic regression models (HLM) were used to risk - adjust for differences in patient case mix and hospital characteristics in order to report hospital - level outcome rates in each region . Hierarchical, or multilevel modelin g, accounts for patient cluster ing within hospitals . 9 8 O ther advantages of multilevel modeling include the ability to profile individual hospitals based on the random effect intercepts and the fact that estimates from smaller hospitals are more reliable . 95 , 9 9 The hierarchical models (HLM) for mortality and readmission in aim 1 were generated using the PROC GLIMMIX procedure in SAS , and model fit was assessed using the Akaike information criterion (AIC). Hospitals were ranked based on their performance with regard to stroke outcomes (in - hospital mortality and 30 - day readmission) by generating hospital - specific predicted over expected (P/E) ratios 9 5 generated by applying the hierarchical CIHI administrative data - based logistic regression model to the Michigan and Ontario data separately. P atient - level predicted 21 and expected probabilities were aggregated and summed at each hospital to create the P/E ratios. This method of multilevel modeling (i.e. random effects model) uses hospital - specific random intercepts to account for hospital - specific effects on the outcome while accounting for patient case mix and cluster ing within each hospital. 95 , 100 The predicted p robability is based on the estimated random intercept from each hospital after accounting for patient case mix. 9 5 The expected probability uses the average of all hospital intercepts within a specific region (e.g. all Michigan hospitals) , and as such repre sents an estimation of the baseline risk at the hypothetical average hospital after accounting for case mix differences between hospitals . The P/E ratio performance in relation to the average performing hospital with the same case mix profile; that of the average hospital within that region (i.e. outcome at this hospital is more favorable than the average hospital) . On the contrary, a P/E ratio gr performance is inferior to that of the average hospital within that region (i.e. outcome at this hospital is less favorable than the average hospital). After hospitals were ranked according to their predicted over e xpected ratios for the mortality and readmission model, good and poor performing outlier hospitals in Michigan and Ontario were identified by the top 10% and bottom 10% respectively in each region ( T ables 6 and 7 ), as previously defined by the American Stroke Association 9 5 . To calculate risk - adjusted outcome rates for each Michigan and Ontario hospital, the hospital - specific P/E ratios were multiplied by the observed outcome rate of all Michigan and Ontario patients , respectively . In order to assess the impact of applying the administrative data - based risk adjustment model s on in - hospital mortality and 30 - day readmission , we examined the 22 correlation between hospital - specific crude and risk - adjusted outcome rates , in both Michigan and O ntario separately, using the Spearman rank correlation. Aim 2: Patient - Level Risk Standardization Upon applying the risk - adjustment model to the Michigan and Ontario patient population separately as in aim 1 , there were differences in risk profile betwe en the Michigan and Ontario patient populations . The differen ces in risk profile occurred due to two different phenomena: first ly , the prevalence rates of the risk factors / comorbidities included in the models were very different ( see T able 2 ) , and secondly, the risk factors in the model s had different magnitudes of effect (i.e. different adjusted odds ratios ) on the outcome s in the two samples (see T ables 7 and 8 ) . To be able to report comparable outcomes rates between Michigan and Ontario patients, we needed to account for the difference in risk profile s between the Michigan and Ontario patient populations . To do this , we applied the model parameter coefficients f rom the Michigan model to the Ontario cohort, to ensure that the effect magnitude (i.e. adjusted OR) of each model parameter was the same in both regions. In the process of applying the Michigan risk - adjustment model coefficients to the Ontario patients, c learly it makes no sense to apply specific hospital - specific intercepts fr om Michigan hospitals to those in Ontario. To alleviate this methodological problem , we used the generalized estimating equations (GEE) procedure to predict patient - specific probabil ities, while account for patient cluster ing within hospitals without the necessity of generating hospital - specific random intercepts 9 5 . The GEE procedure was applied (CIHI model as used in aim 1) to the Michigan patient sample to generat e patient - level predicted probabilities of experiencing the outcome s of interest . We then applied the model parameter coefficients from the Michigan GEE model ( T ables 8 and 9 ) to the Ontario cohort , ensuring that the predicted probability of each Ontario patient was based 23 on Michigan parameter estimates (i.e. same parameter magnitudes of effect in Michigan and Ontario ). Once the procedure was applied, differences in the patient - specific predicted probabilities between Michigan and Ontario would be due to differences i n risk factor prevalence (i.e. risk profile of patients) between Michigan and Ontario. The GEE models for mortality and readmission in aim 2 were generated using the PROC GENMOD procedure in SAS, and fit was assessed using the quasi - likelihood information criterion (QIC) ; fit statistics are s hown in Table 5 . Since we applied the Michigan model coefficients to the Ontario population, we only listed the Mich igan model QIC and c - statistic values. After applying the Michigan - based model to the Ontario data , Mic higan and Ontario p atients were stratified into risk deciles, b ased on their predicted probabilities. The patient risk deciles were determined by the risk distribution of Michigan patients. To further adjust for differences in risk distribution between the Michigan and Ontario study populations, we then directly standardized the Ontario patient population to the risk distribution of the Michigan sample. 6 3 Standardization allows for the controlling of a confounding variable (differences in patien t risk distribution in this situation ) that prevents the outcomes from being comparable between two distinct populations . 101 Directly standardized outcome rates for the Ontario patient population were calculated by multiplying the proportion of Michigan patients within each risk stratum, by the crude Ontario outcome rate (mortality or readmission) in the corresponding risk strat um. 6 3 The stratum - specific standardized Ontario rates were summed to produce the standardized summary rates used for direct comparison to the crude Michigan outcome rate. This procedure allows the calculation of standardized Ontario outcome rates that woul d have been observed if the Ontario patients had the same risk distribution as the Michigan patients . We calculated 95% confidence intervals for the standardized summary outcome rates in both regions 24 and then determine d if a statistical ly significant difference existed between Michigan and Ontario. A step - by - step outline describing the direct risk standardization procedure used in this analysis is shown in Figure 2 . Ultimately, t his direct risk standardization used in a im 2 was only able to standardiz e the magnitude of effect for each model covariate between the two patient samples, but other sources of variation still included the substantial differences in the risk factor frequencies (Table 2 ) , and the fact that Ontario patients are on average stayin g seven days longer in hospital compared to Michigan patients (also shown in Table 2 ) . Ethics Approval This study received institutional review board approval from Michigan State University, as well as the governing bodies of the databases used - Michigan Department of Community Health (Lansing, Michigan) and Institute for Clinical Evaluative Sciences (Toro nto, Ontario). 25 CHAPTER 3: RESULTS Descriptive Characteristics Once the study exclusions were applied ( Figure 1 ), our final cohort included 47,364 stroke patients from Michigan, USA and 35,648 stroke patients from Ontario, Canada. Patient - and hospital - level characteristics for the Michigan and Ontario cohorts are described in Table 2 . Compared to Michigan, the Ont ario cohort contained more male patients (50.8% vs. 49.1%), was older (mean age: 72.4y vs. 69.5y), and stayed longer in hospital (mean length of stay: 12.5d vs. 5.4d). Differences also existed in stroke diagnosis coding, as the Michigan sample had a signif icantly higher frequency of ischemic strokes (80.8% vs. 63.4%), and a significantly lower frequency of unidentifiable strokes (13.2% vs. 0.1%). Additionally, the Michigan sample had slightly lower frequencies of subarachnoid hemorrhages (4.6% vs. 5.5%), in tracranial hemorrhages (9.4 vs. 11.5), and other hemorrhages (5.2% vs. 6.4%). There were large differences in the frequencies of comorbidities between the two datasets (Table 2 ). Ontario stroke patients had higher prevalence rates for acute ischemic heart disease (7.6% vs. 4.0%) and cancer (8.2% vs. 3.4%). In contrast, the Michigan cohort had substantially higher prevalence rates for chronic ischemic heart disease (30.7% vs. 13.8%), diabetes (34.0% vs. 27.0%), heart failure (16.0% vs. 8.7%), and renal failu re (21.5% vs. 9.2%). In terms of hospital - level characteristics (as presented in Table 2 ), the Michigan cohort was more frequently admitted to a stroke - certified center (65.9% vs. 60.0%) and teaching hospital (36.1% vs. 34.5%). The mean acute bed size of Michigan hospitals was also higher than Ontario hospitals (465.8 vs. 289.8). 26 The stroke - specific admission s rate in our patient samples (using the population figures from Table 1, and final sample figures from Figure 1) was 47 .8 per 10,000 population and 26.1 per 10,000 population, respectively for Michigan and Ontario. Michigan had substantially lower unadjusted outcomes rates of in - hospital mortality (7.6% vs. 14.0%) as well as lower 30 - day readmission rates (4.5% vs. 5.1%) ( Table 2 ). Length of Stay The length of stay distributions for the Michigan and Ontario patient samples are shown in Figure 3. The median length of stay (LOS) was shorter in our Michigan sample (4.0 days vs. 7.0 days). Figure 3 shows an upward shift in the Ontario patient LOS distribution. In Michigan, 50% of patients were discharged (dead or alive) between 2 - 6 days after admission , whereas 50% of Ontario patients were discharged (dead or alive) between 4 - 13 days after admission. Discharge patterns for Mich igan and Ontario stroke patients were vastly different especially in the first week after admission ( Figure 4 ). The proportion of Michigan patients discharged alive after two days in hospital was more than double that of Ontario (24.2% vs. 11.6%). By the s eventh day, 75% of Michigan patients had been discharged alive, compared to only 45% in Ontario. Thus, the 7 - day in - hospital mortality rate was lower in Michigan compared to Ontario (6.2% vs. 8.7%). Aim 1: Hospital - Level Risk Profiling Comparison To assess the impact of applying an established administrative data - based risk adjustment model for in - hospital stroke mortality and 30 - day readmission by comparing the distributions of crude (unadjusted) rates and risk - adjusted rates in both Michigan and Ontario h ospitals. 27 The fit statistics for both hierarchical models applied separately to the Michigan and Ontario samples are shown in Table 5 . The Akaike Information Criterion (AIC) for the Michigan models was 21485.3 for the in - hospital mortality model, and 16030.3 for the 30 - day readmission model. AIC values in Ontario were 26401.6 for in - hospital mortality and 14153.1 for 30 - day readmission. The results of the hierarchical logistic regression risk - adjustment model for in - hospital mortality for the two datas ets are shown in Table 3 . Age was shown to increase risk of in - hospital mortality for Michigan and Ontario stroke patients. In other words, for every one - year - hospital mortality increases by 2% (adjus ted odds ratio = 1.02). Similarly in Ontario, each year increase in age constitutes a 4% increase in risk of in - hospital mortality (aOR = 1.04). Among the Michigan sample (Table 3) , subarachnoid, intracranial, and other hemorrhages significantly increased the risk of in - hospital mortality compared to ischemic stroke s (adjusted odds ratios were 7.63, 8.85, and 4.88, respectively). A similar phenomenon was discovered in the Ontario cohort, but the effect magnitudes were not as large for subarachnoid, intracra nial, or other hemorrhages (adjusted odds ratios were 3.96, 3.81, 2.18, respectively) relative to ischemic strokes. Michigan stroke patients, who were diagnosed with unidentifiable (UTD) strokes, had similar risk of in - hospital mortality as ischemic stroke patients (aOR = 0.69, p - value = 0.72). This is also true for our Ontario sample (aOR = 1.00, p - value = 0.9671), but the increased frequency of Ontario UTD stroke patients compared to Michigan (13.2% vs. 0.1%) leads to the assumption that these patients ar e true 28 Acute ischemic heart disease (aOR = 2.25), cancer (aOR = 1.41), heart failure (aOR = 1.38), liver disease (aOR = 1.32), pulmonary edema (aOR = 2.40), renal failure (aOR = 1.30), and shock (aOR = 9.01) were associated with an increased risk of in - hospital mortality in Michigan ( Table 3 ). F or Ontario patients, AIHD (aOR = 1.24), cancer (aOR = 1.72), heart failure (aOR = 1.78), liver disease (aOR = 1.63), renal failure (aOR = 1.41), and shock (aOR = 2.37) increased risk of in - hospital mortality. Common between Michigan and Ontario, chronic ischemic heart disease was not associated with increased risk. Michigan had meaningfully larger adjusted odds ratios for AIHD (2.25 vs. 1 .24), pulmonary edema (2.4 vs. 1.26), and shock (9.01 vs. 2.37). In terms of hospital - level characteristics, patients admitted to a stroke - certified center in Michigan or Ontario was not associated with risk of in - hospital mortality. The results of the tw o hierarchical logistic regression risk - adjusted models for 30 - day readmission are shown in Table 4 . Age was not associated with risk of 30 - day readmission in Michigan and Ontario. Michigan females had an approximately 16% reduced odds of being readmitted to hospital within 30 days of discharge (aOR = 0.84), relative to Michigan males. Gender was not associated with risk of 30 - day readmission among Ontario patients. Relative to ischemic strokes in Michigan, subarachnoid (aOR = 0.54) and intracranial (aOR = 0.55) hemorrhages reduced risk of 30 - day readmission by almost one half. Likewise was true in Ontario for intracranial hemorrhages (aOR = 0.75), although the risk reduction was not as significant as in Michigan. Diagnoses of other hemorrhage in Michigan (a OR = 1.47) and Ontario (aOR = 1.55) increased risk of 30 - day readmission relative to ischemic strokes. Similarly in both regions, unidentifiable strokes presented similar risk of 30 - day readmission as ischemic strokes. 29 Past medical history of acute ischemic heart disease, diabetes, and renal failure were all associated with increased risk of 30 - day readmission in Michigan and Ontario ( Table 4 ). Although not substantially different across regions, the adjusted odds ratios were higher in Ontario, compa red to Michigan, for past medical history of diabetes (1.27 vs. 1.22) and renal failure (1.62 vs. 1.56), but lower for AIHD (1.33 vs. 1.51). Being admitted to a stroke - certified center in Michigan increased the risk of 30 - day readmission (aOR = 1.60), but association of risk among Ontario patients admitted to stroke - certified centers presented a null finding (aOR = 0.98, p - value = 0.7449). There was a total of 78 Michigan and 83 Ontario hospitals included in the hospital - level ormance was quantified using a predicted over expected (P/E) ratio, which if less than 1, means that the hospital is performing better than the hypothetical average hospital with a similar case mix. Hospital - specific P/E ratios are shown in ascending seque nce in Tables 6 and 7 for mortality and readmission, respectively. Stroke - certified centers are indicated by the bold text, while the shaded boxes enclose hospitals in the top and bottom 10%. We compared the distributions of the hospital - specific P/E ratio s, as shown in Figures 5 and 6 . For in - hospital mortality, the range of P/E ratios among Ontario hospitals was narrower than Michigan hospitals (Figure 5 ). In Michigan, there were a total of 8 positive outliers (top 10%), and 8 negative outliers (bottom 10 %). The two worst performing hospitals (i.e. those with the highest mortality after adjusting for case mix) in Michigan and Ontario had P/E ratios of 1.8321 and 1.4323, respectively; while the two best performing hospitals (i.e. those with the lowest morta lity after adjusting for case mix) had a P/E ratio of 0.5316 and 0.6354, respectively in Michigan and Ontario. The range of P/E ratios for the Ontario hospitals were substantially narrower than Michigan hospitals when using 30 - day readmission as a performa nce measure 30 (Figure 6 ). We discuss possible causes of this phenomenon in Chapter 4. P/E ratios for 30 - day readmission are listed in Table 7 . After removing 3 outliers in Michigan, P/E ratios of the worst performing hospitals were 2.0684 and 1.15346, and be st performing hospitals were 0.49828 and 0.86939, respectively for Michigan and Ontario. Figure 7 and 8 alluded to similar findings in Figures 5 and 6 . Figure 7 (in - hospital mortality) shows the similar effect of risk - adjustment on Michigan and Ontario hospitals using the hierarchical logistic regression model; Michigan and Ontario hospitals have similar regression line slopes when plotting observed vs. risk - adju sted outcome rates. Only difference noted is the shift upward of Ontario hospital - specific rates, which is driven by the higher patient length of stay in Ontario. As displayed in Figure 8 , risk adjustment had different effects on Michigan hospitals compare d to Ontario hospitals when using 30 - day readmission as the outcome. Regression line slopes are very different, which is consequence of the substantial narrowing of the hospital - specific P/E distribution among Ontario hospitals (Figure 6 ). This difference is elaborated on in Chapter 4. For both primar y outcomes, the hospital rank correlation between observed (crude) and risk - adjusted rates was higher in Ontario (Figu re 7 and 8 ). The rank correlation between hospital - specific observed and risk - adjusted outco mes rates was quantified using the Spearman rank correlation. A Spearman rank correlation of 1 means that the hospital ranking of observed rates is identical to that of risk - adjusted rates. A lower hospital rank correlation in Michigan means that after ris k - adjustment, there is a greater shift in hospital rankings among Michigan hospitals, compared to Ontario. Relative to Michigan, the Ontario hospital Spearman co efficient was higher for in - hospital mortality (0.95 vs. 0.84) and 30 - day readmission (0.93 vs. 0.86) . Aim 2: Patient - Level Risk Standardization 31 In order to produce comparable outcome rates in Michigan and Ontario, we generated directly standardized outcome rates for Ontario by applying the Michigan administrative data - based risk - adjustment model to the Ontario sample. This allowed us to determine whether the outcome (mortality and readmission) was different between Michigan and Ontario, having accounted for the differences in risk profile between the two health care systems. Justification for using the GEE (generalized estimating equations) procedure to predict Michigan patient - specific probabilities is shown in Tables 8 (in - hospital mortality) and 9 (30 - day readmission). By comparing the HLM (hierarchical logistic mo del) and GEE procedures, it is apparent that the adjusted odds ratios produced from these methods are very similar. The adjusted odds ratios generated from the GEE procedure were then applied to the Ontario patient population to produce Ontario patient - s pecific predicted probabilities t o utilize for the direct standardization in Aim 2. Upon using the Michigan patient risk distribution as the cut points of the risk deciles (Tables 1 0 and 1 1 ), we noticed little change in the patient proportions across all but two risk deciles between Mich igan and Ontario for in - hospital mortality; the largest difference was approximately 2.5% more Ontario patients in the highest two risk deciles (Table 1 0 ). Therefore applying the Michigan model coefficients to the Ontario population only impacted in the hi ghest two risk deciles. The unadjusted in - hospital mortality rate was higher in Ontario (14.0%), compared to Michigan (7.6%), but by directly standardizing the Ontario patients to the risk distribution of Michigan patients, the risk - standardized in - hospita l mortality rate for the Ontario patient sample was 13.3%, exhibiting little change from the crude rate of 14.0%. This direct 32 standardization procedure had little effect on the overall risk - standardized in - hospital mortality rate in Ontario. There was subs tantial variation in stratum - specific observed in - hospital mortality rates among our Michigan sample, ranging from 1.7% in the lowest risk decile, to 29.2% in the highest risk decile. This was similarly the case in Ontario, as the observed rates of the low est and highest risk deciles ranged from 3.4% to 31.0%. The crude Ontario 30 - day readmission rate increased from 5.1% to a risk - standardized rate of 5.3% after standardizing the Ontario patient population to the risk distribution of our Michigan sample (Ta ble 1 1 ). The 5.3% Ontario figure was higher than the crude 30 - day readmission rate of 4.5% in Michigan. There were clear distinctions between the risk distributions of Michigan and Ontario patients for readmission, opposed to the similarities of we found f or in - hospital mortality. The proportion of Ontario patients nested within the risk deciles (cut points set by risk distribution of Michigan patients) ranged from 4.9% to 15.5%. Additionally, there was little variation in the stratum - specific observed read mission rates for Michigan and Ontario, in comparison to in - hospital mortality. There was a 7.5% difference in the observed readmission rate among the highest and lowest risk decile in Michigan; 4.5% was likewise the case in Ontario. 33 CHAPTER 4: DIS CUSSION This comparison of outcomes in hospitalized acute stroke patients between Michigan, USA and Ontario, Canada, provides both hospital - level and patient - level comparison, which highlight differences between the two health systems that complicate our interpretation of the outcomes under study . Our study shows how differences in health care structure between Canada and the United States can create fundamental differences in outcomes that hinder our ability to compar e across the two systems (length of stay and in - hospital mortality, for example) . Because the interpretations of the outcomes in this study are complicated , we cannot definitively determine which system produces more favorable outcomes. Aside from the previous Ontario and North Carolina comparison of processes of care 75 , t o the best of our knowledge, this is the first study that utilizes population - based data sources to compare stroke outcomes between Canada and the United States . To compare outcomes between hospitals in each region ( Michigan and Ontario ) , we calculated hospital - specific predicted over expected (P/E) ratio s . The P/E ratio for each hospital quantifies its performance in relation to the hypothetical average performing hospital in their region with similar case mix . Risk - adjustment models were applied separately in each region, to examine the impact of risk adjustment (using the CIHI administrative model ) on the hospital ranking of Michigan and Ontario hospitals. We found a greater variance in P/E ratios both in - hospital mortality and 30 - day readmission in Michigan hospitals compared to Ontario hospitals (Figures 5 and 6 ) . More specifically, the di stribution was drastically narrower among Ontario hospitals for the 30 - day readmission outcome m easure (Figure 6 ). Although the distribution of Ontario hospitals is shifted upward for in - hospital mortality (Figure 7 ), it is noticeable that risk - adjustment had a similar effect on Michigan and Ontario crude mortality rates (i.e. plots are 34 distributed i n somewhat similar fashion although the Ontario hospitals are more closely plotted along the regression line) . This is not the case for 30 - day readmission (Figure 8 ). For Ontario hospitals, the distribution of P/E ratio s is too sm all, and there is less than two percentage point s between the lowest and highest risk - adjusted readmission rate among Ontario hospitals, as shown in Figure 6 . Furthermore as displayed in Figures 6 and 7, Ontario hospitals had a higher Spearman rank correlation than Michigan hosp itals for both mortality and readmission, respectively. This finding suggests that Ontario hospitals more closely resemble their average hospital, relati ve to Michigan hospitals and their own average. As shown in Figures 5 and 6 , it is quite apparent that the Ontario P/E ratios for the mortality and readmission models are so vastly different, suggesting that the significantly narrower range of Ontario hospital - specific P/E ratios for readmission (Figure 6 ) could be an artifact of the hierarchical logistic regression model, or suggest an error in the Ontario data used for the readmission model . Comparatively, the P/E ratio distributions for mortality between Michigan and Ontario hospitals are similar (Figure 5 ) , but very different for readmission (Figure 6 ) . The significant narrowing of Ontario hospital - specific P/E ratios from the readmission model may also have been the result of systematic differences between the Michigan and Ontario stroke patient populations (i.e. risk profile) , whereby the hierarchical model employed would behave differently in Michigan compared to Ontario . Risk profile differences can be seen in Tables 8 and 9 , listing noteworthy differences in adjusted odds ratios between Michigan and Ontario. The risk profile difference shown between Michigan and Ontario may be a true picture, but this difference could have been influenced by systematic differences in Michigan and Ontario administrative datasets, specifically related to the ascertainment of ca ses and past medical history (i.e. risk factor prevalence). 35 The multilevel logistic regression models we used for the hospital - level analysis quantifies the association of a binary outcome (i.e. mortality or readmission) with a set of known variables 9 7 , in this case patient - and hospital - level characteristics. The prevalence of risk factors included in the model would have a direct impact on the model, and its output of P/E ratios for each hospital. As shown in Table 2 , there is substantial variation in t he prevalence of risk factors included in our mortality and readmission models. This variation is not necessarily the true snapshot of the cardiovascular health state among Michigan and Ontario patients, as there were systematic differences in how past med ical information was acquired in the two datasets. As previously noted, the patient identifiers in Michigan (MRN) and Ontario (IKN) had unique characteristics; because the MRN in Michigan was only traceable within the admitting hospital, we were only able to search for past medical history within a single hospital. This was due to the fact that the Michigan hospitals themselves assign the MRN to patients, thereby not being traceable across multiple institutions; on the contrary the Ontario IKN is traceable across all institutions since the Institute of Clinical Evaluative Sciences centrally assigns it . Aside from patient identifiers, there are other potential driv ers of differences in the prevalence of stroke - related risk factors. The Michigan dataset utilized ICD - 9 diagnosis codes, whereas ICD - 10 was used in Ontario. Published comparability ratios for risk factors included in our mo del are as follows: ischemic (ch ronic and acute) heart disease (0.99), diabetes (1.01), cancer (1.01), renal failure (1.30 ), liver disease (1.04), shock (1.19), heart failure (1.04), and pulmonary circulatory disease s (1.12). 8 3 Since some of these ratios are meaningfully different than 1, this could be another source of variation in risk factor reporting in Michigan and Ontario. Lastly, the different health care structures could contribute to this difference as well. In a multi - payer system as in the United States 1 , there may be incentiv e for hospitals to more completely record medical history 36 on discharge abstracts, before submitting the information to insurance companies for remuneration; this would not apply in a universal health care system as in Canada 5 . Throughout the course of our analyses, we discovered differences for many variables in how patients included in the hospital discharge databases were being coded (i.e. principal diagnoses, admission type, past medical history, etc.). We also found huge differences among some of our e xclusion criteria. Firstly, there were 11,748 more patients excluded in the MIDB (Michigan) compared to the CIHI - DAD (Ontario) for elective admissions; this accounted for about 17.6 % and 1.5% of the starting samples, respectively. There are no clear explan ations for this difference, but a similar code existed within the Michigan and Ontario datasets that 80% of these elective admissions as having elective pro cedure codes (same procedure used to identify elective readmission as listed in Table S3 ), so the majority of these exclusions are truly elective admissions. In Michigan, admission type is acquired from the claims form that the hospital completes at time of discharge for each patient before sending the claim to the insurance company for remuneration. 103 . I n Ontario, admission type is acquired f rom data abstractors examining admission information retrospectively; abstractors base their determination of admission type . 104 The definition in d admission for treatment and/or assessment . The difference in variable definition (and its interpretation) could be a source of the discrepancy in elective admissions. Furthermore, it could also be the source of information to which is being used to dete rmine admission type. Ontario abstractors solely use patient status at admission only to acquire admission type. S ince more information is available in 37 Michigan (i.e. patient information from entire duration of hospital stay ) when the claims form is being The exclusion for elective admissions was the largest discrepancy in our study, but there were other important differences. For example the number of in - hospital strokes that we re identified and excluded varied considerably (3,804 ( 5.4 %) and 743 (1.8%) in - hospital strokes, respectively for Michigan and Ontario). This also could be a difference in coding use between the two regions, because the variable we used to identify an in - h ospital stroke was similar in the sense that it identified a stroke diagnosis as being present at admission or not. If the primary diagnosis was stroke, but was listed as not present at admission, we excluded these patients as having in - hospital strokes. A gain, we can find no clear explanation for these differences, but this variable seems to be utilized more in Michigan , even though an in - hospital stroke was identified using a single variable with similar language in both the CIHI - DAD and MIDB. Potential s ources of the coding usage differences are not obvious (for both elective admission and in - hospital strokes) , but could be a result of how the variables are being interpreted by hospital officials who input the data . The number of patients being discharged to hospice was also different between the two regions, identifying 2,927 ( 4.2 %) Michigan patients and 275 (0.7%) Ontario patients. Reasons for this discrepancy may not be result of administrative coding differences, but instead differences in the respecti ve health care systems. Palliative care is covered in the United States for Medicare patients deemed as terminally ill by their physician, whereas this type of hospice coverage does not exist in Ontario . 102 I n Ontario, palliative care is usually provided by specialized end - of - life care units within a hospital. Lastly, (as shown in T able S2 ), there was a difference in the number of patients excluded by the washout period we employed in this study to identify patients in our cohort that were admitted to hosp ital for a stroke between 2007 and 38 2009. This procedure excluded 150 (0.2%) patients in Michigan and 1,491 (3.6%) patients in Ontario. This difference can be attributed to the inability of the Michigan patient identifier to track past stroke admissions acr oss other hospitals, and because of this, the number of Michigan patients being excluded is an underestimation of the true proportion of patients that have previously suffered from a stroke from 2007 to 2009. The accumulation of systematic differences betw een the Michigan and Ontario patient populations hinders our ability to make direct comparisons of the hospital - level analyses (aim 1), as well as the fact that the models were applied separately in each region. In the event that the datasets could be comb ined, region (i.e. Michigan or Ontario) could be included as a model covariate, and thus be able to directly compare patient - specific risk (predicted probability of experiencing outcome) between Michigan and Ontario patients , because any variation in the o utcome caused by receiving care in different regions would be statistically accounted for in this instance . In order to draw a direct outcome comparison across both patient populations (aim 2) , we needed to a ccount for the difference in risk profile between Michigan and Ontario (Tables 7 and 8) . We applied the Michigan risk - adjust ment model coefficient estimate s from the GEE procedure (described previously in Chapter 2) to the Ontario patient population, so that the magnitude of effect of each risk f acto r was equal in both populations. To further promote comparability, we also used a patient - level direct standardization process 60 that enabled the calculation of a standardized outcome rate, had the Ontario patients possessed the same risk distribution as the Michigan patients. However, s tandardizing the Ontario patients to the risk distribution of the Michigan patients had little effect on the Ontario outcome, because we were unable to control for the fundamental difference in outcomes caused by the variation in patient length of stay (Figure 3), as well as the resounded differences in risk factor frequencies (Table 39 2) . Despite the Michigan sample having a more favorable risk - standardized in - hospital mortality ra te, this measure is not effective in po rtraying a fair comparison until we can effectively account for LOS. However, risk - standardized 30 - day readmission rates were simil ar between Michigan and Ontario . The longer Ontario LOS is a result of the Canadian health care structure 5 ; the longer stay in hospital is of no financial burden to the patient, or financial incentive to the hospital, so the physician is not obliga ted to discharge a patient earlier than necessary . More intensive short - term care may also be a result of the shorter hospital stay in M ichigan, which was also noted for other cardiovascular diseases 60 - 63 . In this study, t o alleviate th e methodological pitfall of different lengths of stay between Michigan and Ontario , we explored the possibility of replacing in - hospital mortality with a time - specific outcome measure (ex. 3 - day or 7 - day in - hospital mortality), but was not possible since the same LOS disparity was found in the first seven days of h ospital stay after stroke (Figure 4 ), as Michigan and Ontario hospitals exemplified unique d ischarge behaviors. Furthermore, because we are unable to track deaths in the MIDB once a patient was discharged from hospital, the 7 - day in - hospital mortality rate may not be a reliable measure for 7 - day post - stroke case fatality since this data does not capture deaths that occur after discharge. Thus 7 - day in - hospital mortality is likely an underestimation of the true 7 - day case fatality although we believe the difference is likely to be small. Additionally, the underestimation of the true 7 - day post - stro ke case fatality will be different because of the different proportions of patients being discharged to hospice ( 4.2% vs. 0.7%) between Michigan and Ontario . In contrast to in - hospital mortality, 30 - day readmission is not directly confounded by patient len gth of stay, and may be a more suitable outcome to compare between Michigan and Ontario. Interpretation of the similarity in 30 - day readmission rates may also prove challenging however . B ecause of the difference in LOS, a Michigan patient will be discharge d earlier in the course of disease 40 compared to an Ontario patient, and therefore experience le ss days under hospitalized care before being discharged , and thus beginning the 30 - day time interval to which a possible readmission is tracked. Canadians experiencing a longer hospital stay would be discharged later, and therefore the starting point of readmission tracking begins later in the natural history of the disease. As a result, we would expect 30 - day readmission rates to be lower in Ontario. To cre ate a fairer arrival date at the hospital. Aside from the fundamental outcome differences between Michigan and Ontario , t here are other limitations that are w orth noting. Firstly, the use of administrative data has well known limitations when used to assess differences in stroke outcomes . The utilization of a stroke registry to ascertain clinical data would allow the inclusion of stroke severity ( i.e . National Institute s of Health Stroke Sc ale ) in our model, predictive ability . 105 Clinical data may also include vital information such as arrival to hospital by ambulance, door - to - needle (DTN) times, more accurate past medical history, and more detailed information on the processes of care once they arrive at the hospital. Secondly , the Michigan and Ontario stroke outcomes reported may not reflect the remainder of their respective countries. By comparing a cute myocardial infarction outcomes and services utilization in Canada and the United States, Ko et al. demonstrated regional - level disparities among different geographic areas of the United States. 64 A comparison of stroke - related mortality and readmissio n outcomes between other American and Canadian regions has not been conducted , so our outcomes may only be applicable to Michigan and Ontario, and not be representative of their nation as a whole. For example, according to national HCUP data from the Unite d States, the 30 - 41 day all - cause readmission rate for US stroke patients in 2011 was 13.7%. 106 Even though this includes elective readmissions, it is still far off from our 4.5% Michigan figure in Table 2 . In conclusion, we were able to demonstrate a cross - s ystems comparison of stroke outcomes, but lack of comparability hinders our interpretation of hospital - and patient - level risk - standardized outcomes. Comparing the Michigan and Ontario health systems using in - hospital mortality was complicated by the differe nce in LOS after stroke; 30 - day mortality rates would be a better comparative measure. Despite the substantial variation in risk profile (i.e. risk factor frequencies and magnitudes of effect) between Michigan and Ontario stroke patients (as shown i n Tables 2 - 3 ) , we found risk - standardized readmission rates to be similar . The fundamental outcome differences between Michigan , USA and Ontario , Canada will hinder our ability to determine which system produced better outcomes until we can successfully account for the existing sources of variation (i.e. length of st ay and risk factor frequencies). 42 APPENDICES 43 APPENDIX A Tables 44 Table 1 : Comparison of demographic, geographic, and health care characteristics between Michigan, USA and Ontario, Canada (2000 - 201 3 ) Characteristics Michigan, USA Ontario, Canada Demographic Population, Total 9,909,877 13,678,700 65 and Older, % 15.0 15.6 G RP , $ 432,573,000 USD 695 , 705 ,000 CDN Rural Population, % 25.3 14.9 Geographic Land Area, km sq. 146,435 917,741 Health Care Hospitals Total, N 120 238 CSC & RSC, N 3 11 PSC & DSC, N 30 18 Stroke Hospitalizations Total, N 27,719 15,623 Unadjusted Stroke Admission Rate* 28 14.4 Adjusted Stroke Admission Rate* 25.5*** 12.8**** Stroke Mortality Total, N 4,451 4,930 Proportion of All Deaths, % 5 5.5 Adjusted Stroke Mortality Rate** 38.7*** 24*** Abbreviations: G R P - gross regional product, CSC - comprehensive stroke center (Mich.), PSC - primary stroke center (Mich.), RSC - regional stroke center (Ont.), DSC - district stroke center (Ont.) *Per 10,000 population **Per 100,000 population ***Age - adjusted ****Age - and sex - adjusted 45 Table 2 : Regional comparison of patient - and hospital - level characteristics in the final Michigan and Ontario study samples. Characteristics Michigan (N = 47364) Ontario (N = 35648) P - Value Patient - Level Gender, N (%) <0.001 Male 23253 (49.1) 18092 (50.8) Female 24111 (50.9) 17556 (49.2) Age, Mean (Median) 69.5 (71.0) 72.4 (75.0) <0.001 Age Category, N (%) <0.001 18 - 24 180 (0.4) 107 (0.3) 25 - 44 2338 (4.9) 1276 (3.6) 45 - 64 14869 (31.4) 8518 (23.9) 65 - 84 21719 (45.9) 18035 (50.6) 8258 (17.4) 7712 (21.6) LOS, Mean (Median), Days 5.4 (4.0) 12.5 (7.0) <0.001 Stroke Type, N (%) <0.001 Ischemic 38246 (80.8) 22611 (63.4) Subarachnoid Hemorrhage 2178 (4.6) 1966 (5.5) Intracranial Hemorrhage 4450 (9.4) 4092 (11.5) Other Hemorrhage 2459 (5.2) 2288 (6.4) Unable To Determine 31 (0.1) 4691 (13.2) Past Medical History, N (%) AIHD 1879 (4.0) 2713 (7.6) <0.001 Cancer 1599 (3.4) 2923 (8.2) <0.001 CIHD 14537 (30.7) 4915 (13.8) <0.001 Diabetes 16097 (34.0) 9610 (27.0) <0.001 Heart Failure 7589 (16.0) 3094 (8.7) <0.001 Liver Disease 812 (1.7) 418 (1.2) <0.001 Pulmonary Edema 200 (0.4) 124 (0.4) 0.089 Renal Failure 10169 (21.5) 3285 (9.2) <0.001 Shock 353 (0.8) 170 (0.5) <0.001 Hospital - Level Stroke - Certified Center, N (%) 31202 (65.9) 21384 (60.0) <0.001 Teaching Hospital, N (%) 17119 (36.1) 12299 (34.5) <0.001 Hospital Bed Size, Mean (SD) 465.8 (280.1) 289.8 (165.9) <0.001 Outcomes Crude Rate, N (%) In - Hospital Mortality 3617 (7.6) 4987 (14.0) <0.001 30 - Day Readmission 2117 (4.5) 1805 (5.1) <0.001 Abbreviations: LOS - length of stay, AIHD - acute ischemic heart disease, CIHD - chronic ischemic heart disease, SD - standard deviation NOTE: Stroke - certified centers in Michigan include comprehensive and primary stroke centers, and regional and district stroke centers in Ontario 46 Table 3 : Regional comparison of adjusted odds ratios from the hierarchical logistic regression model used to profile hospital performance for in - hospital mortality. Characteristics Michigan (95% CI) P - Value Ontario (95% CI) P - Value Patient - Level Age (Years) 1.02 (1.02 - 1.03) <0.0001 1.04 (1.03 - 1.04) <0.0001 Gender Male REF REF Female 1.00 (0.93 - 1.08) 0.9881 1.05 (0.98 - 1.12) 0.1516 Stroke Type Ischemic REF REF Subarachnoid Hemorrhage 7.63 (6.70 - 8.70) <0.0001 3.96 (3.47 - 4.52) <0.0001 Intracranial Hemorrhage 8.85 (8.09 - 9.69) <0.0001 3.81 (3.50 - 4.15) <0.0001 Other Hemorrhage 4.88 (4.32 - 5.52) <0.0001 2.18 (1.94 - 2.45) <0.0001 Unable To Determine 0.69 (0.09 - 5.22) 0.7182 1.00 (0.90 - 1.11) 0.9671 Past Medical History AIHD No REF REF Yes 2.25 (1.94 - 2.62) <0.0001 1.24 (1.10 - 1.40) 0.0006 Cancer No REF REF Yes 1.41 (1.18 - 1.69) 0.0002 1.72 ( 1.56 - 1.90) <0.0001 CIHD No REF REF Yes 0.92 (0.85 - 1.01) 0.0772 1.00 (0.90 - 1.10) 0.9483 Heart Failure No REF REF Yes 1.38 (1.24 - 1.52) <0.0001 1.78 (1.61 - 1.97) <0.0001 Liver Disease No REF REF Yes 1.32 (1.05 - 1.67) 0.0199 1.63 ( 1.26 - 2.11) 0.0003 Pulmonary Edema No REF REF Yes 2.40 (1.67 - 3.45) <0.0001 1.26 (0.81 - 1.97) 0.2916 Renal Failure No REF REF Yes 1.30 ( 1.19 - 1.42) <0.0001 1.41 (1.28 - 1.56) <0.0001 Shock No REF REF Yes 9.01 (7.05 - 11.52) <0.0001 2.37 ( 1.67 - 3.38) <0.0001 Hospital - Level Stroke - Certified Center No REF REF Yes 1.02 (0.83 - 1.27) 0.83 1.03 (0.89 - 1.20) 0.661 Abbreviations: AIHD - acute ischemic heart disease, CI confidence interval, CIHD - chronic ischemic heart disease, REF - reference 47 Table 4 : Regional comparison of adjusted odds ratios from the hierarchical logistic regression model used to profile hospital p erformance for 30 - day readmission. Characteristics Michigan (95% CI) P - Value Ontario (95% CI) P - Value Patient - Level Age (Years) 1.00 (0.99 - 1.00) 0.1787 1.00 (1.00 - 1.01) 0.0512 Gender Male REF REF Female 0.84 (0.76 - 0.92) 0.0003 1.00 (0.91 - 1.11) 0.9314 Stroke Type Ischemic REF REF Subarachnoid Hemorrhage 0.54 (0.40 - 0.72) <0.0001 0.87 (0.68 - 1.12) 0.2814 Intracranial Hemorrhage 0.55 (0.46 - 0.67) <0.0001 0.75 (0.63 - 0.89) 0.0011 Other Hemorrhage 1.47 (1.24 - 1.76) <0.0001 1.55 ( 1.31 - 1.84) <0.0001 Unable To Determine 0.87 (0.12 - 6.54) 0.8946 1.10 (0.95 - 1.27) 0.1867 Past Medical History AIHD No REF REF Yes 1.51 (1.27 - 1.81) <0.0001 1.33 (1.13 - 1.55) 0.0007 Diabetes No REF REF Yes 1.22 (1.11 - 1.34) <0.0001 1.27 (1.14 - 1.41) <0.0001 Renal Failure No REF REF Yes 1.56 (1.40 - 1.72) <0.0001 1.62 (1.41 - 1.86) <0.0001 Hospital - Level Stroke - Certified Center No REF REF Yes 1.60 (1.17 - 2.19) 0.0036 0.98 (0.87 - 1.10) 0.7449 Abbreviation: AIHD - acute ischemic heart disease, CI confidence interval, REF - reference 48 Table 5 : Fit statistics for all models performed in aims 1 and 2. Region Outcome AIC (Aim 1) QIC (Aim 2) C - Statistic Michigan In - Hospital Mortality 21485.33 21848.63 0.784 30 - Day Readmission 16030.26 17040.4 0.626 Ontario In - Hospital Mortality 26401.56 N/A N/A 30 - Day Readmission 14153.11 N/A N/A 49 Table 6 : List of predicted over expected ratios (in ascending sequence) used to quantify hospital performance for in - hospital mortality. Michigan, USA (N = 78) Ontario, Canada (N = 83) 0.53697 0.91581 1.21586 0.63543 0.9334 1.1069 0.54937 0.91606 1.22258 0.69513 0.93484 1.12166 0.60038 0.92451 1.2463 0.69953 0.94597 1.13595 0.60455 0.95912 1.25339 0.70475 0.95035 1.163 0.609 0.97572 1.33597 0.73836 0.95552 1.1754 0.64701 0.98092 1.39002 0.73981 0.9635 1.17742 0.6577 0.99091 1.41329 0.77114 0.97232 1.18183 0.66042 1.0105 1.45983 0.78168 0.98436 1.19054 0.66651 1.01156 1.46377 0.78825 0.99545 1.22257 0.67282 1.01272 1.47857 0.78888 0.99573 1.22985 0.6972 1.01752 1.51295 0.80435 0.99933 1.23697 0.73448 1.03166 1.51307 0.81535 1.00683 1.23943 0.782 1.04245 1.52404 0.82295 1.00782 1.30554 0.79568 1.04796 1.56347 0.83997 1.01278 1.31526 0.80423 1.05319 1.72592 0.84019 1.01598 1.32057 0.80797 1.05916 1.74282 0.8475 1.01858 1.32884 0.8104 1.07923 1.75209 0.84929 1.02634 1.33087 0.82466 1.08201 1.85382 0.85765 1.02854 1.35825 0.82727 1.10248 0.87147 1.02875 1.36426 0.83001 1.11916 0.88436 1.04129 1.36922 0.83042 1.12547 0.88477 1.04557 1.40873 0.83212 1.13116 0.88945 1.04655 1.42028 0.83469 1.14224 0.88977 1.04842 1.43232 0.84804 1.14953 0.89349 1.06274 0.86832 1.16775 0.89667 1.07702 0.88566 1.16841 0.89914 1.08553 0.88849 1.17547 0.90281 1.09514 0.89916 1.17742 0.90713 1.09782 0.90476 1.19011 0.91458 1.10176 0.90847 1.20849 0.91666 1.10513 NOTE: Shaded values indicate positive and negative outliers (top and bottom 10%, respectively), and bolded values indicate stroke - certified centers. 50 Table 7 : List of predicted over expected ratios (in ascending sequence) used to quantify hospital p erformance for 30 - day readmission. Michigan, USA (N = 78) Ontario, Canada (N = 83) 0.49828 0.84523 1.25527 0.86939 0.98568 1.01747 0.53558 0.85274 1.27221 0.88865 0.98632 1.01779 0.54475 0.89919 1.33097 0.92937 0.98704 1.02049 0.56081 0.8995 1.35442 0.92955 0.98774 1.02268 0.566 0.90552 1.38007 0.93498 0.98791 1.02369 0.57724 0.91193 1.38157 0.94808 0.98879 1.02649 0.59477 0.91845 1.46017 0.94912 0.99068 1.02936 0.59613 0.93472 1.4769 0.95177 0.99284 1.0354 0.63054 0.94012 1.51552 0.95318 0.99338 1.03867 0.63348 0.95814 1.57243 0.95792 0.99412 1.04085 0.67537 0.96268 1.60505 0.96455 0.99447 1.04551 0.68285 0.97037 1.66689 0.96535 0.99558 1.04687 0.69258 0.9885 1.72271 0.96711 0.99714 1.05115 0.69872 0.99024 1.93778 0.96824 0.99837 1.05197 0.72755 0.9978 2.0684 0.972 0.99868 1.06573 0.73323 1.00835 5.22967 0.9729 1.00017 1.06794 0.73886 1.01012 5.42608 0.97321 1.00029 1.07331 0.7577 1.04519 7.71385 0.9733 1.00153 1.0757 0.75848 1.04765 0.97491 1.00364 1.09861 0.76927 1.05554 0.97757 1.0049 1.09943 0.77113 1.05654 0.978 1.00496 1.09961 0.77638 1.09923 0.97815 1.00534 1.1061 0.78923 1.10792 0.97893 1.00671 1.15346 0.79105 1.15418 0.98215 1.00717 0.79668 1.16849 0.98236 1.00975 0.80146 1.16995 0.98244 1.01142 0.81546 1.17389 0.98294 1.01356 0.82417 1.17693 0.98335 1.0136 0.83493 1.23083 0.98381 1.01362 0.84342 1.23254 0.98478 1.01653 NOTE: Shaded values indicate positive and negative outliers (top and bottom 10%, respectively), and bolded values indicate stroke - certified centers. 51 Table 8 : Comparison of adjusted odds ratios for the Michigan sample, generated from the hierarchical logistic regression (HLM) and generalized estimating equations (GEE) methods for in - hospital mortality. Characteristics HLM Method (95% CI) P - Value GEE Method (95% CI) P - Value Patient - Level Age (Years) 1.02 (1.02 - 1.03) < 0.0001 1.02 (1.02 - 1.03) <0.0001 Gender Male REF REF Female 1.00 (0.93 - 1.08) 0.9881 1.00 (0.93 - 1.07) 0.9804 Stroke Type Ischemic REF REF Subarachnoid Hemorrhage 7.63 (6.70 - 8.70) <0.0001 7.27 ( 5.52 - 9.58) <0.0001 Intracranial Hemorrhage 8.85 (8.09 - 9.69) <0.0001 8.36 (7.46 - 9.35) <0.0001 Other Hemorrhage 4.88 (4.32 - 5.52) <0.0001 4.63 (3.81 - 5.63) <0.0001 Unable To Determine 0.69 (0.09 - 5.22) 0.7182 0.64 (0.04 - 9.79) 0.7484 Past Medical History AIHD No REF REF Yes 2.25 (1.94 - 2.62) <0.0001 2.17 (1.87 - 2.51) <0.0001 Cancer No REF REF Yes 1.41 (1.18 - 1.69) 0.0002 1.37 (1.17 - 1.62) 0.0001 CIHD No REF REF Yes 0.92 (0.85 - 1.01) 0.0772 0.93 (0.85 - 1.01) 0.085 Heart Failure No REF REF Yes 1.38 (1.24 - 1.52) < 0.0001 1.35 (1.22 - 1.50) <0.0001 Liver Disease No REF REF Yes 1.32 (1.05 - 1.67) 0.0199 1.31 (1.06 - 1.63) 0.0125 Pulmonary Edema No REF REF Yes 2.40 (1.67 - 3.45) < 0.0001 2.33 (1.66 - 3.27) <0.0001 Renal Failure No REF REF Yes 1.30 (1.19 - 1.42) <0.0001 1.29 (1.16 - 1.43) <0.0001 Shock No REF REF Yes 9.01 (7.05 - 11.52) <0.0001 8.51 (6.18 - 11.71) <0.0001 Hospital - Level Stroke - Certified Center No REF REF Yes 1.02 (0.83 - 1.27) 0.83 0.97 (0.78 - 1.21) 0.7769 Abbreviations: AIHD - acute ischemic heart disease, CIHD - chronic ischemic heart disease, REF - reference 52 Table 9 : Comparison of adjusted odds ratios for the Michigan sample, generated from the hierarchical logistic regression (HLM) and generalized estimating equations (GEE) modeling methods for 30 - day readmission. Characteristics HLM Method (95% CI) P - Value GEE Method (95% CI) P - Value Patient - Level Age 1.00 (0.99 - 1.00) 0.1787 1.00 (0.99 - 1.00) 0.2299 Gender Male REF REF Female 0.84 (0.76 - 0.92) 0.0003 0.85 (0.77 - 0.92) 0.0002 Stroke Type Ischemic REF REF Subarachnoid Hemorrhage 0.54 (0.40 - 0.72) <0.0001 0.56 (0.43 - 0.73) <0.0001 Intracranial Hemorrhage 0.55 (0.46 - 0.67) <0.0001 0.57 (0.48 - 0.68) <0.0001 Other Hemorrhage 1.47 (1.24 - 1.76) <0.0001 1.44 (1.13 - 1.83) 0.0035 Unable To Determine 0.87 (0.12 - 6.54) 0.8946 0.88 (0.17 - 4.69) 0.8853 Past Medical History AIHD No REF REF Yes 1.51 (1.27 - 1.81) <0.0001 1.54 (1.32 - 1.80) <0.0001 Diabetes No REF REF Yes 1.22 (1.11 - 1.34) <0.0001 1.21 (1.11 - 1.32) <0.0001 Renal Failure No REF REF Yes 1.56 (1.40 - 1.72) <0.0001 1.55 (1.33 - 1.81) <0.0001 Hospital - Level Stroke - Certified Center No REF REF Yes 1.60 (1.17 - 2.19) 0.0036 1.82 (1.15 - 2.88) 0.0106 Abbreviation: AIHD - acute ischemic heart disease, REF - reference 53 Table 1 0 : Direct standardization procedure to produce comparable in - hospital mortality rates between Michigan, USA and Ontario, Canada Michigan, USA Ontario, Canada* Risk Decile Range* Study Sample, n (%) Mortality, n (%) Study Sample, n (%) Mortality, n (%) Direct Standardization** Standardize d Ontario Rate, % X <= 0.024594 4787 (10.1) 79 (1.7) 3476 (9.8) 119 (3.4) 0.101 x 3.4 0.35 0.024594 < X <= 0.029453 4687 (9.9) 81 (1.7) 3137 (8.8) 181 (5.8) 0.099 x 5.8 0.57 0.029453 < X <= 0.034083 4779 (10.1) 103 (2.2) 3270 (9.2) 216 (6.6) 0.101 x 6.6 0.67 0.034083 < X <= 0.039126 4680 (9.9) 140 (3.0) 3154 (8.9) 276 (8.8) 0.099 x 8.8 0.87 0.039126 < X <= 0.044446 4680 (9.9) 160 (3.4) 3491 (9.8) 328 (9.4) 0.099 x 9.4 0.93 0.044446 < X <= 0.050790 4818 (10.2) 204 (4.2) 3871 (10.9) 507 (13.1) 0.102 x 13.1 1.34 0.050790 < X <= 0.061373 4718 (10.0) 264 (5.6) 3151 (8.8) 532 (16.9) 0.100 x 16.9 1.69 0.061373 < X <= 0.10586 4743 (10.0) 383 (8.1) 3242 (9.1) 647 (20.0) 0.100 x 20.0 2 0.10586 < X <= 0.20310 4735 (10.0) 822 (17.4) 4412 (12.4) 802 (18.2) 0.100 x 18.2 1.82 X > 0.20310 4737 (10.0) 1381 (29.2) 4444 (12.5) 1379 (31.0) 0.100 x 31.0 3.1 Total 47364 (100) 3617 (7.6) 35648 (100) 4987 (14.0) 13.32 Standardized Mortality Rate, % (95% CI****) 7.6 (7.4 - 7.9) 13.3 (13.0 - 13.7)*** *Patient risk was ascertained from the patient - level predicted probability of experiencing the outcome. The Michigan risk distribution was used to calculate the deciles cut points; after using Michigan coefficients to produce Ontario patient probabilities, Ontario patients were also nested in the risk deciles. **Calculated by multiplying the proportion of Michigan patients in a specific stratum with the Ontario observed mortality rat e in the same strata. ***Standardized to the Michigan patient risk distribution. - P))/N], P = proportion, N = total sample size 54 Table 1 1 : Direct standardization procedure to produce comparable 30 - day all readmission rates between Michigan, USA and Ontario, Canada Michigan, USA Ontario, Canada * Risk Decile Range* Study Sample, n (%) Readmission, n (%) Study Sample, n (%) Readmission, n (%) Direct Standardization** Standardized Ontario Rate, % X <= 0.023337 4697 (9.9) 122 (2.6) 5535 (15.5) 244 (4.4) 0.099 x 4.4 0.44 0.023337 < X <= 0.027146 4791 (10.1) 139 (2.9) 4825 (13.5) 225 (4.7) 0.101 x 4.7 0.47 0.027146 < X <= 0.030975 4721 (10.0) 157 (3.3) 3693 (10.4) 138 (3.7) 0.100 x 3.7 0.37 0.030975 < X <= 0.039575 4865 (10.3) 151 (3.1) 4433 (12.4) 238 (5.4) 0.103 x 5.4 0.55 0.039575 < X <= 0.041792 4634 (9.8) 171 (3.7) 3225 (9.1) 153 (4.7) 0.098 x 4.7 0.46 0.041792 < X <= 0.047662 4641 (9.8) 197 (4.2) 3766 (10.6) 185 (4.9) 0.098 x 4.9 0.48 0.047662 < X <= 0.049714 4877 (10.3) 207 (4.2) 2829 (7.9) 113 (4.0) 0.103 x 4.0 0.41 0.049714 < X <= 0.058168 4639 (9.8) 223 (4.8) 2874 (8.1) 161 (5.6) 0.098 x 5.6 0.55 0.058168 < X <= 0.071354 4780 (10.1) 274 (5.7) 2715 (7.6) 192 (7.1) 0.101 x 7.1 0.71 X > 0.071354 4719 (10.0) 476 (10.1) 1753 (4.9) 156 (8.9) 0.100 x 8.9 0.89 Total 47364 (100) 2117 (4.5) 35648 (100) 1805 (5.1) 5.34 Standardized Readmission Rate, % (95% CI****) 4.5 (4.3 - 4.7) 5.3 (5.1 - 5.6)*** * Patient risk was ascertained from the patient - level predicted probability of experiencing the outcome. The Michigan risk distribution was used to calculate the deciles cut points; after using Michigan coefficients to produce Ontario patient proba bilities, Ontario patients were also nested in the risk deciles. **Calculated by multiplying the proportion of Michigan patients in a specific stratum with the Ontario observed mortality rate in the same stratum. ***Standardized to the Michigan patient risk distribution. - P))/N], P = proportion, N = total sample size 55 APPENDIX B Figures 56 57 58 59 60 61 62 63 64 APPENDIX C Supplemental Tables 65 Table S1 : Description of codes used in Michigan (ICD - 9) and Ontario (ICD - 10) to identify strokes. Michigan (ICD - 9) Ontario (ICD - 10) Description 430 I60 Subarachnoid hemorrhage 431 I61 Intracerebral hemorrhage 432 I62 Other hemorrhage 433, 434 I63 Ischemic stroke 436 I64 Unable to determine NOTE: ICD - 9 and ICD - 10 code descriptions were referenced from the Centers for Medicare and Medicaid Services (CMS) and Institute for Clinical Evaluative Sciences (ICES), respectively. 66 Table S2 : List of diagnosis codes used to identify past medical history in the respective discharge databases in Michigan and Ontario. Past Medical History Michigan, USA (ICD - 9) Ontario, Canada (ICD - 10) Comparability Ratios AIHD 410 - 411.99, 413 - 413.99 I20, I21, I22, I24 0.99 (all ischemic heart disease) Cancer 140 - 202.99, V580, V581 C00 - C26, C30 - C44, C45 - C97, Z51.0, Z51.1 1.01 CIHD 414.0 - 414.19, 414.8 - 414.99, 412 - 412.99, 429.2 - 429.29 I25 0.99 (all ischemic heart disease) Diabetes 250 - 250.99 E10.0 - E10.7, E11.0 - E11.7, E13.0 - E13.7, E14.0 - E14.7 1.01 Heart Failure 428 - 428.99 I50 1.04 Liver Disease 456.0 - 456.19, 571 - 571.99, 572.8 - 572.89, 573 - 574.99, V427 B16.1, B16.9, B17, B18, I85, K70.0, K70.2, K70.3, K70.4, K70.9, K72.1, K72.9, K73, K74, K76.0, K76.6, Z94.4 1.04 Pulmonary Edema 514 - 514.99, 518.4 - 518.49 J81 1.12 Renal Failure 584.5 - 584.99, 585 - 586.99 N17, N18, N19 1.30 Shock 785.5 - 785.59 R57 1.19 Abbreviation: AIHD - acute ischemic heart disease, CIHD - chronic ischemic heart disease NOTE: Comparability ratios from reference 83. 67 Table S3 : Description of ICD - 9 procedure codes used to identify elective readmissions in the MIDB. ICD - 9 Code Code Description 0061 Percutaneous angioplasty of extracranial vessel(s) 0063 Percutaneous insertion of carotid artery stent(s) 380.2 Incision of vessel, other vessels of head and neck 381.2 Endarterectomy, other vessels of head and neck 383 Resection of vessel with anastomosis, any site 384.1 Resection of vessel with replacement, intracranial vessels 384.2 Resection of vessel with replacement, other vessels of head and neck NOTE: ICD - 9 code descriptions were referenced from the Centers for Medicare and Medicaid Services (CMS); http://www.cms.gov/Medicare/Coding/ICD9ProviderDiagnosticCodes/codes.html 68 APPENDIX D IRB Approval Letter 69 70 REFERENCES 71 R EFERENCES 1. Rice T, Rosenau P, Unruh LY, et al. United States of America: Health system review. Health Systems in Transition . 2013; 15(3): 1 - 431. 2. Hoffman Jr. ED, Klees BS, Curtis CA. Overview of the Medicare and Medicaid Programs. Health Care Financing Review . 2000; 22: 175 - 193. 3. Kaiser Family Foundation. Medicaid: A Primer , June 2010. Washington, DC: Kaiser Family Foundation. http://www.kff.ord/medicaid/upload/7334 - 04.pdf (Kaiser Family Foundation, 2013c). 4. Centers for Medicare and Medicaid Services. NHE Fact S heet. 2014. http://www.cms.gov/Research - Statistics - Data - and - Systems/Statistics - Trends - and - Reports/NationalHea lthExpendData/NHE - Fact - Sheet.html . Last Accessed May 16, 2015. 5. Marchildon GP. Canada: Health system review. Health Systems in Transition . 2013; 15(1): 1 - 179. 6. US Department of Health & Human Services. Facts & Features. Open Enrollment Week 13: February 7, 2015 February 15, 2015. 2015. http://www.hhs.gov/healthcare/facts/blog/2015/02/open - enrollment - week - thirte en.html . Last Accessed June 1, 2015. 7. Smith JC, Medalla C. US Census Bureau, Current Population Reports, P60 - 250, Health Insurance Coverage in the United States: 2013 , US Government Printing Office, Washington, DC, 2014. 8. Kaiser Family Foundation. State Health Facts. Total Monthly Medicaid and CHIP Enrollment. http://kff.org/health - reform/state - indicator/total - monthly - medicaid - and - chip - enrollment/ . Las t Accessed June 2, 2015. 9. Rice T, Unruh LY, Rosenau P, et al. Challenges facing the United States of America in implementing universal coverage. Bull World Health Organ . 2014; 92: 894 - 902. 10. Pylypchuk Y, Sarpong EM. Comparison of Health Care Utilization: Un ited States versus Canada. Health Serv Res . 2013; 48: 560 - 581. 72 11. Welch WP, Verrilli D, Katz SJ, et al. A Detailed Comparison of Physician Services for the Elderly in the United States and Canada. JAMA . 1996; 275: 1410 - 1416. 12. Katz SJ, McMahon LF, Manning WG. Comparing the Use of Diagnostic Tests in Canadian and US Hospitals. Medical Care . 1996; 34(2): 117 - 125. 13. Li G, Lau JT, McCarthy ML, et al. Emergency Department Utilization in the United States and Ontario, Canada. Academic Emergency Medicine . 2007; 14: 582 - 584. 14. Moses III H, Matheson DM, Dorsey ER, et al. The Anatomy of Health Care in the United States. JAMA . 2013; 310(18): 1947 - 1963. 15. Truffer CJ, Keehan S, Smith S, et al. Health Spending Projections Through 2019. Health Af fairs . 2010; 29: 522 - 529. 16. Canadian Institute for Health Information. Spending and Health Workforce. National Health Care Expenditure Trends, 1975 - 2014. 2014. http://www.cihi.ca/w eb/resource/en/nhex_2014_report_en.pdf . Last Accessed: May 15, 2015. 17. Mozaffarian D, Benjamin EJ, Go AS, et al. Heart Disease and Stroke Statistics - 2015 Update: A Report From the American Heart Association. Circulation . 2015; 131: e29 - e322. 18. Public Health Agency of Canada. Tracking Heart Disease and Stroke in Canada: Stroke Highlights 2011. 2011; http://www.phac - aspc.gc.ca/cd - mc/cvd - mcv/sh - fs - 2011/p df/StrokeHighlights_EN.pdf . Last Accessed April 14, 2015. 19. Statistics Canada. Leading causes of death, by sex (Both sexes) 2011. CANSIM table 102 - 0561. http://www.s tatcan.gc.ca/tables - tableaux/sum - som/l01/cst01/hlth36a - eng.htm . Last Accessed June 14, 2015. 20. CDC/NCHS, National Vital Statistics System, Mortality 2011. LCWK9. Deaths, percent of total deaths, and death rates for the 15 leading causes of death: United St ates and each state, 2011. http://www.cdc.gov/nchs/data/dvs/LCWK9_2011.pdf . Last Accessed June 14, 2015. 21. Statistics Canada. Trends in mortality rates, 2000 to 2011. Chart 2: Age - standardized mortality rates for the 10 leading causes of death, Canada, 2000 and 2011. 73 http://www. statcan.gc.ca/pub/82 - 625 - x/2014001/article/11897 - eng.htm . Last Accessed July 1, 2015. 22. Heart and Stroke Foundation. There Is Life After Stroke. 2013 Stroke Report; http://www.heartandstroke.com/atf/cf/{99452D8B - E7F1 - 4BD6 - A57D - B136CE6C95BF}/StrokeReport2013_ENG.pdf . Last Accessed April 15, 2015. 23. Ovbiagele B, Goldstein LB, Higashida RT, et al. Forecasting the Future of Stroke in the Unite d States: A Policy Statement From the American Heart Association and American Stroke Association. Stroke . 2013; 44: 2361 - 2375. 24. Hall MJ, Levant S, DeFrances CJ. Hospitalization for stroke in US hospitals, 1989 - 2009. NCHS data brief, no 95. Hyattsville, MD : National Center for Health Statistics. 2012. 25. Public Health Agency of Canada. Tracking Heart Disease and Stroke in Canada. Stroke Highlights 2011. 2011. http://www.phac - aspc.gc.ca/cd - mc/cvd - mcv/sh - fs - 2011/pdf/StrokeHighlights_EN.pdf . Last Accessed: May 17, 2015. 26. Norrving B, Adams RJ. Organized Stroke Care. Stroke . 2006; 37: 326 - 328. 27. Cramer SC, Stradling D, Brown DM, et al. Organization of a United State County System for Comprehensive Acute Stroke Care. Stroke . 2012; 43: 1089 - 1093. 28. Alberts MJ, Hademenos G, Latchaw RE, et al. Recommendations for the Establishment of Primary Stroke Centers. JAMA . 2000; 283: 3102 - 3109. 29. LaBresh KA, Reeves MJ, Frankel MR, et al. Hospital Treatment of Patients With Program. Arch Intern Med . 2008; 168(4): 411 - 417. 30. Fonarow GC, Reeves MJ, Smith EE, et al. Charact eristics, Performance Measures, and In - Hospital Outcomes of the First One Million Stroke and Transient Ischemic Attach Admissions in Get With The Guidelines - Stroke. Circ Cardiovasc Qual Outcomes . 2010; 3: 291 - 302. 31. Schwamm LH, Pancioli A, Acker JE, et al. Recommendations for the Establishment of Task Force on the Development of Stroke Systems. Stroke . 2005; 36: 690 - 703. 74 32. The Joint Commissi on. Facts about Joint Commission stroke certification. 2015. http://www.jointcommission.org/facts_about_joint_commission_stroke_certification/defa ult.aspx . Last Accessed May 17, 2015. 33. Alberts MJ, Latchaw RE, Selman WR, et al. Recommendations for Comprehensive Stroke Centers: A Consensus Statement From the Brain Attack Coalition. Stroke . 2005; 36: 1597 - 1618. 34. Mullen MT, Wiebe DJ, Bowman A, et al. D isparities in Accessibility of Certified Primary Stroke Centers. Stroke . 2014; 45: 3381 - 3388. 35. The Joint Commission. Stroke Certification Programs. 2015; http://www.qualitycheck.org/StrokeCertificationList.aspx . Last Accessed April 8, 2015. 36. The Joint Commission. Preparation Essentials for NEW Acute Stroke Ready Hospitals (ASRH) Certification. 2015. http://www.jointcommission.org/assets/1/6/ASRH_Webinar_Mar3120151.PDF . Last Accessed: May 16, 2015. 37. Wilson E, Taylor G, Phillips S, et al. Creating a Canadian stroke system. CMAJ . 2001; 164(13): 1853 - 1855. 38. Heart and Stroke Fou ndation of Ontario. A Guide to Organizing Acute Stroke Care: Coordinated Stroke Strategy. 2001; htt p://www.heartandstroke.on.ca/atf/cf/{33C6FA68 - B56B - 4760 - ABC6 - D85B2D02EE71}/GuidetoOrganizingAcuteStrokeCareManual2001Final[1].pdf ; Last Accessed April 15, 2015. 39. Ontario Stroke Network. The Ontario Stroke System. 2015; http://ontariostrokenetwork.ca/about - the - osn/ontario - stroke - system - oss/ . Last Accessed April 8, 2015. 40. Kapral MK, Fang J, Silver FL, et al. Effect of a provincial system of stroke care delivery on stroke care and outcomes. CMAJ . 2013; 185: E483 - E491. 41. Webster F, Saposnik G, Kapral MK, et al. Organization Outpatient Care: Stroke Prevention Clinic Referrals Are Associated With Reduced Mortality After Transient Ischemic Attack and Ischemic Stroke. Stroke . 2011; 42: 3176 - 3182. 42. United States Census Bureau. State and County Quick Facts: Michigan. http://quickfacts.census.gov/qfd/states/26000.html . Last Accessed April 9, 2015. 75 43. Statistics Canada. Focus on Geography Series, 2011 Census: Province of Ontario. http://www12.statcan.gc.ca/census - recensement/2011/as - sa/fogs - spg/Facts - pr - eng.cfm?Lang=eng&GC=35 . Last Accessed April 10, 2015. 44. Michigan Health and Hospital Association. Michigan Community Hospitals and Health Systems. http://www.mha.org/documents/nonprofit_hospitals.pdf . Last Accessed April 19, 2015. 45. Ontario Mi nistry of Health and Long - Term Care. Hospital Locations and Classifications by LHINs. http://www.health.gov.on.ca/en/common/system/services/hosp/locations.aspx . Last Accessed April 18, 2015. 46. Michigan Department of Community Health. Impact of Heart Disease and Stroke in Michigan: 2008 Report on Surveillance. http://www.michigan.g ov/documents/mdch/Impact_complete_report_245958_7.pdf . Last Accessed April 19, 2015. 47. Michigan Department of Community Health. 2011 Stroke Brief. https://www.michigan.gov/documents/mdch/Stroke_Brief.pub_346836_7.pdf . Last Accessed June 2, 2015. 48. Hall R, Khan F, O for Stroke Prevention and Care. Toronto, ON: Institute for Clinical Evaluative Sciences; 2014. 49. Statistics Canada. Tables by province or territory: Ontario. 201 3 . http://www.statcan.gc.ca/tables - tableaux/sum - som/l01/cst01/econ15 - eng. htm . Last Accessed May 16, 2015. 50. Ontario Ministry of Health and Long - Term Care. Hospitals. Questions And Answers. http://www.health.gov.on.ca/en/common/system/services/hosp/faq.aspx#hospitals . Last Accessed May 15, 2015. 51. Kaiser Family Foundation . State Health Facts. Total Gross State Product. http://kff.org/other/state - indicator/total - gross - state - product/ . Last Accessed July 28, 2015. 52. Michigan Census 2000. Urban and Rural Population for Michigan. http://www.michigan.gov/documents/urban_rural_42109_7.pdf . Last Accessed May 21, 2015. 76 53. Ontario Ministry of Finance, Office of Economic Policy. Economic & Revenue Forecasting & Analysis Branch. Ontario Fact Sheet. May 2015. http://www.fin.gov.on.ca/en/economy/ecupdates/factsheet.pdf . Last Accessed June 7, 2015. 54. Statistics Canada. Table 102 - 0563 Leading causes of death, total population, by sex, Canada, provinces and territories, annual, CANSIM (database). http://www5.statcan.gc.ca/cansim/a26?lang=eng&retrLang=eng&id=1020563&paSer=& pattern=&stByVal=1&p1=1&p2=49&tabMode=dataTable&csid = . Last Accessed July 2, 2015. 55. 2011 Geocoded Michigan Death Certificate Registry. Divis ion of Vital Records & Health Statistics, Michigan Department of Health & Human Services; Population Estimate (latest update 9/2014), National Center for Health Statistics. http:// www.mdch.state.mi.us/pha/osr/deaths/causrankcnty.asp . Last Accessed July 2, 2015 56. Tremblay MS, Connor Gorber S. Canadian health measures survey: brief overview. Can J Public Health . 2007; 98(6): 453 - 456. 57. Centers for Disease Control and Prevention. Nationa l Center for Health Statistics. National Health and Nutrition Examination Survey, 2013 - 2014: Overview. http://www.cdc.gov/nchs/data/nhanes/nhanes_13_14/2013 - 14_o verview_brochure.pdf . Last Accessed: June 17, 2015. 58. Faraco G, Iadecola C. Hypertension: A Harbinger of Stroke and Dementia. Hypertension . 2013; 62: 810 - 817. 59. Robitaille C, Dai S, Waters C, et al. Diagnosed hypertension in Canada: incidence, prevalence, and associated mortality. CMAJ . 2012; 184: E49 - E56. 60. Morbidity and Mortality Weekly Report (MMWR). Vital Signs: Prevalence, Treatment, and Control of Hypertension Unites States, 1999 - 2002 and 2005 - 2008. Weekly . 2011; 60(04); 103 - 108. 61. McAlister FA, Robitaille C, Gillespie C, et al. The impact of Cardiovascular Risk - Factor Profiles on Blood Pressure Control Rates in Adults From Canada and the United States. Canadian Journal of Cardiology . 2013; 29: 598 - 605. 77 62. Kaul P, Reed SD, Hernandez AF , et al. Differences in treatment, outcomes, and quality of life among patients with heart failure in Canada and the United States. JACC Heart Fail . 2013; 1(6): 523 - 530. 63. Ko DT, Tu JV, Masoudi FA, et al. Quality of Care and Outcomes of Older Patients With Heart Failure Hospitalized in the United States and Canada. Arch Intern Med . 2005; 165: 2486 - 2492. 64. Ko DT, Krumholz HM, Wang Y, et al. Regional Differences in Process of Care and Outcomes for Older Acute Myocardial Infraction Patients in the United States and Ontario, Canada. Circulation . 2007; 115: 196 - 203. 65. Kaul P, Armstrong PW, Chang WC et al. Long - Term Mortality of Patients with Acute Myocardial Infarction in the United States and Canada: Comparison of Patients Enrolled in Global Utilization of Strepto kinase and t - PA for Occluded Coronary Arteries (GUSTO) - I. Circulation . 2004; 110: 1754 - 1760. 66. Tu JV, Pashos CL, Naylor CD, et al. Use of Cardiac Procedures and Outcomes in Elderly Patients with Myocardial Infarction in the United States and Canada. N Engl J Med . 1997; 336: 1500 - 1505. 67. Mehta RH, Kaul P, Lopes RD, et al. Variations in practice and outcomes in patients undergoing primary percutaneous coronary intervention in the United States and Canada: insights from the Assessment of Pexelizumab in A cute Myocardial Infarction (APEX AMI) trial. Am Heart J . 2012; 163(5): 797 - 803. 68. Statistics Canada. Health Statistics Division. November 2012. Catalogue no. 82 - 625 - X. Health Fact Sheet; Cholesterol levels of Canadians, 2009 to 2011. http://publications.gc.ca/collections/collection_2013/statcan/82 - 625 - x/82 - 625 - 2012001 - 9 - eng.pdf . Last Accessed June 18, 2015. 69. Muntner P, Levitan E, Brown TM, et al. Trends in the Prevalence, Awareness, Treatment and Control of High Low Density Lipoprotein - Cholesterol among US Adults from 1999 - 2000 through 2009 - 2010. Am J Cardiol . 2013; 112(5): 664 - 670. 70. National Center for Health Statistics. Health, United States, 201 4: With Special Feature on Adults Aged 55 - 64. Hyattsville, MD. 2015. 71. Statistics Canada. CANSIM, table 105 - 0501 and Catalouge no. 82 - 221 - X. Smokers, by sex, provinces and territories (Percent). http://www.statcan.gc.ca/tables - tableaux/sum - som/l01/cst01/health74b - eng.htm . Last Accessed: June 16, 2015. 78 72. Morbidity and Mortality Weekly Report (MMWR). Current Cigarette Smoking Among Adults United States, 2005 - 2013. Weekly . 2014; 63(47): 1108 - 1112. 73. The GUSTO Investigators. An international randomized trial comparing four thrombolytic strategies for acute myocardial infarction. N Engl J Med . 1993; 329: 673 - 682. 74. Haley EC, Kassell NE, Apperson - Hansen C, et al. A randomized, double - blind, vehicle - controlled trial of tirilazad mesylate in patients with aneurysmal subarachnoid hemorrhage: a cooperative study in North America. J Neurosurg . 1997; 86: 467 - 474. 75. Saltman A, Rosamon WD, Fang J, et al. Differences in stroke care delivery in North Carolina and Ontario. Accepted for oral presentation at the 2011 International Stroke Conference; Los Angeles, California . February 11, 2011. 76. Healthcare Cost and Utilization Projec t. Overview of the State Inpatient Database. 2014; https://www.hcup - us.ahrq.gov/sidoverview.jsp#about . Last Accessed April 7, 2015. 77. Institute for Clinical Evaluative Sciences. Data Dictionary, Canadian Institute for Health Information Discharge Abstract Database. 2015; https://data dictionary.ices.on.ca/Applications/DataDictionary/Library.aspx?Library=CIHI . Last Accessed April 7, 2015. 78. Michigan Health and Hospital Association. MHA Data Submission Guide. 2003. http://theidsonline.com/documents/MHA%20Data%20Submission%20Guide.pdf . Last Accessed May 10, 2015. 79. Michigan Department of Community Health (MDCH) Division For Vital Records and Health Statistics (DVRHS). Michigan Inpatient Database. 2012; I nternal MDCH Report. 80. Canadian Institute for Health Information. Discharge Abstract Database (DAD) Metadata. 2014; http://www.cihi.c a/cihi - ext - portal/internet/en/document/types+of+care/hospital+care/acute+care/dad_metadata . Last Accessed April 7, 2015. 81. Canadian Institute for Health Information. Data Quality Documentation, Discharge Abstract Database Current - Year Information, 2013 - 20 14; http://www.cihi.ca/CIHI - ext - portal/pdf/internet/DAD_DATA_QUALITY_13_14_EN . Last Accessed April 8, 2015. 79 82. Institute for Clinical Evaluative Sciences. Canadian Inst itute for Health Information Discharge Abstract Database: A Validation Study. Enhancing the effectiveness of health care for Ontarians through research . 2006; http://www.ices.on.ca/~/media/Files/Atlases - Reports/2006/CIHI - DAD - a - validation - study/Full%20report.ashx . Last Accessed April 8, 2015. 83. Centers for Disease Control and Prevention. National Vital Statistics Reports. Comparability of Cause of Death Between ICD - 9 and ICD - 10: Preliminary Estimates. 2001. http://www.cdc.gov/ nchs/data/nvsr/nvsr49/nvsr49_02.pdf . Last Accessed: May 9, 2015. 84. Statistics Canada. Table 4. Bridge - coding of 1999 deaths: ICD - 10/ICD - 9 comparability ratios. http://www.statcan.gc.ca/pub/84 - 548 - x/2005001/t/4158977 - eng.htm . Last Accessed: June 15, 2015. 85. Canadian Institute for Health Information. Canadian Hospital Reporting Project Technical Notes - Clinical Indicators. 2013; http://publications.gc.ca/collections/collection_2013/icis - cihi/H118 - 86 - 1 - 2013 - eng.pdf . Last Accessed April 8, 2015. 86. Cumbler E, Wald H, Deepak BL, et al. Quality of Care and Outcomes for In - Hospital Ischemic Stroke: Findings from the Get With The Guidelines - Stroke. Stroke . 2014; 45: 231 - 238. 87. Holloway RG, Arnold RM, Creutzfeldt CJ, et al. Palliative Care and End - of - Life Care in Stroke A Statement for Healthcare Professionals From the Ameri can Heart Association/American Stroke Association. Stroke . 2014; 45: 1887 - 1916. 88. Bernheim S, Wang C, Wang Y, et al. Hospital 30 - Day Mortality Following Acute Ischemic Stroke Hospitalization Measures: Methodology Report. Centers for Medicare & Medicaid Services, 2010. 89. Johansen HL, Wielgosz AT, Nguyen K, et al. Incidence, comorbidity, case fatality, and readmission of hospitalized stroke patients in Canada. Can J Cardiol . 2006; 22(1): 65 - 71. 90. Nickles A, Fiedler J, Roberts S, et al. Compliance With the S troke Education Performance Measure in the Michigan Paul Coverdell National Acute Stroke Registry. Stroke . 2013; 44: 1459 - 1462. 91. Staples JA, Thiruchelvam D, Redelmeier DA. Site of hospital readmission and mortality: a population based - based retrospective cohort study. CMAJ . 2014; 2: E77 - E85. 80 92. Krumholz HM, Brindis RG, Brush JE, et al. Standards for Statistical Models Used for Public Reporting of Health Outcomes. Circulation . 2006; 113: 456 - 462. 93. Shahian DM, He X, Jacobs JP, et al. Issues in Quality Measurement: Target Population, Risk Adjustment, and Ratings. Ann Thorac Surg . 2013; 96: 718 - 726. 94. Ding YY. Risk Adjustment: Towards Achieving Meaningful Comparison of Health Outcomes in the Real World. Ann Acad Med Singapore . 2009; 38: 552 - 558. 95. Katzan IL, Spertus J, Bettger JP, et al. Risk Adjustment of Ischemic Stroke Outcomes for Comparing Hospital Performance: A Statement for Healthcare Professionals From the American Heart Association/American Stroke Association. Stroke . 2014; 45: 00 - 00. 96. Can adian Institute for Health Information. Canadian Hospital Reporting Project Clinical Indicators Risk - Adjustment Tables. 2013. 97. Hosmer DW, Lemeshow S, Sturdivant RX, et al. Applied Logistic Regression. Third Edition. Hoboken, NJ: John Wiley & Sons, Inc., 2013. 98. Cohen ME, Dimick JB, Bilimoria KY, et al. Risk Adjustment in the American College of Surgeons National Surgical Quality Improvement Program: A Comparison of Logistic Versus Hierarchical Modeling. J Am Coll Surg . 2009; 209: 687 - 693. 99. Normand SL, Shahian DM. Statistical and Clinical Aspects of Hospital Outcomes Profiling. Statistical Science . 2007; 22(2): 206 - 226. 100. Bouwmeester W, Twisk JW, Kappen TH, et al. Prediction models for clustered data: comparison of a random intercept and stand ard regression model. BMC Medical Research Methodology . 2013; 13: 19. 101. Naing NN. Easy Way to Learn Standardization: Direct and Indirect Methods. Malays J Med Sci . 2000; 7(1): 10 - 15. 102. Warren JL, Barbera L, Bremner KE, et al. End - of - life Care for Lung Cance r Patients in the United States and Ontario. J Natl Cancer Inst . 2011; 103(11): 853 - 862. 103. Department of Health and Human Services. Centers for Medicare and Medicaid. CMS Manual System. Pub 100 - 04 Medicare Claims Processing. 2006. https://www.cms.gov/Regulations - and - 81 Guidance/Guidance/Transmittals/Downloads/R1104CP.pdf . Last Accessed: July 28, 2015. 104. Institute for Clinical Evaluative Sciences. DAD Abstracting Manual. Field 05: Admit Category. Internal Report. 2013. 105. Fonarow GC, Pan W, Saver JL, et al. Comparison of 30 - day mortality models for profiling hospital performance in acute ischemic stroke with vs. without adjustme nt for stroke severity. JAMA . 2012; 308(3): 257 - 264. 106. US Department of Health & Human Services. Agency for Healthcare Research and Quality. All patient readmissions within 30 days. National Statistics, 2011. http://hcupnet.ahrq.gov/HCUPnet.jsp . Last Accessed July 3, 2015.