Direct Costs Of Hip Fractures Among Seniors In Ontario

Transcription

1 Direct Costs Of Hip Fractures Among Seniors In Ontario by Milica Nikitovic A thesis submitted in conformity with the requirements for the degree of Master of Science Graduate Department of Pharmaceutical Sciences University of Toronto Copyright by Milica Nikitovic 2011

2 Abstract Direct Costs of Hip Fractures Among Seniors in Ontario Milica Nikitovic Master of Science Graduate Department of Pharmaceutical Sciences University of Toronto 2011 Osteoporosis is a major public health problem resulting in substantial hip fracture related morbidity. Using healthcare utilization data, we determined the 1- and 2-year direct attributable healthcare costs associated with hip fractures among Ontario seniors in comparison to a matched non-hip fracture cohort. Over a four-year period ( ) we identified 22,418 females and 7,611 males with an incident hip fracture. Approximately 22% of females and 30% of males died in the first year after fracture. The mean attributable cost in the first year was $36,929 ($52,232 vs. $15,503) among females and $39,479 ($54,289 vs. $14,810) among males. Primary cost drivers included acute hospitalizations, complex continuing care, and rehabilitation. Attributable costs remained elevated into the second year, particularly among those who survived the first year ($9,017 females and $10,347 males). Results from this study will aid policy decision makers in allocating healthcare resources and help feed into future health economic analyses. ii

3 Acknowledgements This thesis would not have been possible without the help and support of numerous individuals. First and foremost, I would like to thank my supervisor, Dr. Suzanne Cadarette, for her being there to provide not only your advice and expertise related to my thesis but for always encouraging me to set my goals high and to strive to be a better researcher. To my committee members, Dr. Walter Wodchis and Dr. Murray Krahn, thank you for all of your advice and insightful comments. Your collective expertise was invaluable to this thesis, and your questions taught me to think beyond the details of my work. In addition, I would also like to thank the Institute for Clinical Evaluative Sciences (ICES). In particular, I would like to thank Mike Manno, Brandon Zagorski, Nadia Gunraj and Ximena Camacho for assisting with my data and for helping me navigate the world of administrative data. To my colleagues: Andrea Burden, Mary Elias and Mina Tadrous thank you for always being there to provide feedback with all aspects of my work. Most importantly, thank you for keeping students, and now friends, to work with. I would like to thank those who helped fund and support my graduate research: the Metropolitan Toronto Pharmacists Association, the Toronto Health Economics and Technology Assessment (THETA) Collaborative as well as the Canadian Institutes for Health Research (CIHR) and Osteoporosis Canada. Last, but certainly not least, to my family and friends. To my parents, thank you for continuously providing me with loving support and encouragement throughout all aspects of my academic pursuits. I would not have been able to be where I am today without you. Milica Nikitovic iii

4 Table of Contents Overview... 1 Background... 1 Objectives... 1 Thesis Organization... 3 Chapter Background Osteoporosis Osteoporotic Fractures Hip Fractures Why Are Cost of Illness Studies Important? Chapter Summary... 7 Chapter Literature Review Cost of Hip Fractures in Canada Methods of Hip Fracture Cost Estimation Administrative Data for Cost Estimation Administrative Data to Estimate Hip Fracture Costs Summary of Literature Review Study Objectives Primary Objectives Secondary Objectives Chapter Methods Data Sources Study Design Hip Fracture Cohort Identification Inclusion Criteria Exclusion Criteria Non-Hip Fracture Cohort Selection Matching Hip Fracture Cohort to Non-fracture Cohort Baseline Characteristics iv

5 3.5 Health Outcomes Resource Utilization Processes and Costing Inpatient Hospitalizations Complex Continuing Care Emergency Room and Same Day Surgery Physician Services Rehabilitation Home Care Long Term Care Prescription Drugs Statistical Analysis Descriptive Statistics Fracture Related Costs Total and Attributable Fracture Related Costs Subgroup Fracture Related Costs Survivors and Death Data Access and Ethics Approval Chapter Results Cohort Selection Hip Fracture Cohort Non-Hip Fracture Cohort Cohort Matching Baseline Characteristics Outcomes post index date Second Hip Fracture Death after index date LTC versus community residence at baseline Time to Death after Hip Fracture Health Services Utilization Year One Acute Hospitalizations Emergency Department Visits v

6 Same Day Surgeries Physician Services Complex Continuing Care Rehabilitation Long Term Care Home Care Prescription Medications Year Two Healthcare Costs Total and Attributable Costs (year 1) Attributable Costs by Resource Use Total and attributable costs (year 2) Cost Stratifications LTC Status at baseline and follow-up Second Hip Fracture Survival Status Prior non-hip fractures Sensitivity Analysis Figures and Tables Figures Tables Chapter Discussion Overview Placing Results in Context of Literature First Year Cost of Hip Fractures Second Year Costs Hip Fracture Outcomes and Health Resource Use Hip Fracture Related Mortality Clinical and Policy Implications The Osteoporosis Care Gap Hip Fracture Prevention Hip Fracture Management Strengths and Limitations vi

7 5.6 Future Directions Conclusions References Appendix A Study Design Appendix B Identification of Eligible Hip Fracture Cohort and Non- Hip Fracture Cohort Appendix C Case Costs by Type of Resource Use Appendix D Baseline Covariates and Outcomes Post Index Appendix E - Results Sub- Analyses Appendix F Comparison of Study Cohort and Results to Prior Canadian Literature vii

8 Abbreviations BMD Bone Mineral Density CADTH Canadian Agency for Drugs and Technologies in Health CCAC Community Care Access Centres CCC Complex Continuing Care CCRS Continuing Care Reporting System CIHI Canadian Institutes for Health Information CMG Case Mix Group CMI Case Mix Index CPWC Cost per Weighted Case DAD Discharge Abstract Database DXA Dual-energy X-Ray Absorptiometry HCD Home Care Database ICES Institute for Clinical Evaluative Sciences LTC Long Term Care MOHLTC Ministry of Health and Long Term Care NACRS National Ambulatory Care Reporting System ODB Ontario Drug Benefit OHCAS Ontario Home Care Administrative System OHIP Ontario Health Insurance Plan RIO Rurality Index for Ontario RIW Resource Intensity Weight RPDB Registered Persons Database RUG Resource Utilization Group US United States viii

9 List of Appendices Appendix A Study Design Appendix B Identification of Eligible Hip Fracture Cohort and Non- Hip Fracture Cohort Appendix C Case Costs by Type of Resource Use Appendix D Baseline Covariates and Outcomes Post Index Appendix E - Results Sub-Analyses Appendix F Comparison of Study Cohort and Results to Prior Canadian Literature ix

10 List of Figures Figure 1. Hip and Non-Hip Fracture Cohort Identification (exclusions not mutually exclusive) 55 Figure 2. Total number of incident hip fractures among females and males by fiscal year Figure 3. Distribution of index hip fractures by sex and age group Figure 4. Summary of 1-year direct attributable hip fracture costs for females (A) and males (B) by type of health resource use Figure 5. Summary of second year direct attributable hip fracture costs for females (A) and males (B) who survived year one by type of health resource use x

11 List of Tables Table 1. Costing methodologies of studies evaluating cost of hip fracture using administrative data Table 2. Summary of healthcare utilization databases and key variables used Table 3. Cost of healthcare utilization by resource type and database Table 4. Exposure propensity score * diagnostics Table 5. Number of individuals in the hip fracture cohort and non-hip fracture cohort who were successfully matched Table 6. Balance of matched variables between hip fracture and non-hip fracture cohorts Table 7. Balance of un-matched variables between hip fracture and non-hip fracture cohorts Table 8. Number of individuals in long term care at baseline (index) Table 9. Outcomes post index date (Year 1) Table 10. Outcomes post index date (Year 2) Table 11. Healthcare utilization in year after index date for hip and non-hip fracture cohorts Acute Hospitalizations, Emergency Department Visits and Same Day Surgeries Table 12. Healthcare utilization in year after index date for hip and non-hip fracture cohorts Physician Services (OHIP Claims), Complex Continuing Care, Rehabilitation Table 13. DXA tests received in first year after index date for hip and non-hip fracture cohorts 67 Table 14. Healthcare utilization in year after index date for hip and non-hip fracture cohorts Long Term Care Table 15. Healthcare utilization in year after index date for hip and non-hip fracture cohorts Home Care and Prescription Medications Table 16. Healthcare utilization in year after index date for hip and non-hip fracture cohorts Osteoporosis, Pain and NSAID medication use Table 17. Healthcare utilization in second year after index date for hip and non-hip fracture cohorts DXA tests and Osteoporosis medication use Table 18. Total and attributable direct healthcare costs of hip fracture in first year (365 days). 72 Table 19. Total and attributable 1-year direct healthcare costs of hip by health resource type Acute Hospitalizations, Emergency Department Visits and Same Day Surgeries Table 20. Total and attributable 1-year direct healthcare costs of hip by health resource type OHIP, Complex Continuing Care and Rehabilitation xi

12 Table 21. Total and attributable 1-year direct healthcare costs of hip by health resource type Long Term Care, Home Care, ODB Prescriptions Table 22. Total and attributable direct healthcare costs of hip fracture in second year ( days) Table 23. Total and attributable direct healthcare costs of hip fracture based on cost stratifications Residence Status, Second Hip Fracture and Survival Table B1. Hip Fracture Diagnostic Codes CIHI DAD Table B2. Trauma Diagnostic Codes CIHI DAD Table B3. Non-Osteoporosis Dosing Pharmacotherapy Drug Identification Numbers - ODB. 107 Table B4. Malignant Neoplasm Diagnostic Codes CIHI DAD s Disease Diagnostic Codes CIHI-DAD AND OHIP Table C1. Annual provincial average healthcare costs for Ontario by resource type Table D1. Baseline Diagnoses and Procedures Table D2. Baseline Covariates: Osteoporosis Medications Table D3. Baseline Covariates: Pain-Related Medications Table E1. Frequency of home care services among which costs were not available Table E2. Missing assessment dates in the Continuing Care Reporting System (CCRS) or costs in the Ontario Health Insurance Plan (OHIP) databases Table E3. Total number of visits received, services or prescriptions by the hip and non-hip fracture cohorts by health resource utilization Table E4. Healthcare utilization in second year after index date for hip and non-hip fracture cohorts Acute Hospitalizations, Emergency Department Visits and Same Day Surgeries Table E5. Healthcare utilization in the second year after index date for hip and non-hip fracture cohorts OHIP Claims, Complex Continuing Care, Rehabilitation Table E6. Healthcare utilization in second year after index date for hip and non-hip fracture cohorts Long Term Care, Home Care, Prescription Medications Table E7. Mean and attributable second year costs of hip fracture by health resource type Acute Hospitalizations, Emergency Department Visits and Same Day Surgeries Table E8. Mean and attributable second year costs of hip fracture by health resource type Physician Services, Complex Continuing Care, Rehabilitation Table E9. Mean and attributable second year costs of hip fracture by health resource type Long Term Care, Home Care, ODB Prescriptions xii

13 Table E10. Mean and attributable 1 year costs by Residence Status and Survival at Baseline. 135 Table E11. Mean and attributable costs among hip fracture cohort with a second hip fracture and their matched non-hip fracture control Table E12. Mean and attributable costs among individuals who died in the first and second Table E13. Mean and attributable costs among first and second year survivors Table E14. Sensitivity Analysis Total and attributable costs among hip fracture patients with prior fractures and their matched non-hip fracture control Table E15. Sensitivity Analysis Total and attributable costs (2010 dollars) among non-hip fracture cohort with prior fractures and their matched hip fracture case Table E16. Sensitivity Analysis - Total and attributable costs (2010 dollars) among hip fracture and non-hip fracture cohorts, excluding all individuals with prior non-hip fractures Table F1. Comparison of study cohort to those in the published Canadian Literature xiii

14 Overview Background Osteoporosis related hip fractures result in substantial morbidity and mortality. With an aging population and increasing life expectancy, the risk of osteoporosis and number of hip fractures are projected to increase. To date, few studies have adequately evaluated the costs associated with hip fractures in Canada. Prior studies were conducted over a decade ago and may no longer provide an accurate depiction of current medical care or economic costs related to hip fractures. Recent literature has also suggested that costs of osteoporotic fractures extend past the first year post-fracture, yet no study has examined the impact of fractures on healthcare services utilization and corresponding costs beyond the first year among seniors. In order to successfully evaluate healthcare strategies and pharmaceutical interventions aimed at reducing the burden of hip fractures, an updated assessment of the costs associated with these fractures in Canada was warranted. Objectives Primary Objectives: 1) To determine the mean direct healthcare costs attributable to hip fractures in Ontario seniors occurring between April 1, 2004 and March 31, 2008 over a one-year period when compared to a matched non-hip fracture cohort. 2) To estimate the mean attributable costs of hip fracture stratified by: survival (alive vs. dead at one-year), age group (66-69, 70-74, 75-79, 80-84, 85-89, 90+), sex, and residence status (Long Term Care (LTC) vs. community). 1

15 2 Secondary Objectives: 1) To determine the costs attributable to hip fracture in the second year post fracture. 2) To determine the frequency and costs (total and attributable) of: a. Post hip fracture process measures including: acute hospitalizations, day surgery, emergency department visits, physician services received, entrance into long term care (LTC), rehabilitation, complex continuing care, home care and treatment with prescription medications b. Outcomes post hip fracture i. Second hip fracture ii. Death With these data, policy makers will be better able to make decisions related to healthcare costs and resource allocation. These data will further provide estimates to inform health economic analyses for interventions aimed at reducing hip fracture.

16 3 Thesis Organization This thesis is separated into 5 chapters. Chapter 1 (Burden of osteoporosis and hip fractures in Ontario) sets the stage to this research by reviewing osteoporosis and hip fracture definitions, epidemiology, and overall burden in Canada. Chapter 2 (Literature review of prior studies) describes the importance of cost of illness studies with a review of the literature evaluating methods used and costs associated with hip fractures. Chapter 3 (Methods) outlines the study design, data sources, and analytical approaches used to estimate the costs of hip fractures in this thesis. In Chapter 4 (Results) we present the results of the study including a summary of baseline characteristics, costs associated with hip fractures, and healthcare utilization process measures. The final chapter, Chapter 5, discusses the major findings of the study, along with a description of the strengths and limitations of the research, policy implications, and the future direction of costing research in osteoporosis. Multiple appendices supplement the text providing additional details related to study methodology and results.

17 Chapter 1 Background 1.1 Osteoporosis Osteoporosis is a disease characterized by microarchitectural deterioration of bone resulting in increased bone porosity and subsequently increased susceptibility to fracture [1]. Defined by the World Health Organization as a femoral neck bone mineral density (BMD) T SD [2, 3], it is estimated that osteoporosis affects over two million Canadians, with approximately one in four women and one in eight men over the age of 50 years having the disease [1, 4, 5]. Additionally, 46% of postmenopausal women and 40% of men over the age of 50 are estimated to have osteopenia (bone mineral density between -1 and -2.5 SD below sex-specific young adult values) [6], and are thus at increased risk of developing osteoporosis. Bone loss accelerates substantially among women at the time of menopause, by approximately 2-5% percent per year [7, 8]. This rapid bone loss continues for up to 5 to 10 years at which point it returns to premenopausal levels [9]. Bone density continues to decrease in both elderly men and women at a rate of 0.5%-1.0% per year [10]. With the prevalence of osteoporosis increasing with age and the rapid aging of our population, the incidence of osteoporosis is projected to rise significantly [11, 12]. 1.2 Osteoporotic Fractures Osteoporosis is an asymptomatic disease until it clinically manifests in the form of an osteoporosis related fracture. These fractures are associated with low bone mineral density and typically result from low-impact trauma such as a fall from standing height or less [13]. Fractures most commonly attributed to osteoporosis occur at the hip, wrist and spine;; however nearly all skeletal sites are at increased risk [4, 13]. It has been estimated that 40% of women 4

18 5 and 13% of men aged 50 years and older will experience one or more hip, vertebral, or wrist fracture in their remaining lifetime [14]. Fractures of the wrist and forearm are typically the first osteoporotic fracture sites that come to clinical attention [15, 16]. These fractures are most common among women, with incidence increasing greatly after menopause [16, 17]. Wrist and forearm fractures result in acute pain, but have not been significantly associated with long-term morbidity or mortality [16, 18]. Vertebral fractures are the most common type of osteoporotic fracture, with prevalence increasing greatly with age [19]. These fractures often result in loss of independence, decreased physical abilities, and chronic pain as a result of kyphosis [20-22]. However, identification of vertebral fractures remains a challenge as less than half come to clinical attention [23], with fewer than 10% directly resulting in a hospital visit [16]. Potentially of greatest clinical concern are hip fractures, which have been well documented in the literature as resulting in considerable fracture related morbidity, mortality and decreased quality of life [24, 25] Hip Fractures Approximately 30,000 hip fractures occur annually in Canada [26], with an age-adjusted hip fracture rate of 86.4 per 100,000 person-years (95% CI ) in females and 74.4 per 100,000 person-years (95% CI ) in males [27]. The incidence of hip fracture increases greatly with age, attributed largely to decreases in bone mineral density [7] and increased frailty and thus falls [28]. It is estimated that approximately 80% of all hip fractures are osteoporosis related, with 90% resulting directly from a fall [29]. Individuals sustaining a hip fracture incur a wide range of complications including pain and functional disability [30]. A Canadian study evaluating men and women with an incident hip fracture showed a significant decline in independent living, with 16% of community patients

19 6 admitted to long term care and more than 60% of those who returned to the community receiving home care within the first 12 months post fracture [31]. In addition to the substantial decreases in independence, hip fractures have been shown to result in an absolute rate of mortality of 20-30% [4]. Among individuals who survive their hip fracture, it is estimated that within 1 year, 2.5% will incur a subsequent hip fracture, increasing to 8% within 5 years of fracture occurrence [32]. The risk of subsequent non-hip fractures has also been shown to increase, with a relative risk of 2.5 (95% CI ) for vertebral fracture [33]. With the proportion of Canadians aged 65 years and over projected to increase from 13.7% in 2006 to 25% in 2041 [34], the incidence and burden associated with hip fractures will increasingly become a significant public health concern, and consequently result in increased direct costs to the Canadian healthcare system [11, 12]. 1.3 Why Are Cost of Illness Studies Important? Canadians are publically funded for healthcare under their corresponding provincial health insurance plan. In the province of Ontario, healthcare is funded by the Ontario Health Insurance Plan (OHIP), which provides universal coverage for all medically necessary hospital and physician services [35]. Pharmaceutical drugs prescribed outside of the hospital are not covered under OHIP, however all individuals aged 65 years and older, recipients of social assistance or professional home services, and residences of long term care facilities are covered through the Ontario Drug Benefit Program (ODB) for all drug products listed in the Ontario Drug Benefit Formulary/Comparative Drug Index [36]. Through the Ministry of Health and Long Term Care, seniors are additionally covered for home care services, rehabilitation, as well as partial coverage for residence in a long term care facility. As a result, government spending on

20 healthcare was estimated to reach $135 billion in 2010, with seniors accounting for 44% of all healthcare dollars spent [37]. 7 Cost of illness studies estimate the economic burden that a disease or event places on our healthcare system [38, 39]. These studies provide information regarding the impact of specific diseases, illnesses or events in terms of the total healthcare utilization and costs involved. In a cost constrained healthcare environment, appropriately rationing resources is essential. As a result, costing studies are instrumental in health policy decision-making by determining the magnitude that a disease or event places on the healthcare system and also provide an important source of data that can be used to inform cost effectiveness analyses. In particular, when dealing with health economic models for interventions for fracture prevention, the incremental cost effectiveness ratio has been shown to vary greatly depending on short and long-term assumptions of fracture costs [40]. High quality costing data are therefore vital in assessing healthcare spending, conducting cost effectiveness analyses and ultimately in guiding policy decisions. 1.4 Chapter Summary Osteoporosis and its related hip fractures are increasingly becoming a significant public health issue. Understanding the costs attributable to hip fractures is essential as they will result in considerable fracture related healthcare utilization.

21 Chapter 2 Literature Review 2.1 Cost of Hip Fractures in Canada Numerous American and European studies have estimated the overall cost of illness due to osteoporosis [41-43], and the individual costs related to osteoporotic fractures [44-50]. Recent estimates for mean healthcare costs per hip fracture patient ranged greatly from $15,196 to $38,699 US dollars over 1 year [45, 49]. Although many costing analyses have been conducted world wide to determine the pattern of expenditure related to osteoporotic fractures, these data may not be generalizable to the Canadian population. Country specific analyses are important to account for potential variations in healthcare systems, treatment patterns, population risk levels, and costs. Additionally, the Canadian Agency for Drugs and Technologies in Health (CADTH) has recommended the use of Canadian costing data when evaluating the cost of a disease or event in economic analyses [51]. Studies that assessed the cost of hip fractures in Canada were identified after a thorough search of the MEDLINE (1948-January 2011) and EMBASE (1980-January 2011) databases. A broad search strategy was used with combinations of the search terms: fracture, osteoporosis, cost, and Canada. Additionally, a general search of the grey literature ( was used to capture non-indexed analyses and government reports. Abstracts and literature reviews were excluded. Three studies that met our selection criteria were identified among 48 articles retrieved from the MEDLINE and EMBASE search, with one additional study abstracted from the CADTH website. Only two studies directly examined the costs associated with osteoporosis or hip 8

22 fractures from a Canadian context [31, 52], with the other two studies incorporating alternative cost estimates within economic analyses for osteoporosis treatments in Canada [53, 54]. 9 The first was a burden of illness study that determined the health and economic burden of osteoporosis;; estimating the number of individuals with osteoporosis, number and rate of osteoporosis-related hip, wrist, and vertebral fractures, and cumulative direct healthcare costs in men and women in 1993 [52]. The total burden of osteoporosis was estimated to be $1.3 billion in The majority of cost burden was associated with treating hip fractures, with the total cost of treating 21,457 hip fractures in females and males aged 45 or more years being $260 million and $77 million, respectively. Unlike incidence-based studies which estimate costs associated with an illness from the date of onset forward, this study was a prevalence-based study which estimated the annual costs of osteoporosis, regardless of the date of disease diagnosis or fracture occurrence. Although this study utilized a combination of data sources (literature reviews, Canadian Institutes for Health Information (CIHI), Statistics Canada, Ontario Ministry of Health, Long Term Care Division, Ontario Ministry of Health, OHIP schedule of benefits, Intercontinental Medical Statistics, Ontario Ministry of Health Home Care Program, and physician expert panels) to estimate overall costs of osteoporosis, costs associated specifically with hip fractures were determined solely from hospital resource use and estimated treatment protocols post discharge. Therefore these estimates are an underestimate of total hip fracture costs. Other key service costs such as long term care (LTC) and prescription medications were estimated for all individuals with osteoporosis or osteoporosis-related fracture hospitalizations, but were not stratified on the presence or type of fracture. Additionally, prevalence of disease was based on American and not Canadian data, and resource utilization and costs in the ambulatory setting were based on physician surveys regarding estimated treatment protocols rather than actual patient-level data. Moreover, fractures may have been

23 misclassified, as attribution to osteoporosis was based solely on expert opinion regarding the probability that a fracture at a given age was due to osteoporosis. In addition to limitations related to study design, this analysis was conducted in 1993, prior to the introduction of oral bisphosphonates, and may no longer appropriately represent current healthcare treatment patterns or costs. 10 The second study was a longitudinal cohort study evaluating the 1-year societal cost per hip fracture in a population of men and women over the age of 50 admitted to an acute care facility in the Hamilton-Wentworth region [31]. Costs evaluated included: initial hospitalization, rehospitalization, in-patient rehabilitation/chronic care, day rehabilitation, physician visits, long term care, home care and informal care. The mean 1-year cost of hip fracture was estimated to be $26,527 (95% CI: $24,564-$28,490) in 1997 Canadian dollars, with costs additionally stratified by residence status of patients at- and post-discharge and type of health service resource use. Mean costs were greatest for community residents who transferred to a long term care facility ($44,156) followed by those who were in long term care at baseline ($33,729). Older age, female gender, and survival to 1 year all showed higher costs in univariate analyses. Despite thorough data collection through hospital discharge data and patient follow-up, this study was based on data from only four hospitals in a single region in Ontario and may not be representative of the province or country as a whole. Further, this study may underestimate costs of hip fracture, as medications used in the community were not incorporated. Community fracture relied on patient self-report or a proxy after months of their hip fracture. This may introduce a level of information bias as healthcare utilization is based solely upon patient recollection rather than a record of actual visits made. Prior research has indicated patient-self report of health service utilization to often be an underestimate of true resource use [55]. Patient

24 self-report did, however, allow for indirect costs associated with informal care and assistive devices to be included in the analysis. Potentially the greatest limitation of this study is the lack of a comparison group, which makes it difficult to ascertain the costs attributable to fracture rather than pre-existing health conditions or comorbidities that were secondary to the fracture [56, 57]. 11 Despite the paucity of high-quality costing data for hip fractures, data from these prior studies have been incorporated into recent Canadian economic analyses evaluating the costeffectiveness of various osteoporosis interventions aimed at reducing the burden of fracture [53, 58, 59]. One of the studies [53] stated that cost inputs for hip fracture by age group and in subsequent years post fracture were derived from secondary analyses of the data published by Wiktorowicz et al. [31]. This study provided respective estimates of first and subsequent year costs (2005 dollars) of hip fracture in women of: $17,304 and $0 for those years, $27,014 and $2,707 for years, $32,164 and $4,078 for years, and $33,146 and $9,556 for women aged 85 and over. These estimates were only provided for women, and it was not subject to the same limitations as those discussed for the primary analysis. An economic analysis for nasal calcitonin [54] quoted alternative total 1-year costs (1998 dollars) for hip fracture of $18,874 for females aged and $24,442 for those aged 70 and over. These estimates were stated as being derived from length of stay costs from Canada as a whole and treatment costs from Manitoba, however, the methods utilized to obtain and analyze these costs have not been described or published in the literature.

25 Methods of Hip Fracture Cost Estimation To obtain a comprehensive picture of the breadth of methods used to estimate hip fracture costs, we expanded our search to identify all studies that used longitudinal hip fracture cost estimation methods. Costs of osteoporotic fractures have previously been identified using a number of research methods including decision analytic modeling [42, 60-63] as well as prospective data collection or questionnaires [50, 64, 65]. More recently, a shift towards the use of administrative data to complete fracture-costing analyses has occurred [45-49, 66, 67] Administrative Data for Cost Estimation Healthcare administrative or utilization data are generally classified as routinely collected data for purposes that can span systems involving basic demographic information, clinical data, and data regarding specific health services used and received [68, 69]. These data are often collected for the purpose of documentation of provider claims for prescription and medical services. In Canadian healthcare, administrative databases are used to record the details about persons who are registered with provincial health plans and information on characteristics of inpatient and outpatient hospitalization events, visits to or by physicians or independent health professionals, filled prescriptions, home care visits, rehabilitation and nursing home stays [69]. Although administrative data are primarily collected for billing purposes, they provide a wealth of information to researchers regarding patient demographics as well as health and health service use. These records particularly have important strengths when considering methods to measure healthcare related costs [56, 68, 70] and have been recommended for use in Canada to accurately reflect the true resource usage and costs associated with fractures [51, 71]. Administrative data provide information on large, well-defined populations, and allow for analysis of subgroups that may be difficult to obtain elsewhere (e.g. patients in long term care).

26 Economic measures can be estimated from claims data by applying costs to units of service, weighted resource utilization based on per case or per day, or are often directly available as cost paid per service. These data allow researchers to disaggregate costs based on type of health service use. An important advantage of these databases is that they allow for costs of individual health service use to be linked longitudinally, thus permitting costs to be evaluated in terms of specific episodes of care (e.g. initial treatment and post-discharge). Healthcare administrative data can also be combined with outcome data to determine costs associated with specific outcomes associated with the event, such as having a second hip fracture. As such, these data provide a detailed, convenient and relatively inexpensive source of costing information [56] Administrative Data to Estimate Hip Fracture Costs A summary of methods of recent studies utilizing administrative data in the United States (US) to estimate longitudinal costs of hip fractures are reported in Table 1. Three primary methods have been used to summarize costs related to hip fracture using administrative claims data [38], with some studies utilizing a combination of the methods: all-cause costs [45-49, 67], diagnosisspecific costs [45-47, 66, 67], and attributable costs [46-49, 67]. In all-cause costs, studies sum all costs available for an individual following the date of their diagnosis or event. Although this method captures all costs incurred by a cohort, it is unknown which costs may result from preexisting health conditions or comorbidities and thus this estimate alone is likely to over-estimate the cost associated with the fracture. Diagnosis-specific costs aims to determine costs attributable to the event itself by summing only those treatments either coded or clinically predicted to be related to the diagnosis. Unlike all-cause costs, this method likely underestimates the cost of illness as it fails to account for costs that may be secondary to the event itself, and relies on codes for osteoporosis or fracture that are often not captured for downstream events. It

27 has been suggested in the literature that in order to determine the cost attributable to a disease or event a comparison group should be used. Some studies utilized a pre-/post-fracture cost method, whereby costs in the period prior to the index date are subtracted from the costs incurred after the index date [45, 67]. This method accounts for baseline costs, however may not be appropriate as it fails to account for variables which vary with time and are associated with increased resource use (e.g. age and comorbidites). As such, it has been recommended that costing studies utilize a matched cohort study design whereby individuals with the disease or event of interest are matched to individuals without the disease on variables predictive of the outcome. This method was seen in four of the studies examined [47-49, 67], and allows the summation of all costs after the index date, with the difference in costs representing the cost attributable to the disease [57]. 14

28 Table 1. Costing methodologies of studies evaluating cost of hip fracture using administrative data Author, Year Pike, 2011 [48] Data source Privately insured claims database in US Hip Fx Outcome 1- and 2-year attributable cost Population Design Cost Analysis Hip Fx Cost Estimate Males and females with an OP dx aged Matched Cohort - absolute cost difference between cases and controls 2006 Attributable Cost: $13,334 year 1 $3,930 year 2 Pike, 2010 [47] Christensen, 2010 [45] 1) US Employer claims database (40 companies) 2) Medicare Standard Analytic Files US Commercial and Medicare Advantage enrollees 1-year attributable cost Males and females with OP dx aged and year cost Males and females 45+ with fx of hip as 1 0 dx Matched Cohort Fx Cohort Only (claims with a fx related code or fx-related drug costs during 6mo after index) attributable costs (B) regression-adjusted attributable costs using GLM -related costs (if associated with fx dx in 1 0 position, or a procedure with a fx-coded claim) and follow-up (year after fx) 2006 Attributable Cost: 1) years: $12,562 2) > 65 years: $25, Cost: Pre-Fracture= $10,643 Post-Fracture= $35,898 Kilgore, 2009 [46] Shi, 2009 [49] 5% national sample of Medicare beneficiaries from Chronic Conditions Warehouse in US 1) US Thomson Reuters Marketscan Commercial Claims & Encounters 2) US Medicare supplemental benefit databases 6-month and 1-year costs 1-year attributable costs Males and females >65, with incident closed hip fx Male and Females aged years and 65+ with 1 0 or 2 0 dx of closed hip fx Fx Cohort Only Matched Cohort 1) Total costs - sum from all sources and providers 2) Attributable cost - claim with primary dx code for fx 3) Incremental - difference in 6 months pre and post fx ost difference between cases and controls (B) Regression adjusted attributable cost using GLM (adjusting for comorbidity and pre-period medical expenditures) 2007 Costs: 1) Total = $38,971 2) Attributable = $18,734 3) Incremental = $31, Attributable Costs: 1) years: $31,047 2) > 65 years: $16,823 Ohsfeldt, 2006 [66] Orsini, 2005 [67] US Medstat's MarketScan Commercial Claims & Encounters 1-year attributable costs 1) US Medstat's 1-year costs MarketScan Commercial Claims & Encounters 2)US Medicare Supplemental and Coordination of Benefits Males and females 45+ with 1 0 dx of closed fragility hip fx Males and Females aged 0-64 years and 65+ with an OP dx and/or OP related fx Fx Cohort Only 3 Cohorts 1) Hip fx + OP dx 2) Hip fx + no OP dx 3) Matched to (1): no OP dx or fx codes thought to be fx-related -related (coded with 1 0 dx of OP or specified fx dx in concurrent cohort) and non-op related year before and year after fx Acronyms: Dx = Diagnosis;; Fx = fracture;; GLM = Generalized Linear Regression Model;; OP = Osteoporosis;; US = United States 2003 Mean 1-Year Cost = $26,856 Cohort 1) 2003 Total 1- year Cost = $19,973 (other cohort costs and pre-period costs not stratified by fx type) 15

29 Summary of Literature Review Few studies have estimated the costs associated with hip fractures in Canada. Among two primary costing studies published, estimates were based on poor quality study designs and did not include a comparator group to estimate costs directly attributable to the event. Additionally, cost estimates are over a decade old and may no longer represent current health practices or costs. Current research recommends the use of administrative data to estimate costs associated with disease in comparison to a matched disease-free cohort. This method however, has not been utilized in Canada to estimate costs or healthcare resource use related to hip fractures. As a result of this gap in the literature, updated estimates of the costs associated with hip fractures in Canada are needed. 2.4 Study Objectives Primary Objectives 1) To determine the mean direct healthcare costs attributable to hip fractures in Ontario seniors occurring between April 1, 2004 and March 31, 2008 over a one-year period when compared to a matched non-hip fracture cohort. 2) To determine the mean attributable costs of hip fracture stratified by: survival (alive vs. dead at one-year), age group (66-69, 70-74, 75-79, 80-84, 85-89, 90+), sex, and residence status (LTC vs. community).

30 Secondary Objectives 1) To determine the costs attributable to hip fracture in the second year post fracture. 2) To determine the frequency and costs (total and attributable) of: a. Post hip fracture process measures including: inpatient hospitalizations/day surgery, emergency department visits, physician services received, entrance into LTC, rehabilitation, complex continuing care, home care, treatment with prescription medications b. Outcomes post hip fracture i. Second hip fracture ii. Death

31 Chapter 3 Methods 3.1 Data Sources We utilized a comprehensive collection of healthcare administrative claims and billing data housed at the Institute for Clinical Evaluative Sciences (ICES) to identify and determine healthcare utilization and costs associated with osteoporotic hip fractures in Ontario including: Canadian Institutes of Health Information Discharge Abstract Database (CIHI DAD) Continuing Care Reporting System (CCRS) National Ambulatory Care Reporting System (NACRS) National Rehabilitation System (NRS) Ontario Drug Benefit (ODB) Ontario Health Insurance Plan (OHIP) Ontario Home Care Administrative System (OHCAS) and Home Care Database (HCD) Ontario Registered Persons Database (RPDB) Each database has been well validated and described in the literature (Table 2), and has previously been utilized as a source of data for costing analyses in Ontario [72-74]. 18

32 Table 2. Summary of healthcare utilization databases and key variables used Database Setting Description Key Variables Validity Acute Inpatient Hospitalizatio ns Canadian Institute for Health Information Discharge Abstract Database (CIHI DAD) Continuing Care Reporting System (CCRS) National Ambulatory Care Reporting System (NACRS) National Rehabilitation Reporting System (NRS) Ontario Drug Benefit (ODB) Complex Continuing Care Emergency and Day Surgery Rehabilitation Prescription Drugs The CIHI DAD is a National database that contains demographic, clinical, and administrative data for inpatient hospital admissions. It contains over 1 million inpatient abstracts from 178 acute care facilities in Ontario [75] In 2004 the CCRS replaced the Ontario Chronic Care Patient System (OCCPS). CCRS contains clinical and demographic information on individuals receiving facility based continuing care. Services include medical LTC, rehabilitation, geriatric assessment, respite care, and palliative care, and nursing home care. The NACRS was fully developed in Ontario in 2002 and contains data for all ambulatory care including emergency department visits, outpatient clinics, and day surgeries. The NRS contains National data on rehabilitation facilities and clients, collected from participating adult inpatient rehabilitation facilities and programs The ODB includes all drugs dispensed in community pharmacies and LTC/nursing facilities. The ODB covers all seniors in Ontario (aged 65+) and those on social assistance for all prescriptions listed in the provincial formulary (approximately 3,200) ICD-9-CA ICD-10-CA and CCP Intensity Weight Date Date Case Mix Index ICD-9-CA ICD-10-CA and CCP CACS Resource Intensity Weight Date Date FIM Scores Rehabilitation Client Group Drug Identification Number (DIN) Ministry of Health Care Indicator 19 Nearly 85% agreement between abstractor and most responsible diagnosis code [75] Sensitivity and Positive Predictive Values >95% for codes related to fractures of femur (including neck of femur) [75] Most prevalent conditions in minimum data set exhibited sensitivity of 0.80 or greater in comparison to primary diagnoses coded in acute care Other conditions often lacked key diagnostic information [76] In reabstraction and inter-rater reliability studies, agreement rates in the selection of main problem was >85%, and >73% for reason for visit [77] All eligible rehabilitation facilities in ON are included;; 100% response rate. Discharge record missing for 2.2% of ON episodes [78] and length of stay under-reported [79] Second largest prescription database in Canada. At least 95% of seniors filled 1 Rx in ODB over a 5 year period, however 15-20% filled a Rx from a private insurer [80] High coding reliability, overall error rate of 0.7% (95% CI 0.5%-0.9%) [81] Drugs dispensed during acute hospitalizations are not captured

33 Database Setting Description Key Variables Validity Outpatient and Physician Services Ontario Health Insurance Plan (OHIP) Ontario Home Care Administrative System (OHCAS) and Home Care Database (HCD) Ontario Registered Persons Database (RPDB) Home Care General Population OHIP data cover all services and procedures provided by health care providers who can claim under OHIP (physicians, laboratory services) The OHCAS and HCD provide data on governmentfunded services coordinated by Care Access Centres (CCAC), for individuals requiring home care [83]. The HCD replaced the OHCAS in the 2005/2006 fiscal year. The RPDB contains data on the vital status of all Ontario residents covered under OHIP. service Service Provided Unit Time discharge date Variables health benefits 20 Ontario physicians do not use a direct fee for service practice [82], however alternative funding plans primarily use shadow billing to record services provided. are represented numbers are not always available (e.g. homeless clients, disoriented or elderly patients) and therefore not all individuals in a cohort will be captured some regions Captures all Ontario residents with a valid OHIP number Individuals who move are not required to report to the Ministry and thus may appear to still be in province [84] Individual data were linked across databases using the ICES Key Number (IKN), which is an encrypted code for all individuals with a valid health card number in Ontario. 3.2 Study Design A matched cohort study design was used to determine the 1- and 2- year costs attributable to hip fracture. Prior fracture costing studies have often defined attributable costs based on diseaserelated codes within administrative claims data, however may underestimate the true costs of disease as it often fails to capture costs secondary to the event [85]. Recent literature has recommended the use of matched cohort analyses to better reflect excess costs for patients by subtracting costs of a matched non-diseased/event cohort (e.g. no fracture) from a cohort with the disease or event (e.g. fracture) [38, 57]. With this method, matching on appropriate confounding factors that may impact costs outside fractures improves the accuracy in identifying

34 costs attributable to the disease or event itself. The general study design employed for the current project, including time frame descriptions, is depicted in Appendix A Figure A Hip Fracture Cohort Identification Inclusion Criteria The hip fracture cohort was identified as all men and women in the RPDB between April and March , who turned 66 prior to the end of the accrual period (March 31, 2008). From this group we identified those individuals with an incident fracture of the hip as identified in the CIHI-DAD, with the first date of fracture identified as the index date. The year 2004 was selected to maximize data quality: 1) this start date allowed us to minimize the differences between inpatient and hospital diagnosis coding from the International Classification of Diseases Ninth Revision, Clinical Modification (ICD-9-CM) and procedural codes using Canadian Classification of Diagnostic, Therapeutic, and Surgical Procedures (CCP) to International Classification of Diseases Tenth Revision Canada (ICD-10-CA) and Canadian Classification of Health Interventions (CCI) codes, respectively when determining our outcome measures;; although ICD-9 codes were used when identifying prior hip fractures 2) the OHCAS contains incomplete data for Toronto during the fiscal year, and thus using these data would likely underestimate home care utilization, 3) NACRS has a complete database as of the 2002/2003 fiscal year, and 4) reporting to the NRS became mandatory in 2002/2003 fiscal year in the province of Ontario. The index date was defined as the initial date of hip fracture occurrence based on diagnostic fracture codes within the CIHI DAD (Appendix B Table B1). Identification of hip fractures using diagnostic codes has been highly validated in administrative claims data [86]. A reabstraction study conducted in Canada found 99% agreement for the ICD-9 hip fracture diagnosis code [69]. Additionally, in a re-abstraction study of the DAD using the first three

35 characters of the ICD-10 code for fracture of femur (S72 code, which includes fracture of neck of femur) it was found that agreement for most the responsible diagnosis to have a sensitivity of 0.95 (95% CI 0.94 to 0.97) and positive predictive value of 0.95 (95% CI 0.92 to 0.97) [75]. Thus, we did not require the use of a procedural code in identifying hip fractures. A diagnosis of osteoporosis was not required for inclusion as it is well known that osteoporosis is underdiagnosed in administrative claims databases and therefore is not a specific measure of the disease [87-89]. Hip fractures coded as in-hospital comorbidity were included to capture hip fractures occurring during a hospitalization Exclusion Criteria Individuals who were less than 66 years old were excluded from the study. Age 66+ was chosen for analysis as ODB coverage begins at age 65 years and thus allows us to capture prescription drug utilization with a one-year lookback period prior to the index date. Additionally, osteoporosis related hip fractures are less common in those under the age of 66 years [26]. Individuals coded as having a trauma code within 7 days of the index date were excluded as these fractures are less likely to be osteoporosis related (Appendix B Table B2). Individuals with a prior hip fracture in the CIHI-DAD (up to the year 1991) were excluded to make sure index fractures were incident hip fractures and not a result of follow-up for a prior hip fracture, as well as to differentiate between costs associated with prevalent hip fractures. Caution was taken to ensure that transfers between hospitals were not counted as a prior fracture. To exclude individuals at an increased risk of fracture due to non-osteoporotic conditions, we excluded those with a history (using a 1- other than non-melanoma skin. Additionally, because we are not requiring an osteoporosis diagnosis for inclusion, we excluded individuals who were dispensed osteoporosis medications a). We

36 therefore excluded those with a non-osteoporosis dosing of an oral bisphosphonate (non-cyclical etidronate, 40mg alendronate, 30mg risedronate, 4mg or 4mg/5ml zoledronic acid, clodronate, or pamidronate) or 50U, 100U or injected calcitonin;; identified through the ODB as having a unique drug identification number and dose. Exclusionary codes are depicted in Appendix B Table B3-Table B5. Other exclusion criteria included gender coded as missing, death prior to index date, non-ontario residents, or individuals with a missing OHIP or ICES Key Number (IKN) Non-Hip Fracture Cohort Selection A non-hip fracture cohort was identified by selecting all individuals, including those selected in the hip fracture cohort, from the RPDB between April 1, 2004 and March 31, 2008 who turned 66 by the end of the accrual period (March 31, 2008). Thus, we began with all seniors with a valid health card number in Ontario. A random index date was then assigned to each individual based on the sex-specific distribution of fracture diagnosis dates among the hip fracture cohort. The same exclusion criterion that was applied to the hip fracture cohort was then applied to the non-hip fracture cohort. Individuals were additionally excluded if they had a history of hip fracture prior to the index date (lookback July 1991) or a hip fracture within two years (730 days) after the index date. This allowed for fracture cases to serve as potential non-fracture controls up to two years prior to becoming cases. This was important to consider in order to minimize the potential biases that may arise from excluding these patients [90]. In particular, hip fracture patients may have underlying differences in comparison to individuals without a hip fracture. Although matched on key factors, unmeasured variables such as bone mineral density, frailty, and healthy lifestyle factors may be different;; potentially result if these individuals are excluded. Thus, this method will allow for increased precision in the cost estimates of the non-fracture cohort, and ultimately the attributable costs of hip fracture.

37 Matching Hip Fracture Cohort to Non-fracture Cohort In order to define homogeneous fracture and non-fracture cohorts, cohorts were matched on clinical and demographic factors that have been shown to be associated with resource use. All eligible individuals in the non-hip fracture cohort were first hard matched to those in the hip fracture cohort on: 1) the month and fiscal year of the index hip fracture diagnosis to minimize differences between medical care and treatments available, 2) age at index date (+/- 3 months) and sex to account for potentially increased costs associated with these variables [31, 45], and 3) residence status at baseline (LTC versus community) as the type and quality of care can often differ for those individuals in LTC [31]. These matching criteria allowed us to stratify our costing analysis on each factor without jeopardizing our matches. A propensity score was then calculated for each individual as the probability of having a hip bserved baseline covariates [91]. Propensity scores were on the Adjusted Clinical Group (ACG) system using a 1-year lookback, 2) Rurality Index for Ontario (RIO) using the statistics Canada postal code conversion file and Census Data, and 3) income quintile. The Johns Hopkins ACG system is a case-mix adjustment system that provides a relative measure of an individuals overall health status and risk for use of health services [92]. The ACG system categorizes patients by their ICD-9 and ICD-10 diagnosis codes into one of 32 different aggregated diagnosis groups (ADGs) and then further collapses these groups into one of 12 CADGs (Acute Minor, Acute Major, Likely to Recur, Asthma, Chronic Unstable, Chronic Stable, Specialty Unstable, Specialty Stable, Eye/dental, Psychosocial, and Prevention). A Canadian study evaluated the performance of the ACG system in British Columbia and Manitoba, noting the feasibility of the system with existing data. In particular, the system

38 rejected less than 2% of diagnoses, and was found to explain approximately 50% of physician costs in the same year and 40% of total costs [93, 94]. Additionally, a recent study in British Columbia found drug expenditures to be better predicted by the ACG system in comparison to the Charlson Comorbidity index, with a c-statistic of 0.77 in comparison to 0.68 when predicting costs in the following year, respectively [95]. The ACG system has also previously been used as a matching criteria in a Canadian study predicting osteoporotic fractures [96]. In this study, the ACG software at ICES used claims from both the CIHI-DAD and OHIP in the year preceding the index date to categorize patients into the 12 CADGs, using dummy variables in SAS to match. 25 The RIO is used by the Ministry of Health and Long Term Care and the Canadian Medical Association as eligibility criterion for program incentives, and captures geographical factors such as population count and density as well as access to health services [97]. The index is calculated for each of 585 Ontario Census Subdivisions, which refers to specific municipalities in Ontario. The calculation of the RIO is depicted in Equation 1: Equation 1 Rurality Index for Ontario (RIO) = TIMEb +TIMEa +POPm +GPR + GP + HOSP + AMB + SOC + WTHR + MSS Where, TIMEb measures the time of travel to the nearest basic referral centre relative to the median time for all other Ontario communities;; TIMEa measures the travel time to the nearest advanced referral centre;; POP m is a measure of the community population;; GPR is the measure of population to general practitioner ratio;; GP measures the number of active general practitioners in the community;; HOSP measures the presence of a hospital;; AMB measures the availability of ambulance service;; SOC measures social indicators;; WTHR measures the weather conditions;; and MSS is the measure of selected services available (e.g. anesthesia and

39 obstetrics). Points are awarded based on each individual component of the RIO in a linear fashion, with most factors being a relative measure in comparison to the provincial median. The RIO is an ordinal measure that ranges from 0-100, with higher values indicative of a higher degree of rurality. In Ontario, communities with RIO scores of 45 or greater are considered to be highly rural and therefore eligible for physician incentives [98]. Rurality has been shown to be a particularly important factor for seniors [99] as well as in osteoporosis, with access to dual energy x-ray absorptiometry (DXA) screening mediating access to osteoporosis pharmacotherapy [100]. Similarly, socioeconomic status has been shown to be correlated to healthcare related costs, and thus matching on income quintile was appropriate [101]. Each of these criteria was chosen to ensure comparability between hip fracture and non-hip fracture patients in terms of demographic, socioeconomic, and clinical factors as well as health resource use. 26 Hip fracture patients and non-hip fracture individuals were then further matched 1:1 on their propensity score. Matching on propensity score has been shown to reduce potential selection bias in estimating the effects of a patient group when using observational data [102]. Fracture patients were matched to non-fracture controls on the logit of the propensity score using greedy matching with a maximum caliper width of 0.2 of the standardized difference of the logit of the propensity score, which has been shown to eliminate more bias in comparison to a crude estimator [103] -hip fracture control patient to a hip fracture patient, [91]. This method may result in individuals within the hip fracture cohort who cannot be matched to a non-hip fracture patient, however improves the quality of the matches in comparison to allowing matches without accounting for caliper distance. Once a control was matched to a hip fracture patient, the control was removed from the pool of possible matches, thereby matching without replacement.

40 3.4 Baseline Characteristics not-for-profit corporations separated into one of 14 geographic regions within Ontario. Each LHIN is responsible for working with community members and local health providers to determine priorities related to health services delivered in: long-term care homes, public and private hospitals, community support services, community care access centres, community health centres and mental health ag number of hip fractures is therefore important for guiding health resource allocation. 27 Other un-matched characteristics related to osteoporosis and hip fractures were evaluated in each cohort in the year (365 days) prior to the index date (Appendix D). We assessed the number of individuals with a diagnosis for osteoporosis or who had a DXA test coded in the CIHI-DAD or OHIP datasets. Additionally, the number of individuals with a diagnosis for a prior: (1) vertebral, (2) humerus, radius or ulna, or (3) other osteoporosis-related fracture was identified in the CIHI-DAD and OHIP datasets. Osteoporosis medication use was identified through the ODB dataset as individuals filling one or more prescriptions for an osteoporosisdosing of alendronate, etidronate, calcitonin, raloxifene, risedronate, teriparatide or zoledronic acid (Appendix D). 3.5 Health Outcomes Mortality subsequent to hip fractures was identified through the RPDB. Death was stratified into those who died on or within the first year of index, and those who died in the second year after the index date. Second hip fractures were identified in the hip fracture cohort based on a diagnosis for hip fracture in the CIHI-DAD. A period of 180 days was required between the index hip fracture

41 diagnosis and the second hip fracture diagnosis in order to ensure these were new fractures and not coded as a follow-up for the index hip fracture Resource Utilization Processes and Costing We utilized the Ontario healthcare perspective, and thus only direct costs paid by the Ontario Ministry of Health and Long Term Care were considered in our costing analyses. Costs incurred by patients (e.g. copayment for medications, canes/walkers, transportation) or hospital administrative and support services were not included. An attempt was made to evaluate all costs based on the year that they were incurred and then inflated and reported in 2010 Canadian dollars using the Health Care component of the Ontario Consumer Price Index (CPI) ( Health-related resource use and costs for the hip fracture cohort and matched non-hip fracture cohort were identified over a 1-year follow-up period in the primary analysis and during the second year in secondary analyses. Costs were separated into nine distinct categories based on total resource use: acute inpatient hospitalizations, chronic care, emergency visits, same day surgeries, physician services, prescription drug use, rehabilitation, home care, and long-term care;; based on the healthcare utilization data available and as suggested by CADTH [51, 68]. The summary of methods used to cost resource utilization within each dataset is outlined below and in Table 3. All costs applied in our analyses are outlined in Appendix C Table C1.

42 29 Table 3. Cost of healthcare utilization by resource type and database Resource Use Database Costing Acute Hospitalization CIHI DAD RIW * Provincial average cost per weighted case Emergency Department and Same Day Surgeries NACRS CACS RIW * Provincial average cost per weighted case Rehabilitation NRS RCW * Provincial average cost per weighted case Complex Continuing Care CCRS CMI * Average cost per weighted day * length of stay Physician Services OHIP Cost per service (Schedule of Benefits for Physician Services) Home Care OHCAS (pre April Cost per service 2005) and HCD Long Term Care ODB LTC flag Provincial average cost per day * length of stay Prescription Drugs ODB Cost per prescription Inpatient Hospitalizations The number, type and length of hospitalizations were determined from the CIHI DAD, with hospitalization costs calculated using the Resource Intensity Weight (RIW) method [ ]. Traditional costing data multiply length of stay in hospital by an average hospital cost, thus giving equal status to all patient days. CIHI classifies all patients in the DAD by a Case Mix Group (CMG), which was developed to categorize groups of patients with similar clinical and as procedures and interventions received. An RIW is then calculated by CIHI for each CMG group, and adjusted for based on an individual cases age, atypical length of stay, comorbidity level and number interventions received. As a result, an RIW is assigned to each patient representing the level of resources used relative to the average per case value (RIW = ) [105]. Therefore, the higher an RIW is for a CMG, the more resources they are expected to consume. In order to estimate the specific cost for a hospital admission, the provincial average

43 hospital cost per weighted case for Ontario was multiplied by the RIW for the hospitalization (Equation 2). Year-specific provincial average costs per weighted cases are estimated as a sum of total acute care costs divided by the total weighted cases in a given year, and were obtained from the Ministry of Health (Appendix C Table C1). Costs related to physician services provided in hospital are not included as part of the cost per weighted case, as these are provided directly through the Ministry of Health. Although hospital specific costs are more accurate, the use of provincial costs better reflects the average costs of the province and ensures estimates only reflect the services received rather than costs specific to a location. 30 Equation 2 Case Cost = RIW * Provincial Average Cost per Weighted Case (indexed by year) Complex Continuing Care Hospitalizations in Complex Continuing Care (CCC) were determined from the CCRS. Residents in CCRS are classified into one of 44 distinct Resource Utilization Groups (RUG- IIIs), which are analogous to the CMG groupings for hospitalizations [108]. RUG-III groups are services and treatments received. Each RUG-III group is associated with a Case Mix Index (CMI) that is a weight approximating the relative daily cost of care for a resident within a given RUG in comparison to the average level of resource use in the specific patient population. If a patient falls into more than one RUG-III category, they are assigned to the group with the highest CMI. After obtaining the date of admission/discharge and CMI, the general cost for each patient was calculated by multiplying the CMI for the hospitalization by the estimated average cost per unit and the number of days spent in care (Equation 3) [108].

44 31 Equation 3 Case Cost = CMI * Average Cost per Weighted Day (indexed by year) * Length of Stay Individuals who stay in a CCC facility for longer than 90 days are generally reassessed at a -grouped into a RUG-III category. Due to the high number of missing assessment dates in the CCRS, we calculated an average CMI for each individual visit. Additionally, if a CMI was missing for an individual, the mean CMI for males and females in the hip and non-hip fracture cohorts were applied, respectively Emergency Room and Same Day Surgery The NACRS dataset contains all data describing visits to the emergency room, day procedures, and high-cost ambulatory clinics from the fiscal year onwards. Since April 2003 same day surgeries are derived from NACRS and housed in the CIHI-SDS database at ICES. Records in the CIHI-SDS were analyzed separately from those derived directly from the NACRS dataset in order to determine the specific costs associated with same day surgeries. Similar to the methods described for inpatient hospitalizations, individuals in NACRS are grouped based on the Comprehensive Ambulatory Classification System (CACS) methodology which groups individuals based on their main problem, interventions received, age and gender [109]. These CACS groups are then assigned their own resource intensity weight (RIW) to reflect the average resource utilization for the group relative to the mean acute inpatient cost. Costing data for calculation of RIWs are calculated using data from in Ontario and from Alberta [109]. Because RIWs are not available for all years in NACRS, the RIW2008 was used and applied to years 2004/ /2009, with RIW2009 used thereafter.

45 The costs of visits were estimated by multiplying the provincial average cost per weighted case by the RIW for each case, as described above for hospitalizations [110] (Equation 4). 32 Equation 4 Case Cost = CACS RIW * Provincial Average Cost per Weighted Case (indexed by year) Physician Services All physician services including outpatient visits, laboratory services, and diagnostic tests, were identified through the OHIP claims history database. Costs for physician services were determined based on the documented fee code in the OHIP database as a cost per service or cost per encounter (Equation 5). Physicians paid on a fee-for-service basis are compensated based on the rate for the individual service as outlined in the Schedule of Benefits for Physician Services under the Health Insurance Act [111]. However, as a result of increasing capitation and alternative fee-for-service models in Canada, nearly 30% of physicians in Ontario were not under a direct fee-for-service model in 2006 [82]. These blended models such as family health teams or family health groups, as well as salaried physicians are advised to shadow bill for services. Physicians use a shadow billing process whereby they submit the service code provided, however are not paid directly for a specific service. As a result, the corresponding value for the code in the OHIP database is $0. For fee codes with a zero value, the Median amount reimbursed for the fee code for the specific fiscal year within our cohort was used. Billings by physicians enrolled in health service organizations or community health centres as well as some laboratory, psychiatric, and diagnostic services provided in hospitals are not captured in the OHIP database. Outpatient visits to non-physician healthcare professionals (e.g. physiotherapy) are not available through ICES and therefore were not considered.

46 33 Equation 5 Visit Cost = OHIP Fee (indexed by Billing Code) Rehabilitation Length of stay in rehabilitation was determined from the NRS, which captures submission of data from all Ontario facilities with adult inpatient rehabilitation beds as mandated by the Ontario Ministry of Health and Long Term Care. Costs of rehabilitation were based on the Joint methodology and weighting system, developed to reflect inpatient costs of rehabilitation in Ontario [79, 112]. These methods have previously been used in estimating rehabilitation costs in Ontario [46]. The NRS reports one of 87 Rehabilitation Client Groups (RCG) for each patient according to their primary impairment for admission to the program. These RCGs were then collapsed into one of 21 Rehabilitation Groups (RG) and converted into one of 83 Rehabilitation Patient Groups (RPG) based on a patients age, length of stay, and level of disability as measured by the admission Functional Independence Measure (FIM) motor score and admission FIM cognitive score [79]. The FIM motor and cognitive scores were calculated from the NRS based on a sum of each individual component (ranging from a value of 1 to 7, 7 indicating complete independence) (Equation 6 and Equation 7), resulting in maximum scores of 84 and 35 respectively. Equation 6 FIM motor = Sum FIM score (eating, grooming, bathing, dressing upper, dressing lower, toileting, bladder management, bowel management, bed/chair/wheelchair transfer, toilet transfer, locomotion walk/wheelchair, locomotion stairs) Equation 7 FIM Cognitive = Sum FIM score (comprehension, expression, social interaction, problem solving, memory)

47 Each RPG was then converted to a unique Rehabilitation Client Weight (RCW), based on an individuals trim point and length of stay, representing the average resource use for individuals within an RPG. Weights for years 2004/2005 to 2007/2008 were calculated using the original rehabilitation weighting system developed by the JPPC [79], with updated weights used for years 2008/2009 and 2009/2010. If the length of stay was less than the short term trim, the RCW was set to a fixed rehabilitation cost weight. If the length of stay was greater than the short-term trim number of days and less than the trim value indexed by RPG, then the RCW was set to the RCW indexed by RPG. However, if the length of stay was greater than the trim (indexed by RPG), then the RCW was set to the RCW (indexed by RPG) plus the trim (indexed by RPG) subtracted from the rehabilitation length of stay, then multiplied by the Per Diem Rehabilitation Cost Weight (PDW) (indexed by the RPG). Rehabilitation costs can then be determined as the average cost per weighted visit multiplied by the RCW for the case (Equation 8). 34 Equation 8 Case Cost = RCW for each case * Average Cost per Weighted Rehabilitation Case (indexed by year) Home Care Home care is provided by the Ontario Ministry of Health and Long Term Care and includes services such as visiting health professionals (nurses, physiotherapists, occupational therapists, social workers, speech-language pathologists and dietitians), personal care and support, homemaking, and community health services [113]. Home care services are captured in the OHCAS up to 2005 and the HCD from 2005 on. Each record represented a single service received, except personal/homemaking services and nursing shifts, which represented the time in hours of service received. Costs for home care were determined by applying an average cost per service (or hour) for each service provided (Equation 9) which have previously been

48 obtained from the Community Care Access Centre, Toronto. Service costs for four of the possible services coded were unavailable (placement services, enterostomal therapy, laboratory technology, and other) and were thus considered to have a cost of zero dollars. The cost for case management services was applied only once annually, accounting for the length of time receiving home care services within the specified year. 35 Equation 9 Case Cost = Cost per visit (indexed by year) by type of service provided Long Term Care Patients are required to pay a co-payment for accommodations in a LTC facility;; however the Ministry of Health and Long Term Care covers costs for resident care. The levels of care classification system provides resident classification for LTC facilities, however this is a crosssectional, annual review, and is therefore not appropriate for identifying all individuals in an LTC facility. Length of stay in LTC was determined from the ODB database based on LTC flags for drugs dispensed to LTC patients. Following prior methods [72], the date of entrance was determined as the first of a minimum of 2, similarly determined as the first of 2 non LTC-flagged drug claims or death. The total length of stay was then multiplied by the provincial average cost per day paid by the Ministry of Health and Long Term Care based on basic or standard accommodation costs (Equation 10). Equation 10 Case Cost = Ministry of Health per diem cost (indexed by year) * Length of Stay

49 Prescription Drugs Prescription drug utilization and costs were obtained directly from the ODB based on the total amount paid to the pharmacy from the Ministry of Health, including dispensing fees (Equation 11). Drug utilization and costs were further stratified to evaluate those therapies used to treat/prevent osteoporotic fractures (alendronate, etidronate, risedronate, zoledronic acid, raloxifene, calcitonin, teriparatide) (Appendix D Table D2) as well as drugs used to treat pain (NSAIDs, opioids) (Appendix D Table D3). Denosumab was not approved for use in Canada until September of 2010 and thus was not included in the analysis. Equation 11 Prescription Cost = Fee paid per prescription 3.7 Statistical Analysis Descriptive Statistics All analyses were performed using SAS version 9.1 (SAS Institute Inc., Cary, NC). Balance between the hip fracture cohort and matched non-hip fracture cohort at baseline was assessed using the standardized difference, with a standardized difference < 0.1 indicating appropriate balance [114]. Process measures (entering LTC, rehabilitation, chronic care, or requiring a second hospitalization, receiving a DXA scan (Appendix D Table D1) or osteoporosis/pain medications (Appendix D Table D2 and Table D3) and outcomes (second hip fracture, mortality after fracture) were reported as both the number and percent and compared using the McNemar Chi Square test for categorical variables and the paired t-test for continuous variables.

50 Fracture Related Costs Total and Attributable Fracture Related Costs A total count of hip fractures were tabulated based on sex and age group (66-69, 70-74, 75-79, 80-84, 85-90, 90+). Costs for hip fracture patients and non-hip fracture controls were stratified by: individual costing component (i.e. acute hospitalizations, physician services, chronic care, pharmaceuticals, homecare, LTC, rehabilitation), sex (male/female), age group, and residence status at baseline (LTC vs. community). It is described in the economic literature that in timerestricted costing analyses (e.g. 1 year cost), the total cost is the cost accumulated up to the time of interest or death, whichever occurs first [115]. As a result, the total costs were determined by the summation of each costing component and summarized as the mean cost over 1 year for both hip fracture patients and matched non-hip fracture patients. The attributable hip fracture cost was then determined by subtracting costs from the non hip fracture cohort from the costs of the hip fracture cohort, as has been previously recommended [56, 57]. Variance estimation around attributable mean costs was determined using bootstrapping methods, by drawing multiple samples from each of the populations with replacement, and replicating 1000 times [57]. To estimate the global attributable cost of hip fractures in Ontario, attributable costs were summed across all individuals and an average cost per fiscal year was calculated Subgroup Fracture Related Costs Survivors and Death Because hip fractures result in a high rate of death post fracture, many cases may display excess costs associated with death. On the contrary, controls that survive may have accumulated greater costs simply due to longer survival time. In order to account for this potential issue, prior studies of fracture costs have often excluded individuals who died during the follow-up period. This method, however, results in a loss of data as well as bias costs towards the costs of patients with longer survival times [116]. To account for this, we stratified costs amongst concordant pairs

51 who survived one and two years post index date and concordant pairs who died during the first and second year of follow-up Data Access and Ethics Approval Both the Sunnybrook Hospital Research Ethics Board and the University of Toronto Research Ethics Board provided ethics approval for this study. Drs. Cadaratte, Krahn and Wodchis are ICES scientists and Milica Nikitovic received ethical training by a privacy officer at ICES prior to gaining student privacy law and is thus legally permitted to receive and use personal health information for health services research. ICES maintain policies, practices, and procedures that are approved and regularly audited by the Information and Privacy Commissioner of Ontario ( All data were provided de-identified and data analyses were solely completed by Milica Nikitovic at ICES at Sunnybrook Hospital.

52 Chapter 4 Results 4.1 Cohort Selection Hip fracture patient inclusions and exclusions are outlined in Figure 1A, with identification of seniors in the RPDB who did not experience a hip fracture shown in Figure 1B. Propensity matching statistics and baseline characteristics are depicted in Table 4 Table Hip Fracture Cohort Of 36,253 (73.5% female [n=26,648]) hip fracture patients identified between April 2004 and March 2008, 85.7% (86.3% females [n=23,009], 83.9% of males [n=8,055]) were eligible to match after applying initial exclusion criteria (Figure 1A). Among the 3,639 females and 1,550 males excluded, most were as a result of prior hip fracture (56.4% females [n=2,052], 29.6% males [n=459]) or malignant neoplasm (34.0% females [n=1,238], 52.2% males [n=809]) in the year prior to index Non-Hip Fracture Cohort Among 2,109,161 individuals in the RPDB aged 66 years by March 2008 (55.4% female [n=1,168,713]), 72.4% of females (n=681,173) and 74.4% (n=869,041) of males were eligible as potential non-hip fracture matches (Figure 1B). Of the 299,672 females and 258,665 males excluded, the primary reason was age less than 66 years at index (40.1% females, 12.1% males), being a non-ontario resident (29.8% females, 31.74% males), or having a diagnosis for a malignant neoplasm (14.7% females, 19.9% males). 39

53 Cohort Matching Of those individuals in the hip fracture cohort identified as eligible for matching based on initial exclusion criteria, 1.2% (1.1% females, 1.4% males) were further excluded due to unavailability of matching variables (Figure 1A). Propensity score diagnostics are depicted in Table 4. Of the 30,687 (22,745 females, 7,942 males) eligible hip fracture patients, 97.9% (22,418 females and 7,611 males) were successfully matched (Table 5) Baseline Characteristics Figure 2 illustrates the number of incident hip fractures occurring across each of the four fiscal years (04/ /2005 to 04/ /2008). Total hip fracture numbers were stable across each fiscal year, with a very gradual increase in the number of female hip fracture patients from 2005 through to 2008, and a similar pattern in males from 2004 to The number of hip fractures increased greatly with age, until age 90 years and older where there was a slight decrease in fractures from the previous year (Figure 3). Table 6 presents descriptive characteristics among variables matched at baseline. Based on a standardized difference of <0.10 [103], individuals were well balanced between the hip and non-hip fracture cohorts on all matched variables (Table 6). Mean age at hip fracture was 83.3 (SD 7.1) years for females and 81.3 (SD 7.1) years for males. Among females, 21.4% (n=4,797) resided in a LTC facility at index in comparison to 17.8% (n=1,352) of males. Table 8 further stratifies LTC residence at baseline by age group;; ranging from 5.2% of females and 4.3% of males aged to 32.1% of females and 22.3% of males in those aged 90 years and over. Individuals were fairly evenly distributed across income quintiles among both females and males (Table 6). The average hip fracture patient resided in a non-rural community (RIO <45), with a mean RIO of approximately 17 were acute minor (67% females, 68% males) and acute major (74% females, 75% males). The

54 proportion of individuals in each CADG was similar between sexes, except more men than women fell into the chronic medical unstable category (approximately 55% females and 68% males). Table 7 describes baseline characteristics among unmatched variables. Individuals were well balanced between the two cohorts across LHINs, with the greatest number of fractures and 8 (Central). Cohorts were also well balanced based on a confirmed diagnosis of osteoporosis (approximately 8-9% of females and 3-4% of males) as well as having a DXA test in the year prior to index (approximately 11% of females and 4% of males). However, significantly more individuals in both the female (31.9% vs. 27.7, standardized difference = 0.1) and male (9.9% vs. 5.9%, standardized difference = 0.15) hip fracture cohorts had received an osteoporosis medication as well as had a greater number of non-hip fractures (vertebral, humerus, and in the year prior to index (standardized difference 0.10) Outcomes post index date Table 9 and Table 10 summarize the number of hip fracture patients with a subsequent hip fracture as well as the number of individuals in each cohort who died, stratified by age group and residence status at baseline within the first year (Table 9) and second year (Table 10) after index Second Hip Fracture Within 365 days of index hip fracture, 1.2% (n=278) of females and 0.8% (n=63) of males incurred a subsequent hip fracture (Table 9). This increased into the second year post index fracture date ( days), with 1.7% (n=385) and 1.4% (n=105) of the original hip fracture cohort incurring a second hip fracture (Table 10). When considering only those who survived the first year (77.8% [n=17,441] females and 66.5% [n=5,065] males), 2.2% of females and

55 2.1% of males had a second hip fracture within the second year. No hip fractures were seen within the non-hip fracture cohort, as this was exclusion for cohort entry Death after index date Significantly more females (22.2% vs. 9.3%, p<0.001) and males (33.5% vs. 10.8%, p<0.001) in the hip fracture cohort died during the first year after index in comparison to those in the nonhip fracture cohort (Table 9). Death increased greatly with age, increasing to 35.9% in females in the hip fracture cohort (vs. 17.8% non-hip fracture cohort) and 51.6% among males (vs. 20.5%) aged 90 years and older. Death remained elevated in the hip fracture cohort into the second year post index (9.1% vs. 8.3%, p <0.001 for females and 11.3% vs. 9.4%, p <0.001 for males) (Table 10). There was no statistically significant difference in death among females aged years or males aged 80 years and over (p>0.05). However, when second year deaths were evaluated among only those who survived the first year, the percentage of deaths increased greatly among the hip fracture cohort in both females (12.3% vs. 9.1%, p <0.001) and males (17.0% vs. 10.5%, p<0.001). Over the entire 2-year follow-up period, 31.3% of females (n=7,025) and 44.7% of males (n=3,406) died following their hip fracture in comparison to 17.6% (n=3,945) and 20.2% (n=1,538) in the non-hip fracture cohorts (p<0.001), respectively. This corresponded to a marginal increase in death of 13.7% in females and 24.5% in males incurring a hip fracture over the two-year period LTC versus community residence at baseline Death was significantly greater among individuals in LTC at baseline in comparison to those in the community within both cohorts (Table 9 and Table 10). Among those in LTC, a greater number of deaths occurred in the hip fracture cohort for both females (37.0% vs. 22.6%, p

56 <0.001) and males (53.6% vs. 28.8%, p <0.001). Deaths in LTC increased significantly with age, with approximately 45% of females in the hip fracture cohort aged 90 years and over dying within one year, in comparison to 29% in the non-hip fracture cohort. Similarly, within one year of index, 65% versus 37% of males aged 90 years and over in LTC died within the hip and nonhip fracture cohorts, respectively Time to Death after Hip Fracture When evaluating the entire hip fracture cohort, 5% of individuals died within the first 30 days of hip fracture. Among those who died within the first year, the mean time to death was 104 days (3.5 months) in females and 94 days (3.1 months) in males. Approximately 13% and 14% of deaths among males and females occurred within the first seven days of hip fracture, increasing to 34% and 44% by 30 days, respectively. 4.3 Health Services Utilization Year One Total healthcare resource use in the first year post hip fracture is depicted in Table 11 Table 16, stratified by sex and age group. Total sums of visits and prescriptions in each cohort are summarized in Appendix E Table E Acute Hospitalizations All individuals in the hip fracture cohort were required to have a hospital diagnosis code for cohort inclusion, and thus 100% had an acute care hospitalization (Table 11). This was in comparison to 19.1% of females and 22.6% of males in the non-hip fracture cohort (p<0.001). The hip fracture cohort had an excess of 7,914 (500%) additional hospitalizations among

57 females (14,167 vs. 2,837) and 10,448 (589%) among males (15,459 vs. 2,622) after the initial hospitalization (Table E3). 44 The average length of stay for the initial hip fracture hospitalization was 13.7 days (SD=18.1, median = 9, IQR = 5, 15) for females and 15.6 days (SD=22.4, median = 9, IQR = 6, 17) for males Emergency Department Visits Approximately 86.5% of males and females within each age group in the hip fracture cohort had an emergency department visit within the first year of follow-up (Table 11). This was an absolute increase of nearly a 50% in comparison to the non-hip fracture cohort (p < 0.001), who had 36.6% females and 38.1% with at least one visit Same Day Surgeries Little marginal difference was observed in the number of individuals requiring same day surgeries in the two cohorts (9.5% vs. 12.8% for females and 14.1% vs. 19.4% males) (Table 11). The general trend was towards a greater number of same day surgeries in the non-hip fracture cohort Physician Services All hip fracture patients (99.995%) used physician services, as identified through the OHIP database (Table 12). This was similar, although statistically greater (p<0.001), to the non-hip fracture cohort with approximately 95% of individuals using one of more physician services during the first year of follow-up. Very few individuals in either cohort received a DXA test in the year after index (Table 13). Following a hip fracture, only 4.3% of females (n=956) and 2.5% of males (n=189) had a DXA test in the subsequent year. The greatest proportion of tests was in those aged years of

58 age (12.4% females and 5.8% males) and least in those aged 90 years and over (0.9% females and males). The percentage of non-hip fracture patients receiving a DXA test was similar to that seen in the hip fracture cohort (5.2% females and 1.9% males), despite statistical significance (p<0.05) Complex Continuing Care Among the hip fracture cohort, nearly 18% of both females and males were sent to a complex continuing care hospital;; increasing to approximately 20% in those over the age of 85 years (Table 12). This was significantly greater than the non-hip fracture cohort, where less than 1.5% of individuals were in a chronic care facility during the first year of follow-up;; reflecting the lower number of acute care hospitalizations in this cohort during follow-up Rehabilitation A significant proportion of individuals incurring a hip fracture required rehabilitation within the year after index (32.8% of females, 31.9% of males) in comparison to those in the non-hip fracture cohort (1.4% of females, 1.1% of males) (p<0.001) (Table 12). Among the 19% and 23% of females and males in the non-hip fracture cohort hospitalized within the first year, 7.2% and 4.8% were sent to a rehabilitation facility, respectively Long Term Care At baseline, 21.4% of females and 17.8% of males were in a LTC facility within each of the two cohorts. Within the first year post index, the number of females in LTC within the hip fracture cohort increased to 38.0%, while those in the non-hip fracture cohort only increased to 24.6% (p<0.001) (Table 14). Similarly, the number of males in LTC increased by 12.2% in the hip fracture cohort, and only 2.3% in the non-hip fracture cohort (p<0.001). The total number residing in a LTC facility increased greatly with increasing age. Among those who resided in a

59 LTC facility at index, nearly all individuals who survived returned to LTC in the subsequent year (greater than 99% in both cohorts) (Table 14). However, among those hip fracture patients who resided in the community at index, nearly 24% of females and 19% of men entered a LTC facility in the year after their hip fracture. (Table 14) Home Care Nearly 70% of females and 66% of males received one or more home care services during the first year post index hip fracture (Table 15). This was significantly greater than those in the nonhip fracture cohort (p<0.001), with an absolute difference of 43% and 45% in females and males, respectively. Receipt of home care services was generally greatest in those with hip fractures who were younger, with 80.1% and 77.6% of females and males aged requiring home care in comparison to 57.7% and 58.2% in those over the age of 90, respectively. This likely reflects the greater number of younger patients returning to the community subsequent to hip fracture in comparison to those who are older (Table 8). On the contrary, home care services increased from approximately 10% in those less than 69 years of age to 36% in those 90 years or over in the non-hip fracture cohort Prescription Medications Among both the hip and non-hip fracture cohorts, greater than 93% of women and 85% of men filled at least one prescription through the Ontario Drug Benefit Program in the year following their index date (Table 15). The number filling a prescription was greater among the younger age groups in the hip fracture cohort, shifting to more in the non-hip fracture cohort with increasing age. This may reflect a combination of longer hospitalization periods among older individuals with a hip fracture, where prescriptions are not captured under ODB. Although overall a greater number of individuals in the non-hip fracture cohort filled at least one prescription in comparison to the hip fracture cohort (p<0.05), the number of total prescriptions

60 filled was greater in the hip fracture cohort, with an excess of 583,103 prescriptions of among females and 127,460 among males (Appendix E Table E3). 47 We further stratified by type of prescription (osteoporosis, NSAIDS and opioids) to determine if usage was greater among those who incurred a hip fracture (Table 16). Among those in the hip fracture cohort, less than half (43.7%) of females received an osteoporosis treatment 1, with only one fifth (21.7%) of males filling a prescription. This was an increase of 11.8% from baseline in both the females (31.9% at baseline, 43.7% at one year) and males (9.9% at baseline, 21.7% at one year). Utilization of osteoporosis treatments at 1 year was significantly lower in the non-hip fracture cohort (27.6% of females and 6.6% of males (p<0.001)), which remained stable in comparison to the year prior. Approximately half of individuals in the hip fracture cohort (53.7% females, 48.7% males) received an opioid 2 prescription in the year following hip fracture in comparison to 28.2% of females and 24.9% of males in the non-hip fracture cohort (p<0.001). Non-steroidal anti-inflammatory drug 3 use was greater among the non-hip fracture cohort, although absolute differences were small (4.5% in females and 5.4% in males) Year Two Health service resource use was substantially lower in the second year post index date in the hip fracture cohort in comparison to the first year;; however resource use generally remained elevated in comparison to the non-hip fracture cohort (Appendix E Table E4 -Table E6). 1 Osteoporosis medications = alendronate, risedronate, etidronate, raloxifene, calcitonin, teriparatide, zoledronic acid (Appendix D, Table D2) 2 Opioids = codeine, fentanyl, hydromorphone, levorphamol tartrate, meperidine, morphine and oxycodone combinations (Appendix D, Table D3) 3 NSAID = ASA, diclofenac, diflunisal, etodolac, floctafenine, indomethacin, ketoprofen, ketorolac tromethamine, mefenamic acid, meloxicam, naproxen, oxaprozin, piroxicam, sulindac, tenoxicam, tiaprofenic acid (Appendix D. Table D3)

61 The number of DXA tests increased in the second year in comparison to the first year in both the hip- and non-hip fracture cohorts (Table 17), although this remained at less than 10% of females and 4% of males. 48 Among those in the hip fracture cohort who survived the first year (n= 17,411 females [78%], n= 5065 [67%]), 49.8% of females and 25% of males were on an osteoporosis medication in the second year, an absolute increase of 6.1% and 3.3%, respectively (Table 17). The use of osteoporosis medications did not increase among survivors of the non-hip fracture cohort (27.8% in year 1 to 28.8% year 2 among females;; 6.6% year 1 to 6.9% year 2 among males). 4.4 Healthcare Costs Healthcare costs were evaluated across the entire continuum of health resource use. All costs were reported as total mean costs among the hip fracture cohort and non-hip fracture cohort, as well as the mean attributable (incremental) hip fracture cost. Confidence intervals were calculated using bootstrapping with replacement. Bootstrapped attributable mean costs did not vary greatly from estimated attributable mean costs, and therefore the mean attributable costs are reported as exact differences (data not shown). Missing values and cost imputations are summarized in Appendix E Table E1 and Table E Total and Attributable Costs (year 1) Among all individuals in the hip fracture cohort, the total direct cost in the first year post index was $52,232 (95% CI $51,748, $52,716) for females and $54,289 (95% CI $53,295, $55282) for males (Table 18). After accounting for total costs among matched controls the mean 1-year direct attributable cost of hip fracture was $36,929 (95% CI $36,380, $37,466) for females and $39,479 (95% CI $38,331, $40,677) for males (Table 18). Attributable hip fracture costs

62 accounted for 71% of the total cost among females and 73% of the total cost among males. Although total costs among those in the hip fracture cohort increased with age, total costs of matched controls demonstrated a corresponding increase in costs. As a result, attributable costs generally decreased over time after the age of 75 years in females and 70 years in males. 49 Across the four fiscal years evaluated (04/ /2005 to 04/ /2008), the cumulative mean annual attributable cost of hip fractures in Ontario was estimated at $206.9 million for females and $75.1 million for males Attributable Costs by Resource Use Table 19 - Table 21 outline the total and attributable 1-year direct healthcare costs of hip fractures by each of the nine types of resource use;; stratified by sex and age groups. Figure 4 depicts the proportion of attributable costs driven by each resource type among females (4A) and males (4B). Acute hospitalizations accounted for the greatest proportion of attributable hip fracture cost ($18,792 [51%] for females and $21,289 [53.9%] for males) (Table 19). Among the total acute hospitalization costs, the greatest cost resulted from the index hospitalization for hip fracture, accounting for 75.6% ($14,210) and 75.9% ($16,158) of all acute hospitalization costs among females and males, respectively (Figure 4). This cost increased with age, ranging from $12,550 and $16,185 in females and males aged 66-69, to $14,734 and $16,771 in those aged 90 years and older. Overall, the initial hospitalization accounted for 39% and 41% of the total attributable hip fracture cost (Figure 4). After accounting for increased acute hospitalization costs, the greatest drivers of attributable hip fracture costs were as a result of complex continuing care ($5,588 [15%] females, $6,468 [16%] males), rehabilitation ($5,251 [14%) females, $5,453 [14%] males), and physician services

63 billed through OHIP ($3,103 [9%] females, $3,265 [8%] males) (Table 20 and Figure 4). The total mean cost of long term care was considerable among the hip fracture cohort ($9,419 females, $6,746 males), however costs were also high among the matched cohort ($6,949 females, $5,494 males);; resulting in attributable costs of approximately $2,470 in females and $1,252 in males Table 21. Long term care costs did not have a significant impact on attributable costs in males over the age of 90 ($609 (95%CI -$263, $1,492)). 50 Home care services accounted for approximately 3%-4% of attributable costs in both males ($1,345) and females ($1,135). Although there were attributable costs associated with emergency department visits (Table 19) and prescription medications among females (Table 21), these played a very small role (<1%) in the overall 1-year attributable hip fracture cost. Prescriptions among males (Table 21) and same day surgeries among both females and males (Table 19) were not found to be attributable to the cost of hip fractures;; with overall costs greater among the non-hip fracture cohort (Prescriptions = -$43 among males;; Same day surgeries = -$33 females and -$58 males) (Table 19) Total and attributable costs (year 2) Costs remained elevated into the second year post fracture (Table 22), with an estimated mean attributable cost of $4,597 (95% CI $4,233, $4,972) for females and $3,038 (95% CI $2,334, $3,764) for males. However, the number of individuals alive in the non-hip fracture cohort at the beginning of the second year was substantially greater than the hip fracture cohort (n=287 more females [16,653 hip fracture cohort vs. 20,328 non-hip fracture cohort] and 1,722 males [5,065 hip fracture cohort vs. 6,787 non-hip fracture cohort]). This was reflected among males over the age of 85 years, where no significant attributable hip fracture costs were seen (Table 22). The greatest proportion of attributable costs in the second year was as a result of long term care ($3,213 [70%] females, $1,575 [52%] males), acute hospitalizations ($574 [12%] females, $785

64 [26%] males), and home care ($364 [8%] females, $531 [17%] males) (Appendix E, Table E7 to Table E9). 51 Among the 30,029 hip fracture patients, the mean cumulative direct attributable hip fracture cost in the second year was $31.5 million Cost Stratifications A summary of mean total and attributable hip fracture costs stratified by residence status, number of hip fractures and survival status are summarized in Table 23. Costs further stratified by age groups are reported in Appendix E Table E10-Table E LTC Status at baseline and follow-up Residence greatly influenced hip fracture costs. Among those individuals in the community at time of fracture, 1-year attributable costs increased to $43,930 ($53,187 vs. $9,258) in females and $46,025 ($55,878 vs. $9,853) in males (Table 23). On the contrary, LTC residence at baseline resulted in lower attributable costs ($11,213 in females and $9,133 in males) as a result of the high total costs incurred by both the hip and non-hip fracture patients in LTC ($48,733 vs. $37,510 in females and $46,918 vs. $37,758 in males) (Table 23). Attributable costs among those in LTC at baseline further decreased with increasing age in both sexes (Appendix E Table E10). When evaluating residence status after hip fracture, hip fracture patients who were in the community at baseline and were subsequently transferred to a LTC facility (24% females and 19% males) had total costs increase to $83,447 among females and $89,920 among males. These costs were substantially greater than those who returned to the community (total cost of $43,830 in females and $47,918 in males).

65 Second Hip Fracture Second hip fractures (1.2% of females and 0.8% of males in year one;; 1.7% of females and 1.4% of males in year two) (Table 9 and Table 10), resulted in substantially higher attributable costs (Table 23). Patients who incurred a second hip fracture in the year after their index fracture had costs increase nearly double that of those of the entire hip fracture cohort ($70,621 in females and $73,638 in males). Fractures that occurred in the second year post index resulted in an attributable cost of $39,895. This was slightly higher than observed in the first year for index hip fracture in females ($36,929), with the attributable cost in males increasing by $12,979 ($39,479 year 1 to $52,458 year 2) Survival Status Costs associated with death and survival were evaluated only among concordant pairs, whereby both the hip fracture patient and their matched non-hip fracture control survived or died. Overall, 91.9% of females and 91.4% of males in the hip fracture cohort who survived the first year were matched to an individual who also survived. Among matched survivors, the mean 1- year attributable costs were $41,149 ($54,218 hip fracture cohort, $13,069 non-hip fracture cohort) and $45,742 ($57,390 hip fracture cohort, $11,648 non-hip fracture cohort) in females and males respectively (Table 23 and Table E13). This reflected a corresponding cost increase of $4,220 and $6,262 in comparison to the attributable cost estimate among all individuals in the first year. Costs were further evaluated in the second year among only those individuals who survived the first year (Table 23 and Figure 5). Second year attributable costs among first year survivors were estimated at $9,017 ($22,983 hip fracture cohort, $13,966 non-hip fracture cohort) in females and $10,347 ($22,909 hip fracture cohort, $12,563 non-hip fracture cohort) in males. Cost drivers in the second year varied from those in the first year, with the greatest cost

66 proportion of attributable costs shifting from acute hospitalizations, complex continuing care and rehabilitation to LTC, acute hospitalizations and home care (Figure 5). LTC costs accounted for 62% of the attributable cost in females ($5,641) and 44% in males ($4,536). Acute hospitalization costs remained elevated, accounting for 14% of costs in females ($1,295) and 25% in males ($2,584). Rehabilitation and continuing care costs decreased significantly from the first year, where combined they accounted for less than 5% of the total cost in females and 9% in males in the second year. Overall, first year survivors of hip fractures added an additional direct attributable healthcare cost of $64.5 million to the province of Ontario in the second year. 53 Among the 4,977 females and 2,546 males who died in the first year after hip fracture, 13.8% and 15.2% of matched individuals in the non-hip fracture cohort were identified who also died within the year. Among these deaths, the mean attributable cost was $10,935 (95% CI $8,347, $13,364) in females and $14,451 (95% CI $10,062, $18,826) in males (Table 23). When extended into the second year post index, only 11% of concordant pairs who died were identified. Attributable hip fracture costs were minimal among these individuals ($256 (95%CI $-4,297, $4,939) for females and $469 (95%CI $-5,073, $6,383) for males);; with confidence intervals indicating uncertainty in costs. These same individuals had similar total costs in the year prior (first year after index date) as the overall cohort that survived the first year for both females ($38,466 (95%CI $36,376, $39,487)) and males ($46,232 (95%CI $38,285, $49,470)) Prior non-hip fractures Sensitivity Analysis Costs were further stratified among individuals with prior non-hip fractures in order to assess the impact of this imbalance between the hip and non-hip fracture cohorts (Appendix E Table E14 Table E16). Both cases and controls with prior non-hip fractures had greater total costs, particularly among males. Males were balanced between cohorts when evaluating prior vertebral fractures (1.1% vs. 0.5%, standardized difference <0.1), however, significantly more females in

67 the hip fracture cohort had a prior clinical vertebral fracture (1.5% vs. 0.49%, standardized difference = 0.1)). Vertebral fractures resulted in the greatest total cost increase among both cohorts. When only those females with prior vertebral fractures in the hip fracture cohort were considered, attributable costs increased by $4,049. Similarly, when evaluating female hip fracture patients matched to a control with a prior vertebral fracture, attributable costs decreased 2.2% of females and 11.9% vs. 1.8% of males in the hip and non-hip fracture cohorts, respectively). Attributable costs were not greatly affected when only those in the hip fracture 2 for females and no increase ($-221) among males). However, attributable costs decreased when (decrease of $4,030 (females) and $8,347 (males)). 54 Most notably, when all individuals with prior non-hip fractures were excluded from the analysis, attributable hip fracture costs were not affected ($36,929 vs. $36,762 in females and $39,479 and $39,267 in males) (Table E16). Therefore prior fractures had little effect on the overall attributable hip fracture cost estimates in our study, but may result in a slight overestimate of costs among a small sub-population of individuals.

68 Figures and Tables Figures A. Hip Fracture Cohort B. Non-Hip Fracture Cohort A. All individuals in RPDB Aged 66+ by Mar08 N = 1,168,713 females N = 940,448 males A. All individuals in RPDB Aged 66+ by Mar08 N = 1,168,713 females N = 940,448 males B. Hip Fracture Diagnosis (Apr04-Mar08) N = 26,648 females N = 9,605 males Age <66 years at index N = 341 (1.3%) females N = 261 (2.7%) males Prior Hip Fracture N = 2,052 (7.7%) females N = 459 (4.8%) males Trauma Code (+/-7d of index) N = 26 (0.1%) females N = 52 (0.5%) males Non-OP dispensing BP/CCT N = 12 (0.1%) females N = 81 (0.3%) males Malignant Neoplasm N = 1,238 (4.7%) females N = 809 (8.4%) males N = 27 (0.1%) females N = 19 (0.2%) males Missing OHIP number/ikn N = 0 (0%) females N = 0 (0%) males Non Ontario Residents N = 12 (0.%) females N = 3 (0%) males Death prior to index N = <6 (0%) females N = 0 (0%) males Gender coded as missing N = 0 (0%) females N = 0 (0%) males Age <66 years at index N = 120,262 (10.3%) females N = 114,178 (12.1%) males Prior Hip Fracture N = 44,891 (3.8%) females N = 14,068 (1.5%) males Trauma Code (+/-7d of index) N = 32 (0.00%) females N = 46 (0.00%) males Non-OP dispensing BP/CCT N = 1331(0.1%) females N = 387 (0.04%) males Malignant Neoplasm N = 44,030 (3.8%) females N = 51,558 (5.5%) males N = 852 (0.1%) females N = 698 (0.1%) males Missing OHIP number/ikn N = 0 (0%) females N = 0 (0%) males Non Ontario Residents N =89,188 (7.6%) females N = 82,115 (8.7%) males Death prior to index N = 69,221 (5.9%) females N = 63,597 (6.8%) males Gender coded as missing N = 0 (0%) females N = 0 (0%) males Hip fracture within 2-years of follow-up N = 13,768 (1.2%) females N = 5,158 (0.6%) males Eligible for matching based on exclusion criteria N = 23,009 (86.3%) females N = 8,055 (83.9%) males Unavailability of Matching Variables N = 264 (1.1%) females N = 113 (1.4%) males Eligible for matching based on exclusion criteria N = 869,041 (74.4%) females N = 681,173 (72.4%) males Unavailability of Matching Variables N = 9,484 (1.1%) females N = 8,998 (1.3%) males Hip Fracture Cohort eligible for potential match to non-hip fracture cohort N = 22,745 (85.3%) females N = 7,942 (82.7%) males Non-Hip Fracture Cohort eligible for potential match to hip fracture cohort N = 859,557 (73.5%) females N = 672,175 (71.5%) males Figure 1. Hip and Non-Hip Fracture Cohort Identification (exclusions not mutually exclusive)

69 56 Figure 2. Total number of incident hip fractures among females and males by fiscal year Figure 3. Distribution of index hip fractures by sex and age group

70 57 A. Females (Attributable Cost = $36,929 (95%CI $36,380, $37,466)) B. Males (Attributable Cost = $39,479 (95%CI $38,331, $40,677)) Figure 4. Summary of 1-year direct attributable hip fracture costs for females (A) and males (B) by type of health resource use

71 58 A. Females (Attributable Cost = $9,017 (95% CI $8,578, $9,471)) B. Males (Attributable Cost = $10,347 (95% CI $9,417, $11,275)) Figure 5. Summary of second year direct attributable hip fracture costs for females (A) and males (B) who survived year one by type of health resource use