1 COMPARATIVE SAFETY AND EFFECTIVENESS OF TOTAL HIP ARTHROPLASTY ARTICULATING SURFACES By JONATHAN SCHELFHOUT A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMEN TS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2014
2 Â© 2014 Jonathan Schelfhout
3 To my mother Teresa, my father Steve, and my sister Jennifer for always being there for me
4 ACKNOWLEDGEMENT S I thank my advisor, Abraham Har t zema, for the guidance and supervision he has given me during my doctoral education . He has been a better mentor than I could have ever hoped for. I would also like to thank my supervisory committee members Teresa Kauf, Ri chard Segal and Jill Boylston Herndon for their generous advice and continual support. I would also like to express my gratitude to the students, faculty, and staff of Department of Pharmaceutical Outcomes and Policy for their support and friendship. In pa rticular I would like to give special thanks to Jill Hunt for always going the extra mile in everything she did for me. I would like to thank my project supervisor, Anna Ghambaryan, for her contributions and continual support. I would also like to acknowl edge Danica Marinac Dabic and Nilsa Loyo Berrios for their contributions in the planning and interpretation of the analysis. For funding throughout the various phases of my doctoral education I would like to thank the Pharmaceutical Outcomes and Policy Dep artment, the American Foundation for Pharmaceutical Education (AFPE) and the Oak Ridge Institute for Science and Education (ORISE) . Neither this work nor my graduate education would have been possible without their generous support.
5 TABLE OF CONTENTS p age ACKNOWLEDGEMENTS ................................ ................................ ............................... 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 LIST OF ABBREVIATIONS ................................ ................................ ............................. 9 ABSTRACT ................................ ................................ ................................ ................... 11 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 13 Background ................................ ................................ ................................ ............. 13 Need for Study ................................ ................................ ................................ ........ 18 Research Questions and Hypotheses ................................ ................................ ..... 21 Research Objective 1 ................................ ................................ ....................... 22 Research Objective 2 ................................ ................................ ....................... 23 2 LITERATURE REVIEW ................................ ................................ .......................... 24 Epidemiology of THA ................................ ................................ .............................. 24 Epidemiology of Study Outcomes ................................ ................................ ........... 27 Revision ................................ ................................ ................................ ............ 27 Causes of Revision and Other Complications ................................ .................. 34 Periprosthetic osteolysis and aseptic loosening ................................ ......... 34 Dislocation ................................ ................................ ................................ . 34 Increased met al ion levels, metalosis, and cancer ................................ ..... 35 Fracture ................................ ................................ ................................ ...... 37 Periprosthetic joint infection (PJI) ................................ ............................... 37 Deep vein thrombosis (DVT) ................................ ................................ ...... 38 3 METHODOLOGY ................................ ................................ ................................ ... 39 Nationwide Inpatient Sample (NIS) ................................ ................................ ......... 39 Multi Payer Claims Database (MPCD) ................................ ................................ .... 40 Research Objective 1: Utilization of Articulating Surface Codes ............................. 42 Research Objective 2: Comparative Safety Analysis ................................ .............. 43 Risk of Revision ................................ ................................ ................................ 44 Secondary Outcomes ................................ ................................ ....................... 46 Early Revision Risk ................................ ................................ .......................... 46 Sensitivity Analysis ................................ ................................ ................................ . 47 Abbreviated Follow up ................................ ................................ ...................... 47
6 Propensity Scores ................................ ................................ ............................ 47 Propensity score calculation ................................ ................................ ...... 48 Propensity score stratifica tion ................................ ................................ .... 48 Matching ................................ ................................ ................................ ........... 50 Multiple Imputation ................................ ................................ ........................... 51 4 RESULTS ................................ ................................ ................................ ............... 54 Description of the Cohort ................................ ................................ ........................ 54 MPCD Cohort ................................ ................................ ................................ ... 54 NIS Cohort Characteristics and Com parison ................................ .................... 55 Research Objective 1: Utilization of Articulating Surface Codes ...................... 57 Research Objective 2: Comparative Safety Analysis ................................ .............. 59 Risk of Revision ................................ ................................ ................................ 59 Secondary Outcomes ................................ ................................ ....................... 61 Dislocation ................................ ................................ ................................ . 62 Periprosthetic Joint Infection (PJI) ................................ ............................. 62 DVT ................................ ................................ ................................ ............ 63 Mechanical loosening ................................ ................................ ................ 63 Sensitivity Analyses ................................ ................................ ................................ 64 Effect of Extended Follow up ................................ ................................ ............ 64 Propensity Score s ................................ ................................ ............................ 65 Multiple Imputation ................................ ................................ ........................... 67 5 DISCUSSION ................................ ................................ ................................ ......... 93 Baseline Demographics a nd Articulating Codes ................................ ..................... 93 Comparison of the Cohorts ................................ ................................ ............... 94 Trends in THA ................................ ................................ ................................ .. 95 Research Objective 1: Trends in Articulating Surface Code Utilization ............ 96 Research Objective 2: Comparative Safety ................................ ............................ 99 Sensitivi ty Analysis ................................ ................................ ............................... 101 Propensity Score Calculation ................................ ................................ ......... 102 Propensity Score Stratification ................................ ................................ ....... 103 Propensity Score Matching ................................ ................................ ............. 104 Multiple Imputation ................................ ................................ ......................... 105 Study Limitations ................................ ................................ ................................ .. 106 Conclusions and Future Research ................................ ................................ ........ 107 LIST OF REFERENCES ................................ ................................ ............................. 110 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 119
7 LIST OF TABLES Table page 4 1 Baseline characteristics of MPCD by payer ................................ ............................ 68 4 2 Characteristics of cohort by articulating surface code ................................ ............. 69 4 3 Characteristics of NIS Cohort (weighted estimates) ................................ ................ 71 4 4 Baseline characteristics of NIS by p ayer ................................ ................................ . 72 4 5 Hospital volume logistic regression ................................ ................................ ......... 73 4 6 Risk of revision ................................ ................................ ................................ ........ 73 4 7 RMST hazard ratio for MoM MoP revision ................................ .............................. 73 4 8 Risk of dislocation ................................ ................................ ................................ ... 74 4 9 Risk of joint infection ................................ ................................ ............................... 74 4 10 Risk of DVT ................................ ................................ ................................ ........... 75 4 11 Risk of mechanical loosening ................................ ................................ ................ 75 4 12 1 year restricted follo w up of outcomes ................................ ................................ 76 4 13 Abbreviated baseline period by payer ................................ ................................ ... 76 4 14 Propensity score stratification ................................ ................................ ............... 77 4 15 Pooled results of propensity score stratification ................................ .................... 77 4 16 Propensity score matching ................................ ................................ .................... 77 4 17 Propensity score matching by year ................................ ................................ ....... 78 4 18 MCMC Imputation ................................ ................................ ................................ . 78
8 LIST OF FIGURES Figure page 3 1 Illustration of data elements comprising MPCD. ................................ ..................... 53 4 1 Articulating surface ICD 9 CM code utilization from 2006 to 2011 in the NIS ......... 79 4 2 Articulating surfaces in the NIS ................................ ................................ ............... 79 4 3 Articulating surface ICD 9 CM code utilization from 2006 to 2011 in the MPCD by payer ................................ ................................ ................................ .............. 80 4 4 Articulating surfaces in the MPCD ................................ ................................ ........... 80 4 5 Kaplan Meier survival plot of THA revision by MoM and MoP articulating surfaces ................................ ................................ ................................ .............. 81 4 6 Kaplan Meier survival plot of THA revision by CoP and MoP articulating surfaces ................................ ................................ ................................ .............. 82 4 7 Kaplan Meier survival plot of THA revision by CoP and MoM articulating surfaces ................................ ................................ ................................ .............. 83 4 8 Kaplan Meier survival plot of THA dislocation by MoM and MoP ............................ 84 4 9 Kaplan Meier survival plot of THA PJI by MoM and MoP ................................ ........ 85 4 10 Kaplan Meier survival plot of THA DVT by MoM and MoP ................................ .... 86 4 11 Kaplan Meier survival plot of THA DVT by CoP and MoP ................................ ..... 87 4 12 Distribution of overlapping propensity scores. ................................ ....................... 88 4 13 Standardized differences for propensity score quintiles: MoM MoP ...................... 89 4 14 Standardized differences for propensity score quintiles: CoP MoP ....................... 90 4 15 Standardized differences for propensity score quintiles: CoP MoM ...................... 91
9 LIST OF ABBREVIATIONS CAD C oronary artery disease CCI Charlson comorbidity index CCW Chronic Condition Warehouse CIF C umulative i ncidence f unction CMS Centers for Medicaid and Medicare Services CoC C eramic on ceramic CoP C eramic on polyethylene CPT Current Procedural Terminology DRG D iagnosis Related Group DVT D eep vein thrombosis FFS F ee for service FDA F ood and Drug Administration HoH H ard on H ard HR Hazard ratio HVH High volume hospitals ICD 9 CM International Class ification of Diseases, 9 th Edition, Clinical Modification KM Kaplan Meier LVH Low volume hospitals MI M yocardial i nfarction MoM M etal on metal MoP M etal on polyethylene MPCD M ulti Payer Claims Database PAD P eripheral arterial disease PLoS P rolonged length of stay
10 NHI N ormative Health Information NIS Nationwide Inpatient Sample OA O steoarthritis PJI P eriprosthetic joint infection ROC R eceiver operating characteristics SD S tandard deviation THA T otal hip arthroplasty UHMWPE U ltra high molecular weight polye thylene US United States VHVH Very high volume hospitals VLHV Very low volume hospitals
11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Do ctor of Philosophy C OMPARATIVE SAFETY AND EFFECTIVENESS OF TOTAL HIP ARTHROPLASTY ARTICULATING SURFACES By Jonathan Schelfhout August 2014 Chair: Abraham Hartzema Major: Pharmaceutical Sciences There have been few observational studies to explore arti culating surfaces in Total Hip Arthroplasty (THA), and little is known regarding the use or safety of the surfaces in the US. The objective was to explore trends in the use of articulating surfaces, determine the comparative safety of these surfaces, and p ropose methods best suited for future comparative safety analyses of articulating surfaces in large administrative claims databases. Data from the Nationwide Inpatient Sample (NIS) were used to test for changing patient and institutional characteristics o f different articulating surfaces from 2007 to 2011 . T he Multi Payer Claims Database ( MPCD ) was used to perform a series of comparisons regarding the safety of metal on polyethylene (MoP), metal on metal (MoM), and ceramic on polyethylene (CoP) hip prosthe ses. Sensitivity analyses explore d the use of propensity score matching and stratification, the effect of shorter follow up periods, and the effect of missing articulating surface codes using multiple imputation .
12 The analysis demonstrated that the number o f CoP articulating surfaces used in THA have been increasing over time, and that MoM articulating surfaces have become less common. The comparative safety analysis revealed no statistically significant difference in rate of revision for a ny hip combination , however, CoP articulating surfaces had a moderately significant increased risk of revision when compared to MoM (p=0.059) , and MoM hips were associated with lower rates of dislocation compared to MoP (p=0.033 ). Propensity score matching appeared to be an effective method to control for measured confounding in the analysis, and the results of the multiple imputation sensitivity analysis indicate that articulating surface codes may not be missing completely at random. This analysis suggests few substantial differences exist between THA articulating surfaces when controlling for patient characteristics available in administrative claims data. It demonstrated several techniques for the analysis of THA in claims data, and explored how different methods may affe ct the results. The proper use and thorough understanding of how these methods is necessary to control for the changing population of THA in the US, and will become even more important as claims data becomes an integral part of the evaluation of THA articu lating surfaces.
13 CHAPTER 1 INTRODUCTION Background Total hip arthroplasty (THA) , also referred to as total hip replacement, has been has revolutionized the management of painful and debilitating arthriti s affecting the hip . 1 There were 327,000 THA procedures performed in the United States (US) in 2009, projected to rise to 527,000 b y 2030 . 2 This growth is due to several factors, including an aging population, increased prevalence of osteoarthritis, and advancements in THA implants and surgery that provide options and increas ed benefit for a broader patient base. The goal of THA is to reduce pain and improve mobility in patients suffering from degenerative joint diseases of the hip . 3 It differs from partial joint replacement in that both the femoral and pelvic component are replaced , typically b y inserting a stem with a rounded head into the femur and an acetabular cup attached to the pelvis. The two components relieve the friction b etween bone surfaces and allow for greater stability. Patients are typically referred by a physician to a surgeon based on levels of pain, joint damage, functional impairment, and radiographic evidence . 4 There are 3 primary indications for THA , osteoarthritis, avascular necrosis, and rheumatoid arthritis . THA may also or traumatic a rthritis , 3 although these indications are rare r . Osteoarthritis is the most common form of arthritis . 5 It is second only to heart disease as the predominant cause of fun ctional decline in the elderly , 6 and is the primary cause of h ip replace ments in the US . 7 Age, gender , ethnicity, joint injury, obesity, and genet ics are all risk factors for osteoarthritis . 6 , 8 Hip osteoarthritis is typically
14 diagnosed when pain, morning stiffness , and hip internal rotation are present in the absence of other plausible diag noses. Radiographic criteria such as osteophytes, joint space narrowing, and erythrocyte sedimentation rate can also diagnose hip osteoarthritis . 9 , 10 Progression of hip osteoarthritis is thought to be caused by loss of cartilage and is associated with joint damage, hip pai n, and decreasing hip function . 8 Treatment of osteoarthritis typically includes a combination of pharmacologic and non pharmacologic interventions. Exercise, weight loss, and lifestyle modifications are recomme nded. Many patients benefit from regular physical therap y . Walking aids may also reduce pain and increase mobility. Pharmacologic treatment options are aimed primarily at reducing pain associated with osteoarthritis . Acetaminophen, NSAIDs, corticosteroids, and weak opiates may be used depending on severity of osteoarthritis . THA is considered in cases where a combination of pharmacologic and non pharmacologic interventions ha s failed to reduce pain and improve functional impairment . 6 , 11 , 12 Avascular n ecrosis, also referred to as osteonecrosis, results from the interru ption of the blood supply to the bone , specifically the femur for cases in which THA may be indicated . The femoral head becomes damaged aft er the blood flow is restricted and eventually collapses in absence of treatment. There are several risk factors for avascular necrosis including trauma to the femur or hip , use of corticosteroids, smoking, alcohol, hemoglobinopathies, radiation, pregnancy, HIV, and genetic factors . 13 Non operative treatments, like canes or walkers, may be sufficient for minor cases of avascular necrosis containing only small bone lesions. Potential
15 pharmacologic treatments include anabolic steroids and bisphosphonates used to halt disease progression, but a signifi cant proportion of cases require operative treatment . 14 THA has proven effective in halting disease progression and preventing negative outcomes in patients who have progressed to later stages of rheumatoid arthritis , the third primary indication for THA . 15 Younger patients generally have worse prognoses, and addi tional methods may be taken to preserve bone mass for future revisions that are likely to occur. P atients with rheumatoid arthritis typically have increased rates of revision and dislocation than similar patients who were instead diagnosed with osteoarthri tis . 16 The modern era of THA began in 1962 when John Charnley implanted the first Ultra Hig h Molecular Weight Polyethylene femoral head, vastly improving upon the glass, metal , and other designs used at the time . 17 Charnley also began using acrylic bone cement to imp rove the fixation of the prosthesis . These importa nt modifications t o the THA prosthesis resulted in a 25 year implant survival without revision of 77% to 81% . 18 , 19 Further development in THA implants has been driven by the need to decrease wear and improve longevity. The incremental nature of the evolution of hip prostheses has led to a diversity of products in the US and world markets. THA prostheses can be described by the material comprising the articulating surfaces that form a contact between the femoral head and acetabular cup component . Articulating surfaces typically come in a number of combinations : metal on polyethylene (MoP), metal on metal (MoM), ceramic on po lyet hylene (CoP) and ceramic on ceramic (CoC). T he choice of articulating surface explains only a minority of the variation seen in hip prostheses . Prostheses may have a range of femoral head sizes that can greatly
16 affect both positive and negative outcomes o f THA . 20 Implants may be fixated with a cemented or cementless procedure . 21 Surgical processes may also vary, from the type and placement of incisions to various imaging and/or computer based devices that may be used to assist the procedure . 22 Also, t he size, shape, and other small but important details of the implant may vary by manufacturer . Metal on p olyethy lene (MoP) . Also referred to as ultra high molecular weight polyethylene (UHMWPE), MoP impla nts hav e been available since the 1960 s and are the most com monly used implants in the US and Europe . It is typically recommended that the average patient who is 65 years or older and who is of low to normal activ ity levels receive a MoP prosthesis . 21 , 23 . These devices represent a predictably safe and cost effective option where the additional cost of other articulating surfaces has yet to be justified . 23 Although the analysis that t his recommendation was based on is over a decade old, newer articulating surfaces brought to market have continued to increase in price and have yet to consistently demonstrate better outcomes. Metal on m etal (MoM) . MoM prostheses were introduced in the 1960 s with the McKee Farrar THA implant, but initially these devices resulted in high failure rates due to design flaws and a propensity for early cup loosening . 24 MoM implants have since undergone many changes, and made a resurgence in the 1980 s when they were redesigned with increased metal hardness and larger femoral heads . 25 MoM articulating surfaces are in a class sometimes referred to as hard on hard (HoH) articulating surfaces and have several proposed advantages relative to MoP. First, MoM artic ulating surfaces allow for the use of a larger femoral head. This should reduce the risk of dislocation, as well as increasing the surface area of the contact between the femoral
17 head and acetabular cup components. Interestingly, this increased surface are a has reported lower rates of wear due to a better lubricated friction surface . 26 MoM implants have shown lower rates of in vitro wear compared to MoP , 27 and may cause less osteolysis due to a reduction in the amount and size of particulate debris generated through extended wear . 28 However, there is some evidence that suggests MoM articulating surfaces may have an issue with a unique type of osteolysis resu lting from the release of very small metal particles and metal ions . 29 Ceramic on p olyethylene (CoP) and ceramic on c eramic (CoC) . Ceramic implants were introd uced in the 1970 s as a potential solution to polyethylene wear in CoP articulating surfaces . 1 Early ceramic implants had increased rates of failure and were especially prone to fracture due to the reduced hardness of ceramics compa r ed to the metal femoral comp o nents used at the time. Recent advances have improved the hardness of the ceramics and the fracture concern has largely been m itigated. Modern implants typically use alumina or zirconia ceramics and in many ways are similar to MoM articulating surfaces. CoC prostheses are considered HoH surfaces along with MoM , and have many similar purported benefits . Proponents of ceramics clai m they have superior lubrication properties, smoother surfaces and reduced friction, improved hardness, and no concerns of metal ion release when compared to metal articulating surfaces . 33 Implants using CoC articulating surfaces have shown lower rates of annual wear compared to MoM implants in some settings , and have arguably the best tribological performance of any current implant . 34 One common complaint of CoC bearings is that they have a
18 sound s . This noise has been reported in as many as 10.7% of CoC implants . 35 Articulating surfaces in claims data . Articulating surface codes were introduced into the International Classifications of Disease, 9 th Edition, Clinical Modifications (ICD 9 CM) coding language in October 2005. CoP codes were added at the beginning of 2007. Articulating surface codes are not associated with billing, meaning that their in clusion in a medical claim is requested but not required because reimbursement for THA is the same regardless of whether a code is used. Little research has been conducted on the factors associated with the inclusion of these codes in THA claims, and the a ccuracy of articulating surface codes in medical claims d a ta has not been validated . Need for Study THA has shown very high levels of efficacy in reducing pain and increasing mobility in patients with osteoarthritis 36 with implant survival of as great as 80% after 20 years of follow up . 37 THA is being performed in a wider r ange of patients due to these successes, and improvements to THA prostheses aim to further improve the ir longevity and functionality. The rapid changes in characteristics of the patient population and the increases in the number and types of THA options av ailable have outpaced efforts to fully understand THA safety and effectiveness . Clinical trials have traditionally not been required by the FDA, and are rarely sufficiently long term or comparative in nature. This has resulted in knowledge being deri ved fr om observational studies. The US lacks large, high quality medical device and orthopedic registries like those maintained in other countries. Many registries have been started over the past decade, but they lack the follow up necessary to study important T HA outcomes. It is
19 difficult to apply results from foreign registries to the US population due to differences in medical practice and patient populations. Primary among these are that US physicians tend to use a greater proportion of MoM artic ulating surfa ces, and fewer CoC . 38 There is some evidence that US physicians use a grea ter proportion of HoH articulati ng surfaces in elderly patients . 39 The US differs on key patient characteristics known to influence both the choice of THA articula ting surface and risk of negative outcomes like age, sex, and race. Finally clinical follow up, surgical techniques, and cement use are likely to vary between countries. Because data from these large, well controlled registries ha ve poor generalizability t o the US population, much of what we know about the safety and effectiveness of THA in the US comes from analyses using medical claims data . Administrative claims data h ave the potential to contribute to our knowledge of THA safety. First, large claims dat abases can be more generalizable to the US population than clinical trials or registries. Claims databases represent actual utilization, which may differ from the population of patients that takes part in clinical trials. Registries are often regional, and often include hospitals or clinics that specialize in THA and may not be reflective of general clinical practice. Both registries and clinical trials may have physicians with a greater degree of specialization or expertise in THA than is typical of the US population. Second, claims data has the potential to study a larger number of patients than registries or clinical trials. This increased power allows for the analysis of less frequent or unexpected adverse events. Finally, claims data may include importa nt information that is often lacking or less reliable in registries or clinical trials. Administrative claims data can follow patients for long periods of time and reliably captures aspects of medical care that occur outside of the registry or trial enviro nment.
20 This data can include prescription medications, visits to other clinics or hospitals, and the detailed medical history of the patient. Some of this information may be collected in clinical trials or registries through questionnaires or other methods , but claims data has the ability to consistently capture this information over long periods of time. There is need for the advancement of methods used in the comparative safety analysis of THA and other orthopedic devices in claims data. It is likely tha t the demographics and composition of THA patients and the articulating surfaces use d have evolved over time. The most effective methods to control for patient characteristics may be dependent on this evolution , and therefore may also be changing . Addition ally, ICD 9 CM procedure modifying codes for THA articulating surface were introduced in October 2005. Previous analyses have been limited to relatively short patient follow up, and the choice of optimal methods may differ for analyses with longer observat ion periods. Finally, a nalyses in the US have also had difficulty including both older patients eligible for Medicare and younger THA patients in the same analysis due to a lack of mul t i payer claims databases, resulting in poor generalizability of study r esults to the overall US population. The purpose of this study is to assess the utilization and comparative risk of complications for each of the four main articulating surfaces (MoP, MoM, CoC, and CoP ). It assess how the patient and institutional charact eristics of THA and THA articulating surfaces have been changing over time, and explore potential effects of these trends. It is able to use both younger and older THA patients through the inclusion and linking of Medicare and commercially insured patients . It further provides estimates
21 of the comparative safety of THA articulating surfaces, and performs a number of sensitivity analyses to determine how choice of methodology can influence the results. Research Questions and Hypotheses The first research ob jective evaluate s the patient and institutional characteristics associated with utilization of THA and THA articulating surfaces in the US, and how these characteristics have changed over time . ICD 9 CM procedure modifying codes were introduced October 200 5, and there has been little evidence regarding the utilization of THA articulating surfaces in the US . Patient and institutional characteristics are an important consideration in the design of a comparative safety analysis, and it is necessary to understa nd how they have been changing over time. Research question 1a explores trends in the articulating surface codes in the US, and research question 1b explores the patient and institutional characteristics associated with the inclusion of an articulating sur face code in a THA claim. Research question 1c investigates factors related to the inclusion or missingness of an articulating surface code. Specifically, it is interested in whether codes are correlated with patient characteristics or if they are more lik ely in institutions more accustomed to performing THA. Research question 1d explores the possibility that patient and institutional factors may serve as a confounding factor by evaluating the relationship between hospital procedure volume and in hospital c omplications. R esearch O bjective 2 explore s the comparative safety of each articulating surface while controlling for key patient, provider, and institutional characteristics. The only analysis of the comparative safety of THA in the US population was cond ucted in patients older than 65 from 2005 until 2009. Concerns over increased rates of revision with the use of MoM arose in 2008 .T herefore , it is important to provide more recent
22 evidence. The second research objective also analyzes comparative safety in a more diverse cohort of commercially insured and Medicare insured participants, expanding on the previous results. The risk of important short term outcomes of THA were analyzed, specifically risk of revision, dislocation, deep vein throm bosis (DVT), peri prosthetic joint infection (PJI), and mechanical loosening. Research Objective 2 also includes a sensitivity analysis exploring how the choice of methodology may have affected the results. Specifically, the use of extended follow up, propensity scores, and multiple imputation were explored. Research Objective 1 Explore changes in utilization of the four main articulating surfaces from 200 7 through 2011 . Research Question 1a: Ha s use of MoP, MoM, CoC, and CoP articula ting surfaces changed over time? RQ1a H 0 : The percentage of THA claims with MoP, MoM, CoP, or CoC articulating surface codes has not changed from 2007 to 2011. RQ1a H a : The percentage of THA claims with MoP, MoM, CoP, or CoC articulating surface codes has increased from 2007 to 2011. Research Qu estion 1b: Has the association between THA articulating surface and patient, provider, and institutional characteristics changed over time? RQ1b H 0 : The association between any one THA articulating surface and patient characteristics (a ge , sex, race, Char lson comorbidity score) or provider characteristics (surgeon volume, procedure duration hospital size, volume, region, teaching status, and ownership) has remained constant over time. RQ1c H a : The association between any one THA articulating surface and pa tient characteristics (a ge , sex, race, Charlson comorbidity score) or provider characteristics (surgeon volume, procedure duration hospital size, volume, region, teaching status, and ownership) has changed over time. Research Question 1c: Has the use of th e optional ICD 9 CM procedure code for THA changed over time?
23 RQ1 c H 0 : The percentage of THA procedures with an optional bearing surface code has remained constant from 2007 to 2011 . RQ1c H a : The percentage of THA procedures with an optional bearing surfa ce code has increased from 2007 to 2011. Research Question 1d : Is there an association between increased hospital procedure volume and in hospital mortality or prolonged length of stay (PLoS)? RQ1d H 0 : Increased hospital volume is associated with decreased likelihood of in hospital mortality or PLoS. RQ1d H a : Increased hospital volume is associated with decreased likelihood of in hospital mortality or PLoS. Research Objective 2 Investigate the associat ion between THA articulating surface and outcomes. Research Question 2a : Is MoM associated with an increased risk of revision, dislocation, PJI, DVT, or mechanical loosening when compared to MoP articulating surfaces? RQ2a H 0 : MoM articulating surfaces are not associated with an increased risk of revision when compared to MoP articulating surfaces. RQ2a H a : MoM articulating surfaces are not associated with an increased risk of revision when compared to MoP articulating surfaces. Research Question 2 b : Is CoP associated with an increased risk of revision, dis location, PJI, DVT, or mechanical loosening when compared to MoP articulating surfaces? RQ2b H 0 : CoP articulating surfaces are not associated with an increased risk of revision when compared to MoP articulating surfaces. RQ2b H a : CoP articulating surfaces are not associated with an increased risk of revision when compared to MoP articulating surfaces. Research Question 2 c : Is CoP associated with an increased risk of revision, dislocation, PJI, DVT, or mechanical loosening when compared to MoM articulating s urfaces? RQ2c H 0 : CoP articulating surfaces are not associated with an increased risk of revision when compared to Mo M articulating surfaces. RQ2c H a : CoP articulating surfaces are not associated with an increased risk of revision when compared to Mo M arti culating surfaces.
24 CHAPTER 2 LITERATURE REVIEW The literature review is divided into 3 sections. The first section explains the epidemiology of THA and identification of THA in claims data. The second section outlines current knowledge of THA in the US. T he final section reviews the current knowledge regarding the safety of THA. Epidemiology of THA Approximately one fourth of a dults who live to the age of 85 develop osteoarthritis . 40 The lifetime risk of THA has been estimated at 11.6% for women and 7.1% for men, with rates steadily increasing over time . 41 This growth has been attributable to an increased population of eld erly patients suffering from osteoarthritis and improvements to THA that have made benefits available to a wider base of patients. New surgical techniques have made it easier to replace both left and r ight hip components. Larger femoral heads, improved fixation of the acetabular component, and articulating surfaces with greater life expectancy have made THA a viable and successful treatment for younger patients . 42 There is less information on the trends in utilization of specific articulating surfaces used in THA. While the overall incidence of THA has been steadily increasing, trends in articulating surface use tend to be more variable due to the high frequency of improvements to THA prostheses, changing populations of patients who can benefit from these improvements, and practice differences between countries. Bozic et al. 38 explored THA utilization in Medicaid from 2005 2009 and found the MoP ICD 9 code was the most commonly include d in THA claims, (63.1%), followed by MoM (33.3% ) , and CoC (3.5%) . The analysis also highlighted the importance of population
25 characteristics in the prevalence of each articulating surface. MoM and CoC THA tended to be used more frequently in younger patie nts, males, and in the S outh and W est census regions. This is not surprising as younger males are thought to benefit the most from HoH THA. However, the rates of HoH use in older patients remained substantial. MoM THA was used in 30.9% of patients 75 85 ye ars of age, and 29.8% older than 85. The benefits of MoM THA are thought to occur primarily in younger patients, and it is unclear whether the theorized benefits of MoM would result in lower rates of negative outcomes in older patients. 43 and similar levels of utilization in these age groups are not seen in other countries. For example, MoM comprised 6. 8% of total THA procedures from 2003 to 2011 in England and Wales , 44 with utilization increasing from 2003 until a peak in 2008 and a rapid reduction from 2008 until 2010 . The reduction in use was likely due to concerns over met al ions and metalizes ; however, rates were low compared to the US prior to these concerns. Osteoarthritis is the most frequent underlying cause of THA, with prevalence estimates for other indications of THA varying based on the population and sample. Non o steoarthritis diagnoses are under reported in some settings, and using administrative claims data to identify THA has shown poor sensitivit y in cases where avascular necrosis and rheumatoid arthritis are listed as the primary diagnosis in Medicare data 45 These diagnoses were found to have a sensitivity of 0.65 and 0.54, respectively, when evaluated against a chart review. Conversely, osteoarthritis has a sensitiv ity and positive predictive value of 0.96 and 0.86, respectively, in Medicaid patients undergoing THA. There is also some evidence that an association exists between the reporting of AVN and RA and certain negative outcomes in claims data. 45 It
26 was found that these diagnoses may be more likely to occur when the disease is severe, creating an artificial association with poor outcomes. Codes for THA were revised bas ed on a request from a group of orthopedic surgeons and health s ervice researchers in order to (1) facilitate quality improvement through a better understanding of the mechanisms of failure following Total Joint Arthroplasty ( TJA ) , (2) to provide more accu rate and descriptive d a ta inputs for the American Joint Replacement Registry project and (3) to enable more appropriate prospective payments (e . g . Diagnosis Related Groups) to hospitals that reflect actual resource utilization related to specific types of revisio n THJ procedures . 46 The changes to ICD 9 CM codes focused on providing better information on the failure mechanisms of THA including codes for specific reas on for revision, and articulating surface. While these codes have a great potential to facilitate the study of THA safety in claims data, dependent on the ability of hospit al coders to abstract relevant information from charts and other medical documentation. The identification of THA and revision THA in administrative claims data has been validated by Daneshvar et al. in a random sample of 637 patients at two tertiary care university associated hospitals in Canada. Diagnostic codes in administrative claims data were compared to data from a chart abstraction performed by certified health records analysts. While this analysis was not performed in the US or among US based paye rs, it supports a high level of concordance between ICD 9 CM codes and THA procedures documented in medical records . The authors found a 99% negative predictive value and 91% specificity for identifying THA, and a 99% sensitivity and 91%
27 positive predictiv e value in identifying revision THA. The ICD 9 CM procedure modifying codes for articulating surface have not been validated in a US based administrative claims database. These codes are not mandatory nor tied to reimbursement, and a high level of variabil ity is seen in their use. Epidemiology of Study Outcomes The potential complications related to THA are closely tied to the intended outcomes, and decisions regarding THA articulating surface may be driven by the desire to avoid negative outcomes . Complica tions during operation are relatively rare, and surgery mortality rates are low (0.18%) with 0.44 events per 1,000 inpatient days . 47 O ther complications may occur over the lifetime of the prosthesis and are correlated with activity levels an d age of patient and prosthesis . T hese are further discussed below . Revision A THA revision typically refers to the removal or exchange of at least one prosthetic component. Generally, revisions are necessary to correct any degeneration of the prosthesis o r surrounding tissue, or to correct a physical implant failure such as dislocation or fracture. Revisions may involve anything from replacement of a friction liner to replacement of the entire prosthesis. Revision is the most important risk to consider in THA. Revisions can be costly, painful, and increase future rat es of complications in patients. Furtherm ore, they are relatively common among younger THA recipients who are likely to have at least one revision during the life of the implant . 4 8 Rates of THA revisions are highly variable and estimates in modern THA range greatly depending on characteristics of the population, prostheses, and study design. Evidence regarding the safety of THA and likelihood of revision comes primarily from one of three
28 sources: clinical trials, registries, and claims data. Data from these sources is often conflicting and understanding why and how characteristics of the population and data source can influence results is important. Clinical trials are typically considered the gold standard for eff icacy estimates in health research. However, there have been few large clinical trials of THA prosthesis or procedures . This is largely approval process that only re cleared devices . 49 Substantial equivalence may be demonstrated for individual aspects of the device, as would be the case if prosthesis adopted a femoral head and acetabular cup from two separate previously cleared devices. The efficacy and safety these devices , in turn, may have been based on previously cleared devices, and so on. In an analysis of a MoM device recalled from the market, various components of the prosthesis could be traced back through 95 distinct previously cleared devices . 49 Of the trials that have been performed, few appe ar to have been comparative in nature, of high methodological quality, and had sufficient power and follow up to analyze important safety endpoints. Sedrakyan et al . 50 performed a review of safety and effectiveness summaries for all pre market studies containing comparative information on THA articulating surfaces. The authors identified 16 randomized studies with q uality with a mean follow up ranging from 3 months to 8 years. The study was unable to find consistent evidence regarding the comparative safety of articulating surfaces, and results could not be aggregate d due to differences in methodology and study design.
29 The 501(K) approval process has come under scrutiny following evidence of increased rates of revision for the DuPuy ASR XL MoM THA prosthesis. Analyses in the Australian Orthopedic Association National Joint Re placement Registry showed fourfol d increase in rates of revision compared to other articulating surfaces . 51 This finding was later replicated in the National Regis try of England and Wales . 44 National orthopedic registries are viewed as important sources of safety and efficacy data because they include all or nearly all THA procedures performed in a given country, have important patient and prostheses characteristics, and have the ability to follow patents for a many years with low rates of dropout. However, registries have also provided inconclusive and inconsistent info rmation on THA safety. Sedrakyan et al . 50 analyzed registries from Australia, New Zealand, England & Wales, Ital y, and the US, and reported inconsistent findings regarding the risk of revision of each articulating surface. All three articulating surfaces studied (MoM, CoC, and CoP) were associated with an increased occurrence of revision relative to MoP in a t least one registry. T hese findings were not consistent across registries, with MoM associated with an increased risk in three of six registries analyzed , and CoC in only one of six. CoP was associated with an increased occurrence of revision in one registry, but a red uced occurrence in two others. These findings raise important cautions about the use of registry data. First, generalizability between registries is likely poor as characteristics of the population, patient selection, and treatment vary across countr ies. Second, there is no harmonization in methods or reporting of outcomes across registries. Results cannot be aggregated or easily compared, and this variation opens the door to the possibility that
30 differences in study design and measurement of outcomes may greatly influence the results. Finally, even though registries may contain a large percentage of the THA procedures performed in a particular country, the re is no control over patient characteristics or provision of care and therefore registries are n ot immune to the same confounding and selection biases that are seen i n other observational studies. The same issues regarding inter registry generalizability may negatively affect An analysis fro m the National Joint Registry of England and Whales extended the increased failure rates of ASR MoM prosthesis to other MoM designs . 44 O nly 8% (31,171) of the 402, 051 total implants studied were MoM THA or MoM resurfacing. The percentage of MoM THA differs greatly from the US (where closer to one third of THA is MoM over the same period ) and it is likely other important characteristics of patient s and clinical pract ice differ . The US has no national joint or orthopedic registries to compare to results from other national registries. The few registries that exist in the US are smaller, more regional, or were created more recently than national registries in other coun tries . Much of the information on THA safety in the US comes from medical claims data, which has historically has provided limited and inconclusive results. One primary reason is that ICD 9 CM procedure modifying codes for articulating surface were only in troduced in October 2005, meaning recent studies have a relati vely short length of follow up. A matched cohort analysis of Medicare patients from 2006 to 2007 investigated short term risk of revision and found no si gnificant differences by articulating sur face . 46 The analysis found that MoM articulating surfaces were associated with an increased risk of periprosthetic joint infection when compared to CoC articulatin g surfaces after
31 adjusting for patient and hospital factors. There is no clear underlying biological or clinical rationale for these findings, and the absolute risk of infection was low (0.59% vs. 0.32%). The analysis also found an unadjusted increased ris k of DVT, although this effect was not significant after adjusting for patient risk factors . Th e Bozic et al. analysis highlights many of the issues inherent in using claims data to study THA articulating surfaces. First, the average and maximum length of follow up was very short. This means that only short term outcomes could be assessed, when many theorized benefits of HoH articulating surfaces occur in the long term. Second, the study only included patients who had an ICD 9 CM code for articulating surf ace. The authors claim the matched design mitigated the bias introduced by the missing instances of this code, although this cannot be verified. The claims data were also missing key covariates including surgical approach, surgeon experience, implant desig n, and femoral head size. Lastly, the study could only include older patients eligible for Medicare , and it is unclear how the results may be different if the study population was broadened . Bozic et al. later expanded upon this study by including data fro m patients enrolled in Medicare from 2005 to 2009. 38 The longer follow up of th e subsequent study allowed for the exploration of more medium term outcomes and obtained similar results. This is not surprising given the overlap in datasets and time periods. Participants receiving MoM and CoC articulating surfaces were more likely to be younger, male, and live in the south. Unadjusted results indicated that MoM articulating surfaces were associated with increased rates of PJI, mechanical loosening, and DVT, but these findings did not remain statistically significant when adjusted fo r patient and hospital
32 characteristics. The effect of adjusting for potential confounding factors highlights the differences in risk inherent in the population differences related to the choice of articulating surface. Katz et al. followed Medicare benefi ciaries who received a THA procedure between 1995 and 1996 until 2008 to determine the cumulative risk of revision. 52 The long follow up of this study highlighted sev eral important issues in performing THA studies in Medicare data. The risk of revision was 2% over the first eighteen months, and then approximately 1% per year over the next decade. Possibly the most illuminating finding of this study was that this rate o f revision was low compared to the rate of mortality in the sample, which was 29% of participants between 65 and 75 years of age, and 59% for those older than 75 years. Th e study was unable to provide comparative evidence regarding articulating surface, bu t provided valuable information on the long term risk of revision and the importance of accounting for loss to follow up and competing risks. A review of the literature by Zywiel et al. 53 further demonstrated the difficulty in performing evaluations of THA survival by aggregating published analyses. The authors found seventeen prospective clinical based studies of MoM survival. Only four of these studies achieved a level of evidence of I or II, meaning they had proper use of study design elements such as randomization, allocation concealment, blinding, and adequate follow up time and minimal loss. All of the reviewed studies failed at least one of these criteria. T he review found inconsistent and wide ranging results regarding the survival of MoM and CoC implants. The s urvival of MoM articulating surfaces was 71% to 100% with mean follow up times ranging from 36 to 336 months. Survival of CoC articulating
33 surfaces w as 73% to 100% at mean follow ups ranging from 31 to 240 months. When results were limited to studies with level I or II of evidence, the survival at mean follow up rose and the length of follow up declined. Although the ranges of survival varied, both MoM and CoC consistently had short and mid term survival rates comparable to MoP articulating surfaces . Th e Zywiel et al. analysis was unable to aggregate results between trials due to differences in study methodology and data collection. Each study had diff erent patient populations, and different indications for the articulating surface chosen. Analyses used different implant types that varied in acetabular component design and fixation, femoral head diameter, and implant positioning. The surgeons who partic ipated in the trials may have different levels of skill and experience performing THA compared to other surgeons . Finally, many of the well controlled designs took place early in the implant design marketing phase. These studies are often used to assess sa fety prior to the implant being released to market, or shortly thereafter. The length of follow up in these studies was much shorter than the less rigorously contr olled population based studies. This unfortunately means that much of the evidence on the lon g term complications of THA may also be the least reliable and most biased. Wylde and Blom make the interesting observation that it is improper to evaluate the failure of a prosthesis solely by the occurrence of revision . 54 Their rationale is that many THA procedures may produce pain and disability but may not lead to a revision. Their argument is based largely on evidence that rates of successful THA after 7 years decline from close to 100% when u sing revision as an outcome to rates near 70% when severe pain is included in the definition of failure. This theory has two important
34 implications to current research. First, certain types of prosthesis may improve risk of revision and levels of pain diff erently. It has been noted that while rates of revision may be increased with MoM resurfacing, it may also have better functional outcomes than THA . 55 Second, thre sholds for revision may be different for certain types of prosthesis . 56 This may indica te that different types of prosthesis operating with identical levels of pain and flexibility may have variable rates of revision depending on when and why physicians choose to correct complications. Causes of Revision and Other Complications Revision may be necessary due to a number of causes. Often, these causes may not lead to a revision and represent important negative outcomes to prevent. Several of the most common complications and causes of revisions are reviewed below . Periprosthetic osteolysis and aseptic loosening The primary cause of failure of early generation THA implants was periprosthetic osteolysis . 57 when particulate matter cause by wear on the articulating surface enters the tissue surrounding the implant . 58 These particles interact with microphages in the tissue to cause bone resorption that can lead to aseptic loosening and fixation failure requiring a revision operation . 33 , 59 Osteolysis is thought to be a greater problem with MoP bearings, while the hard on hard bearings results in fewer and smaller particulate matter that elicits less of a response from microphages . 34 Dislocation Dislocation is the second leading cause of THA revision , 60 with an estimated 4.8% of hips having at least one dislocation over the life of the prosthesis . 61 Around 10% of these disl ocations occur in the first month, and close to 40% occur in the first
35 year. The cumulative risk of dislocation is around 2% at one year, and increases at a rate of 0.2% per year. These estimates were taken from a study of early Charnley MoP THA procedures performed between 1969 and 1984, but have the benefit of a long follow up (mean 23.2 years for those not censor ed, 13.7 for those who were), and extremely low dropout rate (99.7% of eligible patients at 7 years, 94.9% at 20 years. Dislocation occurs when the femoral head becomes dislodged from the acetabular cup. D islocations after the first five years are believed to be associated with polyethylene wear (in the case of MoP and CoP articulating surfaces) , component malposition, or increased soft tissue com pliance . 62 The risk of dislocation is a consideration in the choice of THA articulating surface. Younger, more active patients are thought to be at an increased risk of dislocation due to the increased range of motion and wear placed on the prosthesis. The primary method to reduce dislocations in these patients is to use a larger femoral head (typically >36mm) which has the dual benefit of increased arc movement and an increased distance the femoral head must tra vel before it becomes dislocated . Traditionally, a larger femoral head size was only possible with MoM or CoC ar ticulating surface, however, there is some evidence that femoral head sizes have been increasing in use with MoP articulating surfaces . An analysis of Me dicare data fro m 1998 until 2007 found that risk of dislocation fell as calendar year increased, leadin g the authors to theorize that the effect was due to an increased use of larger femoral heads over time . 63 Increased metal ion levels , metalosis , and cancer Earl y MoM prosthesis had well documented issues with metal sensitivity and increased levels of cobalt and chromium in the blood and urine. 64 , 65 These problems, along with other complications, led to unacceptably high short term revision rates for
36 MoM prostheses when compared to MoP and their use diminished. MoM THA made a resurgence in the 19 80s in part due to increased hardness of the metals and improved friction characteristics that were thought to mitigate previous concerns over metalosis. Reports of increased rates of revision for MoM articulating surfaces arose in the Australian Orthopedic Association National Joint Replacement Registry 51 and National Registry of England and Wales 44 prompting the recall of the Articular Surface Replacement (ASR) hip prosthesis manufactured by DuPuy Orthopedics , Inc . T he rate of revision at five years was four times greater for the ASR compare d to all other prosthes e s, and two and a half times greater than other MoM articulating surfaces. The ASR was voluntarily withdrawn from the market in August 2010 with an estimated 9 3,000 prostheses used worldwide . The suspected cause of the increased rate s of revision was an increase in metal ions displaced by friction that caused inflammation in the tissue surrounding the prosthesis. The widespread use and increase in risk associated with ASR prosthesis has increased scrutiny of other types of MoM implant s. Reports of increased levels of metal ions have raised additional concerns due to previous research associating exposure to metal ions and certain types of cancer . 58 These concerns have not been substantiated in either th e National Joint Registry of England and Wales , 66 n or the Finnish Arthroplasty Register , 67 , 68 when linked to cancer registries. The follow up time in these studies was not exceptionally long (7 and 4 years respectively) ; however, they represent the highest quality estimate s of cancer risk in THA to date. These studies have also shown some unexpected protective effects that call into question the effect bias and confounding (such as selection bias) may have on the results.
37 Fracture Periprosthetic fractures can be categorized as either intraoperative fractures which occur during operation or postoperative fractures which occur after . The prevalence of intraoperative fractures is typically less than 1% for cemented femoral stems, and close to 5% in cementless stems . 69 Estimates of postoperative fractures are more difficult to determine, but are estimated at ar ound 1%. These estimates should be taken with caution and are heavily dependent on patient characteristics, type of prosthesis, length of follow up, surgical technique, and other factors. Generally, prosthesis f racture was a relatively common complication of early generation hip implants, especially first generation CoC implants . 70 Advances in the strength of materials used has result ed in femoral stems that are less likely to fracture . 71 A recent estimate in the Swedish National Hip Arthroplasty Registry estimated that the 10 year cumulative risk of periprosthetic femoral fracture was 0.64% . 72 Periprosthetic joint infection (PJ I) P JI , sometimes a lso referred to as septic joint failure, occurs when bacteria enters the wound caused by THA or revision resulting in an infection in the tissue surrounding the prosthesis. PJI can occur immediately after surgery , or i n rare cases years later . PJI occurs i n an estimated 1 2% of THA, resulting in increased ant ibiotic use, pain and extended rehabilitation for the patient, reoperations, an d increased healthcare resource use . 73 There is some evidence that rates of PJI have been increasing over time, possibly doubling between 1990 and 2004 . 74 A case control study of THA patients f rom 1990 until 2011 revealed that depression, obesity, cardiac arrhythmia, and male gender are associated with increased rates of PJI . 75
38 Deep vein thrombosis (DVT) Deep vein thrombosis (DVT) is a condition where a blood clot forms in a deep vein, typically the femoral artery in THA. It occurs in an estimated 23% of THA patients in the absence of thromboprophylaxis . 76 If untreated, DVT may cause pain, swelling, edema, ulcers, and restricted blood flow to the heart and brain. Various forms of thromboprophylaxis are available and have proven effective in preven ting DVT when admin istered following surgery . In extreme cases of DVT, surgery may be required to clear the blood clot and restore venous patency. Patients undergoing THA have a relatively higher risk of DVT due to their increased age and characteristics o f the operation. DVT is thought to be more common in lengthy surgeries and when followed by a lengthy bed rest period . 77 THA may also result in trauma to the femo ral or iliac veins from the process of inserting and securing the prosthesis, or from heat involved in the cementing process . 78
39 CHAPTER 3 METHODOLOGY The methodology chapter is divided into four sections. The first section discuss es the creation of the MPCD and NIS cohorts. The second section details the methods used in the NIS for the ana lysis of trends in THA from 2001 to 2011 , and in articulating surface ICD 9 CM codes from 2006 to 2011 . This section also uses the NIS to explore the generalizability of the MPCD cohort. The third section present s methods for analyzing the comparative safety of articulating surfaces in the MPCD . T he final section includes sensitivity analyses designed to test assumptions in the comparative safety a nalyses . All analyses were performed using SAS statistical software ( version 9.3, SAS Institute, Cary NC ) , and all statistical tests were conducted at a significance level of p=0.05 unless otherwise stated. Nationwide Inpatient Sample (NIS) The N ationwide I npatient S ample (NIS) is collected as part of the Healthcare Cost and Utilization Project (HCUP), managed by the Agency for Healthcare Research and Quality (AHRQ). The data consist of an annual 20% stratified sample of US community hospitals and contains discharge information for all patients at the hospital regardless of payer. Sample weights and strata provided in the NIS were used to calculate an estimated annual incidence of THA operations. Th ese data w ere used to determine trends in the utilization o f articulating surfaces in the US in specific patient populations and to assess characteristics correlated with the use of these ICD 9 CM codes i n THA claims. All claims with an International Classification of Diseases, Ninth Revision, Clinical Modificati on (ICD 9 CM) procedure code for THA (81.51) were evaluated. The
40 participant in each claim was defined by categorical age group (<45, 45 54, 55 64, 65 74, 75 84, >85), sex (male, female), race (white, black/African American, other), primary diagnosis (oste oarthritis, other), primary payer (Medicare, commercial , other), and length of stay. Osteoarthritis was identified using ICD 9 CM diagnosis code 715.xx. Hospital teaching status and ownership (public, private) were available from the American Hospital Asso ciation Annual Surve y of Hospitals database. Annual hospital procedure volume was assessed by dividing hospitals into quartiles of very high volume hospitals (VHVH), high volume hospitals (HVH) low volume hospitals (LVH) and very low volume hospitals (VLVH ) based on the number of procedure s performed at the institution in a given year. Multi Payer Claims Database (MPCD) The Multi Payer Claims Database (MPCD) contains publicly and privately insured individuals from 2007 through 2011. Data on publically insu red participants is provided from the Centers for Medicaid & Medicare Services (CMS) Chronic Condition Warehouse (CCW) which contain s a 15% sample of all Medicare patients from 2007 to 2009 combined with a 15% sample of all Medicaid patients from 2007 to 2 011. Information (NHI) database containing primarily UnitedHealth Group ( > 80%) fee for service patients. The combined database contains 7.64 million Medicare beneficiar ies, 9.35 million Medicaid beneficiaries, and 63.4 million privately insured participants. The MPCD is similar to the U.S. population demographically and geographically, and links patients across multiple insurance providers to allow for observation of pa tients as they change insurance coverage . This linkage is useful when addressing research objectives where exposure may o ccur prior to gaining Medicare eligibility and
41 the outcome after. The MPCD contains 6,116,851 (12.6%) individuals in NHI data that have transitioned into the CCW enrollment files. These individuals likely qualified for Medicaid or Medicare over the observation period. Of these, 919,215 (1.89%) are present in the 15% sample of the CCW included in the MPCD. Figure 3 1 provides an illustrati on of the overlap in the insurance providers contained in the MPCD . Cohort Description The cohor t for this study was abstracted from the MPCD and contain s all patients aged 21 and older with at least 6 months continuous coverage and a THA procedure , identi fied using ICD 9 CM procedure code 81.51 when present in the inpatient setting . Articulating surfaces are identified by procedure modifying ICD 9 CM codes 00.74 for MoP, 00.75 for MoM, 00.76 for CoC, and 00.77 for CoP where present during the index hospita lization for THA . The diagnosis for THA was determined by the primary diagnosis code listed on the claim for THA. Each THA procedure will be classified as osteoarthritis or other. O steoarthritis will be identified by the presence of ICD 9 CM diagnosis code 715.XX. Covariates were obtained from the claims data and will include available known confounders for negative outcomes of THA. Demographic covariates are age, sex, race, region, and year of THA. Participant comorbidities for THA are chronic lung disease (CLD) , congestive heart failure (CHF) , coronary artery disease (CAD), diabetes mellitus, history of thromboembolism, morbid obesity, p rior myocardial infarction (MI), peripheral arterial disease (PAD) , and an adjusted Charlson comorbidity index . 79 , 80 Hospital region is coded as North e ast, South, Midwest, or West. The Charlson comorbidity index score was calculated using diagnoses and claims contained
42 in the 6 months preceding index hospitalization date for THA and participants were categorized as having zero , one to two, three to four, or five or more comorbidities . Descriptive stat istics are presented for both the NIS and MPCD overall , and stratified by insurance provider and articulating surface . An additional set of descriptive statistics are reported for participants in the MPCD who switched between insurance p roviders during the study period compared to those who did not. Continuous variables are summarized as mean s with standard deviation (SD), and median s with inter quartile range s . Categorical variables are summarized as frequenc ies and percentage s . Descript ive statistics are compared between the NIS and MPCD to assess generalizability of the study population to the US THA population. Research Objective 1: Utilization of Articulating Surface Codes Research Objective 1 e xplore s changes in utilization of the f our main articulating surfaces from 2007 through 2011 . Guidelines suggest the use of secondary codes identifying articulating surface whenever a primary code for THA, resurfacing, or revision is used. Articulating surface codes specified in the ICD 9 CM co ding language are 00.74 for MoP, 00.75 for MoM, 00.76 for CoC, and 00.77 for CoP. Participants without an ICD 9 CM procedure modifying code for articulating surface are considered a separate group to analyze changes in coding. Changes in the estimated numb er of claims with each articulating surface code is presented graphically, and changes in the 2 tests for each participant and institutional factor. To test for cohort charact eristics related to the use or absence of an articulating surface code, each THA claim was given a binary variable for the inclusion of any
43 and institutional characteri stics in the NIS and the inclusion of an articulating surface code was then analyzed with multiple logistic regression. The dependent variable was the presence or absence of an articulating surface code, and independent variables are age, sex, race, region , THA diagnosis, hospital volume quartile, payer, and year of THA. Interactions between year and other covariates were considered, and included in the model if significant at a p<0.10 level. Ordinal variables were converted to a series of independent binar y variables. The overall rate of in hospital complications, mortality and prolonged length of stay (PLoS), and the relationship between these outcomes annual procedure volume was explored in the NIS. Mortality was determined from patient discharge data. A participant was considered to have a PLoS if they were in the top 10% of hospital stays for THA procedures in a given year. PLoS has been used previously as a marker for complications that may not be reported in claims data. 81 The association between hospital procedure volume and in hospital complications was explored with a logistic regression. The analysis controlled for potential confounding patient factors and changes in mortality over time. Two models were created for in hospital mortality and PLoS that controlled for patient demographics age, sex, race, region, and year of operation while assessing the difference in risk betwe en VLVH, LVH, or HVH and VHVH. Re search Objective 2: Comparative Safety Analysis The primary comparative safety analysis explored risk of revision, and secondary analyses investigated the difference s in risk of dislocation, joint infection, DVT, and mechanical loosening. Only the first oc currence of an outcome was included in the analysis due to the inability to determine which of the two prosthesis failed or was
44 revised. Each analysis consisted of distinct 1:1 comparisons of the three articulating surfaces, MoM MoP, CoP MoP, and CoP MoM. CoC articulating surfaces were not included in these comparisons because there were too few claims with codes for this articulating surface combination in the MPCD . For each comparison, we refer to the refers to the second articulating surface mentioned (e.g. MoM MoP would have MoM as a case and MoP as a control). Risk of Revision The first research question of Research Objective 2 analyzed the comparative safety of art iculating surface with regards to risk of THA revision . The index date for each participant was the first THA claim appearing in the cohort. THA revision was identified with ICD 9 CM procedure codes 00.70 (revision of hip replacement, both acetabular and f emoral components) , 00.71 (revision of hip replacement, acetabular component) , 00.72 (revision of hip replacement, femoral component) , 00.73 (revision of hip replacement, acetabular liner and/or femoral head only) , or 81.53 (revision of hip replacement, no t otherwise specified) . Participants were excluded if enrolled in a non f ee for s ervice (FFS) plan in any of the included health insurance programs or if there is evidence of complications from a previous THA or of a non elective THA procedure. A procedure is considered non elective if there is evidence of hip or femur fracture, conversion of previous hip surgery to THA, infection of the pelvic region or thigh, or mention of metastatic or bone cancer during the 90 days preceding THA discharge date . These ex clusion criteria have been validated in Medicare data and have a positive predictive value of 0.99 for primary THA and 0.92 for revision. 82 Participants enrolled in
45 Medicare were excluded if they were eligible through disability or end stage renal disease. Kaplan Meir (KM) curves were used to illustrate the cumulative risk of each outcome for each articulating surface . 83 Hazard ratios for all analyses were reported at 6 months, 1 year, 2 years, and 3 years. Differences in the revision rates of each articulating surface were calculated using log rank tests. The effect of patient, provider, and hospital charact eristics on the risk of revision were estimated using a Cox P roportional H azards model specified in equation 3 1 . (3 1 ) i (t) = 0 (t) = baseline hazard function at time t (unspecified) 1 = parameter indicating effect of THA articulati ng surface x i1 = THA articulating surface group 2 i ) = parameter indicating effect of covariates ( x i2 ik ) = covariates Covariates for the Cox model were time to revision as the dependent variable and age, sex, race, region, obesity, year of THA, reg ion, CLD, CHF, CAD, diabetes mellitus, history of thromboembolism, morbid obesity, prior MI, PAD, and Charlson comorbidity index were included as independent variables . Participants were censored at death, end of study period, fracture, or arthron omy . Viol ations of the proportional hazard assumption were tested by the inclusion of a time dependent interaction variable in the Cox model . Th e Cox model was also performed stratified by insurance provider to analyze differences in parameters in the Medicare and commercially insured cohorts.
46 Secondary Outcomes The remainder of research questions in Research Objective 2 was addressed using a series of analyses similar to the base case Cox model, performed with the secondary outcomes dislocation, DVT, PJI, and mech anical loosening. Dislocation was identified using ICD 9 CM codes 718.35, 996.42, and 835.xx , DVT was identified using ICD 9 CM codes 451.11, 451.19, 451.81, 451.2, 451.9 453.1, 453.2, 453.4, 453.8, 453.9, 4534.0, 453.41, and 453.42. PJI was identified usi ng ICD 9 CM code 996.66 and mechanical loosening with code 996.41. Participants were required to meet the same inclusion and exclusion criteria as for the primary outcome . KM curves and hazard rates are presented for each articulating surface combination. The Cox model adjusted for the same set of independent variables as for the primary outcome and the proportional hazards assumption was tested using scaled Schoenfeld residuals. Each model was stratified by insurance provider; however, these results are on ly presented where significant differences exist. Early Revision Risk Cox models for primary and secondary outcomes were used to analyze the risk of revision during the 1 year following index hospitalization date for THA , regardless of procedure date. Ear ly revision risk is important for a number of reasons. First, a THA revision can increase the risk of future revisions, and therefore first year risk may be especially important. Second, factors that affect risk of revision soon after surgery may be differ ent than those that affect risk of revision later on. Early revisions may be due to errors related to surgery or a familiarity with the prosthesis, while later revisions may be due to causes such as wear. Finally, it is important to explore h ow differential length of follow up may affect the analysis. Previous research has suggested
47 that there may be few er CoP THA during the first year or years of the analysis. This could lead to fewer CoP participants with long follow up, and contrasting earl y revision risk with the previous analyses may provide valuable information. Early revision risk was analyzed using Cox models identical to those described in the previous section with respect to the outcome and comparison. The only difference is that part icipants were censored at 365 days follow up if they had yet to have an outcome or other cause of censoring. Sensitivity Analysis A series of sensitivity analyses explore s the influence of key assumptions and methods on the results of the base case analys is. All sensitivity analyses will assess risk of revision for the three articulating surface comparisons unless otherwise stated. Each sensitivity analysis is described in detail below. Abbreviated Follow up The MP C D contains information on participants a fter they switch insurance providers. The effect of this additional information was explored by comparing results from a Cox model without the additional follow up to the base case results. The Cox model performed was identical to the base case with the ex ception that participants additionally censored at the end of the insurance coverage they had during the index hospitalization. This analysis was performed only for risk of revision. Propensity Scores The propensity score sensitivity analysis uses two me thods to adjust for confounding factors, propensity score stratification and individual matching. The
48 following sections describe these processes along with the creation and assessment of the propensity scores. Propensity score calculation The previously described base case method controls for these characteristics by means of multivariate regression modeling. Another me thod is to use a propensity score , which is a single summary measure that represents a participant s likelihood of receivin g a particular type of exposure (in this case, a particular articulating surface) . 84 The propensity score for each participant was cal culated using a logistic regression model which estimated the odds of being a case as opposed to a control using age, sex, race, region, obesity, year of THA, hospital volume, region, CLD, CHF, CAD, diabetes mellitus, history of thromboembolism, morbid obe sity, prior MI, PAD, and Charlson comorbidity index as independent variables . The distribut ion of propensity scores was plotted t o review for sufficient overlap of scores between exposure categories . Non overlapping propensity sc ores were trimmed to simula te clinical equipoise. This propensity score model was applied to MoM MoP, CoP MoP, and CoP MoM comparisons separately using only the two articulating surface codes listed for the comparison, and resulted in three distinct datasets with unique propensity s cores for each combination. Propensity s core s tratification P articipants were matched based on propensity scores and other covariates using frequency matching (also known as interval matching or stratification) , which involves matching the entire stratum of participants. Frequency matching allows for the inclusion of all eligible participants, and allows for transparent and easy to describe effect differences within the strata. By analyzing the risk within each strata, the potential
49 for a non linear relati onship between propensity score and risk can be evaluated . Covariate balance can also be explored for different levels of propensity score, revealing if the model resulted in a poor fit for participants who were more or less likely to receive a particular articulating surface. P ropensity score strata w ere constructed as mutually exclusive quintiles and participants were sorted according to their estimated propensity score for the articulating surface comparison being evaluated . Previous research has indica ted that five groups is sufficient to remove the majority of bias from the association of the covariates included in the propensity score . 85 Each strata was analyzed for sufficient balance of propensity scores and covariates usi ng the standardized difference, as illustrated in equation 3 5 for continuous variables and equation 3 6 for dichotomous v ariables . The standardized difference is a convenient measure for assessing the difference between matched groups because it is not influenced by sample size. While there is no generally accepted value of d that represents a meaningful imbalance in covaria tes, the present analysis used 0.10 as it has been suggested as a general threshold. 86 (3 5) (3 6 ) A Cox m odel wa s performed on each matched cohort of data. The primary difference is that these models do not include the baseline covariates that were included in the propensity score. Results of the Cox models are pooled across strata after weighting by the size of eac h strata. Strata specific and pooled hazard ratios are
50 presented alongside those of the base case results to provide insight into the effect of covariate balancing. A second propensity score stratification was performed where participants are placed into q uintiles based on propensity score for each year of the analysis, as opposed to five strata that include participants from multiple years of the analysis. To facilitate this method, propensity scores were re estimated for each year of the analysis and the year was not included in the logistic regression. This method has the limitation of a smaller sample size in each strata (each year quintile contain s roughly a quarter of the participants in the previous quintile ), but provides valuable information regardi ng whether propensity scores are changing over time. Each strata was analyzed for sufficient balance of propensity scores and covariates using the standardized difference (equation 3 5). Hazard ratios are again pooled across strata and presented in contras t to the first propensity score stratification results. Matching A n additional sensitivity analysis using propensity scores employed greedy matching to create pairs or groups of participants with similar propensity scores. Briefly, greedy matching selects a random case from the dataset. The nearest control patient is then selected as a match, and the process is repeated until all controls are assigned as matches to cases. The matching algorithm used a 1:n matching criteria, allowing for up to 5 controls to be matched to each case. This type of variable matching has been shown to reduce bias in a greater extent than matching on a fixed number of controls. 87 Once a control was matched to a case, it was not eligible to be matched to any other cases, commonly referred to a s matching without replacement.
51 The matching algorithm was used for each articulating surface comparison with the comparison specific propensity s cores calculated previously. A Cox proportional hazards model determined the hazard ratio for each articulating surface combination while conditioning out the effect of matched pairs. This modeling technique allows for each matched set to have a different baseline risk of revision, but a constant proportional risk across attributable to articulating surface across all matched sets. Multiple Imputation Previous studies have indicated that a large percentage of THA claims are missing data on articulating surf ace. The base case Cox model (which uses only claims with an articulating surface code) implicitly assumes that the claims with an articulating surface (complete cases) are essentially identical to those without (incomplete cases), and therefore are ignora ble. This strong assumption is commonly referred to as being missing completely at random (MCAR). However, it is possible that systematic differences exist between complete and incomplete cases. In this situation the results of the previous analyses may on ly be generalizable to individuals with articulating surface codes in the population. Multiple imputation is one method for dealing with data containing missing information. 88 Multiple imputation and similar methods attempt to input a mean value for missing data in incomplete cases by using data from complete cases. One strength of m ultiple imputation over similar methods is that it uses the association between the missing variable and other covariates to provide a set of values that represent the uncertainty about the right value to impute . The result is a range of values for each in and the relationship between those variables and non missing data in complete cases.
52 This estimation assumes that the probability data is missing is determined by the v ariables included in the dataset, an assumption commonly referred to as missing at random (MAR). Specifically, Markov Chain Monte Carlo (MCMC) method of multiple imputation was used to create a set of plausible values for missing articulating surface codes . The MCMC method uses Bayesian inference and simulation methods to sample from multidimensional probability distributions via Markov chains. The result is a series of datasets with multiple sets of simulated values for the missing variable that when combi ned, provide an estimate for the missing value and a variability that reflects the uncertainty due to missing data. Cox models identical to the base case scenario were performed in 20 imputed datasets and results were pooled to provide a combined paramete r estimate for the hazard rate in each articulating surface comparison. It is important to note that MCMC methods assume the missing data are based on a normal distribution and it provides non integer values for missing data. The Cox proportional hazards m odel requires an integer binary exposure. To correct for these discrepancies we restricted the imputation of articulating surface a lower bound of 0 and an upper bound of 1, and each imputed value was rounded to the nearest integer. This method has been sh own to introduce some bias, but the level of bias is generally very low. 89
53 Figure 3 1. Illustration of data elements comprising MPCD.
54 CHAPTER 4 RESULTS Description of the Cohort The description of the cohort is divided into three sections. This first section describes the baseline characteristics of the MPCD in de tail. In the second section, the NIS cohort is described and used to check the generalizability of the MPCD cohort. The third section describes trends over time using data from both the NIS and MPCD cohorts in accordance with the research hypotheses presen ted for Research Objective 1 . The final section contains a n exploration of potential unmeasured confounding in the MPCD . MPCD Cohort After applying all inclusion and exclusion criteria, the final cohort consisted of 236,319 subject years from 64,028 partic ipants. Participants enrolled in Medicaid were not included in the cohort due to the relatively small number and abbreviated follow up (limited to 2 years). Similarly, there were few CoC articulating surface codes in the MPCD, and therefore the characteris tics of these claims will be described but they are excluded from the comparative safety analyses. The insurance provider for each participant was classified based on the payer indicated on the index THA claim. For a small number of participants, claims we re submitted to both Medicare and a private insurance with THA listed as the primary procedure during the index hospitalization. These participants are described as their own cohort in the baseline demographics. However, they were included in the Medicare sub cohort in the comparative safety and sensitivity analyses because they most closely resembled Medicare only participants.
55 Baseline characteristics of the MPCD cohort are presented in Table 4 1. The cohort consisted of 41,408 (64.7%) Medicare insured, 2 1,709 (33.9%) commercially insured, and 911 (1.4%) participants with a claim listing THA as the primary procedure submitted to both Medicare and a commercial insurance during the index hospitalization. The cohort was primarily white with slightly more fema les than males, and a majority of participants were over the age of 70. The primary indication for THA was osteoarthritis in more than 90% of claims. Participants with commercial insurance were more likely to be non white, had fewer comorbidit ies , and were more likely to have a diagnosis of obesity compared to those with Medicare as the primary payer . There were 26,493 participants (41.4%) with an articulating surface code for THA during the index hospitalization. These codes consisted of 13,228 (49.9%) MoP , 8,604 MoM (32.5%), 916 CoC (3.5%), and 3,745 CoP (14.1%). Participants with MoP articulating surface codes tended to be older and more female than participants with MoM or CoP articulating surface codes , and were less likely to live in the south and have an osteoarthritis diagnosis . Table 4 2 contains baseline characteristics of the cohort by articulating surface . NIS Cohort Characteristics and Comparison The overall weighted composition of the NIS from 2001 to 2011 is provided in T able 4 3 . Briefly, est imates for the US population showed that THA recipients were more likely to be female, between 55 and 74 years of age, whit e, and to have osteoarthritis as the primary indication for THA. The NIS estimates that the primary payer for THA is Medicare in 52.7 % of cases and commercially insured in 40.2% of cases. This would suggest that the MPCD consists of a lower proportion of
56 commercially insured and a greater proportion of Medicare insured participants than the US population as a whole. The composition of the NIS was similar to that of the MPCD from 2007 through 2010 when stratified by insurance provider and region . The NIS cohort contained estimates of THA articulating surface code use from a longer time period (2006 to 2011) than did the MPCD, and Table 4 4 illustrates the composition of the NIS over this time period when stratified by insurance provider. The commercially insured MPCD cohort contained a slightly greater proportion of non white, male participants, and participants were more likely to live i n the W est and Midwest than the average US THA patient. Medicare participants were very similar to those in the NIS, indicating the random 15% sampling of Medicare participants for the MPCD raises little concern over ge neralizability to the overall populat ion of Medicare patients in the US . There were an estimated 2,316,623 THA procedures performed in the US between 2001 and 2011 according to the NIS. An estimated 305,175 of these procedures were performed in 2011; a 34% increase from 2006 and an 86% incr ease from 2001. THA became disproportionality more common in participants who were younger, male, commercially insured, non white, and who did not have a diagnosis of osteoarthritis as t he primary indication for THA. The incidence of THA increased in every major demographic group, although the growth was not uniform. In particular, THA in participants aged 45 to 64 years old increased 250% from 2000 to 2011 compared to a 61% increase in participants older than 65 over the same period. The average number of procedures per hospital increased during the study window while the number of hospitals with at least one claim for THA in the NIS
57 decreased. Unadjusted in hospital mortality and the average length of sta y decreased from 2000 to 2011. VHVH and HVH hospita ls differed from low volume hospitals in that they typically had a higher proportion of privately insured patients, higher proportion of osteoarthritis primary diagnoses, shorter average length of stay, and lower in hospital mortality. Larger volume hospit als were typically located in higher average income zip codes. Time trends in institutional characteristics and patient demographics were similar between large and small volume hospitals. Research Objective 1: Utilization of Articulating Surface Codes MoP was the most frequently observed articulating surface code in the NIS. Estimates from the NIS suggest the incidence of MoP articulating surfaces in the US has been steadily increasing from 2006 through 2011. There was a 53 % increase in MoP THA over this ti me period, rising from 43,639 in 2006 to 66,665 in 2011. The incidence of MoM peak at year 2008. The incidence of MoM codes rose by 62 % from 2006 to 2008, reaching a similar overall incide nce as MoP. MoM incidence then decreased by 290 % from 2008 to 2011. CoP articulating surfaces had the greatest absolute and relative increase in incidence, rising from 11,108 in 2007 (the year of their introduction) to 34,984 in 2011, a greater than 3 fold increase . There were an estimated two CoP articulating surface codes for each MoM code in 2011, a large change from 2007 when there were three MoM codes for each CoP articulating surface code . A visual representation of articulating surface c odes in the N IS is provided in F igure 4 1. Trends in articulating surface code utilization within Medicare and commercial payers differed in the NIS. Generally, trends in Medicare had similar direction but a reduced magnitude compared to the overall sample. MoM codes were more frequent
58 from 2006 to 2008, and then decreased below CoP codes by 2011. The THA claims for commercially insured participants differed to a greater degree from the overall cohort. MoM codes were more frequently used than any other articulating sur face in each year from 2006 to 2008. The decrease in MoM codes was steeper ; MoM code use in 2011 was 30% of the total during 2008. The increase in CoP codes was also more pronounced, with nearly a four fold increase from 2007 to 2011. Figure 4 2 contains a graphical representation of the annual incidence of THA in the Medicare and commercially insured cohorts. Trends in the utilization of THA articulating surface codes in the MPCD were similar to the trends estimated by the NIS over the same time period ( 2007 2010). The MPCD revealed a gradual increase and decrease in MoM codes, and relatively constant increases in the number of MoP and CoP codes. These trends had a close resemblance to the trends estimated in Medicare participants in the NIS. Figure 4 4 c ontains a representation of articulating surface code utilization in the MPCD. Trends in Medicare and commercially insured participants in the MPCD were similar to trends in the respective cohorts in the NIS. Specifically, the commercially insured MPCD coh ort saw more drastic increases and decreases in MoM and a greater increase in CoP when compared to the Medicare insured cohort in the MPCD. Figure 4 5 contains an illustration of articulating surface code use in Medicare and commercial cohorts in the MPCD. Articulating surface codes were missing in 54% to 60% of all THA claims in the NIS for each year of the analysis. Codes were more likely to be missing in female, non -
59 white older, and Medicare insured participants. Additionally, the likelihood a THA claim included an articulating surface code increased from 34.5% in VLVH to 52.0% in VHVH. Results of the logistic regression revealed that VHVH hospitals had significantly lower rates of in hospital mortality and PLoS when compared to the other three quintile s of hospital procedure volume. In particular, VLVH had an 82% increased risk of mortality and a 2.37 times greater risk of PLoS. Results of the adjusted and unadjusted logistic regre ssion are presented in T able 4 5 . Research Objective 2: Comparative Safet y Analysis The following section contains the hazard ratios and plots of hazard ratios for the primary and secondary analyses in the MPCD . All analyses are then stratified by insurance provider to show the hazard for outcome in the specific subpopulation. Each outcome contains a description of the comparative risk between the three articulating surface comparisons, the hazard ratio ( HR ) for each payer (Medicare or commercial), and HR when follow up is limited to 1 year after index hospitalization . Risk of R evision The average length of follow up prior to censoring was 1.5 years (SD: 1.1 year ) ranging from 1 day to a maximum of 3.9 years . The primary reason for censoring was end of study window (85%), followed by end of enrollment (10%) and the appearance of a second THA claim (5%). The average length of follow up differed for each censoring reason, with participants reaching the end of the study window being the longest (574 days), followed by end of coverage (330 days) and occurrence of a second THA claim (221 days). There were 1,459 revisions recorded in participants that met inclusion and exclusion criteria. The average time to revision was 254 days (SD: 278 days). A total of 79 revisions occurred after the occurrence of one or more censoring criteria.
60 Ta ble 4 6 and Table 4 7 contai n the unadjusted and adjusted hazard ratios and 95% confidence intervals for the outcome of revision by articulating surface categories. The unadjusted analyses revealed a marginally significant increase in hazard for CoP when c ompared to MoM. Neither the MoM MoP nor CoP MoP comparisons were statistically significant. The adjusted model contained similar results with a marginally significant increase in risk for the CoP MoM comparison and a failure to reach statistical significan ce in the other comparisons. When the Cox proportional hazards model was performed in each insurance provider, hazard ratios were differed slightly from the base case model. The hazard ratios for CoP compared to MoP was larger in the commercially insured strata than the Medicare strata . MoM articulating surfaces were also associated with an increased hazard in the commercially insured when compared to MoP , but the difference was small and not statically significant. The confidence intervals were wider, lar gely due to the reduced sample size of each strata when compared to the sample as a whole. Results for the strati fied analys e s are presented in T able 4 9. There was evidence supporting a possible violation of the proportional hazards assumption in the C ox model comparing MoM to MoP , as determined visually by scaled Schoenfield residuals and quantitatively by the inclusion of a time effect modeled in the C ox regression ,(p<0.01 ) . S urvival of MoP articulating surfaces is initially lower than MoM, with MoM cros sing the MoP survival cure at around 2.5 years ( survival curve illustrated in F igure 4 4 ). T herefore , results for this comparison are also report ed with the restricted mean survival time (RMST) estimate as it provides a clear er understanding of the changin g hazard over time leading to the violation of the
61 proportional hazards assumption (Table 4 8 ) . The RMST provides piecewise estimation of the hazard ratios, valua ble when the overall hazard ratio averages different effects over time. Dividing the hazard r atio into 1 year segments revealed a slightly decreased risk of revision for MoM compared to MoP soon after the index hospitalization , but this rate rose steadily over time until it was 3.6 times the hazard between years three and four. Restricting follow up to one year resulted in a statistically significant increase in risk of revision for CoP articulating surfaces when compared to MoM (HR 1.56, p<0.01). CoP approached statistical significance when compared to MoP (HR 1.32, p=0.06). Results from the MoM MoP comparison are identical to the RMST as they are calculated in the same way. We restricted the length of follow up to analyze the early revision risk for each articulating surface comparison. Risk of revision for MoM compared to MoP was lower than in the base case scenario. The results are identical to those presented in the RMST estimates, as they are the same calculation. The risk of revision for CoP compared to MoP was also greater than in the base case scenario, with a 32% increased risk over the p reviously reported 20% increase. CoP also had a greater risk in the first year than over all 4 years when compared to MoM, with a 56% increase compared to a 29% increase over the entire 4 year period. Secondary Outcomes The remaining research questions in Research Objective 2 are explored with analyses similar to the base case analysis , but performed on a series of secondary outcomes in each articulating surface comparison . These analyses report that HR for each articulating surface comparison, KM and log r ank tests for equality between articulating surfaces, HR for each payer (Medicare or commercial) and HR when follow -
62 up is limited to 1 year after index hospitalization . Results for each secondary outcome are described separately below. Due to the large num ber of comparisons performed and outcomes analyzed, KM curves are only presented where log rank tests indicate a statistically significant difference. Dislocation There were 1,626 dislocations recorded in the MPCD cohort . The majority of dislocations occur red relatively soon after the index hospitalization with an average time to dislocation of 133 days (SD=224 days ). The KM curves and log rank test suggested that there was a significant difference in risk of dislocation between MoM and MoP articulating sur faces (p<0.01). The results of the adjusted Cox model indicated that MoM articulating surfaces had a 19% reduced risk for dislocation when compared to MoP (p=0 . 03). The majority of the reduction in risk for this comparison was in commercially insured parti cipants, where the HR was 0.63 (p<0.01) compared to 0.93 (p=0.54) for Medicare. CoP articulating surfaces generally had an increased risk of dislocation relative to MoM, however this increase in risk was not statistically significant (p=0.08). Full results of articulating surface comparisons for ris k of dislocation are listed in T able 4 8. When the follow up was limited to one year after index hospitalization , there was a statistically significant decreased risk for MoM when compared to both CoP (p=0.04) and MoP (p<0.01) Periprosthetic Joint Infection (PJI) There were 1,155 cases of periprosthetic joint infection recorded in the MPCD cohort. The average time to infection was 170 days (SD: 238 days). KM curves and log rank tests suggested a significant reduct ion in risk of PJI for the MoM MoP comparison. This difference was no longer statically significant after adjusting for potential
63 confounding factors in the Cox model (p=0.06).Cox models failed to reject the null hypothesis (no difference in articulating s urfaces) for all articulating surface combinations. CoP articulating surfaces were more likely to lead to a joint infection in Medicare (HR=1.32) rather than commercial participants (HR=0.92), however neither ratio reached statistical significance. Full re sults of articulating surface comparisons for risk of PJI are listed in T able 4 9. Results were largely unaffected by limiting the follow up period to one year in all articulating surface comparisons, and none of the comparisons reached statistical signifi cance over the one year follow up. DVT There were 5,617 cases of deep venous thromboembolism recorded in the MPCD cohort. The average time to DVT was 136 days (SD: 232 days). KM curves and log rank tests suggested an increased risk of MoP when compared to both MoM (p<0.01) and MoP (p<0.01) (F igures 4 10 and Figure 4 11). The Cox model did not support this difference when controlling for potential confounding variables, and the comparison was not statistically significant with a HR was closer to 1. When the Cox model was stratified by payer, CoP articulating surfaces approached a statistically significant reduction in risk of DVT when compared to Mo P (HR 0.86, p=0.10) and MoM (HR 0.80, p=0.06) in the Medicare cohort. Full results of all models for risk of DV T are located in T able 4 10. Results were similar when restricting follow up to one year for all articulating surface comparisons. Mechanical l oosening There were 785 cases of mechanical loosening recorded in the MPCD cohort. The average time to infection was 341 days (SD: 303 days). Log rank tests of the KM curves failed to reject the hypothesis of a difference between any two articulating
64 surfaces, as did the comparisons in the Cox models. The commercially insured cohort appeared to have an increased risk of loosening with MoM and CoP articulating surfaces; however the small number of outcomes and long average time between index hospitalization and outcome made it difficult to draw any conclusions from these findings. Full results for risk of mechanical lo osening are available in T able 4 11. Limiting the length of follow up to one year did not affect the results to a large degree. Sensitivity Analyses T he first sensitivity analysis explores the effect of allowing participants to switch insurance provider s during follow up. The second sensitivity analysis describes the creation of propensity scores and the analyses performed by propensity score strata to explore differences in risk for participants most likely to receive a particular articulating surface c ode. The final sensitivity analysis explored the effect of imputing data for missing artic ulating surface revision codes. Effect of Extended Follow up The base case analysis included data from the entire enrollment history included in the MPCD. This sensi tivity analysis explores the effect of restricting the analysis to the incident insurance enrollment episode that included the THA index hospitalization. By comparing the results from this restricted analysis to the base case, the benefit of connecting enr ollment periods across different insurance providers in the MPCD can be evaluated. There were 2,878 participants (5.8%) that made a change in their insurance and contributed data to the MPCD. Mean follow up time increased to 531 days (SD: 374 days) if part icipants were observed after an insurance switch, which equated to an extra 494 days for those who switched. The insurance enrollment a participant had during the time of the THA operation comprised 96% of the full enrollment window in the entire
65 cohort, a nd 62% of the full enrollment window in those who switched. There were an addition 33 revisions identified when participants were followed through additional enrollment periods. Limiting the enrollment period had a minimal effect on the results of the adju sted C ox models. Results for the limited baseline period model stratified by insur ance provider are presented in T able 4 1 3 . Propensity S cores Propensity scores for all articulating surface combinations had sufficient overlap with little trimming necessary . Overall sample sizes for each comparison were similar to the base case analysis due to the small number of trimmed participants , with fewer than 20 participants trimmed in each model . The distribution of propensity scores is provided in F igure 4 12 . In e ach pane l of F igure 4 12 , the bottom propensity score distribution represents the case and the top distribution represents the control. Propensity scores for cases were in general greater than controls, and the shape of the propensity score distribution wa s similar for cases and controls in each comparison . The CoP MoP comparison had the least similar distribution of propensity scores, with a greater proportion of the controls having very low estimated propensity scores. The standardized difference for eac h variable used in the propens ity score is presented in F igure 4 13 , Figure 4 14 , and Figure 4 15 by quintile along with a reference standardized difference for the unmatched cohort. The propensity score model was generally able to balance measured covaria tes within propensity score quintiles to achieve a standardized difference less than 0.10. The MoM MoP comparison performed well, with standardized difference values under the threshold for several variables in which the reference category had large imbala nces. The quintile containing the largest propensity scores was the least balanced with year, insurance
66 type and age all falling above 0.10 but below the reference category. The CoP MoP comparison resulted in a lower proportion of quintiles achieving a sta ndardized difference of 0.10 or lower, however, this comparison had a greater imbalance to begin with. The CoP MoM model also performed well but was relatively balanced in the base case scenario. Results of Cox proportional hazards models showed no clear linear trend in risk of revision by strata for any articulating surface comparison. O nly one strata contained a statistically significant difference in risk (CoP MoP, Q2) . Confidence intervals were wider than the base case analysis because each strata has 20% of the overall sample size for each comparison . Full results of the analysis by propensity score quintile are presented in T able 4 14 . Pooled results of the propensity score strata are presented in T able 4 15 and were similar to the base case model. Mo M was associated with a small but non significant reduction in risk when compared to MoP. CoP appeared to have a relatively larger, but still not significant increase in risk when compared to both MoP and MoM. Figure 4 16 contains a depiction of quintiles and 95% confidence intervals for each articulating surface comparison. Participants were also matched by propensity score. The greedy matching algorithm selected 3 to 5 controls for each case in the analysis, with the average number of matches per case dep ending on the ratio of sample size for cases and controls. Generally, results of the propensity score matching analysis were similar to the base case scenario for risk of revision. The CoP MoP comparison benefited from a slightly smaller confidence interva l, likely because this type of matching provides greater efficiency when a rare case can be matched to a frequent control. Although CoP
67 articulating surfaces were not rare in the MPCD, this comparison did have the greatest different in the number of cases and controls. Results of the propensity score matched model are presented in T able 4 16. A second matching strategy provided more strict control for the effect of THA index hospitalization year in the analysis. Participants were again matched on propensit y score with Greedy matching, but this matching strategy matched strictly within year of procedure. Under this matching strategy, CoP was associated with a statistically significant increase in risk when compared to MoM (HR 1.37, p=0.03). MoM was associate d with a lower risk of revision relative to MoP than in the base case scenario, but failed to reach statistical significance. Results from the propensity score matching within year analysis are reported in T able 4 17. Multiple Imputation Missing articulat ing surface codes were imputed using M arkov chain Monte Carlo (MCMC) simulation. The simulation process produced 20 samples with values for missing articulating surfaces using the covariance matrix between articulating surface and all other variables model ed in the base vase analysis to sample a variance adjusted mean. The results of the MCMC imputation model differed from the results obtained from the base case scenario. The hazard ratio of the MoM MoP comparison was lower than the base case scenario. Both comparisons with CoP articulating surfaces revealed a reduced hazard ratio indicating CoP was associated with a smaller increase in risk than in the base case scenario. Results from the multiple imputation mo del are presented in T able 4 23 .
68 Table 4 1 . B aseline characteristics of MPCD by payer Variable Total (n=64,028) Medicare (n=41,408) Private (n=21,709) Multiple (n=911) Age Group <56, % 10.0 0.0 29.5 0.0 56 65, % 14.7 1.6 40.1 4.5 66 75, % 41.7 53.1 19.1 59.4 >75, % 33 .6 45.3 11.2 36.1 Female, % 57.5 61.5 49.9 57.1 Race a White, % 78.4 93.8 48.4 93.2 Black, % 4.5 4.5 4.6 5.5 Other, % 1.9 1.6 2.4 1.2 Missing, % 15.2 0.0 44.6 0.1 Coronary Artery Disease, % 22.3 26.9 13.3 25.1 Congestive Hea rt Failure, % 5.4 7.0 2.5 4.8 Chronic Lung Disease, % 19.5 21.6 15.5 20.6 Diabetes, % 20.2 22.5 15.8 21.7 Myocardial Infarction, % 4.6 5.6 2.6 5.0 Obesity, % 12.8 11.6 14.9 11.9 Peripheral Artery Disease, % 11.4 14.2 6.0 11.5 Thromboembolism, % 2.8 3 .4 1.6 2.9 Articulating Surface MoP, % 20.7 23.2 15.7 23.4 MoM, % 13.4 12.4 15.5 12.3 CoC, % 1.4 0.9 2.5 1.5 CoP, % 5.8 4.7 8.1 4.7 Missing, % 58.6 58.9 58.1 58.1 Region Northeast, % 16.7 18.8 12.8 13.5 M idwest, % 28.7 29.0 28.2 26.9 South, % 36.4 33.5 41.6 38.0 West, % 18.3 18.7 17.3 21.6 Primary Indication for THA Osteoarthritis, % 93.3 95.3 89.5 95.9 Aseptic necrosis, % 4.0 2.5 6.9 1.9 Other, % 2.7 2.2 3.6 2.2 Charlso n Comorbidity Index Score Zero, % 49.4 43.4 61.0 46.5 One or Two, % 34.1 37.1 28.5 36.0 Three or Four, % 11.6 13.7 7.7 12.1 Five or more, % 4.8 5.8 2.8 5.4 Year of THA 2007, % 14.1 13.8 14.8 13.3 2008, % 27.8 28 .2 27.0 26.8 2009, % 29.4 28.8 30.7 30.4 2010, % 28.7 29.3 27.5 29.5 *values are means(SD) or percentages. *values of polytomous variables may not sum to 100% due to rounding a self reported race identification b as determined by claim on which THA was submitted
69 Table 4 2 . Characteristics of cohort by articulating surface code Variable MoP (n=13,228) MoM (n=8,604) CoC (n=9,16) CoP (n=3,745) Not Missing (n=26,493) Missing (n=37,535) Age Group <56, % 5.4 14.7 27.8 17.5 10.9 9.4 56 65, % 10.7 17.3 26.6 23.1 15.1 14.4 66 75, % 44.1 40.5 31.3 40.9 42.1 41.4 >75, % 39.8 27.5 14.2 18.5 31.9 34.8 Female, % 59.9 52.5 52.1 55.0 56.5 58.2 Race a White, % 82.6 77.5 70.0 73.2 79.2 77.9 Black, % 3.5 4.1 3.5 4.4 3.8 5.1 Other, % 1.7 1.7 4.3 2.0 1.8 1.9 Missing 12.2 16.7 22.2 20.4 15.2 15.1 Coronary Artery Disease, % 23.6 20.4 14.0 17.1 21.3 23.0 Congestive Heart Failure, % 5.9 4.5 2.6 3.4 5.0 5.7 Chronic Lung Disease, % 19.8 18.6 16.9 17.4 19.0 20.0 Diabetes, % 20.9 18.7 15.2 17.0 19.4 20.8 Myocardial Infarction, % 5.0 4.0 2.4 3.4 4.3 4.8 Obesity, % 12.7 14.4 13.4 14.3 13.5 12.2 Peripheral Artery Disease, % 11.9 9.9 6.9 7.9 10.5 12.0 Thromboembolism, % 3.3 2.7 2.2 2.6 3.0 2.7 Region Northeast, % 20.8 13.1 19.7 16.9 17.7 16.0 Midwest, % 30.0 27.8 23.9 23.2 28.1 29.0 South, % 27.8 36.8 34.7 35.0 32.0 39.5 West, % 21.4 22.3 21.7 24.9 22.2 15.5 Primary Indication for THA Osteoarthrit is, % 94.8 93.3 90.2 93.3 94.0 92.8 Aseptic necrosis, % 3.2 4.3 7.6 4.4 3.9 4.1 Other, % 2.0 2.4 2.2 2.3 2.2 3.1 Charlson Comorbidity Index Score Zero, % 47.6 53.3 59.9 57.6 51.3 48.1
70 Variable MoP (n=13,228) MoM (n=8,604) CoC (n=9,16) CoP (n=3,745) Not Missing (n=26,493) Missing (n=37,535) One or Two, % 35.2 32.3 27.7 30.3 33.3 34 .7 Three or Four, % 12.4 10.2 8.7 8.4 11.0 12.1 Five or more, % 4.9 4.3 3.6 3.7 4.5 5.0 Primary Payer for THA Medicare, % 72.6 59.5 39.0 51.8 64.2 65.0 Private, % 25.8 39.2 59.5 47.0 34.3 33.6 Multiple, % 1.6 1.3 1.5 1. 1 1.4 1.4 Year of THA 2007, % 13.9 16.8 19.4 10.3 14.5 13.9 2008, % 25.9 29.2 24.3 19.5 26.0 29.0 2009, % 30.0 31.3 28.6 31.1 30.5 28.7 2010, % 30.2 22.7 27.6 39.2 28.9 28.5 *values of polytomous variables may not sum to 100 % due to rounding a self reported race identification T able 4 2, Continued
71 Table 4 3 . Characteristics of NIS Cohort (weighted estimates) Variable 2000 2006 2011 Males, n (%) 68,488 (41.7) 98,943 (43.5) 134,482 (44.1) Age, n (%) <45 12,360 (7.5) 14,464 (6.3) 14 ,983 (4.9) 45 54 20,348 (12.4) 34,071 (14.9) 43,763 (14.3) 55 64 31,323 (19.1) 55,741 (24.4) 85,361 (27.9) 65 74 49,195 (29.9) 62,061 27.2) 87,176 (28.5) 75 84 42,016 (25.6) 50,742 (22.2) 59,866 (19.6) >85 9,216 (5.6) 11,302 (5.0) 14,60 6 (4.8) White, n (%) 109,520 (89.2) 138,993 (86.9) 234,789 (85.8) Osteoarthritis primary diagnosis 131192 (79.8) 192696 (84.4) 262735 (85.9) Medicare, n (%) 95,401 (58.2) 123,120 (54.1) 160,107 (52.7) Private, n (%) 59,183 (36.1) 89,607 (39.4) 122,052 (40.2) Length of stay, days 4.6 4.0 3.3 Number diagnoses on record 4.4 5.8 7.5 Number of procedures 1.6 2.0 2.0 Total charges (not adjusted for inflation) 25,368 41,600 55,748 Annual hospital volume 48.8 67.1 87.0
72 Table 4 4 . Baseline characterist ics of NIS by payer Variable Total (n= 1,372,814 ) Medicare (n= 557 , 625 ) Commercial (n= 444,432 ) Age Group <56, n (%) 242 , 328 (22.5) 25 , 026 (4.5) 177 , 439 (39.9) 56 65 , n (%) 289 , 110 (26.9) 52 , 669 (9.4) 211 , 079 (47.5) 66 75 , n (%) 297 , 3 29 (27.6) 251 , 186 (45.0) 40 , 382 (9.0) >75 , n (%) 247 , 725 (23.0) 228 , 744 (41.0) 15 , 533 (3.5) Female , n (%) 601 , 344 (56.1) 346 , 859 (62.3) 217 , 591 (49.2) Race a White , n (%) 744 , 382 (69.2) 401 , 029 (71.9) 301 , 945 (67.9) Black , n (%) 58 , 1 36 (5.4) 25 , 473 (4.6) 23,016 (4.1) Other , n (%) 51 , 794 (4.8) 23 , 041 (4.1) 21,271 (4.8) Missing , n (%) 222 , 156 (20.6) 108 , 081 (19.4) 98,182 (22.1) Articulating Surface MoP , n (%) 212 , 086 (19.7) 128 , 124 (23.0) 73 , 506 (16.5) MoM , n (%) 159 , 727 (14.8) 69 , 790 (12.5) 78 , 128 (17.5) CoC , n (%) 27 , 368 (2.5) 7 , 293 (1.3) 17 , 618 (4.0) CoP , n (%) 75 , 020 (7.0) 27 , 654 (5.0) 41 , 073 (9.2)) Missing , n (%) 60 2,291 (55.9) 324 , 764 (58.2) 234,107 (52.7) Region Northeast , n (%) 221 , 884 (20.6) 113 , 243 (20.3) 94693 (21.3) Midwest , n (%) 286 , 181 (26.6) 147 , 991 (26.5) 121164 (27.3) South , n (%) 349 , 176 (32.4) 187 , 504 (33.6) 134338 (30.2) West , n (%) 219 , 250 (20.4) 108 , 887 (19.5) 94237 (21.2) Primary Indication for THA Osteoarthritis , n (%) 918 , 594 (85.3) 483 , 883 (86.7) 382 , 343 (86.0) Aseptic necrosis , n (%) 88,109 (8.1) 32,356 (5.8) 41,358 (9.3) Other , n (%) 69,789 (6.5) 41,285 (7.4) 20,732 (4.7) Year of THA , n (%) 2007 , n (%) 242 , 446 (22.5) 128 , 315 (23.0) 98 , 955 (22.3) 2008 , n (%) 266 , 432 (24.8) 137 , 194 (24.6) 109 , 657 (24.7) 2009 , n (%) 275 , 265 (25.6) 143 , 984 (25.8) 112 , 683 (25.4) 2010 , n (%) 292 , 348 (27.2) 148 , 131 (26.6) 123 , 137 (27.7) a self reported race ident ification
73 Table 4 5. Hospital volume logistic regression Hospital Quartile Unadjusted OR and 95% CI vs. VHVH Adjusted OR and 95% CI vs. VHVH Mortality VLVH 2.65 (2.21 3.17) 1.82 (1.48 2.23) LVH 1.73 (1.43 2.09) 1.40 (1.13 1. 71) HVH 1.62 (1.33 1.96) 1.46 (1.19 1.77) PLoS VLVH 2.13 (2.07 2.18) 2.37 (2.32 2.44) LVH 1.40 (1.36 1.44) 1.50 (1.47 1.55) HVH 1.15 (1.12 1.19) 1.14 (1.11 1.17) Table 4 6 . R isk of revision Articulating Surfa ce H azard R atio 95% Confidence Interval p value Unadjusted MoM vs MoP 0.957 0.791 1.157 0.651 CoP vs MoP 1.207 0.943 1.546 0.136 CoP vs MoM 1.288 0. 990 1. 675 0.059 Adjusted MoM vs MoP 0.9 72 0.799 1.182 0. 770 CoP vs M oP 1.2 30 0.951 1.592 0.1 15 CoP vs MoM 1.291 0.990 1. 684 0.059 Medicare (adjusted) MoM vs MoP 0. 952 0.750 1.209 0. 688 CoP vs MoP 1.159 0.830 1.619 0.387 CoP vs. MoM 1.283 0.895 1.838 0.175 Commercial (adjusted) MoM vs MoP 1.024 0.714 1.47 0.896 CoP vs MoP 1.322 0.858 2.037 0.206 CoP vs. MoM 1.257 0.836 1.888 0.272 Table 4 7 . RMST hazard ratio for MoM MoP revision Year Interval n at risk Hazard Ratio 95% Confidence Interval p value Year 0 to 1 21832 0. 823 0.649 1. 042 0. 106 Year 1 to 2 13874 1.276 0.821 2.811 0.279 Year 2 to 3 7448 1.736 0.893 4.157 0.104 Year 3 to 4 2409 3.608 0.304 42.84 0.310
74 Table 4 8 . R isk of dislocation Articulating Surface H azard R atio 95% Confidence Interval p value Unadjust ed MoM vs MoP 0.78 0 0.647 0.941 0.00 1 CoP vs MoP 0.952 0.746 1.214 0.6 90 CoP vs MoM 1.239 0.949 1.616 0.115 Adjusted MoM vs MoP 0.811 0.669 0.983 0.033 CoP vs MoP 0.954 0.741 1.23 0 0.718 CoP vs MoM 1.272 0.972 1 .664 0.079 Medicare (adjusted) MoM vs MoP 0.931 0.738 1.173 0.543 CoP vs MoP 1.006 0.722 1.401 0.972 CoP vs. MoM 1.153 0.807 1.648 0.434 Commercial (adjusted) MoM vs MoP 0.628 0.443 0.891 0.009 CoP vs MoP 0.877 0.586 1.312 0.522 CoP vs. MoM 1.396 0.913 2.135 0.123 Table 4 9 . Risk of joint infection Articulating Surface H azard R atio 95% Confidence Interval p value Unadjusted MoM vs MoP 1.231 1.009 1.503 0.041 CoP vs MoP 1.128 0.851 1.494 0.402 CoP vs MoM 0.924 0.692 1.232 0.589 Adjusted MoM vs MoP 1.216 0.991 1.493 0.062 CoP vs MoP 1.128 0.842 1.511 0.420 CoP vs MoM 0.939 0.702 1.256 0.671 Medicare (adjusted) MoM vs MoP 1.215 0.94 1.57 0.136 CoP vs MoP 1.317 0.914 1.9 0 0.14 0 CoP vs. MoM 1.079 0.737 1.579 0.696 Commercial (adjusted) MoM vs MoP 1.245 0.872 1.778 0.227 CoP vs MoP 0.924 0.573 1.491 0.748 CoP vs. MoM 0.791 0.504 1.242 0.308
75 Table 4 10 . Risk of DVT Articula ting Surface H azard R atio 95% Confidence Interval p value Unadjusted MoM vs MoP 0.883 0.805 0.968 0.008 CoP vs MoP 0.804 0.703 0.918 0.001 CoP vs MoM 0.913 0.793 1.052 0.208 Adjusted MoM vs MoP 1.016 0.925 1.116 0.741 CoP vs MoP 0.938 0.817 1.076 0.360 CoP vs MoM 0.943 0.818 1.088 0.423 Medicare (adjusted) MoM vs MoP 1.035 0.929 1.152 0.534 CoP vs MoP 0.864 0.727 1.026 0.095 CoP vs. MoM 0.804 0.701 1.007 0.06 Commercial (adjusted) MoM vs MoP 0.964 0.786 1.183 0.727 CoP vs MoP 1.136 0.891 1.447 0.303 CoP vs. MoM 1.180 0.929 1.500 0.176 Table 4 11 . Risk of mechanical loosening Articulating Surface H azard R atio 95% Confidence Interval p value Unadjusted MoM vs MoP 1.096 0.864 1.391 0.451 CoP vs MoP 1.124 0.805 1.57 0 0.492 CoP vs MoM 1.059 0.748 1.498 0.748 Adjusted MoM vs MoP 1.107 0.866 1.414 0.418 CoP vs MoP 1.048 0.741 1.482 0.790 CoP vs MoM 1.047 0.738 1.486 0.796 Medi care (adjusted) MoM vs MoP 1.041 0.774 1.399 0.792 CoP vs MoP 0.741 0.449 1.221 0.239 CoP vs. MoM 0.788 0.468 1.328 0.372 Commercial (adjusted) MoM vs MoP 1.178 0.745 1.863 0.483 CoP vs MoP 1.632 0.951 2.799 0.075 CoP vs. MoM 1.388 0.846 2.277 0.194
76 Table 4 1 2 . 1 year restricted follow up of outcomes Comparison Hazard Ratio 95% Confidence Interval p value MoM vs MoP Revision 0.823 0.649 1.042 0.106 Dislocation 0.729 0.591 0.899 0.003 P JI 1.168 0.930 1.466 0.181 DVT 0.993 0.897 1.100 0.898 Mechanical Loosening 0.827 0.601 1.139 0.246 CoP vs MoP Revision 1.320 0.993 1.756 0.056 Dislocation 0.938 0.719 1.224 0.640 PJI 1.125 0.818 1.545 0.46 9 DVT 0.900 0.776 1.044 0.163 Mechanical Loosening 1.13 0.765 1.673 0.537 CoP vs MoM Revision 1.56 1.153 2.116 0.004 Dislocation 1.361 1.022 1.813 0.035 PJI 0.950 0.692 1.304 0.75 2 DVT 0.916 0.786 1.068 0.263 Me chanical Loosening 1.465 0.966 2.222 0.07 3 Table 4 13 . Abbreviated baseline period by payer Population/ comparison Hazard Ratio 95% Confidence Interval p value Full cohort MoM vs MoP 0.952 0.75 1.209 0.688 CoP vs MoP 1.159 0.83 1.6 19 0.387 CoP vs. MoM 1.233 0. 938 1. 622 0.133 Medicare MoM vs MoP 0.947 0.745 1.203 0.655 CoP vs MoP 1.139 0.813 1.596 0.449 CoP vs. MoM 1.268 0.882 1.823 0.200 Commercial MoM vs MoP 1.086 0.742 1.59 0.670 CoP vs MoP 1.253 0.786 1.999 0.343 CoP vs. MoM 1.115 0. 724 1.716 0.621
77 Table 4 14 . Propensity score stratification Comparison Hazard Ratio 95% Confidence Interval p value MoM vs MoP Strata 1 0.696 0.389 1.246 0.223 Strata 2 1.03 0 0.663 1.601 0.895 Strata 3 1.106 0.724 1.689 0.641 Strata 4 0.902 0.6 00 1.356 0.621 Strata 5 0.942 0.627 1.416 0.774 CoP vs MoP Strata 1 0.829 0.384 1.792 0.634 Strata 2 2.114 1.243 3.595 0.006 Strata 3 1.151 0.627 2.113 0.651 Strata 4 0.901 0.528 1.538 0.702 Strata 5 1.526 0.881 2.644 0.132 CoP vs MoM Strata 1 1.125 0.611 2.071 0.706 Strata 2 1.651 0.969 2.812 0.065 Strata 3 1.218 0.676 2.194 0.513 Strata 4 0.916 0.505 1.659 0.772 Strata 5 1.78 0.833 3.804 0.137 Table 4 15 . Pooled results of propensity score stratification Comparison Hazard Ratio 95% Confidence Interval MoM vs MoP 0.9 35 0. 600 1. 462 CoP vs MoP 1. 304 0. 733 2.336 CoP vs. MoM 1.338 0. 712 2.508 Table 4 16 . Propensity score matching Comparison Hazard Ratio 95% Confidence Interval p value MoM vs MoP 0.985 0.812 1.195 0.87 8 CoP vs MoP 1.263 0.977 1.634 0.075 CoP vs. MoM 1.286 0.986 1.677 0.064
78 Table 4 1 7 . Propensity score matching by year Comparison Ha zard Ratio 95% Confidence Interval p value MoM vs MoP 0.938 0.759 1.161 0.5581 CoP vs MoP 1.225 0.927 1.62 0.1541 CoP vs. MoM 1.371 1.028 1.827 0.0316 Table 4 1 8 . MCMC Imputation Comparison Hazard Ratio 95% Confidence Interval p value MoM vs MoP 0.91 4 0.777 1.075 0.273 CoP vs MoP 1.188 0.811 1.741 0.371 CoP vs. MoM 1.166 0.930 1. 461 0. 179
79 Figure 4 1 . Articulating surface ICD 9 CM code utilization from 2006 to 2011 in the NIS A B Figure 4 2. Articulating surfaces in the NIS by A) Medicare and B) commercial
80 Figure 4 3 : Articulating surface ICD 9 CM code utilization from 2006 to 2011 in the MPCD by payer A B Figure 4 4. Articulating surfaces in the MPCD by A) Medicare and B) commercial
81 Figure 4 5 . Kaplan Meier survival plot of THA revision by MoM and MoP articulating surfaces
82 Figure 4 6 . Kaplan Meier survival plot of THA revision by CoP and MoP articulating surfaces
83 Figure 4 7 . Kaplan Meier survival plot of THA revision by CoP and MoM articulating surf aces
84 Figure 4 8 . Kaplan Meier survival plot of THA dislocation by MoM and MoP
85 Figure 4 9 . Kaplan Meier survival plot of THA PJI by MoM and MoP
86 Figure 4 10 . Kaplan Meier survival plot of THA DVT by MoM and MoP
87 Figure 4 11 . Kaplan Meier survi val plot of THA DVT by CoP and Mo P
88 Figure 4 12 . Distribution of overlapping propensity scores for MoM MoP (a), CoP MoP (b), and CoP MoM (c).
89 Figure 4 13 . Standardized differences for propensity score quintiles : MoM MoP
90 Figure 4 14 . Standardized differences for propensity score quintiles: CoP MoP
91 Figure 4 15 . Standardized differences for propensity score quintiles: CoP MoM
92 Figure 4 16 . Results of propensity score stratification into quintiles
93 CHAPTER 5 D ISCUSSION T he present study is the first to explore the utilization of CoP articulating surface codes in medical claims data, and to report on the comparative safety of CoP in terms of revision, dislocation, DVT, PJI, and mechanical loosening in a large and diverse sample of US patients. T he present analysis is also one of the few to explore the comparative safety of MoM and MoP articulating surfaces in US claims data, and the first to analyze the safety of THA in claims data from non Medicare payers. While the present analysis provides sev eral interesting and previously unreported findings, the clinical applicability of these results is severely restricted by the limitations inherent in the data, methods, and validity of identifying articulating surfaces via claims data. While the results s hould be interpreted with caution, the thorough application of a number of diverse methods reveals many important considerations in the analysis of THA and medical devices in observational studies . Results of the present study are divided into three sectio ns, exploration of THA claims in the US, results of the comparative safety analysis, and impact of various sensitivity analyses on the results . All three sections are described in detail below. Baseline Demographics and Articulating Codes The NIS provided the ability to analyze ten years of data to determine trends in THA that are important to the present study. It also allowed for an in depth analysis of the factors related to articulating surface ICD 9 CM code use, including detailed information regardin g characteristics of hospital size and location. Possibly the greatest contribution of the NIS to the current study was the ability to compare the generalizability of the MPCD to the US population. The MPCD provided little information
94 as to the specifics o f the commercial insurance plans it contained, and consisted of only a 15% sample of Medicare patients. Estimates from the NIS suggest the sample contained in the MPCD differed in some ways from the US population, but that much of the variance could be exp lained by over representation by insurance provider and region. In this section we discuss the baseline characteristics of the cohorts, the similarities between the two, and analyze factors related to the use of articulating surface codes. Comparison of t he Cohorts The population of the MPCD tended to be older, more likely to be female, and more likely to have a primary diagnosis of osteoarthritis compared to the NIS . This was due in part to over representation of Medicare patients in the MPCD . Medicare w as listed as the primary payer for 65% of the THA claims in the MPCD, while the NIS estimated that around 53% of THA claims in the US over the same time period have Medicare as the primary payer. When both datasets were stratified by insurance provider, th e baseline demographics were similar. The Medicare cohort in the MPCD appeared nearly identical to the estimates from the NIS , which is logical given that the MPCD was constructed in part from a 15% random sample of all Medicare patients. There is little r eason to believe the random sampling failed with respect to THA. The commercially insured participants in the MPCD were less similar to NIS estimates of the US population. Some of this variability may originate from the over representation of the South an d Midwest regions in the MPCD . Further analysis revealed that participants in the South and Midwest were more likely to be younger, less likely to be white or to have non osteoarthritis diagnoses, mirroring the differences seen between the MPCD and nationa l estimates from the NIS. A large amount of the
95 variability between datasets was removed when both samples were stratified by payer and region , which is consistent with the overrepresentation of Medicare individuals and an underrepresentation of commercial ly insured participants from the Northeast and West in the MPCD . Trends in THA Analyses in the NIS revealed that there were a number of trends in the utilization of THA over time that warrant consideration in a comparative safety analysis. The incidence and rate of THA operations has been growing rapidly in the US over the past decade. The rate of THA has increased more than 7.8% annually and more rapidly than anticipated . Previous projections indicated that the number of THA procedures would ris e to more than 253,000 by 2010. 2 Our analysis found the actual number of THA performed was more than 300,000 in 2010. THA has become relatively more common in younger individuals, especially those aged 55 64. These patients comprised only 19% of all THA procedures in 2000, and increased to 30% in 2011. Rates of THA have also been incre asing in male s . Together these results may be indicative of advances that have made THA a more attractive option for younge r, more active patients. Improvements to implant design such as larger femoral heads, improved hard on hard articulating surfaces, and more wear resistant polyethylene liners may have led to increased use in patients that would have previously waited to un dergo THA. These findings are somewhat consistent with previous projections of THA use in younger patients. An analysis by K ur tz et al . 42 predicted patients under the age of 65 would comprise 50% of THA procedures by 2011. Our analysis found 47% of THA procedures were performed in this age group during 2011. If these trends in THA continue past 2011 as they are projected, it may
96 have an important effect on the gener alizability of the results of the comparative safety analyses . Research Objective 1: Trends in Articulating Surface Code Utilization Our analysis discovered that the use of MoM articulating surface codes peaked in 20 08, around the time safety concerns aro se. We also found that rates of CoP codes increased steadily throughout the study period. To our knowledge, no other study has reported a similar increase in the use of CoP articulating surface codes or increases i n the rate of CoP utilization. It was also determined that articulating surface codes were included in less than half of THA claims in the NIS and MPCD. With such a large percentage of missing codes, small associations between appearance of a code and baseline characteristics could have an importa nt impact on the analysis. In preparation for the present study, a specialist familiar with chart abstraction and coding processes for orthopedic procedures was consulted and the process for the inclusion or non inclusion of a code was clarified. The speci alist could think of no specific reason any particular claim would not have an articulating surface code attached, other than difficulty in matching the data in the medical chart or the device name to the appropriate articulating surfaces. This led to a n umber of hypotheses regarding potential correlations between inclusion of a code and patient or institutional characteristics. The first hypothesis was that it may be more difficult to determine the articulating surface of certain types of models of hip pr ostheses. If the types of hip prostheses that had difficult to determine articulating surface codes were also more frequently used in certain type s of patient s , associations between patient characteristics and articulating surface would present. For exampl e, new generation polyethylene articulating surfaces may have names that are
97 unfamiliar to a medical claims coder. If prostheses with new generation polyethylene are more likely to be used in younger and male patients then these patients would be underrepr esented in the overall distribution of patients with articulating surface codes. THA procedu res are more familiar with the devices used and therefore are more likely to link the device name to an articulating surface code, henceforth referred to as the The plausibility of these hypotheses was investigated in the N IS. The analysis focused on whether each hypothesis could be considered a confounder in a comparative safety analysis. In the present analysis, a confounder is any variable that is associated with both choice of articulating surface and the risk of outcome . For the variation between prostheses hypothesis, participant characteristics known to be associated with negative outcomes were compared across THA claims including an articulating surface code and those without. The NIS estimated that participants witho ut an articulating surface code were slightly older (65.8 years vs. 64.3 years), more likely to be female (57.4% vs. 54.6%), and more likely to be white (67.0% vs. 68.9%) than those with an articulating surface code. Similar differences were seen in the MP CD (Table 4 3). All of these differences were statistically significant, but the potential level of impact is less clear given the small magnitude of the differences. The greatest difference in use of articulating surface codes occurred across regions. The Northeast and West included articulating surfaces codes in more than half of claims (55.0% and
98 53.6%, respectively) while the Midwest and South included codes in less than half of claims (46.3% and 38.6%, respectively). T he coding proficiency hypothesis was explored in two steps. First, the association between hospital volume and the proportion of THA claims including an articulating surface code was measured. Second, the association between hospital volume and negative outcomes of THA was assessed with a logistic regression model. The annual procedure volume of a hospital was associated with both the likelihood an articulating surface claim was included in a THA claim, and a lower likelihood of negative in hospital outcomes. Hospitals in the lowest quarti le of annual procedure volume coded 34.5 % of articulating surfaces, compared to 52 % in the highest quartile. The lowest volume hospitals had a 1.82 times greater risk of in hospital mortality and a 2.37 times greater risk of PLoS when compared to the highe st quartile hospitals. These findings support previous studies that have shown decreased rates of negative outcomes, specifically dislocation and VTE, after THA. 91 , 92 It is impossible to rule out the possibility that confounding factors or patient selection is responsible for these associations, but hospital pro cedure volume appears to have an important role in the analysis of THA. High volume centers have been recommended by the Institute of Medicine as a key strategy for improving the quality and outcomes of care, 93 and it is possible that such a switch could benefit patients undergoing THA. These findings raise important concerns regarding the use of articulating surface codes in claims data. If hospitals with lowe r rates of complication are more likely to use THA articulating surfaces codes, it could be that rates of these complication s are being underestimated. This would have a negative effect on the generalizability of study
99 results . Moreover, if the effect was not similar for each articulating surface, variatio n in coding practices could bias a comparative safety analysis. Further, t he difference between the number of patients treated at the lowest and highest volume hospitals has been growing over time. VHVH ho spitals performed around 60% of all THA procedures in 2001, rising to nearly 80% in 2011. Hospital procedure volumes have been linked to lower rates of mortality in a number of surgical procedures, including open heart surgery, vascul ar surgery, and corona ry bypass, 90 and it is plausible that the reduction seen in in hospital outcomes of THA may be reflective of a lower quality of THA procedure that also increases rates of longer term negative outcomes. Research Objective 2: Comparative Safety The results of the comparative safety base case analysis failed to find any statistically significant differences in risk of revision between articulating surfaces at the p=0.05 level . CoP articulating surfaces were associated with an elevated risk of revision when compared to MoP and MoM, with the CoP MoM comparison statistic ally significant at the p<0.10 level. The risk of dislocation for CoP compared to MoM was also significant at a p<0.10 level. There is some published evidence in the literature supporting increased risk for CoP articulating surfaces One study in younger pa tients found that 33% of CoP hips were revised after 10 years, compared to 0 revisions for a particular type of MoM prostheses . 94 An increased risk of dislocation for CoP hips compared to MoM has also been suggested in published literature . 95 Both of these studies were performed in a single institution and did not include a large number of patients. T he present analysis is the first to find an increased risk for dislocatio n and revision for CoP THA in a large observational analysis.
100 The C ox model comparing MoM and MoP articulating surfaces showed evidence of a violation of the proportional hazards assumption. The h azard for MoM hips was initially lower but eventually becam e greater than MoP after approximately 2.5 years. This highlights the importance of thinking about THA revision a s a multifaceted outcome with many distinct causes, each potentially having a different temporal association with risk. MoM THA has a reduced r isk of dislocation due to use of larger femoral heads than MoP and CoP articulating surfaces . Most dislocations occur relatively soon after the THA procedure, with around 40% of dislocations occurring in the first year. 61 Joint infection and DVT are similarly thought to occur in greater frequency soon after surgery, and mechanical loosening generally occurs over longer time periods. Segmenting the risk of revision by year in the MoM MoP model revealed a small di fference in the hazard early, eventually rising to a 1 69% increase between 2 and 3 years, and a 361% increase between years 3 and 4. MoM articulating surfaces were associated with a 19% decrease in risk of dislocation (p=0.03) when compared to MoP . This r eduction in risk was greatest for younger participants with commercial insurance, where the hazard was 0.62 compared to 0.93 for Medicare insured participants (p<0.01). The hazard rate for dislocation in Medicare participants is nearly identical to a previ ous estimate for Medicare patients in US claims data. 38 It is important to consider the generalizability of results in the interpretation of any analysis. If results from the Medicare population were considered generalizable to the entire population, the risk of dislocation would be overestimated. There was a significant unadju sted increase in risk of PJI for MoM articulating surfaces whe n compared to MoP (HR 1.23, p =0 .04). The covariate adjusted analysis
101 resulted moderate evidence of increased risk, however it was not statistically significant (HR=1.22 p=0.06). The analysis fa iled to find statistically significant difference s in the other articulating surface comparisons. These results are similar to results obtained by Bozic et al . , who found a statistically significant 19% increase in hazard for MoM hips when compared to MoP. Our findings also suggested that, unlike dislocation, the increased hazard ratio was similar in both Medicare and c ommercially insured participants. The average time to infection was also similar in Medicare and commercially insured participants (170 days and 167 days respectively). We found no significant differences in the hazard of DVT or mechanical loosening between any two articulating surface s. A previous analysis revealed significant differences between MoM and MoP on both DVT and mechanical loosen ing. The results of our analysis had similar hazard ratios in direction and magnitude, but lacked statistical significance due to small sample size. In particular, the present analysis was not well suited to analyze the risk of mechanical loosening. The av erage time to mechanical loosening was 341 days, longer than the other outcomes analyzed. The relatively large number of censored participants by this time, in addition to the small number of mechanical loosening outcome s included in the MPCD (785), result ed in an underpowered analysis. Sensitivity Analysis The sensitivity analysis revealed several important considerations when comparing claims with different articulating surface codes in observational data. The propensity score stratification method indica ted that there is no linear relationship between propensity score and difference in hazard. We discovered through the propensity score matching sensitivity analysis that matching is an effective method at
102 increasing sample size and conditioning out the eff ect of confounding factors. Finally, the multiple imputation analysis, when combined with the previous analysis on the missingness of articulating surface codes, indicates that imputation may play an important role in reducing the effect of data missing at random. Propensity Score Calculation The estimated propensity scores were largely successful in balancing the measured baseline covariates. Participants were divided into quintiles to represent the likelihood they would have received a particular articul ating surface based on the observed covariates available in the MPCD. The first quintile had a mixture of covariates that was the least associated with a particular articulating surface, and the covariate distribution of the fifth quintile had the greatest association. Grouping participants in this way led to an acceptable level of balance between cases and controls for most covariates. Prior to propensity score stratification, cases and controls differed greatly in respect to age, insurance type, and year of procedure in all three articulating surface comparisons. The difference in age/insurance type were particularly large for the comparisons involving MoP articulating surfaces, which are typically used in older/Medicare individuals. Stratification was lar gely successful in balancing age/insurance type for the MoM MoP comparison, but participants in the quintile with the largest estimated propensity scores (quintile 5) still showed some signs of imbalance. Comparisons with CoP articulating surfaces had a re latively large degree of imbalance in year of THA between cases and controls. This could be related to the large increase in use of CoP over the study period that was discovered in the first section of the present analysis. The exact cause of these imbalan ces is unclear,
103 however it is an important consideration that age, insurance provider, and year of THA may be more difficult to balance between cases and controls in each articulating surface comparison. Propensity Score Stratification The results of the propensity score stratified model revealed interesting information about the relationship between predictors of articulating surface selection and risk of revision. The purpose of this sensitivity analysis was to determine if there was a differential risk of revision between participants that were most likely to receive a particular articulating surface and those who were less likely. We failed to find any association between propensity score (when divided into quintiles) and an increase or decrease in haza rd. There are a number of interesting observations that may help explain or understand this finding. First, the propensity score may have done an acceptable job balancing the risk of revision in each quintile. There was only one quintile that resulted in a statistically significant HR for risk of revision (Q2 CoP vs MoP). With 15 total quintiles, it is plausible that this is due to the large number of tests performed. Differences between HR for quintiles within each comparison fluctuated at an apparent ran dom pattern and there was a large degree of ove rlap in confidence intervals. It should also be considered that the propensity score did a poor job balancing the true baseline risk of revision. The purpose of the propensity score is to balance measured base line covariates between treatment groups. In this respect, there was little evidence the propensit y scores performed poorly (see T able 4 13, Table 4 14, Table 4 15 ). However, the propensity score is not able to balance unmeasured confounding factors. The v ariables in the propensity score models were not able to explain a large
104 degree of the variability in choice of articulating surface. If there were other factors that were not included in the model, it could explain the fluctuation in HR between quintiles and the wide confidence intervals. Propensity Score Matching Results of the propensity score matched models were similar to the covariate adjusted base case model. The propensity score matched model used 1:n matching which improved power slightly, with a greater effect in the CoP MoP comparison which had the greatest discrepancy between the number of exposures in each group. When participants were matched based on year and propensity score calculated separately within each strata, the results differed sli ghtly from the base case model and previous propensity score matching method. The hazard for MoM when compared to MoP was reduced by a small amount , while the hazard for CoP compared to MoM rose. It is possible that this difference is due to the changing composition in MoM articulating surfaces over time. MoM articulating surfaces had the most drastic shift in utilization over the study window in both the MPCD and NIS. In general, MoM participants in the MPCD were more likely to be older and female, and le ss likely to be commercially insured as time passed. Older age and female gender have been suggested as risk factors for revision. 97 , 98 It is also possible there were unmeasured characteristics changing over this period as well. This may cause problems when the follow up period is limited because participants with a long follow up (3 to 4 years) are systematically different than those whose follow up time is right censored by the end of the observation window. To partially mitigate the effect of this difference, participants were matched by year, ensuring comp arisons between articulating surfaces and the effect of the covariates included in the model can have the same effect regardless of follow up time.
105 When propensity scores were calculated and matched within each of the four years of the analysis, the CoP Mo M analysis had the greatest change when compared to the results of the previous propensity score matching scenario. This difference is likely related to the changing distribution of CoP and MoM articulating surface codes over time. Research Objective 1 rev ealed that these articulating surfaces had the greatest annual changes in incidence, with CoP rising rapidly from 2007 to 2010, and MoM rising until 2008 and falling after. It is possible that year of THA was such an important predictor of articulating sur face in the propensity score, that removing it and matching based propensity scores within year allowed for a better balance of other covariates between cases and controls. It would appear the standardized differences for the comparisons containing CoP ach ieved less balance than the MoM MoP comparison, possibly due to the influence of a rapidly changing incidence of CoP over the study window. Multiple Imputation Results of the multiple imputation analysis were similar to those obtained in the base case sce nario, but were associated with a smaller increase in risk for CoP articulating surfaces. The MoM MoP comparison resulted in a lower risk of revision for MoM articulating surfaces than the base case analysis. These results, along with the results of object ive 1 that found differences in patient and institutional characteristics for articulating surface code use, suggest that articulating surface codes may not be missing completely at random. Imputing missing codes using covariates in the data correlated wit h included codes had some effect on the hazard ratios of the comparisons. In the multiple imputation analysis, as well as with the rest of the sensitivity analyses, it
106 strengths and limitations of the data in different ways, and produced slightly different results. Each comparative analysis of orthopedic devices in administrative claims data requires a thorough exploration of the patients and data being included, and careful asses sment of which method best fits the study questions and research goals. Study Limitations There were several limitations that should be considered when interpreting the results of the present analysis. First, articulating surface codes have not been valida ted, and the relationship between these codes and the actual articulating surface used in THA is unclear. We consulted specialists and designed s everal exploratory analyses to examine the use of these codes. The appearance of each articulating surface code s demonstrated some level of face validity as they appeared to follow historical trends in THA , for example, the increase and decrease of MoM THA corresponded with safety concerns that originated in 2008 . Also, a rticulating surface codes were prevalent in populations that are expected to benefit from them most, and those which have been previously associated with their use. However, given the large percentage of missing codes and the relatively small hazard rates reported, even small issues with the validit y of these codes could have a substantial impact on the results. Second, the present analysis was unable to control for several i mportant characteristics of THA that are not contained in administrative claims data. The specific type of prosthesis used may be important. For example, MoP and CoP prostheses may contain conventional polyethylene or newer cross linked polyethylene liners, which is important because the latter is thought to have less wear over time. The specific type of implant is important as w ell. In the studies that found a higher rate of revision for ASR XL MoM prostheses, the rate of revision was between 2 and 3 times higher than other
107 similar MoM prostheses. There is variability within each articulating surface combination in the size of th e femoral heads used, a specification that is thought to be closely related to risk of dislocation. The present analysis was also unable to determine information regarding the surgical procedure. The fixation method, cemented or cementless, may be an impor tant consideration. The MPCD did not contain hospital or surgeon identifiers and therefore could not control for the effect of increased volume of procedures. Given the statistically significant associations discovered in the first section of this project, this may be a serious limitation. The potential for unmeasured confounders is a concern for most retrospective observational analyses. However, the present analysis lacks information on several confounding factors that are known to be important and could have an effect on risk equal to or greater than what was demonstrated in the comparative safety analysis. Third, there were a large number of statistical analyses performed in the present study, and statistical significance through multiple testing cannot be ruled out. With the alpha level of 0.05, one false positive out of every twenty tests would be expected. While there were not twenty tests performed, it is important to keep in mind the relationship between the number of tests and the marginal statisti cal significance for some of the findings. Conclusions and Future Research Using claims data to analyze the comparative safety of THA is a difficult task. There are a number of important limitations that cannot be ignored. However, the potential benefits are numerous . A dministrative claims data ha ve the ability to include a far larger number of patients than any registry. The large sample size opens the door to rare outcomes that cannot be explored in clinical trials or registry data. Claims data tend
108 to b e more generalizable and applicable to the current population, as it is less regional and center specific than registries. Claims data may contain information that is difficult to collect in a registry setting, such as medication use and provision of care outside of the medical network contained in the registry. Th e present analysis highlights many of these benefits. It has demonstrated in a nationally representative sample that incidence of articulating surfaces codes have been changing over time , and tha t the US differs from several of the large registries that previous safety data has been drawn from. . Claims data also made it possible to use a relatively large sample size to explore specific causes of revision by articulating surface The NIS allowed fo r the generalizability of these results to the US population to be assessed, and provided insight into the trends that may be contributing to these results. In particular, age and insurance provider may greatly affect the generalizability of a comparative safety analysis. Previously reported estimates of risk have been conducted primarily with Medicare data. This analysis has demonstrated that results in Medicare patients may not be reflective of younger, commercially insured and therefore lacks generalizab ility to the US as a whole. The present study also revealed that choice of methods is an important consideration in the comparative analysis of THA. There is evidence to support that the inclusion of articulating surface codes in a THA claim may not be in dependent of patient or institutional factors. We showed that propensity scores are an effective method for balancing baseline covariates between articulating surfaces, but that age and time are important considerations in the structure of how propensity s cores are created. Finally, revision of THA is a multifaceted outcome that may be better viewed as a composite
109 outcome. Many causes of revision have different risk profiles over time, and the implication of differential risks over time is an important cons ideration for any analysis. Analysis methods should have due diligence in determining the temporality of risk associated with the study outcomes and using appropriate modeling strategies to best reflect a potentially changing risk over time. T his is the f irst comparative safety analysis in claims data that has explored the risk of revision and other negative outcomes in CoP articulating surfaces in the US. While the hazards ratios may suggest an increased risk with CoP articulating surfaces across a number of methodical approaches, there are a number of limitations which inhibit the clinical relevance of these findings. Until they are reported or repeated in data with more detailed information on the important aspects of THA and better control for potential confounding factors, these results should be interpreted with caution. Additional research is needed to explore exactly what is causing this increased risk . The FDA is currently working on unique device identification codes, which will substantially incr ease the ability for comparative safety analyses of THA to be performed in claims data. The present study was designed to chronicle the current state of THA in the US, explore trends that may help understand how this state may change in the future, and to provide insight into methodology that may be useful for analyzing the comparative safety of articulating surfaces in the future.
110 LIST OF REFERENCES 1. Learmonth ID, Young C, Rorabeck C. The operation of the century: total hip replac ement. Lancet. Oct 27 2007;370(9597):1508 1519. 2. Kurtz S, Ong K, Lau E, Mowat F, Halpern M. Projections of primary and revision hip and knee arthroplasty in the United States from 2005 to 2030. The Journal of bone and joint surgery. American volume. Apr 2007;89(4):780 785. 3. Faulkner A, Kennedy LG, Baxter K, Donovan J, Wilkinson M, Bevan G. Effectiveness of hip prostheses in primary total hip replacement: a critical review of evidence and an economic model. Health Technol Assess. 1998;2(6):1 133. 4. D reinhofer KE, Dieppe P, Sturmer T, et al. Indications for total hip replacement: comparison of assessments of orthopaedic surgeons and referring physicians. Ann Rheum Dis. Oct 2006;65(10):1346 1350. 5. Lawrence RC, Helmick CG, Arnett FC, et al. Estimates of the prevalence of arthritis and selected musculoskeletal disorders in the United States. Arthritis and rheumatism. May 1998;41(5):778 799. 6. Haq I, Murphy E, Dacre J. Osteoarthritis. Postgraduate medical journal. Jul 2003;79(933):377 383. 7. Ravi B, Escott B, Shah PS, et al. A systematic review and meta analysis comparing complications following total joint arthroplasty for rheumatoid arthritis versus for osteoarthritis. Arthritis and rheumatism. Dec 2012;64(12):3839 3849. 8. Wright AA, Cook C, Abbot t JH. Variables associated with the progression of hip osteoarthritis: a systematic review. Arthritis and rheumatism. Jul 15 2009;61(7):925 936. 9. Altman R, AlarcÃ³n G, Appelrouth D, et al. The American College of Rheumatology criteria for the classificat ion and reporting of osteoarthritis of the hip. Arthritis & Rheumatism. 1991;34(5):505 514. 10. Busija L, Bridgett L, Williams SR, et al. Osteoarthritis. Best practice & research. Clinical rheumatology. Dec 2010;24(6):757 768. 11. Zhang W, Moskowitz RW, Nuki G, et al. OARSI recommendations for the management of hip and knee osteoarthritis, Part II: OARSI evidence based, expert consensus guidelines. Osteoarthritis and cartilage / OARS, Osteoarthritis Research Society. Feb 2008;16(2):137 162. 12. Sinusas K . Osteoarthritis: diagnosis and treatment. American family physician. Jan 1 2012;85(1):49 56.
111 13. Babis GC, Sakellariou V, Parvizi J, Soucacos P. Osteonecrosis of the femoral head. Orthopedics. Jan 2011;34(1):39. 14. Beaule PE, Amstutz HC. Management of Ficat stage III and IV osteonecrosis of the hip. The Journal of the American Academy of Orthopaedic Surgeons. Mar Apr 2004;12(2):96 105. 15. Grecula MJ, Thomas JA, Kreuzer SW. Impact of implant design on femoral head hemiresurfacing arthroplasty. Clinical orthopaedics and related research. Jan 2004(418):41 47. 16. Ortiguera CJ, Pulliam IT, Cabanela ME. Total hip arthroplasty for osteonecrosis: matched pair analysis of 188 hips with long term follow up. The Journal of arthroplasty. Jan 1999;14(1):21 28. 1 7. Gomez PF, Morcuende JA. A historical and economic perspective on Sir John Charnley, Chas F. Thackray Limited, and the early arthoplasty industry. The Iowa orthopaedic journal. 2005;25:30 37. 18. Callaghan JJ, Albright JC, Goetz DD, Olejniczak JP, Johns ton RC. Charnley total hip arthroplasty with cement. Minimum twenty five year follow up. The Journal of bone and joint surgery. American volume. Apr 2000;82(4):487 497. 19. Kavanagh BF, Wallrichs S, Dewitz M, et al. Charnley low friction arthroplasty of th e hip. Twenty year results with cement. The Journal of arthroplasty. Jun 1994;9(3):229 234. 20. Peters CL, McPherson E, Jackson JD, Erickson JA. Reduction in early dislocation rate with large diameter femoral heads in primary total hip arthroplasty. The J ournal of arthroplasty. Sep 2007;22(6 Suppl 2):140 144. 21. NIH consensus conference: Total hip replacement. NIH Consensus Development Panel on Total Hip Replacement. JAMA : the journal of the American Medical Association. Jun 28 1995;273(24):1950 1956. 22. DiGioia AM, 3rd, Jaramaz B, Colgan BD. Computer assisted orthopaedic surgery. Image guided and robotic assistive technologies. Clinical orthopaedics and related research. Sep 1998(354):8 16. 23. Bozic KJ, Morshed S, Silverstein MD, Rubash HE, Kahn JG. Use of cost effectiveness analysis to evaluate new technologies in orthopaedics. The case of alternative bearing surfaces in total hip arthroplasty. The Journal of bone and joint surgery. American volume. Apr 2006;88(4):706 714.
112 24. Zahiri CA, Schmalzrie d TP, Ebramzadeh E, et al. Lessons learned from loosening of the McKee Farrar metal on metal total hip replacement. The Journal of arthroplasty. Apr 1999;14(3):326 332. 25. Dorr LD, Wang ZI, Longjohn DB, Dubois B, Murken R. Total hip arthroplasty with use of the metasul metal on metal articulation Four to seven year results. Journal of Bone and Joint Surgery American Volume. Jun 2000;82A(6):789 798. 26. Dowson D, Hardaker C, Flett M, Isaac GH. A hip joint simulator study of the performance of metal on me tal joints: Part II: design. The Journal of arthroplasty. Dec 2004;19(8 Suppl 3):124 130. 27. Silva M, Heisel C, Schmalzried TP. Metal on metal total hip replacement. Clinical orthopaedics and related research. Jan 2005(430):53 61. 28. Harris WH. Osteoly sis and Particle Disease in Hip Replacement a Review. Acta Orthop Scand. Feb 1994;65(1):113 123. 29. MacDonald SJ, McCalden RW, Chess DG, et al. Metal on metal versus polyethylene in hip arthroplasty: a randomized clinical trial. Clinical orthopaedics a nd related research. Jan 2003(406):282 296. 30. Mont MA, Schmalzried TP. Modern metal on metal hip resurfacing: important observations from the first ten years. The Journal of bone and joint surgery. American volume. Aug 2008;90 Suppl 3:3 11. 31. McMinn DJ, Daniel J, Ziaee H, Pradhan C. Indications and results of hip resurfacing. International orthopaedics. Feb 2011;35(2):231 237. 32. Vail TP, Mina CA, Yergler JD, Pietrobon R. Metal on metal hip resurfacing compares favorably with THA at 2 years followup . Clinical orthopaedics and related research. Dec 2006;453:123 131. 33. Jazrawi LM, Kummer FJ, DiCesare PE. Alternative bearing surfaces for total joint arthroplasty. The Journal of the American Academy of Orthopaedic Surgeons. Jul Aug 1998;6(4):198 203. 34. Fisher J, Jin Z, Tipper J, Stone M, Ingham E. Tribology of alternative bearings. Clinical orthopaedics and related research. Dec 2006;453:25 34. 35. Jarrett CA, Ranawat AS, Bruzzone M, Blum YC, Rodriguez JA, Ranawat CS. The squeaking hip: a phenomeno n of ceramic on ceramic total hip arthroplasty. The Journal of bone and joint surgery. American volume. Jun 2009;91(6):1344 1349.
113 36. Rahman WA, Greidanus NV, Siegmeth A, Masri BA, Duncan CP, Garbuz DS. Patients report improvement in quality of life and s atisfaction after hip resurfacing arthroplasty. Clinical orthopaedics and related research. Feb 2013;471(2):444 453. 37. Berry DJ, Harmsen WS, Cabanela ME, Morrey BF. Twenty five year survivorship of two thousand consecutive primary Charnley total hip repl acements: factors affecting survivorship of acetabular and femoral components. The Journal of bone and joint surgery. American volume. Feb 2002;84 A(2):171 177. 38. Bozic KJ, Lau EC, Ong KL, Vail TP, Rubash HE, Berry DJ. Comparative effectiveness of metal on metal and metal on polyethylene bearings in medicare total hip arthroplasty patients. The Journal of arthroplasty. Sep 2012;27(8 Suppl):37 40. 39. Bozic KJ, Kurtz S, Lau E, et al. The epidemiology of bearing surface usage in total hip arthroplasty in the United States. The Journal of bone and joint surgery. American volume. Jul 2009;91(7):1614 1620. 40. Murphy LB, Helmick CG, Schwartz TA, et al. One in four people may develop symptomatic hip osteoarthritis in his or her lifetime. Osteoarthritis and ca rtilage / OARS, Osteoarthritis Research Society. Nov 2010;18(11):1372 1379. 41. Culliford DJ, Maskell J, Kiran A, et al. The lifetime risk of total hip and knee arthroplasty: results from the UK general practice research database. Osteoarthr Cartilage. Ju n 2012;20(6):519 524. 42. Kurtz SM, Lau E, Ong K, Zhao K, Kelly M, Bozic KJ. Future young patient demand for primary and revision joint replacement: national projections from 2010 to 2030. Clinical orthopaedics and related research. Oct 2009;467(10):2606 2612. 43. Bozic KJ, Pui CM, Ludeman MJ, Vail TP, Silverstein MD. Do the potential benefits of metal on metal hip resurfacing justify the increased cost and risk of complications? Clinical orthopaedics and related research. Sep 2010;468(9):2301 2312. 44. Smith AJ, Dieppe P, Vernon K, Porter M, Blom AW. Failure rates of stemmed metal on metal hip replacements: analysis of data from the National Joint Registry of England and Wales. Lancet. Mar 31 2012;379(9822):1199 1204. 45. Losina E, Barrett J, Baron JA, Katz JN. Accuracy of Medicare claims data for rheumatologic diagnoses in total hip replacement recipients. Journal of clinical epidemiology. Jun 2003;56(6):515 519.
114 46. Bozic KJ, Ong K, Lau E, et al. Risk of complication and revision total hip arthroplast y among Medicare patients with different bearing surfaces. Clinical orthopaedics and related research. Sep 2010;468(9):2357 2362. 47. Memtsoudis SG, Pumberger M, Ma Y, et al. Epidemiology and risk factors for perioperative mortality after total hip and k nee arthroplasty. Journal of orthopaedic research : official publication of the Orthopaedic Research Society. Nov 2012;30(11):1811 1821. 48. McAuley JP, Szuszczewicz ES, Young A, Engh CA, Sr. Total hip arthroplasty in patients 50 years and younger. Clinic al orthopaedics and related research. Jan 2004(418):119 125. 49. Ardaugh BM, Graves SE, Redberg RF. The 510(k) ancestry of a metal on metal hip implant. N Engl J Med. Jan 10 2013;368(2):97 100. 50. Sedrakyan A, Normand SL, Dabic S, Jacobs S, Graves S, Ma rinac Dabic D. Comparative assessment of implantable hip devices with different bearing surfaces: systematic appraisal of evidence. BMJ. 2011;343:d7434. 51. de Steiger RN, Hang JR, Miller LN, Graves SE, Davidson DC. Five year results of the ASR XL Acetabu lar System and the ASR Hip Resurfacing System: an analysis from the Australian Orthopaedic Association National Joint Replacement Registry. The Journal of bone and joint surgery. American volume. Dec 21 2011;93(24):2287 2293. 52. Katz JN, Wright EA, Wrigh t J, et al. Twelve year risk of revision after primary total hip replacement in the u.s. Medicare population. The Journal of bone and joint surgery. American volume. Oct 17 2012;94(20):1825 1832. 53. Zywiel MG, Sayeed SA, Johnson AJ, Schmalzried TP, Mont MA. Survival of hard on hard bearings in total hip arthroplasty: a systematic review. Clinical orthopaedics and related research. Jun 2011;469(6):1536 1546. 54. Wylde V, Blom AW. The failure of survivorship. J Bone Joint Surg Br. May 2011;93(5):569 570. 55. Smith TO, Nichols R, Donell ST, Hing CB. The clinical and radiological outcomes of hip resurfacing versus total hip arthroplasty: a meta analysis and systematic review. Acta orthopaedica. Dec 2010;81(6):684 695. 56. Goodfellow JW, O'Connor JJ, Murray DW. A critique of revision rate as an outcome measure: re interpretation of knee joint registry data. J Bone Joint Surg Br. Dec 2010;92(12):1628 1631.
115 57. Harris WH. Osteolysis and particle disease in hip replacement. A review. Acta Orthop Scand. Feb 1994 ;65(1):113 123. 58. Keegan GM, Learmonth ID, Case CP. Orthopaedic metals and their potential toxicity in the arthroplasty patient: A review of current knowledge and future strategies. J Bone Joint Surg Br. May 2007;89(5):567 573. 59. Aspenberg P, Herbert sson P. Periprosthetic bone resorption. Particles versus movement. J Bone Joint Surg Br. Jul 1996;78(4):641 646. 60. Morrey BF. Results of reoperation for hip dislocation: the big picture. Clinical orthopaedics and related research. Dec 2004(429):94 101. 61. Berry DJ, von Knoch M, Schleck CD, Harmsen WS. The cumulative long term risk of dislocation after primary Charnley total hip arthroplasty. The Journal of bone and joint surgery. American volume. Jan 2004;86 A(1):9 14. 62. V on Knoch M, Berry DJ, Harms en WS, Morrey BF. Late dislocation after total hip arthroplasty. The Journal of bone and joint surgery. American volume. Nov 2002;84 A(11):1949 1953. 63. Malkani AL, Ong KL, Lau E, Kurtz SM, Justice BJ, Manley MT. Early and late term dislocation risk afte r primary hip arthroplasty in the Medicare population. The Journal of arthroplasty. Sep 2010;25(6 Suppl):21 25. 64. Benson MK, Goodwin PG, Brostoff J. Metal sensitivity in patients with joint replacement arthroplasties. British medical journal. Nov 15 1975 ;4(5993):374 375. 65. Coleman RF, Herrington J, Scales JT. Concentration of wear products in hair, blood, and urine after total hip replacement. British medical journal. Mar 3 1973;1(5852):527 529. 66. Smith AJ, Dieppe P, Porter M, Blom AW. Risk of cancer in first seven years after metal on metal hip replacement compared with other bearings and general population: linkage study between the National Joint Registry of England and Wales and hospital episode statistics. BMJ. 2012;344:e2383. 67. Haddad FS. Prima ry metal on metal hip arthroplasty was not associated with increased cancer risk. The Journal of bone and joint surgery. American volume. Feb 20 2013;95(4):364. 68. Makela KT, Visuri T, Pulkkinen P, et al. Risk of cancer with metal on metal hip replacement s: population based study. BMJ. 2012;345:e4646.
116 69. Berry DJ. Epidemiology: hip and knee. The Orthopedic clinics of North America. Apr 1999;30(2):183 190. 70. Charnley J. Fracture of femoral prostheses in total hip replacement. A clinical study. Clinical o rthopaedics and related research. Sep 1975(111):105 120. 71. Learmonth ID. Total hip replacement and the law of diminishing returns. The Journal of bone and joint surgery. American volume. Jul 2006;88(7):1664 1673. 72. Lindahl H. Epidemiology of periprosth etic femur fracture around a total hip arthroplasty. Injury. Jun 2007;38(6):651 654. 73. Bozic KJ, Ries MD. The impact of infection after total hip arthroplasty on hospital and surgeon resource utilization. The Journal of bone and joint surgery. American v olume. Aug 2005;87(8):1746 1751. 74. Kurtz SM, Lau E, Schmier J, Ong KL, Zhao K, Parvizi J. Infection burden for hip and knee arthroplasty in the United States. The Journal of arthroplasty. Oct 2008;23(7):984 991. 75. Bozic KJ, Ward DT, Lau EC, et al. Risk Factors for Periprosthetic Joint Infection Following Primary Total Hip Arthroplasty: A Case Control Study. The Journal of arthroplasty. May 20 2013. 76. Turpie AG, Levine MN, Hirsh J, et al. A randomized controlled trial of a low molecular weight heparin (enoxaparin) to prevent deep vein thrombosis in patients undergoing elective hip surgery. N Engl J Med. Oct 9 1986;315(15):925 929. 77. Powers PJ, Gent M, Jay RM, et al. A randomized trial of less intense postoperative warfarin or aspirin therapy in the pr evention of venous thromboembolism after surgery for fractured hip. Arch Intern Med. Apr 1989;149(4):771 774. 78. Weinmann EE, Salzman EW. Deep vein thrombosis. N Engl J Med. Dec 15 1994;331(24):1630 1641. 79. Deyo RA, Cherkin DC, Ciol MA. Adapting a clini cal comorbidity index for use with ICD 9 CM administrative databases. Journal of clinical epidemiology. Jun 1992;45(6):613 619. 80. Murray SB, Bates DW, Ngo L, Ufberg JW, Shapiro NI. Charlson Index is associated with one year mortality in emergency departm ent patients with suspected infection. Academic emergency medicine : official journal of the Society for Academic Emergency Medicine. May 2006;13(5):530 536.
117 81. Kalish RL, Daley J, Duncan CC, Davis RB, Coffman GA, Iezzoni LI. Costs of potential complicati ons of care for major surgery patients. American journal of medical quality : the official journal of the American College of Medical Quality. Spring 1995;10(1):48 54. 82. Katz JN, Losina E, Barrett J, et al. Association between hospital and surgeon proced ure volume and outcomes of total hip replacement in the United States medicare population. The Journal of bone and joint surgery. American volume. Nov 2001;83 A(11):1622 1629. 83. Kaplan EL, Meier P. Nonparametric Estimation from Incomplete Observations. J Am Stat Assoc. 1958;53(282):457 481. 84. Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70(1):41 55. 85. Imbens GW. Nonparametric estimation of average treatment effects under exogeneity: A review. Rev Econ Stat. Feb 2004;86(1):4 29. 86. Normand S LT, Landrum MB, Guadagnoli E, et al. Validating recommendations for coronary angiography following acute myocardial infarction in the elderly: a matched analysis using propensity scor es. Journal of clinical epidemiology. 2001;54(4):387 398. 87. Ming K, Rosenbaum PR. Substantial gains in bias reduction from matching with a variable number of controls. Biometrics. 2000;56(1):118 124. 88. Rubin DB. Multiple imputation for nonresponse in s urveys. Vol 81: John Wiley & Sons; 2004. 89. Bernaards CA, Belin TR, Schafer JL. Robustness of a multivariate normal approximation for imputation of incomplete binary data. Statistics in medicine. 2007;26(6):1368 1382. 90. Luft HS, Bunker JP, Enthoven AC. Should operations be regionalized? the empirical relation between surgical volume and mortality. The New England Journal of Medicine. 1979;301(25):1364 1369. 91. Battaglia TC, Mulhall KJ, Brown TE, Saleh KJ. Increased surgical volume is associated with low er THA dislocation rates. Clinical orthopaedics and related research. Jun 2006(447):28 33. 92. Singh JA, Kwoh CK, Boudreau RM, Lee GC, Ibrahim SA. Hospital Volume and Surgical Outcomes After Elective Hip/Knee Arthroplasty A Risk Adjusted Analysis of a Larg e Regional Database. Arthritis and rheumatism. Aug 2011;63(8):2531 2539.
118 93. The Leapfrog Group Website . http://www.leapfroggroup.org/. Accessed April 4, 2014. 94. Migaud H, Jobin A, Chantelot C, Giraud F, Laffargue P, Duquennoy A. Cementless metal on meta l hip arthroplasty in patients less than 50 years of age: comparison with a matched control group using ceramic on polyethylene after a minimum 5 year follow up. The Journal of arthroplasty. 2004;19(8):23 28. 95. Clarke M, Lee P, Villar R. Dislocation afte r total hip replacement in relation to metal on metal bearing surfaces. Journal of Bone & Joint Surgery, British Volume. 2003;85(5):650 654. 96. Ong KL, Lau E, Suggs J, Kurtz SM, Manley MT. Risk of subsequent revision after primary and revision total joint arthroplasty. Clinical Orthopaedics and Related ResearchÂ®. 2010;468(11):3070 3076. 97. Huddleston JI, Wang Y, Uquillas C, Herndon JH, Maloney WJ. Age and obesity are risk factors for adverse events after total hip arthroplasty. Clinical orthopaedics and r elated research. Feb 2012;470(2):490 496. 98. Inacio MC, Ake CF, Paxton EW, et al. Sex and risk of hip implant failure: assessing total hip arthroplasty outcomes in the United States. JAMA internal medicine. Mar 25 2013;173(6):435 441.
119 BIOGRAPHICAL SKETCH Jonathan Schelfhout was born in Panama City, Florida, a son to Steven Schelfhout and Teresa Schelfhout. He has one sister, Jennifer Schelfhout. He received p sychology and e conomics from the University of Florida in 2005. He began a research fellowship with the Pharmaceutical Outcomes and Policy Department during the summer of 2005, and was accepted to the same program the next year. He received his Ph.D. from the University of Florida in the summer of 2014. During h is graduate education Jonathan has served as a Research Fellow at the FDA and an economics intern at GlaxoSmithKline. Jonathan has authored and coauthored several peer reviewed publications and presented at national and international conferences. He has i nterest and experience working in both pharmacoeconomics and pharmacoepidemiology and he is interested in applying these topics to the study of chronic infections and the safety of medical devices.