Longitudinal analysis of historical payments and utilization in patients with hip and knee osteoarthritis

Background Osteoarthritis (OA) is a highly prevalent condition associated with substantial clinical and economic burden. Value-based payment reform requires detailed understanding of care utilization. However, previous analyses of OA care have limitations, such as constraining analysis to a single year or to surgical patients. We aimed to more comprehensively characterize health services utilization and payment for hip and knee OA through a 3-year longitudinal analysis, including both operative and non-operative services, using Medicare claims data. Methods We utilized Medicare Standard Analytic Files available from PearlDiver, Inc. The target population for analysis was patients with osteoarthritis in the hip, knee, or both, identied by ICD-9 diagnosis codes. Patients were limited to those whose 1) rst instance of hip or knee OA diagnosis in the payer dataset occurred between 2008 and 2011 and 2) who were continuously active in the payer dataset for one year prior to and three years following diagnosis. Results Payments for relevant hip and knee OA services were highest during the rst year after diagnosis, representing 55.6%, 65.3%, and 51.2% of total payments for patients with knee, hip, or knee and hip OA, respectively. Payments and utilization of services in the second and third treatment years were lower, and similar to the year prior to OA diagnosis. Service-level analysis revealed that total payments were driven by utilization of high-intensity services like surgery, which accounted for 57.7%, 60.6%, and 63.6% of payments in the rst year for knee, hip, and knee and hip OA patients, respectively. High utilization of Medical Services like physician-administered medications, arthrocentesis, and physical therapy drove high payments as well, especially later in treatment. Conclusion Hip and knee OA treatment intensity is highest in the year immediately following a new diagnosis and decreases considerably in the second- and third-years following diagnosis across all relevant service categories. This analysis supports the identication of specic, time-sensitive opportunities to transform hip and knee OA care and payment models to optimize patient-centered while costs across the episode of

OA care episodes. Unfortunately, there are no large-scale examples of condition-based bundled payment models for OA or other chronic conditions. New payment models are necessary to incentivize high-value care for chronic OA and must consider appropriateness of high-intensity services like surgery. Payment reform for chronic hip and knee OA requires a detailed understanding of care utilization and payments. Current analyses have substantial limitations, lacking longitudinal investigation and cohorts inclusive of all patients with OA. Most published literature has focused on spending for patients undergoing total joint arthroplasty (TJA) and short procedural follow-up windows, typically 90 days.
In an effort to address these gaps in the literature surrounding OA, and to more thoroughly understand the nature of OA care in the United States, we attempted to comprehensively characterize hip and knee OA utilization and payment among the Medicare population through a 3-year longitudinal analysis following initial diagnosis, and a 1year lookback period prior to diagnosis. The purpose of this study was to evaluate payment and utilization of operative and non-operative services associated with the management of hip and knee OA among Medicare bene ciaries. By characterizing payments and utilization of care at the individual service level, we hope to identify salient trends and opportunities for targeted care and payment model transformation beyond procedural bundles for joint replacement and inform effective care and payment model construction and associated policy.

Methods
The data for this study are available through PearlDiver, Inc. (PearlDiverInc.com) using the Medicare National Sample Administrative Database (SAF5, which includes 5% random sample of physician billing records) of the PearlDiver Patient Record Database. The PearlDiver database contains de-identi ed and Health Insurance Portability and Accountability Act-compliant patient records and billing codes. The SAF5 subset contains records from 55 million Medicare bene ciaries between 2005 and 2015.
Patients with either osteoarthritis (OA) in the hip, knee, or both, without co-existing rheumatoid arthritis were identi ed using International Classi cation of Diseases, 9 th Revision (ICD-9) diagnosis codes (Appendix A). For these patients, we identi ed the rst instance of a hip or knee OA diagnosis and an Evaluation and Management Current Procedure Terminology (CPT) code on the same date. For a patient to be included in the analysis cohort, this index date had to occur in the calendar years 2008-2011. Our nal inclusion criteria for the analysis cohort was remaining active in the payer dataset for one year prior (washout period) and three years following the index date. Filtering criteria served to ensure completeness of records over 4+ years and capture only the rst episode of hip or knee OA on the payer record by ensuring patients were either newly diagnosed with OA or new to payer records. Our process of identifying and tracking patients before and after their index date is described in Figure 1.
Utilization and payment data for these patients were analyzed for medical services relevant to the management of hip or knee OA ( Figure 2). These relevant medical services were informed by expert opinion and clinical practice guidelines (12,13). In selecting these medical services, the clinical experience of several of the authors was considered (KB, WJ, RM). Conservative decisions were made to include the broadest array of medical services that might be utilized for OA patients.
While limited Medicare Part D data were available in the PearlDiver database, we chose to exclude calculations of oral drug expenditure in order to focus on inpatient and outpatient services and procedures. Further, calculation of these expenditures would have introduced signi cantly variability into our results, as we could not be certain of the relationship between many medications and speci c OA symptoms (rather than concurrent ailments) utilizing only administrative claims data.
Medical service utilization and payment data include unique patient count, service date count, total payments, payment per total patients (or per population), service dates per total patients (or per population). Data summaries and analysis were performed using SAS (SAS Institute; Cary, NC).

Results
Applying our trigger and ltering criteria to the SAF5 database, we identi ed a total of 92,931, 26,797, and 34, 315 Medicare bene ciaries that met our criteria for osteoarthritis of the knee, hip, or both, respectively. Table 1 describes each lter used to select patients for analysis, and Table 2 provides additional demographic information on the different cohorts.
Knee Osteoarthritis (Tables 3 and 4) Year 1 In the rst year of care following diagnosis of knee OA, 92,931 unique Medicare bene ciaries accounted for 737,640 service dates, or 41.6% of service dates in the three years following diagnosis. These patients accounted for $202M in payments in the rst year of care (55.6% of payments in the three-year post-diagnosis analysis period).
In total, 15.3% of patients underwent surgical procedures within one year of diagnosis, with 8.2% and 7.0% of patients undergoing Total Knee Arthroplasty and Arthroscopy, respectively. A substantial proportion of patients received other services, including Radiology (78.7% of patients) with 72.9% of patients receiving an X-Ray, 9.2% a CT scan, 20.9% an MRI, and 4.7% ultrasound imaging. Medical Services were used by 69.0% of patients, including behavioral health services (0.02%), hyaluronic acid injection (12.1%), corticosteroid injection (46.0%), nutrition services (0.8%), arthrocentesis (45.8%), stimulation (10.1%), PM&R services (13.0%), physical or occupational therapy (32.3%), and chiropractor services (8.9%). Evaluation and Management services were driven by ambulatory visits (94.1%), inpatient professional services (11.4%), emergency services (14.9%), and skilled nursing facilities Medical Services accounted for the most service dates (35.7% of all dates in the rst year of care) at 2.8 service dates per patient in the cohort. Physical therapy (18.6%, 1.5 dates per patient), arthrocentesis (10.6%, 0.8 dates per patient), and corticosteroid injection (8.1%, 0.6 dates per patient). Radiology accounted for 16.5% of service dates (1.3 dates per patient), driven by X-Ray (13%, 1.0 dates per patient) and MRI (2.9%, 0.2 dates per patient). Evaluation and Management services accounted for 21.6% of service dates in the rst year of care (1.7 dates per patient), and Surgery accounted for just 2.5% of dates and 0.2 dates per patient. Additional breakdown of service dates and payments per patient is provided in Table 4.

Evolution of Care
In addition to the 36-month post-diagnosis period, we analyzed payments and utilization for a period of one year prior to initial diagnosis. Utilization and payments for relevant services in the rst year following diagnosis increased dramatically from the prior year (56% more service dates, 163% greater payments). Total payments fell by 55% from Year 1 to Year 2 (to $90.8M), driven by a 30.8% decrease in utilization (service dates). Evaluation and Management, Surgery, and Radiology services drove this decrease in utilization and payment (72.5%, 65.2%, and 56.1% fewer service dates, respectively). Payments fell an additional 22.0% from Year 2 to Year 3 (to $70.8M), driven by an increased proportion of spending on Medical Services as compared to Surgery, Radiology, and Evaluation and Management. Total payments ($77.1M) and service dates (472,594) were similar in the year prior to diagnosis and second and third years after diagnosis ($90.8M, 509,910 service dates, and $70.8M, 524,589 service dates, respectively). The average payment per patient in the cohort was $2,182, $978 and $762, in Years 1, 2, and 3 of care following diagnosis, respectively.
Hip Osteoarthritis (Tables 5 and 6) Year 1 In the rst year of care following diagnosis of hip OA, 26,797 unique patients accounted for 202,836 service dates, or 43.6% of service dates in the three years following diagnosis. These patients accounted for $62.7M in payments in the rst year of care (65.3% of payments in the three-year post-diagnosis analysis period).
Medical Services accounted for the most service dates (32.1% of all dates in the rst year of care) at 2.4 service dates per patient in the cohort. Physical therapy (20.5%, 1.6 dates per patient), stimulation (4.4%, 0.3 dates per patient), chiropractor services (6.4%, 0.5 dates per patient), and corticosteroid injection (4.8%, 0.4 dates per patient). Radiology accounted for 21% of service dates (1.3 dates per patient), driven by X-Ray (16.8%, 1.3 dates per patient) and MRI (3.5%, 0.3 dates per patient). Evaluation and Management services accounted for 18.3% of service dates in the rst year of care (1.4 dates per patient), and Surgery accounted for just 2.4% of dates and 0.2 dates per patient. Additional breakdown of service dates and payments per patient is provided in Table 6.

Years 2 and 3
Surgery and Medical Services accounted for 54.1% ($10.4M) and 21.2% ($4.1M) of payments in Year 2, respectively. Total Hip Arthroplasty accounted for 52.1% ($10M) of payments. Physical Therapy drove the majority of Medical Services payments, at 16.4% ($3.2M) of payments. Evaluation and Management services accounted for just 4.5% ($0.9M) of payments. Medical Services and Radiology services accounted for the most patients seen in Year 2, at 39.9% and 33.8% of total, respectively, while 3.5% of patients underwent a surgical procedure in Year 2. Medical Services made up 40.1% of service dates, while Evaluation and Management, Radiology, and Surgery service dates decreased from Year 1 to 4.8%, 11.9%, and 0.9%, respectively.
In Year 3, Surgery accounted for 46.7% ($6.6M) of total payments (44.9%, $6.4M Total Hip Arthroplasty), while the proportion of payments toward Medical Services rose to 26.9% ($3.8M), driven by an increased proportion of payments for physical therapy (to 20.1%, $2.8M). The proportion of patients receiving any care decreased from Year 2, with 2.4% of patients receiving undergoing surgery, 28.3% receiving Radiology services, and 10.0% utilizing Evaluation and Management services.

Evolution of Care
Utilization and payments for relevant services in the rst year following diagnosis increased from the prior year (42% more service dates, 127% greater payments). Total payments fell by 69% from Year 1to Year 2 (to $19.2M), driven by a 36% decrease in utilization (service dates). Evaluation and Management, Surgery, and Radiology services drove this decrease in utilization and payment (83.3%, 77.1%, and 63.6% fewer service dates, respectively). Payments fell an additional 26.4% in Year 3 (to $14.1M), driven by an increased proportion of spending on Medical Services as compared to Surgery, Radiology, and Evaluation and Management. Total payments ($27.6M) and service dates (142,760) were greater in the year prior to diagnosis and second and third years after diagnosis ($19.2M, 129,917 service dates, and $14.1M, 132,747 service dates, respectively). The average payment per patient in the population was $2,340, $717, and $528 in Years 1, 2, and 3 of care following diagnosis, respectively.
Hip and Knee Osteoarthritis (Tables 7 and 8) Year 1 Additionally, we identi ed 34,315 Medicare bene ciaries with both a hip and a knee ICD-9 diagnosis code, at least one of which occurred for the rst time on the payer record between 2008 and 2011. This cohort accounted for 693,912 service dates and $226M in payments over the three year post-diagnosis analysis period, as well as 119,895 service dates and $37M in the year prior to diagnosis.
Medical Services accounted for the most service dates (50.2% of all dates in the rst year of care) at 3.3 service dates per patient in the cohort. Physical therapy (29.6, 2.0 dates per patient), arthrocentesis (11.6%, 0.8 dates per patient), chiropractor services (7.8%, 0.5 dates per patient), and corticosteroid injection (10.1%, 0.7 dates per patient). Radiology accounted for 28.1% of service dates (1.9 dates per patient), driven by X-Ray (22.7%, 1.5 dates per patient) and MRI (4.8%, 0.3 dates per patient). Evaluation and Management services accounted for 30.0% of service dates in the rst year of care (2.0 dates per patient), and Surgery accounted for 4.4% of dates and 0.3 dates per patient. Additional breakdown of service dates and payments per patient is provided in Table 8. In Year 3, Surgery accounted for 61.8% ($30.8M) of total payments (29.3%, $14.6M Total Knee Arthroplasty, 29.5%, $14.7M Total Hip Arthroplasty), while the proportion of payments toward Medical Services rose to 19.5% ($9.7M), driven by an increased proportion of payments for physical therapy (to 12.7%, $6.3M). The proportion of patients receiving any care decreased from Year 2, with 7.9% of patients receiving undergoing surgery, 50.3% receiving Radiology services, and 39.0% utilizing Evaluation and Management services.

Utilization Trends
We observed several trends in service utilization and payments among patient cohorts with osteoarthritis in the knee, hip, or both. Figure 3 provides a comparison of service utilization and payment across all years. Notably, Year 1 accounted for greater utilization and payments than either Years 2 or 3 across all joints. The distribution of payments was heavily skewed toward Surgery services, which were more common in the rst year of treatment, though utilization of all services declined over time.

Discussion
Transformative care and payment innovation require a detailed, actionable understanding of utilization and payment patterns. Through this analysis of OA service utilization and payment, we aimed to more comprehensively characterize the longitudinal nature of chronic arthritis care, in both a public and private payer setting.Our analysissuggests that there is an early, discrete, previously uncharacterized critical period in both hip and knee OA disease course to which care and payment model reform could be targeted to make an impact on relevant utilization and associated payments in order to appropriately incentivize use of evidence-based clinical practice guidelines and appropriate use criteria that drive higher-value patient-centered care. This opportunity occurs soon after initial diagnosis and management.
Service utilization and intensity relevant to hip or knee OA is signi cantly and broadly increased in the Medicare population following diagnosis of OA, as compared to the year preceding and the subsequent two years of management. The observed service utilization can be compared to accepted hip and knee OA clinical practice guidelines (13) or appropriate use criteria (14) to identify and characterize speci c opportunities for higher value care. For example, a signi cant proportion of patients with knee OA received hyaluronic acid injections, which do not re ect evidence-based practice, and were not endorsed at the time these data were collected (12,13,(15)(16)(17). New care delivery and payment models could incentivize a reduction in hyaluronic acid injections, and savings in time and money could be re-allocated to support the delivery of services that are decidedly indicated. For example, nutrition services are evidenced-based for hip and knee osteoarthritis (12,13,15), yet represented just 0.01% of total payments in this analysis. Moreover, a disproportionate amount of total joint arthroplasties occurs in the rst year following diagnosis. In light of recent research suggesting that some portion of total knee arthroplasties are not appropriate (6), new care delivery and payment models could incentivize careful consideration of when lower extremity joint replacement is most appropriate. Some have even suggested that total hip arthroplasty may be overutilized (18). Potential savings related to reduction in utilization could be re-allocated to support alternative conservative measures, care coordination, new team structures, or communication and education tools.
Limitations of this study include those that stem from our administrative claims data sources, which depend on the accuracy and completeness of coding and billing. Relatedly, our analysis sought to identify the rst instance of hip or knee OA through Medicare claims, but these may not be true rst instances for some patients; these patients may have experienced inconsistency in historical coding or may have been previously diagnosed while on another payer plan. It is worth noting that many patients do not see a doctor when they initially experience OA symptoms, but rather once their symptoms become severe or they feel prepared for an operation. The frequent event of receiving an operation in the rst year after diagnosis could be more related to delayed diagnosis than a premature operative decision. Further, our analysis assumes minimal variation in healthcare costs and global utilization across the calendar years analyzed. In future work, we plan to consider patient variation within each cohort, including age, geographic region, timing of diagnosis and services, and other factors, though these are somewhat limited in claims data. Incorporation of Medicare Part D claims data would allow for calculation or oral drug expenditure, but we do not include this due to the limited availability of this data.
Current and historical payment reform has focused on hip and knee osteoarthritis as one condition, e.g. Medicare Acute Care Episode Demonstration (ACE) (19), Bundled Payments for Care Improvement Initiative (BPCI) (20), Comprehensive Care for Joint Replacement Model (CJR) (21), Bundled Payments for Care Improvement Initiative-Advanced (BPCI-A) (22) which are triggered by either hip or knee OA-related procedures. However, because hip and knee osteoarthritis have different clinical courses; we analyzed patients with these conditions separately, as well as patients who had OA in both joints.
In addition to these limitations, our analysis does not evaluate the timing of relevant services in relation to one another. In the literature, there has been commentary on timing of services with relation to total knee arthroplasty (23,24). While our study focused on total costs, we foresee including service sequencing in future analysis.Our list of relevant services represents a potential area for bias, although these services were selected from a thorough discussion among subject matter experts and review of practice guidelines. Additionally, the cost of treatment certainly varies with intensity of service, geographic location, age and comorbidities, service location and other factors, which we hope to address in future analyses. Lastly, our analysis focused on costs and utilization associated with the management of hip and knee OA, with no consideration of patient-centered outcomes, which are not available through administrative claims data sets. Future efforts should include evaluation of both outcomes and costs in order to understand the value of services delivered to patients with hip and knee OA.

Conclusion
Our work demonstrates that hip and knee OA-related utilization and payments are signi cantly greater across virtually all relevant services in the year immediately following initial diagnosis and are signi cantly lower in subsequent years. Additionally, this analysis demonstrates that the greatest proportion of total cost of care for hip and knee OA can be attributed to increased utilization of operative procedures and selective outpatient services.
Efforts to reform payments for OA would be most impactful if they address the utilization that drives total cost of care. Alternative payment models focused on reducing unnecessary or low-value services could be impactful for the Medicare population analyzed and potentially for other insurers, should the same trends be present. This analysis also sets a foundation for the identi cation of speci c, time-sensitive opportunities to transform hip and knee OA care, and could inform future payment model innovation that could optimize patient-centered outcomes while controlling costs across the episode of care.

AVAILABILITY OF DATA AND MATERIAL
The data that support the ndings of this study are available through Pearl Diver, Inc. (a subscription online repository of healthcare data and information) but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. The claims data analyzed for this study through Pearl Diver, Inc, are from Centers for Medicare and Medicaid Services (CMS) Limited Data Set les.

COMPETING INTERESTS
No external funds were used to conduct this study.    If the number of service dates or total payments was less than 10 but greater than 0, PearlDiver reports the value as 5 throughout. Percentages re ect proportion of service dates or payments for total analysis period (Years 1, 2, and 3) Other Arthroplasty= revision and partial arthroplasty PM&R= physical medicine & rehabilitation PT/OT= physical and occupational therapy If the number of service dates or total payments was less than 10 but greater than 0, PearlDiver reports the value as 5 throughout.   Primary Service Categories and speci c procedures identi ed from expert opinion and clinical practice guidelines.