Treatment Patterns of Newly Diagnosed Rheumatoid Arthritis Patients from a Commercially Insured Population

This observational cohort study queried medical and prescription claims from the HealthCore Integrated Research Database (HIRD)^SM, a repository of longitudinal claims data for approximately 40 million health plan enrollees across the US. The study population of interest was newly diagnosed RA patients. The study cohort consisted of patients identified during the period from July 1, 2006 through August 31, 2014 based on a combination of claims-based algorithm [20, 21]. The index date was defined as the earliest date on which a patient fulfilled one of the three algorithm criteria. Study data were handled in compliance with all applicable Health Insurance Portability and Accountability Act (HIPAA) rules. Patient anonymity and confidentiality were preserved with data de-identification throughout. Investigational Review Board (IRB) informed consent requirements were waived for this non-experimental study, which was conducted under the research exception provisions of the Privacy Rule, 45 CFR 164.514(e).

Study patients were ≥ 18 years on their index date and had continuous enrollment with both medical and pharmacy benefits during the 6-month period preceding the index date and for ≥ 1 year post-index. To be considered newly diagnosed patients, no RA diagnoses were allowed during the 6-month pre-index period. Patients meeting any one of three claims-based criteria were included: (1) presence of ≥ 1 medical claim for an RA diagnosis (ICD-9 714.0x, 714.1x, or 714.2x) by a rheumatologist; (2) ≥ 2 medical claims with an RA diagnosis by a non-rheumatologist occurring at least 30 days apart; or, (3) presence of ≥ 1 medical claim with an RA diagnosis occurring concurrently with or followed by a medical or pharmacy claim for any DMARD therapy within 1 year. Patients were followed beyond 1 year until they reached the end of the study period, left the health plan, or died, whichever occurred first. Patients with ≥ 1 claim of a cancer diagnosis (ICD-9 code 140.xx–209.3x, 230.xx–234.xx) during the 6-month pre-index period were excluded from the study.

Patterns of DMARD treatment—including csDMARDs, biologic tumor necrosis factor inhibitors (TNFi; etanercept, adalimumab, infliximab, certolizumab, and golimumab), biologic non-TNFi (abatacept, rituximab, tocilizumab, anakinra), and JAKi (tofacitinib)—were assessed during follow-up. Additionally, use of prescription pain medications (opioids, nonsteroidal anti-inflammatory drugs [NSAIDs], muscle relaxants, and anticonvulsants) and steroids (oral and injectable glucocorticoids) were examined. All-cause chronic opioid use was defined as receiving ≥ 180 days supply of opioid medication during the follow-up period. Discontinuation of DMARD treatment was considered as the absence of a claim for a DMARD within 90 days after the last day supplied of the most recent medication claim. The discontinuation date was defined as the last medication claim date plus the days supplied for that claim. Switching was defined as the filling of a DMARD class that was different from any medication in the current regimen or most recent prior medication while not continuing the original therapy; stopping and restarting the same medication was not considered a switch. Add-on therapy included the filling of a second DMARD class while maintaining continuous therapy with a previously prescribed DMARD. A maximum of three treatment changes (adding on or switching therapy) were captured for each patient, resulting in the capture of up to four lines of DMARD treatment. Medication dosage was not captured in this study.

Two sensitivity analyses were conducted: (1) limiting the analysis to patients identified during January 1, 2010 through August 31, 2014 to account for the potential evolution in therapeutic management and treatment recommendations in recent years, and (2) requiring patients to have three distinct RA diagnosis on distinct dates when the diagnostic workups or tests (e.g., imaging of joint, laboratory test for C-reactive protein, synovial fluid analysis rheumatoid factor, cyclic citrullinated peptide, erythrocyte sedimentation rate) used for diagnosing/ruling out RA were not included, AND visited a rheumatologist at time of first diagnosis.

Prespecified comorbid conditions were defined as the presence of ≥ 1 diagnosis code for the condition of interest during the 6-month pre-index period. The Quan–Charlson Comorbidity Index (QCI) was calculated to quantify the burden of comorbid disease in this population [22]. The specialty of the physician associated with the RA diagnosis was identified on the index date medical claim. If physician information was not available on the index date, the claim closest to the index date (either pre- or post-index) with a diagnosis for RA was used to determine diagnosing specialty.

This single-cohort study of RA patients was descriptive in nature, and treatment patterns were analyzed within the entire cohort with no comparison group. Descriptive statistics (means, standard deviations, medians, frequencies, and percentages) were reported for each outcome of interest. All analyses were conducted using SAS Enterprise Guide v7.1 (Cary, NC, USA).

After applying inclusion and exclusion criteria (Fig. 1), there were 63,101 RA patients with mean ± SD age of 57 ± 15.6 years, and 72.7% were female (Table 1). Most patients were enrolled in preferred provider organization (PPO) health plans (72.5%), and 27.8% had Medicare Advantage or Medicare Supplemental plus Part D coverage. Nearly half of the study population (44.2%) was diagnosed or treated by a rheumatologist at index. Hypertension (40.5%), dyslipidemia (34.0%), and osteoarthritis (31.1%) were among the most commonly observed comorbid conditions.

The use of a large, geographically diverse, administrative claims database provided this study with several key strengths. First, the data allowed us to retrospectively observe the longitudinal treatment patterns in newly diagnosed RA patients over a period of 1–9 years (mean, 3.5 years) after their initial RA diagnosis. Second, a sensitivity analysis limiting the observation period to 2010 onward showed consistent results, and thus the findings are a reflection of historical treatment patterns over the past decade as well as recent treatment practices. Finally, these data provide real-world evidence of disease management rather than being limited to the strict guidelines of a clinical trial or other controlled experiment. This study includes a contemporary patient population and presents a view of the current treatment landscape.

While there were clear strengths of this study, these findings must be viewed against some important limitations. First, we identified patients newly initiating RA medication after a 6-month period in which there were no claims for RA medications or RA diagnoses before the index RA medication claim/RA diagnosis code. This approach may not have captured the true incident RA population, i.e., prevalent RA cases may be included, and as a result, our results may be conservatively biased. Second, the minimum continuous enrollment required was 12 months; however, patients were followed beyond the 12-month timeframe until the end of the study period, and their patterns later into enrollment may be affected by loss to follow-up. The mean observation period of 3.5 years from index may offer insufficient time to fully observe the actual use of more advanced DMARD therapies as biologics or JAKi. Finally, our analysis focused on commercially insured patients, possibly under-representing elderly patients enrolled in Medicare.

Furthermore, diagnosis codes are good indicators but not guarantees of the presence of disease as they may be subject to miscoding or may be included as a rule-out criterion. As noted, the reliance on administrative claims (a combination of ICD-9 codes from medical claims and claims for DMARD therapy), without access to medical records, is potentially limiting in identifying RA patients. Studies examining the validity of administrative claims to identify RA patients have found mixed results depending on the algorithm used. Typically the use of multiple claims with RA diagnoses improved the positive predictive value of the definitions, and requiring a diagnosis from a rheumatologist improves the algorithm even further [42‐44]. Studies that utilized a more focused code list of RA diagnoses, such as the one used in this study, tended to have higher validity compared with those that used all ICD-9 codes starting with 714 [44]. While our definition of RA may include some patients who do not truly have RA, the algorithm was developed with the goal of minimizing this occurrence. We also conducted a sensitivity analysis using a much more restrictive (i.e., specific) definition of RA, and study results were not fundamentally different.

In general, claims databases are subject to common limitations including the inability to account for medications from other sources such as over-the-counter purchases or samples from providers. Claims data provide limited insight into the clinical status of patients, particularly their severity or activity of their disease and the process surrounding moving onto DMARD therapy—for example, whether DMARD was offered then declined; prescribed and never filled; filled and not used; or not offered at all. Similarly, it is unknown whether patients not seen by a rheumatologist were referred but declined, or not referred. Reasons for initiation of one DMARD class over another, such as the selection of a non-TNFi or a JAKi as the index DMARD therapy, are unknown, and these patients may differ from the general RA population.