Study Overview

This was a retrospective observational cohort study using medical and prescription claims from the Optum Clinformatics Data Mart (CDM) linked to rheumatology electronic health records (EHR) in the Illumination Health data warehouse. The Illumination Health repository includes data from a national rheumatology practice-based research network (Bendcare, part of the OneFlorida+ Clinical Research Network Consortium [9], one of the eight Clinical Research Networks that comprise PCORnet [10]). Data residing in the Illumination Health data warehouse are curated from more than 700 community rheumatology providers practicing throughout the United States. In aggregate, it contains longitudinal patient-level health plan claims data in addition to disease activity scores and other clinical measures for RA. The study period spanned January 1, 2010, through March 31, 2020, or as the intersection of the claims and EHR data (described in subsequent sections) allowed.

Cohort Selection

Adults (i.e., ≥ 18 years old) with a diagnosis of RA as identified by at least two or more rheumatologists’ diagnosis codes for RA using International Classification of Diseases (ICD)–9 codes (714.0, 714.2, and 714.81) and ICD-10 codes (M05.*, M06.*, ignoring M06.1 and M06.4) between January 1, 2010, and December 31, 2019, were eligible for the study. They were also required to have at least 6 months of continuous coverage with pharmacy and medical benefits in Optum CDM and at least two or more rheumatologist visits in the EHR data with a valid RA disease activity measurement in the Illumination Health data warehouse. Although there are several metrics by which clinical RA disease activity can be measured [11], the Clinical Disease Activity Index (CDAI) was preferred, as it incorporates data from both patients and rheumatology providers. Given that the CDAI is typically only assessed at clinic visits, and based on some quality metrics, may only be checked once or twice a year [12], CDAI values in the Illumination Health data warehouse were considered valid for up to 18 months after each recorded measurement before censoring occurred. Lastly, patients had to have at least one filled/administered medication claim for an RA medication, defined as the index drug, and they had to be continuously eligible in the health plan for at least 6 months following the index date (Optum CDM). Thus, patients were required to have data from both sources to be eligible for inclusion: CDAI and RA diagnoses from the EHR, and RA diagnosis, health coverage and pharmacy claims data from Optum CDM. Patients were excluded if they had any diagnosis of other autoimmune and connective tissue diseases (e.g., psoriatic arthritis, ankylosing spondylitis, and other spondyloarthropathies, systemic lupus erythematosus, scleroderma, dermatomyositis, polymyositis, and primary systemic vasculitis) in the year prior to the index date, and were censored if there was a diagnosis for these during follow-up. The study schema is described in Supplementary Fig. 1. The study protocol was approved by the Advarra Institutional Review Board (Pro00043329) and was conducted in accordance with the Health Insurance Portability and Accountability Act compliance requirements.

Index Date and Follow-up Periods

To establish a baseline period for covariate assessment, the start of follow-up was anchored at the index date, defined as the calendar date that the patient met all inclusion/exclusion criteria detailed above. Data in the 12 months prior to the index date were used to establish baseline demographic and clinical characteristics, including disease activity state and costs, with certain covariates (e.g., number of prior b/cs/tsDMARDs used) assessed using all available prior data, to a maximum of 36 months. Patients were censored at the time of loss of enrollment in the health plan, the end of the study period (March 31, 2020), or a gap of more than 18 months since the most recent CDAI value.

Data Sources

The study used data from the Optum CDM and linked it to the Illumination Health Real-World Evidence Platform. Optum CDM is derived from a database of administrative health claims for members of a large national managed care company affiliated with Optum. The database includes approximately 17–19 million annual covered lives. The CDM data comprise both commercial and Medicare Advantage health plan data, including individuals over the age of 65 years.

The Illumination Health platform is an EHR-based repository of real-world rheumatology data from community rheumatology practices representing more than 700 community rheumatology providers utilizing one of two commonly used EHR systems. The rheumatology practices are spread across the US, and the study population included in the study represents RA patients with commercial health insurance. It captures clinical elements, ICD-10 diagnoses, procedures, current and past medications, and laboratory results. Patient-generated data were also included, including patient-reported outcomes captured both in office and out-of-office via mobile technologies (tablet app and smartphone apps). Follow-up required concurrent enrollment in the commercial health plan that temporally overlapped with the first and last visit in the EHR data.

Linking Data Sources

Linking of Optum CDM and Illumination Health data elements was enabled by third-party software designed for this purpose (Datavant). Data were tokenized and de-identified, consistent with Expert Determination Certifier recommendations. Records across data components (e.g., medical claims, pharmacy claims, and EHR) were then linked using the unique token for each patient before the transfer of the data file to the analytic research team. Patients in the Illumination Health data warehouse were linked to the Optum database, after which the cohort selection criteria were applied. In addition to requiring overlapping enrollment in the provider’s practice and the health plan, an additional step to confirm the validity of the linkage was applied. This step required at least two RA diagnosis codes, at least one of which had to occur on the exact same calendar date in both datasets for the linkage to be considered valid.

Exposure Variables

RA disease activity as the main exposure variable was identified in the EHR data and was assessed using the CDAI and classified as remission (CDAI ≤ 2.8), low (LDA, CDAI > 2.8 and ≤ 10), moderate (MDA, CDAI > 10 and ≤ 22), and high (HDA, CDAI > 22) [13, 14]. Exposure variables included the most recent therapy prescribed to the patients for RA, prior to the index date. The therapy groups were as follows: csDMARDs: methotrexate, sulfasalazine, hydroxychloroquine, and leflunomide as monotherapy or in combination with other csDMARDs; TNFis: etanercept, adalimumab, certolizumab, golimumab, and infliximab as monotherapy or in combination with csDMARDs; non-TNFis: abatacept, rituximab, sarilumab, and tocilizumab as monotherapy or in combination with csDMARDs; tsDMARDs: tofacitinib, upadacitinib, and baricitinib (i.e., JAKi) as monotherapy or in combination with csDMARDs; and “none” was defined as none of the above treatments prescribed in the baseline period, even though they met the RA cohort inclusion criteria. National drug code and healthcare common procedure coding system codes were used to identify the medications prescribed to these patients in the claims data.

Outcome Assessment

The primary outcome was healthcare costs for different disease activity states. Costs were obtained directly from the health plan and were standardized prior to being made available to the study team to avoid disclosure of proprietary information. This procedure was done by the health plan prior to analysis to establish standard costs that reflect the allowed payments across all provider services. For example, professional service rates were assigned using a resource-based relative value scale approach.

Costs reflected those paid by the payer and the patient, and included the estimated paid amount, patient co-insurance, copayments, and deductibles. Costs were grouped as medical costs (outpatient physician visits, diagnostic and laboratory services), inpatient costs (relating to hospitalization), and pharmacy costs, which included both filled prescription medications and intravenous (IV) infusions (which typically are categorized as medical costs, but were reclassified so as to group RA medications that bill under the medical benefit as drug-related costs). Total costs represented the sum of these costs. To account for inflation, the study team used the cost factors table provided by the health plan [15]. The annual cost factor was multiplied by the cost data for each specific year and type of service to normalize to 2020 costs. The secondary outcomes include the mean time (in days) to achieve remission or LDA and the mean time patients remained in remission or LDA in a subgroup with M/HDA at the index date.

Covariates

Covariates were assessed at the index date and were used to characterize patients overall and by the disease states described previously. Covariates of interest included demographics, clinical characteristics, and comorbidities thought to potentially influence treatment response (e.g., diabetes mellitus, chronic obstructive pulmonary disease (COPD), fibromyalgia, and depression) and those common to RA (e.g., cardiovascular disease and osteoporosis). RA-related medications including nonsteroidal anti-inflammatory drugs, opioids, and glucocorticoids were also described.

Statistical AnalysisOverview

The primary analysis described the proportion of patients in remission or LDA, including those who started in a higher disease category and achieved remission or LDA over time. The study also assessed costs, overall and by baseline RA therapy groups. Standardized mean differences (SMDs) were used to compare exposure groups, with SMDs > 0.10 generally considered as being clinically relevant [16]. Because disease activity varies over time, disease activity as measured by the CDAI was updated during follow-up daily, as often as recorded in the EHR by the rheumatology provider. Costs were attributed to the CDAI disease activity category (remission, low, moderate, and high) associated with each person-day of follow-up, yielding time-based cost intervals. For example, an RA patient examined by a rheumatology provider and recorded in the EHR as being in moderate disease activity would have all costs for that day, and all subsequent costs until the next CDAI measurement, accrue to the moderate disease activity category. If at a subsequent rheumatologist visit they moved to low disease activity, then the subsequent costs would be attributed to the low disease activity category. The CDAI was updated, and healthcare costs were accrued, on a person-day basis.

Subgroup Analyses

An analysis of a subgroup of patients who were in M/HDA at the index date and who subsequently started a new RA therapy described the mean time in days to achieve remission or LDA. Among those who achieved LDA or remission, we also assessed the mean time in days in remission or LDA before censoring. Finally, because the associations between cost and disease activity did not provide direct evidence that costs for any given individual might be reduced if they attained LDA, we calculated a within-person difference in costs by comparing costs in the 6 months prior to initiating a new therapy versus the costs 4–10 months post initiation. The 4- to 10-month time frame was selected given that the onset of action of most RA therapies yields a near-maximal benefit by approximately 4 months [17,18,19].

Analytic Methods

Descriptive statistics were used to depict the demographic and clinical characteristics of the patients by disease activity and therapy group. Frequencies and percentages were reported for categorical variables. Means and standard deviations, and medians and percentiles were provided for continuous variables. Given the skewness of the data, log transformation was considered, but given the nontrivial frequency of zero costs, we bootstrapped confidence intervals. Two hundred resamples using replacement created a synthetic sample of the same size as the original dataset. Confidence intervals were estimated using the percentile intervals of the mean value of the 200 repetitions. Multivariable adjustment was done to control for confounding by age, sex, and comorbidities (ischemic heart disease, diabetes mellitus, osteoporosis, COPD, depression, and fibromyalgia), selected based on clinical knowledge and avoiding factors that might be causally related to RA disease activity or function, which might lead to overadjustment. Adjusted costs were modeled using negative binomial regression, with remission as the referent category. All analyses were performed using R Statistical Software (version 4.1.0) [20].