Abstract
Background
The opportunity for new biomarker-based tests to predict treatment response and the appropriate use of emerging rheumatoid arthritis (RA) therapies may depend on patients’ prior treatment history. We examined rheumatoid arthritis (RA) treatment sequences and biologic/JAKi initiation overall and by line of therapy (LoT) to estimate the size of the eligible RA patient population in the U.S. for a new predictive treatment response test using a population-based RA inception cohort.
Methods
We utilized an augmented health plan claims database to create an inception cohort of RA patients and estimated the rate of advanced treatment (biologic or JAKi) addition or switch. Results were stratified by advanced treatment naïve vs. treatment-experienced status and also described specific RA treatment sequences and combinations over time.
Results
Among 37,656 RA patients, 59,557 new RA treatment initiations were identified. Most patients (85.2%) initiated biologics; the remainder initiated JAKi. Of these, 40.2% used biologics/JAKi as monotherapy. The overall biologic/JAKi addition/switch rate was 25.7 per 100 patient-years (22.7/100py in biologic-naïve patients, and >30/100py in treatment-experienced patients). Rates varied substantially between prescribers, with an observed add/switch rate in the practices of prescribers in the highest decile (53.8/100py) more than 4-fold greater than rates in the practice of prescribers in the lowest decile (12.1/100py).
Conclusions
These findings highlight the size and characteristics of the eligible RA population in the U.S. initiating a new RA biologic or JAKi treatment each year, thus providing valuable insights for stakeholders developing or marketing new RA therapies or biomarker-based diagnostic tests to predict future treatment response.
Keywords: rheumatoid arthritis, line of therapy, biologics, JAK inhibitors
Introduction
Patients with rheumatoid arthritis (RA) have an increasing array of treatment options available. After use of conventional synthetic disease-modifying antirheumatic drugs (csDMARDs) such as methotrexate (MTX), the medications available to RA patients include biologic disease-modifying antirheumatic drugs (bDMARDs) including tumor necrosis factor inhibitors (TNFis), other cytokine inhibitors (e.g., IL-6R, IL-1R), B-cell depleting therapies, and T-cell modulators; as well as Janus kinase inhibitors (JAKis), in the class of targeted synthetic DMARDs (tsDMARDs). However, the success of these medications to achieve treatment goals (i.e., remission or a minimal clinically important difference) in RA patients is highly variable and challenging to predict. Given the uncertainty about the best RA treatment for individual patients, a trial-and-error method is frequently used to select treatment.
Unfortunately, failure to achieve RA treatment goals not only diminishes patients’ quality of life due to the disease itself but may worsen disease-related comorbidities, lead to irreversible radiographic damage, result in adverse events associated with ineffective medications, and be costly.1–3 Moreover, as RA patients try and fail an increasing number of advanced therapies, treatment response typically falls. For example, the ACR20, ACR50, and ACR70 response rates in patients who are biologic naïve are approximately 60%, 40%, and 20%, respectively, whereas in patients who have failed 1 TNFi, the corresponding treatment response is closer to 50%, 30%, and 15%, respectively4 5. Treatment failure may undermine the patient-provider relationship, discourage patients, lead to poor persistence and compliance, and result in future reticence to switch therapies6,7, leading to under-controlled inflammation. Continued disease activity leads to excess mortality, cost, and burden on health care services.8 For these reasons, maximizing the response to patients’ initial and subsequent RA treatments is critical.
New blood-based diagnostic tests to predict treatment response are commercially available or are in development. However, the performance of these biomarker tests may vary based on patients’ treatment history. These tests may be specific to individual drugs or mechanisms of action (MOA) and may perform differently in those who are biologic-naïve compared to those who are biologic-experienced, particularly in those who have multiple prior treatment failures with difficult-to-treat RA9. Identifying the most common treatment sequences used in early RA and examining the rate of switching may be helpful in guiding the development and optimization of biomarker-based response prediction tests and understanding the frequency such tests might be used in clinical practice.
Using national augmented health plan claims data applied to an inception cohort of RA patients, our objectives were to 1) describe the sequencing pattern of b/tsDMARDs, by specific medications and by LoT [i.e., the lifetime number of distinct biologics and JAKis that a patient previously received], overall and between providers; and 2) examine the rate of RA medication initiation and treatment switching. The rationale underpinning these objectives was to quantify the opportunity for a predictive biomarker test to inform treatment response and estimate the volume of tests that might be used by RA patients in the U.S. Additionally, assuming that not all classes of RA medications would be predicted in a first generation of such a test, the most common sequences of RA treatments would inform which drug classes should be prioritized. The focus of our analysis was on b/tsDMARDs given that biomarker-based tests to predict future effectiveness of csDMARDs were considered unlikely to be cost-effective, and that methotrexate would remain the first line therapy used by the vast majority of RA patients.
Methods
Data Source and Study Period
We used the Komodo Healthcare Map® dataset, a data schema designed for Real-World Evidence (RWE) studies, from January 2016 to March 2024 to create an inception cohort of RA patients. Komodo’s Healthcare Map is composed of event-based, patient-level healthcare data from 60+ sources. These sources can be grouped into provider-derived (including pharmacy-derived) and payer-derived sources. Together, they comprise healthcare encounters for over 330 million medically insured patients, including those on Medicare, Medicaid, or commercial insurance, to provide a longitudinal view of a patient journey. We extracted disease characteristics, treatment history, and medication filling patterns information, and focused on use of b/tsDMARDs. Komodo contains data about diagnoses (ICD-10-CM), medications (identified using both pharmacy claims and medical claims coded via Healthcare Current Procedure Coding System [HCPCS]), and health care resource utilization.
Newly Diagnosed RA Study population
Cohort inclusion criteria for the RA cohort required ≥2 outpatient ICD10-CM encounters with a diagnosis of RA from rheumatology providers occurring ≥30 days apart and within 365 days in patients ≥18 years of age. A diagnosis of RA was defined by ICD-10 codes M05.* and M06.* (ignoring M06.1 and M06.4). Continuous medical and pharmacy coverage was required for ≥12 months preceding the first RA diagnosis code (‘baseline’) and through the second RA diagnosis code. Using information in the 12-month baseline and all available prior data, RA patients with any prior diagnoses of psoriatic arthritis, ankylosing spondylitis, axial spondyloarthritis, reactive arthritis, inflammatory bowel disease, polymyalgia rheumatica, hidradenitis supparativa, or malignancy (excluding non-melanoma skin cancer) were excluded to avoid confounding related to other diseases that might influence treatment switching. Patients with any prior biologic or JAKi use before or during the 12-month baseline period were also excluded to assemble an inception cohort of patients with new onset RA. Prior csDMARD use was allowed given the non-specific nature of some RA therapies (e.g., hydroxychloroquine for patients with undifferentiated arthritis later diagnosed with RA; use of MTX in other inflammatory arthritis such as psoriatic arthritis). Given that the accuracy of identifying RA in large health plan data sources is greatly enhanced when coupled to prescriptions for DMARDs13, patients in the RA cohort were required to use ≥1 cs/b/tsDMARD between the first RA diagnosis date and the end of the observation period (3/31/2024).
RA Medication Exposure, Time, and Rate of Switching
The treatment sequence of RA medications was examined both by mechanism of action and individual drugs, identifying the sequence of therapy utilization. This approach informed the frequency of ‘TNFi cycling’ (i.e., switching from one TNFi medication to another) and JAKi cycling. The time to switch from one biologic/JAKi to the next was estimated using Kaplan Meier (KM) curves, overall and stratified by LoT. Patients were censored if they lost coverage or if the data contained more than one biologic or JAKi initiated on the same day. To complement the KM results, the rate of adding or switching to a new biologic/JAKi (hereafter described as the ‘switch rate’) was estimated per 100 patient years (py), stratified by LoT and separately, by prescribing provider. Subcutaneous and intravenous biologic and JAKi use were identified using various coding systems including the HCPCS, national drug codes (NDC), and generic/brand name. Biosimilar medications were treated as equivalent to their corresponding reference (bio-originator) products for the purposes of analysis. Co-medications commonly given to RA patients (e.g., glucocorticoids, NSAIDs, opioids) were identified in a similar fashion.
Prescriber Clustering
The individual prescribing provider was examined as a potential factor influencing the rate of RA treatment switching. Prescribers were identified using National Provider Identification (NPI) numbers identified for each index RA b/tsDMARDs claim. Patients were assigned to the prescribing provider (available in the raw data) at the time of each RA medication initiation, regardless of whether that provider continued to refill the medication. In the subgroup analysis that reported treatment switch rates by provider, we required providers have ≥10 eligible RA patients in the dataset to provide stable estimates.
Statistical Analysis
Descriptive statistics were used to summarize patient characteristics, both at the time of initial bDMARD and JAKi initiation, as well as at the time of each subsequent switch. Age and all other covariates were re-computed at each switch date. Time to switch by LoT was compared using Wilcoxon rank sum test, with 95% CI for p values comparing line of therapy reported from Cox proportional hazards models that accounted for the clustering of treatment episodes within patients (using a robust sandwich estimator) and also adjusted for age and sex. Sankey plots were used to describe the sequence of treatments by LoT and between specific RA medications. We also created Sankey plots among the subgroup of patients with ≥3 years of follow-up. A violin plot was used to show the distribution of the switch rates across each prescriber’s practice.
Analyses were conducted using SAS 9.4 (SAS, Cary, NC) and R version 4.3. This study was reviewed and approved by central institutional review boards (Advarra Pro00062070 and Pro00070131).
Results
Cohort characteristics
After applying inclusion and exclusion criteria (Supplemental Table 1) to the health plan data, we identified 110,528 RA patients that initiated cs/b/tsDMARDs. After restricting to those who initiated ≥1 b/tsDMARDs during the study period (34.8%), the sample size was 37,656 unique RA patients initiating 59,557 unique b/tsDMARDs. As shown in Table 1, the cohort was median (IQR) age of 54.3 (45.1, 51.4) years and 73% women. Of those with known race/ethnicity (71.5% of total), 68.5% were White, 11.3% were Black, and 13.7% were Hispanic. The median (IQR) disease duration at the time of first b/tsDMARDs initiation was 0.5 (0.2, 1.2) years, and patients’ previous amount of health plan enrollment was 3.5 (2.3, 3.9) years.
Table 1:
Characteristics of RA patients measured at initiation of the first and each subsequent biologic or tsDMARD
| First Initiation of b/tsDMARD | Second and Subsequent b/tsDMARDs initiations | All Initiations of b/tsDMARD | |
|---|---|---|---|
| N | 37,656 | 21,901 | 59,557 |
| Age, years | 54.3 (45.1, 61.4) | 54.0 (45.4, 60.4) | 54.3 (45.3, 61.1) |
| >=65 years | 5,121 (13.6) | 2,235 (10.2) | 7,461 (12.5) |
| Female | 27,499 (73.0) | 16,812 (76.8) | 44,311 (74.4) |
| Race/ethnicity * | |||
| White | 18,440 (68.5) | 11,172 (71.8) | 29,612 (69.7) |
| Black | 3,030 (11.2) | 1,528 (9.8) | 4,558 (10.7) |
| Hispanic | 3,685 (13.7) | 1,911 (12.3) | 5,596 (13.2) |
| Other (e.g., Asian, Pacific Islander) | 1,767 (6.6) | 954 (6.1) | 2,721 (6.4) |
| Insurance ** | |||
| Commercial | 29,563 (79.0) | 18,015 (82.3) | 47,578 (80.4) |
| Medicare | 4,856 (13.0) | 2,085 (9.6) | 6,907 (11.7) |
| Medicaid | 2,985 (8.0) | 1,782 (8.1) | 4,702 (7.9) |
| Disease duration, years | 0.5 (0.2, 1.2) | 1.4 (0.8, 2.5) | 0.9 (0.4, 1.9) |
| Health plan enrollment prior to first RA diagnosis, years | 2.5 (1.6.3.9) | 2.3 (1.5, 3.5) | 2.4 (1.5, 3.8) |
| Health plan enrollment prior to b/tsDMARDs initiation, years | 3.5 (2.3, 5.1) | 4.3 (3.1, 5.7) | 3.9 (2.6, 5.5) |
| Health plan enrollment after b/tsDMARDs initiation | 1.9 (0.9, 3.4) | 1.9 (0.9, 3.2) | 1.8 (0.8, 3.2) |
| Calendar year of initiation | |||
| 2017 | 2,434 (6.5) | 334 (1.5) | 2,768 (4.6) |
| 2018 | 4,416 (11.7) | 1,547 (7.1) | 5,963 (10.0) |
| 2019 | 5,632 (15.0) | 2,685 (12.3) | 8,317 (14.0) |
| 2020 | 5,451 (14.5) | 3,353 (15.3) | 8,804 (14.8) |
| 2021 | 6,182 (16.4) | 3,863 (17.6) | 10,045 (16.9) |
| 2022 | 6,308 (16.8) | 4,278 (19.5) | 10,586 (17.8) |
| 2023 | 6,180 (16.4) | 4,823 (22.0) | 11,003 (18.5) |
| 2024 | 1,053 (2.8) | 1,018 (4.6) | 2,071 (3.5) |
| Line of therapy *** | |||
| 1 | 37,656 (100) | 37,656 (63.2) | |
| as monotherapy | 13,689 (36.4) | 13,689 (36.4) | |
| 2 | 13,316 (60.8) | 13,316 (22.4) | |
| as monotherapy | 6,058 (45.5) | 6,058 (45.5) | |
| 3 | 5,267 (24.0) | 5,267 (8.8) | |
| as monotherapy | 2,533 (48.1) | 2,533 (48.1) | |
| 4 | 2,064 (9.4) | 2,064 (3.5) | |
| as monotherapy | 1,027 (49.8) | 1,027 (49.8) | |
| 5 | 798 (3.6) | 798 (1.3) | |
| as monotherapy | 389 (48.7) | 389 (48.7) | |
| 6+ | 456 (2.1) | 456 (0.8) | |
| as monotherapy | 218 (47.8) | 218 (47.8) | |
| RA Therapy Initiated | |||
| bDMARDs | 34,183 (90.8) | 16,535 (75.5) | 50,718 (85.2) |
| Abatacept | 1,458 (3.9) | 3,046 (13.9) | 4,504 (7.6) |
| Anakinra | 31 (0.1) | 53 (0.2) | 84 (0.1) |
| Rituximab | 802 (2.1) | 486 (2.2) | 1,288 (2.2) |
| Sarilumab | 72 (0.2) | 605 (2.8) | 677 (1.1) |
| Tocilizumab | 565 (1.5) | 1,826 (8.3) | 2,391 (4.0) |
| Adalimumab | 17,009 (45.2) | 3,329 (15.2) | 20,338 (34.1) |
| Certolizumab pegol | 1,645 (4.4) | 941 (4.3) | 2,586 (4.3) |
| Etanercept | 9,361 (24.9) | 3,781 (17.3) | 13,142 (22.1) |
| Golimumab | 2,007 (5.3) | 1,386 (6.3) | 3,393 (5.7) |
| Infliximab | 1,233 (3.3) | 1,082 (4.9) | 2,315 (3.9) |
| tsDMARDs (JAKi) | 3,473 (9.2) | 5,366 (24.5) | 8,839 (14.8) |
| Baricitinib | 62 (0.2) | 175 (0.8) | 237 (0.4) |
| Tofacitinib | 2,013 (5.3) | 2,371 (10.8) | 4,384 (7.4) |
| Upadacitinib | 1,398 (3.7) | 2,820 (12.9) | 4,218 (7.1) |
| tsDMARDs (by calendar year) †† | |||
| 2017–2018 | 478 (7.0) | 325 (17.3) | 803 (9.2) |
| 2019–2020 | 1,441 (13.0) | 1,694 (28.1) | 3,135 (18.3) |
| 2021–2022 | 1,184 (9.5) | 1,923 (23.6) | 3,107 (15.1) |
| 2023–2024 | 370 (5.1) | 1,424 (24.4) | 1,794 (13.7) |
| Prior csDMARDs † | 54,046 (90.7) | 20,580 (94.0) | 33,466 (88.9) |
| Methotrexate | 45,331 (76.1) | 17,579 (80.3) | 27,752 (73.7) |
| Hydroxychloroquine | 24,026 (40.3) | 10,090 (46.1) | 13,936 (37.0) |
| Sulfasalazine | 9,434 (15.8) | 4,255 (19.4) | 5,179 (13.8) |
| Leflunomide | 12,086 (20.3) | 6,086 (27.8) | 6,000 (15.9) |
| Opioids**** | 10,463 (27.8) | 6,639 (30.3) | 17,102 (28.7) |
| Oral Glucocorticoids**** | 25,654 (68.1) | 14,467 (66.1) | 40,121 (67.4) |
bDMARD = biologic disease-modifying anti-rheumatic drug; tsDMARDs = targeted synthetic DMARD (i.e., JAKi); JAKi = Janus kinase inhibitor
Data shown are n(%) or median (25th, 75th) unless specified
among the 72% with available race/ethnicity
among the 99% with available insurance group
Monotherapy is defined as no csDMARD prescription on the day of b/tsDMARD initiation or in the subsequent 90 days
in the 6 months prior to initiation
using all available prior data
denominator is the total number of initiations during each time period
Biologics (90.8%), rather than JAKi (9.2%) were the most common first line advanced RA treatment initiated (90.8%). Most patients (88.9%) had evidence for prior csDMARD use, the majority of which was methotrexate. Patient characteristics and treatments of patients initiating a first b/tsDMARD (Table 1) were generally similar to those initiating a subsequent b/tsDMARD, except that non-TNFi biologics and JAKi were increasingly prescribed as patients advanced to later lines of therapy. The proportion using monotherapy (no concomitant csDMARD use) increased from 36.4% (first line therapy) to almost 50% once patients were on their third, or subsequent, b/tsDMARD. Use of JAKi peaked in 2019–2020 (18.3% of all b/tsDMARD initiations) after which it began to decline (e.g., 13.7% of all initiations in 2023–2024).
Treatment sequencing patterns
Figure 1 describes the pattern of treatment switching between RA medication groups. Not only was TNFi the most common first line therapy (83.1%), but also it was the most common second line therapy (48.0%), as many patients switched from a first to a second TNFi (i.e., TNFi cycling). The next most common second line therapy was JAKi, followed by abatacept. The details of specific medication switching is shown in Figure 2. Adalimumab and etanercept remained the dominant TNFi used as both first and second line advanced therapies. Abatacept was consistently used by a minority of RA patients at each line of therapy. Among the lesser-used IL-6 treatments, tocilizumab was used more frequently as an earlier line of therapy and sarilumab used later. B cell depleting therapies were infrequently used. Results from the subgroup analysis restricting patients with ≥3 years of follow-up (n=11,332) revealed similar patterns (Supplemental Figures 1a and 1b). Among the JAKi, use for all three JAKi peaked in 2020, with a subsequent precipitous drop in tofacitinib use (Figure 3, orange line), whereas upadacitinib used dipped briefly in 2022 but then rebounded and then remained relatively flat (green line). Baricitinib use remained low and flat across all years.
Figure 1:
Treatment Switching among Biologics and JAKi by Mechanism of Action for Lines of Therapy 1–4 (n=37,656 unique RA patients)
Figure 2:
Treatment Switching among Biologics and JAKi for specific RA medication for Lines of Therapy 1–4 (n=37,656 patients)
Figure 3:
Trends in JAKi use over Calendar Time* among Newly Diagnosed RA Patients
*data from 2024 reflects only the first three months
Time to switch and switch rate
Time to switch from one b/tsDMARDs to another is shown in Figure 4a. The overall switch-free survival at 1 year was 70.1% and differed according to LoT. The switch-free survival curve was steeper earlier during treatment, and the overall median time to switch was 38.2 (95% CI 36.9, 39.5) months. The corresponding rate of switch was 25.7/100 py overall and was lower in b/tsDMARDs naïve patients (22.8/100py) compared to those with 1, 2, or 3+ prior b/tsDMARDs (switch rate: 31.5, 32.6, and 34.8/100py, respectively). After accounting for the clustering of treatment episodes within patients and adjusting for age and sex, the time to switch was approximately 1.4 – 1.5 times shorter in patients who had previously received 1 or more b/tsDMARDs (Figure 4b, green, blue, and purple curves) compared to patients who were b/tsDMARDs naïve (red curve), p < 0.001 for each, with no other pairwise comparisons significant.
Figure 4:
(A) Time to switch to a new b/tsDMARD among new onset RA patients (n=37,656) and (B) stratified by line of therapy*
Note: dotted lines denote the 50th percentile (median) for time to switch to a new biologic/JAKi
*p value for comparison of biologic naïve patients to patients with exactly 1, 2, or 3+ prior therapies is significant (adjusted p < 0.0001), but no other comparisons between line of therapy were significant. Analysis also adjusted for age, sex, and clustering of treatment episodes within patients.
Variability in switch rate by rheumatology provider
Based on results in the subgroup analysis of prescribers with at least 10 RA patients (n=1,726 providers and n=36,406 initiations, 96.7% of all treatment initiations in the main analysis), the violin plot in Figure 5 shows substantial variability between providers in the rate of switch. As shown in the upper portion of the curve, the providers with the greatest likelihood of starting or switching b/tsDMARDs did so at a rate of approximately 100/100py (i.e., on average, every RA patient in their practice switched once per year). This rate was approximately 3- to 5-fold greater than the switch rate of the much larger number of providers in the lower portion of the plot (20–30/100py). Switch rates for providers in the highest vs. lowest decile were 12.1/100py and 53.8/100py, respectively.
Figure 5:
Variability in the Rate of Biologic and JAKi Treatment Switches, by Prescriber (n=1,726 prescribers with at least 10 RA patients represented in the analysis*; n=36,406 initiations)
Discussion
In this analysis of a contemporary inception cohort of RA patients with new onset disease, we found that the rate of treatment switching of biologics and JAKi varied appreciably according to LoT. Biologic and JAKi-naïve patients were less likely to switch than those who previously failed ≥1 b/tsDMARDs. We also observed substantial variability between providers in the rate of switching, with a more than 4-fold difference in the switch rate between the lowest (10th percentile) and highest (90th percentile) of rheumatology providers. We also observed that despite literature and a conditional recommendation from the ACR suggesting that after failure of one TNFi, switching MOA is likely advisable15, TNFi cycling remains common, even contemporarily. Several explanations may underlie this observation, including requirements from pharmacy formularies or an initial good response to a first TNFi, with drug-specific side effects. We also found JAKi cycling was common. However, use of JAKi fell after 2020, likely affected by new safety results from the Oral Surveillance trial (disclosed in 2021, published in 2022)16,17, and the associated change in JAKi labeling. The frequency of advanced RA treatment changes likely is partly influenced by lack of a diagnostic test that can predict response to specific RA medications or even mechanisms of action, leaving providers and patients to select their initial and subsequent RA therapy empirically, with the goal of finding the most optimal RA medication for each patient using trial-and-error, leading to repeated switching over time.
Assuming that there was an accurate and predictive biomarker test that could guide RA treatment selection for biologics and JAKi,19,20the number of biomarker tests to predict treatment response that would be required to manage a population of RA patients can be estimated. Based on the observed RA treatment switching rate of 25.7/100py applied to an inception cohort of biologic/JAKi users like the one we assembled, if patients were tested each time they initiated a new RA treatment, a total of 257 RA tests per 1000 RA patients would be ordered annually. In a more established RA cohort with prevalent disease, the rate of treatment switching is greater, and thus the corresponding number of annual tests ordered would be somewhat higher. These estimates invoke the assumption that the biomarker test results change over time and that patients should be re-tested at the time of each new treatment switch. At the other extreme, and invoking the maximally conservative assumption that RA patients only ever needed to be assayed once by a diagnostic test that could predict the efficacy of all RA treatments over a patient’s lifetime, then the number of tests ordered reflects the total number of prevalent RA patients in the U.S. In the U.S., there are approximately 1.39 million RA patients (using an estimated prevalence of 0.54%18) plus the annual incidence of newly diagnosed RA (approximately 13 new RA cases per 100,000 individuals20, which given the U.S. population size of 340 million people, yields between 4,000 and 5,000 newly diagnosed RA patients per year). As another consideration, it might be reasonable to consider testing only patients who have an inadequate response to methotrexate (about 70% of all RA), although predictive biomarker testing for MTX response might be envisioned if it could be made sufficiently cost effective. In terms of which RA medications might be most useful to predict response/non-response based on their frequency of use, TNFi, JAKi and abatacept would seem to reflect the greatest opportunities given that these are more commonly used RA treatments in newly diagnosed patients. However, their prevalence of use must be balanced against the accuracy of prediction for lesser-used RA treatments.
Strengths of our study include a large, contemporary cohort with data current through 2024 and application of rigorous methods to create an inception cohort of RA patients with new onset disease. While we recognize the potential for misclassification of new onset RA, there was a median of 2.5 years of claims data prior to the first RA diagnosis, increasing the certainty of identifying patients with new onset RA22. Additionally, we included patients with a diversity of health plans and insurance types including commercial and government insurance, increasing the generalizability of our results.
Nevertheless, our results must be contextualized considering our study design. We recognize the potential for misclassification of line of therapy in EHR and claims data. However, we conducted a separately reported validation study (manuscript under review) in the Excellence Network in Rheumatology to Innovate Care and High-impact research (ENRICH) practice based network where patients had claims data linked to EHR and patients’ self-reported lifetime use of RA therapies collected on an in-office table app23. This validation study showed excellent agreement between claims and EHR and self-reported data as to line of therapy (crude agreement 84.5%). As another consideration that may impact our results, we only had a median of 1.9 years of follow-up after the first b/tsDMARDs during which to ascertain rates of subsequent treatment switching and LoT, and there were relatively few switches with LoT ≥5 (n=1245, 2.1% of total). This led us to combine lines of therapy 3 and above and report the switch rate in this combined group; the switch rate for people having failed more RA treatments may be even higher than 35/100py. Finally, we examined RA treatment addition and switching irrespective of the reason for changing therapies. Howe er, we do not expect the reason for changing treatments to impact the potential utility of a predictive biomarker to inform which advanced therapy might next be optimal.
In conclusion, results from this analysis of line of therapy, switching rates of biologics and JAKi, and treatments patterns in this population-based inception cohort of RA patients are likely to be useful to a diverse audience of payers, policymakers, and developers of diagnostics and pharmaceuticals. Our findings provide a contemporary look at the sequence of treatments used in RA patients and may help guide development of new therapies and diagnostic biomarkers to predict treatment response. More specifically, these results may help estimate both the market opportunity for new treatments and diagnostic tests to predict therapy response as well as to inform the potential impact for payers on coverage policies by estimating the size of the potential RA population and the potential health care utilization and financial impact of test coverage.
Supplementary Material
Significance and Innovation.
Improving personalized therapy through development of biomarkers to predict treatment response to rheumatoid arthritis (RA) therapies is of major interest, but the value of such biomarkers depends on the frequency of treatment switching, and performance may be impacted by a patient’s treatment history. Using population-based data 2016–2024, we assembled an inception cohort of RA patients with new onset disease to assess the rates of RA treatment switching overall, by line of therapy (LoT), and between provider practices.
We observed that the rate of RA treatment switching (25.7/100 person-years [py] overall) was lowest in biologic-naïve patients (22.8/100py) and increased (>30/100py) as RA patients tried and failed additional LoT. There was dramatic variability between prescribers of RA medications in the rate at which they switched their patients’ RA medications.
Cycling from one TNFi to another, as well cycling from one JAKi to another, was relatively common. JAKi and abatacept use was the most common therapy used after TNFi and thus may be the most fruitful opportunities for predictive biomarkers. As anticipated given recently-recognized safety concerns, JAKi use decreased appreciably after 2020.
Acknowledgements:
This study was supported by Aqtual. Partial support for the infrastructure of the Excellence Network in RheumatoloGY to Innovate Care and High-impact research (ENRICH) practice-based network was provided by NIH P30AR072583. Some coauthors receive support from NIH (P30AR072583). Publication was not dependent on approval by the funding agency.
References
- 1.Meehan RT, Amigues IA, Knight V. Precision medicine for rheumatoid arthritis: the right drug for the right patient—companion diagnostics. Diagnostics (Basel). 2021;11(8):1362. doi: 10.3390/diagnostics11081362 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Wang Z, Huang J, Xie D, He D, Lu A, Liang C. Toward overcoming treatment failure in rheumatoid arthritis. Front Immunol. 2021;12:755844. doi: 10.3389/fimmu.2021.755844 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Martins A, Oliveira D, Rocha TM, et al. What is the association of depression with clinical response to therapy in patients with psoriatic arthritis treated with biologic disease-modifying antirheumatic drugs? Clin Rheumatol. Nov 14 2023;doi: 10.1007/s10067-023-06806-2 [DOI] [PubMed] [Google Scholar]
- 4.Mokbel A, Movahedi M, Philippopoulos E, Ojani P, Keystone EC. The Proportion of Patients With Rheumatoid Arthritis Achieving ACR20/50/70; Consistent Patterns of a 60/40/20 as Demonstrated by a Systematic Review and Meta-analysis. J Clin Rheumatol. Jun 1 2023;29(4):183–189. doi: 10.1097/rhu.0000000000001945 [DOI] [PubMed] [Google Scholar]
- 5.Felson DT, Anderson JJ, Boers M, et al. The American College of Rheumatology preliminary core set of disease activity measures for rheumatoid arthritis clinical trials. The Committee on Outcome Measures in Rheumatoid Arthritis Clinical Trials. Arthritis Rheum. Jun 1993;36(6):729–40. doi: 10.1002/art.1780360601 [DOI] [PubMed] [Google Scholar]
- 6.Radawski C, Genovese MC, Hauber B, et al. Patient perceptions of unmet medical need in rheumatoid arthritis: a cross-sectional survey in the USA. Rheumatol Ther. Sep 2019;6(3):461–471. doi: 10.1007/s40744-019-00168-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Gavigan K, Nowell WB, Serna MS, Stark JL, Yassine M, Curtis JR. Barriers to treatment optimization and achievement of patients’ goals: perspectives from people living with rheumatoid arthritis enrolled in the ArthritisPower registry. Arthritis Res Ther. Jan 7 2020;22(1):4. doi: 10.1186/s13075-019-2076-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Maniadakis N, Toth E, Schiff M, et al. A targeted literature review examining biologic therapy compliance and persistence in chronic inflammatory diseases to identify the associated unmet needs, driving factors, and consequences. Adv Ther. Sep 2018;35(9):1333–1355. doi: 10.1007/s12325-018-0759-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Mellors T, Withers JB, Ameli A, et al. Clinical validation of a blood-based predictive test for stratification of response to tumor necrosis factor inhibitor therapies in rheumatoid arthritis patients. Network and Systems Medicine. 2020/08/01 2020;3(1):91–104. doi: 10.1089/nsm.2020.0007 [DOI] [Google Scholar]
- 10.Beukelman T, Mudano A, Stewart P, et al. Generating real-world evidence from the Excellence Network in Rheumatology. Pharmacoepidemiol Drug Saf. Dec 2024;33(12):e70067. doi: 10.1002/pds.70067 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Martin-Mola E, Balsa A, García-Vicuna R, et al. Anti-citrullinated peptide antibodies and their value for predicting responses to biologic agents: a review. Rheumatol Int. 2016/08/01 2016;36(8):1043–1063. doi: 10.1007/s00296-016-3506-3 [DOI] [PubMed] [Google Scholar]
- 12.Courvoisier DS, Chatzidionysiou K, Mongin D, et al. The impact of seropositivity on the effectiveness of biologic anti-rheumatic agents: results from a collaboration of 16 registries. Rheumatology (Oxford). Feb 1 2021;60(2):820–828. doi: 10.1093/rheumatology/keaa393 [DOI] [PubMed] [Google Scholar]
- 13.Chung C, Rohan P, Krishnaswami S, McPheeters M. A systematic review of validated methods for identifying patients with rheumatoid arthritis using administrative or claims data. Vaccine. December/30 2013;31S10:K41–K61. doi: 10.1016/j.vaccine.2013.03.075 [DOI] [PubMed] [Google Scholar]
- 14.Curtis JR, Xie F, Zhou H, Salchert D, Yun H. Use of ICD-10 diagnosis codes to identify seropositive and seronegative rheumatoid arthritis when lab results are not available. Arthritis Res Ther. Oct 15 2020;22(1):242. doi: 10.1186/s13075-020-02310-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Fraenkel L, Bathon JM, England BR, et al. 2021 American College of Rheumatology guideline for the treatment of rheumatoid arthritis. Arthritis Care Res (Hoboken). 2021;73(7):924–939. doi: 10.1002/acr.24596 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Ytterberg SR, Bhatt DL, Mikuls TR, et al. Cardiovascular and Cancer Risk with Tofacitinib in Rheumatoid Arthritis. New England Journal of Medicine. 2022-01-27 2022;386(4):316–326. doi: 10.1056/nejmoa2109927 [DOI] [PubMed] [Google Scholar]
- 17.Pfizer. Pfizer shares co-primary endpoint results from post-marketing required safety study of XELJANZ® (tofacitinib) in subjects with rheumatoid arthritis (RA). Pfizer Inc. https://www.pfizer.com/news/press-release/press-release-detail/pfizer-shares-co-primary-endpoint-results-post-marketing [Google Scholar]
- 18.Hunter TM, Boytsov NN, Zhang X, et al. Prevalence of rheumatoid arthritis in the United States adult population in healthcare claims databases, 2004–2014. Rheumatology International 2017 37:9. 2017-April-28;37(9)doi: 10.1007/s00296-017-3726-1 [DOI] [PubMed] [Google Scholar]
- 19.Almutairi KB, Nossent JC, Preen DB, Keen HI, Inderjeeth CA. The Prevalence of Rheumatoid Arthritis: A Systematic Review of Population-based Studies. The Journal of Rheumatology. 2021-05-01 2021;48(5):669–676. doi: 10.3899/jrheum.200367 [DOI] [PubMed] [Google Scholar]
- 20.Cai Y, Zhang J, Liang J, et al. The Burden of Rheumatoid Arthritis: Findings from the 2019 Global Burden of Diseases Study and Forecasts for 2030 by Bayesian Age-Period-Cohort Analysis. J Clin Med. Feb 6 2023;12(4)doi: 10.3390/jcm12041291 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Curtis JR, Xie F, Chen L, et al. The comparative risk of serious infections among rheumatoid arthritis patients starting or switching biological agents - PubMed. Ann Rheum Dis. 2011. Aug;70(8)doi: 10.1136/ard.2010.146365 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Curtis JR, Xie F, Chen L, Greenberg JD, Zhang J. Evaluation of a methodologic approach to define an inception cohort of rheumatoid arthritis patients using administrative data - PubMed. Arthritis Care Res (Hoboken). 2018. Oct;70(10)doi: 10.1002/acr.23533 [DOI] [PubMed] [Google Scholar]
- 23.J C, F X, Y S, P S, A M. Line of Therapy of Biologics and JAK Inhibitors in RA, PsA and AxSpA: Implications for Design and Uptake of New Drugs and Diagnostics. Arthritis Rheumatology. 2023;75(Suppl 9) [Google Scholar]
- 24.Beukelman T, Su Y, Xie F, et al. Using electronic health records and linked claims data to assess new medication use and primary nonadherence in rheumatology patients - PubMed. Arthritis Care Res (Hoboken). 2024. Apr;76(4)doi: 10.1002/acr.25269 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.