Disease Burden, Patient Experiences, and Unmet Needs in Those With Rheumatoid Arthritis Initiating a Third Advanced Therapy: Insights From 20 Years of Real‐World Data

Abstract

Objective

Despite advances in rheumatoid arthritis (RA) treatment, a considerable proportion of patients exhibit refractory disease, prompting the need for a comprehensive understanding of refractory RA. We aimed to analyze the burden and patient experiences associated with initiation of a third biologic or targeted synthetic disease‐modifying antirheumatic drugs (b/tsDMARDs) (BT3‐RA) in a large observational cohort.

Methods

Data were obtained from participants with RA in the FORWARD Databank from 1999 to 2019. Participants, stratified into BT3‐RA and a comparator BT1‐RA (first initiation of a b/tsDMARD) cohorts, were matched based on key demographic and disease‐specific parameters. Demographics, patient‐reported outcomes (PROs), comorbidities, and health care interactions were assessed at initiation of first advanced therapy and at the time of meeting BT3‐RA criteria.

Results

After matching, 1,384 participants were included in the study (692 each for BT3‐RA and BT1‐RA). The BT3‐RA cohort had worse PROs, greater comorbidity burden, and lower health satisfaction than BT1‐RA controls. Those with BT3‐RA had significantly higher odds of having a greater number of rheumatology visits in the previous six months than controls (>4 visits, 0–2 visit reference, odd ratio [OR] 3.8 [95% confidence interval (CI) 2.7–5.4], P < 0.001; 3–4 visits, 0–2 visit reference, OR 1.9 [95% CI 1.5–2.5]; P < 0.001). Those with BT3‐RA also had higher odds of concomitant glucocorticoid use (OR 1.5 [95% CI 1.2–2.0], P < 0.001) and gastrointestinal disorders (OR 1.5 [95% CI 1.1–1.9], P < 0.01).

Conclusion

Exposure to three advanced RA therapies was associated with significant disease burden and unmet health care needs. These findings underscore the importance of well‐defined refractory criteria and the need for further investigation into this RA phenotype to identify targeted treatment strategies and ultimately improve outcomes.

INTRODUCTION

Despite major advances in rheumatoid arthritis (RA) treatment and the improved outcomes that have been associated with the expanding number of disease‐modifying antirheumatic drugs (DMARDs) available, a substantial number of patients are refractory to multiple biologics, highlighting a persistent unmet need in patients with RA. ¹ , ² , ³ , ⁴ The term “refractory RA” (reRA) characterizes a state of persistent inflammation, symptoms, or disease activity despite exposure to multiple advanced therapies, presenting a complex therapeutic challenge. However, the definition of reRA is not standardized and is often used interchangeably with broader terms such as “difficult‐to‐treat RA” (D2T RA) or “treatment‐resistant RA,” leading to a lack of consistency across studies and an absence of universally accepted criteria for its identification. ⁵

SIGNIFICANCE & INNOVATIONS.

• This study highlights the need for a standardized refractory rheumatoid arthritis (RA) definition that incorporates patient‐reported outcomes (PROs) and real‐world data.

• This study leverages two decades of real‐world data to provide a foundational understanding of PROs and RA disease burden in the context of multiple advanced therapy exposures.

• The foundational understanding of PROs and burden in this population may guide ongoing efforts to formally define refractory RA and differentiate it from the broader difficult‐to‐treat RA.

Broader definitions of reRA sometimes include patients whose switch from one treatment to a subsequent line of therapy was driven by factors related to safety, side effects, nonadherence, comorbidities, or “nonmedical” reasons such as availability of other drugs, biosimilars, or costs. ³ , ⁶ More recently, reRA has been considered a more specific subset of these patients, comprising those for whom several advanced therapies have proven ineffective. ⁶ , ⁷ Recently, an EULAR task force formalized the definition of D2T RA, ⁸ which has three criteria, but the specific criteria used to define reRA have been arbitrary, and there is no universally accepted definition. reRA definitions are generally based on a specified number of failed DMARDs and a measure of disease activity. ¹ , ² Depending on the definition used and the cohort studied, the frequency of reRA in the RA population is estimated at 6% to 21%. ¹ , ² , ⁴ The heterogeneous nature of D2T RA and reRA underscores the need for standardized definitions and a thorough understanding of patient burden to facilitate accurate comparisons, at‐risk patient identification, and the development of more targeted interventions. ³ , ⁵

Due to the multifaceted interplay among tolerability, economic factors, comorbidities, varying mechanisms of DMARD inefficacy, and other considerations, personalized approaches to treatment are often required, as is the need to combine these individual cases together to gain a more comprehensive understanding of the factors contributing to reRA. ³ , ⁶ , ⁹ The EULAR task force's endeavor to define D2T RA emphasized the scarcity of data and the need for further research in this area. ¹⁰ , ¹¹ Real‐world data are a valuable resource for this purpose and could inform the pursuit of personalized therapeutic approaches. ¹¹

When definitions of reRA include measures of disease activity, typically Disease Activity Scores in 28 joints (DAS28) or EULAR criteria for poor responders are used, and a recent systematic review of reRA definitions indicated that only 5% of reRA studies include patient‐reported outcomes (PROs) in the definition of reRA. ⁵ Outcome data are limited in this area ¹ and capturing a variety of measures beyond disease activity, such as fatigue and other patient‐relevant outcomes, has been a challenge. ² This is despite the fact that even among patients who achieve low disease activity with treatment, quality of life measures are often more difficult to improve, and patients continue to experience persistent pain and fatigue, complicated by both physical and mental health comorbidities. ⁵ Due to differences in perceptions of treatment efficacy and achievement of treatment targets between patients and medical professionals, patient experiences need to be incorporated into reRA definitions in order to consider and address their unmet needs. ⁵

Calls to develop a standardized definition of reRA have been made, highlighting the importance of including additional factors besides exposure to a specified number of DMARDs, such as PROs and other real‐world data. ⁵ Without such a systematic approach to identify and evaluate refractory disease, the impact and mechanisms cannot be determined. In order to better understand reRA and ultimately improve therapies tailored to individuals, a better understanding of PROs and other factors associated with reRA is necessary. ¹ , ² For these reasons, there is a need for more thorough analyses of the burden of RA among those with multiple advanced therapy exposures in a real‐world setting, as well as an exploration of associated baseline characteristics, PROs, comorbidities, and health care interactions.

By employing a multifaceted characterization of individuals with RA and exposure to multiple advanced therapies with multiple mechanisms of action, our investigation aimed to shed light on the distinct burden and patient experiences in this population compared to individuals with persistent use of a single advanced therapy. Expanding our baseline knowledge of PROs in this area, including the magnitude of the differences as well as identifying the highest impact measures, may provide insight as to how these measures can be incorporated into a formal definition of reRA in the future. The objective of this study was to look at the burden and patient experience of those adults with RA initiating their third biologic with at least two different mechanisms of action (BT3‐RA) in comparison to those initiating their first biologic (BT1‐RA). The BT3‐RA is the first of three criteria for EULAR's D2T RA definition and does not require clinical examination, making it an ideal candidate for study by leveraging over two decades of real‐world data.

METHODS

Study design

This study was approved by the Via Christi Hospitals Wichita, Inc. Institutional Review Board (IRB00001674), and all participants provided signed informed consent. Data for this retrospective observational study were provided by adults with RA enrolled in FORWARD, The National Databank for Rheumatic Diseases. ¹² FORWARD is a patient‐centered, multidisease registry, and one of the largest and most comprehensive rheumatology data collection efforts in the world. Starting in 1998, FORWARD data are obtained initially from patients and validated from hospital and physician sources and from national death records. Participants are surveyed by questionnaire at six‐month intervals on a broad range of topics including health care use, medications, comorbidities, and PROs.

Study population

The study population included FORWARD participants with a primary diagnosis of RA with observations between 1999 and 2019. The 1999 start was due to biologics first being available in the fall of 1998. The 2019 cutoff was applied to address the strong potential for confounding due to substantial disruptions and medication changes as a result of the COVID‐19 pandemic that are known to have occurred in this population. ¹³ , ¹⁴

Participants were included if they had no history of biologic DMARD (bDMARD) or targeted synthetic DMARD (tsDMARD) use at study entry and had subsequent exposure to one or more of those advanced therapies. Observations before initiation of the first advanced therapy were excluded. Participants were included in the two comparison cohorts based on the number of bDMARD or tsDMARD exposures and their duration. The b/tsDMARD three‐exposure RA (BT3‐RA) cohort included participants with exposure to three or more advanced therapies during observation, with at least one of those therapies being a tumor necrosis factor inhibitor (TNFi) and at least one being a non‐TNFi (nTNFi) biologic or a JAK inhibitor (JAKi). Due to the current lack of knowledge around PROs in the reRA population, the BT3‐RA definition for this study was based only on advanced therapy exposures to help gain a foundational understanding of PROs and other real‐world experiences in this cohort that may inform a future standardized reRA definition that includes appropriate PROs. Use of three or more advanced therapies was selected rather than two (as in EULAR's D2T definition) as a more stringent requirement to improve the likelihood of including refractory individuals despite not using a PRO for a disease activity cutoff. The comparison group included participants with exposure to only one or two advanced therapies during observation, with continued use of the first therapy for at least two years. For those with exposure to a second therapy, observations from the initiation of that therapy onward were excluded. This control cohort was called b/tsDMARD one‐exposure RA (BT1‐RA).

Missing questionnaires or gaps in observation were allowed. When gaps occurred, if upon a participant's return, they reported the same treatment, it was considered to be an ongoing use of the same therapy even during the time of the missed questionnaire. When a new treatment was reported upon a participant's return, it was considered a change in treatment at that time (at the earliest point the change was reported).

Participants in each of the two groups were matched 1:1 without replacement using a stratified matching approach by age (10‐year intervals), sex, RA duration (fewer than 2 years, 2–4 years, 5–9 years, 10–19 years, 20 or more years), calendar year at baseline (pre‐2005, 2005–2009, 2010–2014, and 2015–2019), and time to meeting BT3‐RA criteria or follow‐up time (for the BT1‐RA group). That is, if it took three years for a participant to meet BT3‐RA criteria, their matched BT1‐RA control was required to have persistent use of their initial advanced therapy for at least three years. Following matching, the total sample size in each group was 692, for a total of 1,384 included participants (Figure 1).

Figure 1.

Study cohort sample size flow chart. BT1‐RA, first biologic or targeted synthetic disease‐modifying antirheumatic drugs; BT3‐RA, third biologic or targeted synthetic disease‐modifying antirheumatic drugs; RA, rheumatoid arthritis.

Variables

Demographics, concomitant medications, PROs, comorbidities, and health care interactions were assessed at two time points: baseline (initiation of first biologic) and follow‐up (for BT3‐RA, the observation at which the individual first meets BT3‐RA criteria; for controls, the matched observation time of their BT3‐RA counterpart). Demographic variables included age, sex, race, education level, rural or urban residence, history of smoking, body mass index (BMI), and disease duration. Race and ethnicity were determined by self‐report from a fixed set of categories (White [not of Hispanic origin], Black [not of Hispanic origin], Asian or Pacific Islander, American Indian or Alaskan Native, Hispanic, and others). Concomitant medications (taken in the prior six months) included glucocorticoids, conventional synthetic DMARD (csDMARD) use, nonsteroidal anti‐inflammatory drug (NSAID) use, and opioid use. PROs included pain visual analog scale (VAS), fatigue VAS, patient global VAS, Health Assessment Questionnaire II (HAQ‐II), and Patient Activity Scale II (PAS‐II). Comorbidities included a history (ever) of pulmonary disorder, cardiac disorder, fracture, depression, diabetes, cancer, and gastrointestinal (GI) disorder. Rheumatic Disease Comorbidity Index (RDCI), ¹⁵ polysymptomatic distress (PSD), and fibromyalgia criteria were also assessed. ¹⁶ Health care interaction variables were based on the prior six‐month time period and included the number of rheumatology visits, family medicine visits, gastroenterology visits, other specialist visits, emergency department visits, hospitalizations, hospitalizations for infections, and a health satisfaction Likert scale (0–4).

Initial advanced therapies were described for both the BT3‐RA and BT1‐RA cohorts. For the BT3‐RA cohort, medication trajectories leading to refractory status were assessed by advanced therapy category (TNFi, nTNFi, and JAKi). Those with a baseline observation before the year 2005 were excluded from the drug trajectory analysis due to almost exclusive TNFi use during that period (little use of nTNFi and no available JAKi).

Statistical analysis

Stata version 14.2 and version 17.0 were used for analyses. Descriptive statistics were calculated for the BT3‐RA cohort and for BT1‐RA controls at baseline and at follow‐up. For univariable analyses comparing demographics, medications, PROs, comorbidities, and health care interactions, significance was assessed with chi‐square and t‐tests, as appropriate.

Least Absolute Shrinkage and Selection Operator (LASSO) models were used to select covariates for multivariable analyses. LASSO is a model selection methodology that estimates regression coefficients by maximizing the log‐likelihood function with a constraint imposed on the sum of the absolute values of all regression coefficients. The constraint was estimated via cross‐validation. Consequential covariates identified in LASSO models were included in multivariable logistic regression models at baseline (to identify factors associated with future incidence of BT3‐RA) and at follow‐up (to identify factors associated with BT3‐RA), with each model limited to the covariates measured at the associated time point. The selection of the sink model included all variables listed in the variable section, including PAS‐II and its three individual composite variables. However, to assess overfitting of the model, sensitivity analyses were conducted, including only the three individual components (pain, patient global, and HAQ‐II) at the two time points. A final sensitivity analysis was conducted in which the models were rerun excluding the health satisfaction variable, as this variable may be influenced by prior treatments or experience and may introduce reverse causality.

RESULTS

Sample size

The FORWARD Databank included 37,607 individuals with a primary rheumatic diagnosis of RA. Of the 6,575 participants with RA who had no advanced therapy exposure at study entry but had subsequent exposure during observation, 718 (10.9%) met BT3‐RA criteria. Of those, 692 were matched 1:1 to BT1‐RA controls (from a pool of 3,555), for a total of 1,384 participants included in the study (Figure 1). Prematching characteristics are presented in Supplementary Table 1, and characteristics of the 26 unmatched individuals are presented in Supplementary Table 2. Before matching, the BT3‐RA cohort was younger (56.1 vs 60.4 years; P < 0.001), more likely to be female (87.2% vs 80.7%; P < 0.001), and had higher BMI (28.5 vs 27.8; P = 0.01).

Univariable analyses

Characteristics of the BT3‐RA and BT1‐RA cohorts at baseline and at follow‐up are presented in Table 1. The matching parameters produced comparison groups with no statistically significant demographic differences. The mean age at baseline was 57 years, and the study population was 88% female and 92% White. Median time since RA symptom onset was 12 years at baseline. Median time from baseline to follow‐up was six years.

Table 1.

Characteristics of FORWARD participants at baseline (initiation of first advanced therapy) and follow‐up (the point of meeting BT3‐RA criteria; matched time point for BT1‐RA controls) by BT3‐RA status^*

	Baseline			Follow‐up
	BT1‐RA	BT3‐RA	P	BT1‐RA	BT3‐RA	P
Demographics
Age, mean (SD), y	56.7 (11.4)	56.4 (10.9)	0.57	62.7 (12.0)	62.3 (11.5)	0.55
Female, %	88.2	88.2	1	88.2	88.2	1
White, %	92.0	91.2	0.60	92.0	91.2	0.60
Education, mean (SD), y	13.9 (2.4)	14.1 (2.2)	0.08	13.9 (2.4)	14.1 (2.2)	0.08
Rural residence, %	23.1	26.6	0.13	23.1	26.6	0.13
Hx of smoking, %	49.7	50.3	0.83	51.9	53.3	0.59
BMI, mean (SD)	28.0 (7.2)	28.4 (6.4)	0.30	28.1 (7.4)	28.8 (6.7)	0.08
Duration, mean (SD), y	14.5 (10.3)	14.4 (10.7)	0.87	20.5 (11.3)	20.4 (11.6)	0.87
Concomitant medications, %
Glucocorticoid	40.1	46.1	0.03	33.0	47.0	<0.001
csDMARD	84.6	85.7	0.58	70.8	71.8	0.68
NSAID	64.8	61.6	0.21	43.3	43.8	0.85
Weak opioid a	22.0	28.2	<0.01	25.8	32.2	0.01
Strong opioid b	3.8	5.6	0.10	5.9	9.4	0.02
Patient‐reported outcomes, mean (SD)
Pain VAS, 0–10	3.5 (2.7)	4.3 (2.8)	<0.001	3.7 (2.8)	4.5 (2.6)	<0.001
Fatigue VAS, 0–10	4.2 (2.9)	4.8 (3.0)	<0.001	4.5 (3.1)	5.1 (2.9)	<0.001
Patient global VAS, 0–10	3.4 (2.5)	4.0 (2.5)	<0.001	3.7 (2.5)	4.4 (2.3)	<0.001
HAQ‐II, 0–3	1.0 (0.7)	1.1 (0.7)	<0.01	1.0 (0.8)	1.2 (0.7)	<0.001
PAS‐II, 0–10	3.4 (2.2)	4.0 (2.1)	<0.001	3.6 (2.3)	4.3 (2.0)	<0.001
Comorbidities
RDCI (0–9), mean (SD)	1.4 (1.4)	1.7 (1.5)	<0.01	1.9 (1.6)	2.1 (1.7)	<0.01
Hx of pulmonary disorder, %	28.3	26.4	0.45	43.2	45.1	0.49
Hx of cardiac disorder, %	17.3	19.4	0.32	32.2	32.7	0.83
Hx of fracture, %	16.2	16.8	0.79	27.5	31.6	0.01
Hx of depression, %	38.8	41.1	0.37	52.2	55.1	0.28
Hx of diabetes, %	9.5	10.4	0.61	18.7	18.4	0.87
Hx of cancer, %	16.2	14.5	0.36	29.9	25.1	0.05
Hx of GI disorder, %	50.5	57.9	<0.01	66.8	74.4	<0.01
PSD (0–31), mean (SD)	10.3 (7.2)	11.8 (7.5)	<0.001	10.7 (7.7)	12.6 (7.7)	<0.001
Fibromyalgia criteria, %	22.5	28.2	0.02	23.2	32.9	<0.001
Health care interactions c
Rheumatology visits, mean (SD)	3.4 (1.9)	3.6 (2.0)	0.13	2.5 (1.7)	3.4 (2.0)	<0.001
Family medicine visits, mean (SD)	2.3 (1.9)	2.3 (1.7)	0.56	2.2 (1.8)	2.3 (1.8)	0.40
Gastroenterology visits, mean (SD)	0.8 (1.3)	0.8 (1.0)	0.83	0.6 (1.1)	0.7 (1.0)	0.21
Other specialist visits, mean (SD)	2.1 (2.3)	2.0 (2.2)	0.53	2.2 (2.2)	2.1 (2.1)	0.44
ED visits, mean (SD)	0.7 (0.8)	0.7 (0.9)	0.62	0.6 (0.7)	0.7 (0.8)	0.55
Any hospitalization, %	10.0	11.3	0.44	10.0	13.3	0.06
Hospitalized for infection, %	2.7	1.7	0.20	2.5	3.3	0.34
Health satisfaction (0–4), mean (SD)	2.4 (1.2)	2.0 (1.3)	<0.001	2.3 (1.2)	1.9 (1.2)	<0.001

^*Status is based on treatment exposures at follow‐up. Hx of comorbidities indicates any previous reported occurrence (ie, “ever”). BMI, body mass index; BT1‐RA, first biologic or targeted synthetic disease‐modifying antirheumatic drugs; BT3‐RA, third biologic or targeted synthetic disease‐modifying antirheumatic drugs; csDMARD, conventional synthetic disease‐modifying antirheumatic drug; ED, emergency department; GI, gastrointestinal; HAQ‐II, Health Assessment Questionnaire II; Hx, history; NSAID, nonsteroidal anti‐inflammatory drug; PAS‐II, Patient Activity Scale II; PSD, polysymptomatic distress; RDCI, Rheumatic Disease Comorbidity Index; VAS, visual analog scale.

^a Codeine, tramadol, hydrocodone.

^b Morphine, fentanyl, methadone, hydromorphone, oxycodone, oxymorphone.

^c Health care interactions and concomitant medications were in the six‐month time period preceding the observation.

In univariable comparisons, all assessed PROs (pain, fatigue, global, HAQ‐II, and PAS‐II) at both time points were significantly worse among the BT3‐RA group. Mean differences ranged from 0.6 to 0.8 higher among the BT3‐RA cohort for the PROs on a 0 to 10 VAS. PROs in both groups worsened from baseline to follow‐up, but the mean differences between BT3‐RA and BT1‐RA remained consistent over time.

Glucocorticoid use, opioid use, RDCI, GI disorder, and PSD (and consequently, meeting fibromyalgia criteria) were also all higher for the BT3‐RA group at both time points, and health satisfaction was lower. Glucocorticoid use appeared to remain consistent over time for the BT3‐RA group (46% to 47%) and appeared to decrease in the control group (40% to 33%). The history of GI disorders increased over time for both groups, with a higher prevalence in the BT3‐RA group at both time points (51% and 58% at baseline, 67% and 74% at follow‐up). The history of fracture did not differ between the groups at baseline but was significantly higher in the BT3‐RA cohort at follow‐up (32% vs 28%). Health satisfaction appeared to be stable over time and consistently lower in the BT3‐RA group (on a 0–4 Likert scale, 2.4 and 2.0 at baseline, 2.3 and 1.9 at follow‐up). There was no significant difference between the BT3‐RA group and controls in the number of rheumatology visits in the previous six months at baseline; however, the number of visits at follow‐up was significantly higher in the BT3‐RA cohort (3.4 visits on average in the previous six months, compared to 2.5 visits in controls; P < 0.001). There were no significant differences in other health care visit types between the two groups at either time point.

Drug trajectories

The first advanced therapy received by participants in the BT3‐RA and BT1‐RA groups is presented in Table 2. The most common first advanced therapy in both groups was a TNFi (88% for BT3‐RA and 91% for controls), usually etanercept, infliximab, or adalimumab. An nTNFi was the first advanced therapy for 11% of BT3‐RA and 8% of controls, whereas a relatively small proportion (1% of each cohort) started on a JAKi.

Table 2.

Initial advanced therapy by BT3‐RA or BT1‐RA status^*

Medication (category)	BT1‐RA	BT3‐RA
Etanercept (TNFi), n (%)	265 (38.3)	297 (42.9)
Infliximab (TNFi), n (%)	237 (34.3)	192 (27.8)
Adalimumab (TNFi), n (%)	113 (16.3)	97 (14.0)
Abatacept (nTNFi), n (%)	30 (4.3)	33 (4.8)
Rituximab (nTNFi), n (%)	16 (2.3)	8 (1.2)
Tofacitinib (JAKi), n (%)	9 (1.3)	7 (1.0)
Tocilizumab (nTNFi), n (%)	7 (1.0)	10 (1.5)
Golimumab (TNFi), n (%)	7 (1.0)	8 (1.2)
Certolizumab (TNFi), n (%)	6 (0.9)	15 (2.2)
Anakinra (nTNFi), n (%)	2 (0.3)	25 (3.6)

^*BT1‐RA, first biologic or targeted synthetic disease‐modifying antirheumatic drugs; BT3‐RA, third biologic or targeted synthetic disease‐modifying antirheumatic drugs; JAKi, JAK inhibitor; nTNFi, non‐TNFi; TNFi, tumor necrosis factor inhibitor.

For the BT3‐RA cohort, drug trajectories leading to refractory status are presented in Figure 2. The most frequently occurring pattern was TNFi➔TNFi➔nTNFi (27%). The second and third most common trajectories were TNFi➔nTNFi➔nTNFi (12%) and TNFi➔TNFi➔JAKi (11%), respectively. Data on self‐reported reason for discontinuation were very limited due to a high degree of missingness (Supplementary Table 3). Summary statistics on treatment duration for each therapy are presented in Supplementary Table 4.

Figure 2.

Treatment trajectories to BT3‐RA (2005–2019). BT1‐RA, first biologic or targeted synthetic disease‐modifying antirheumatic drugs; BT3‐RA, third biologic or targeted synthetic disease‐modifying antirheumatic drugs; JAKi, JAK inhibitor; nTNFi, non‐TNFi biologic disease‐modifying antirheumatic drug; TNFi, tumor necrosis factor inhibitor.

Multivariable analyses

Based on the LASSO results, the following covariates were selected for inclusion in the baseline multivariable model, ranked from most consequential: health satisfaction (Likert scale, 0–4), pain, GI disorder (ever), education (years, continuous), glucocorticoid use (prior six months), hospitalized for infection (prior six months), pulmonary disorder (ever), rural or urban residence, csDMARD use (prior six months), cancer (ever), NSAID use (prior six months), depression (ever), rheumatology visits (prior six months), RDCI, opioid use (prior six months), and White race. The following were selected for the follow‐up model: rheumatology visits (prior six months), glucocorticoid use (prior six months), GI disorder (ever), PAS‐II, education (years, continuous), cancer (ever), health satisfaction (Likert scale, 0–4), BMI, and rural or urban residence.

Although some of these covariates varied significantly in univariable analyses, many differences were attenuated in the multivariable models (Figure 3). Covariates significantly associated with future incidence of BT3‐RA in the baseline model included health satisfaction (odds ratio [OR] 0.8 [95% confidence interval (CI) 0.7–0.9]; P < 0.01) and years of education (OR 1.1 [95% CI 1.0–1.1]; P = 0.02). In the model at the follow‐up time point, more than four rheumatology visits in the previous six‐month time period (0–2 reference; OR 3.8 [95% CI 2.7–5.4]; P < 0.001), three to four rheumatology visits (0–2 reference; OR 1.9 [95% CI 1.5–2.5]; P < 0.001), concomitant glucocorticoid use (OR 1.5 [95% CI 1.2–2.0]; P < 0.001), GI disorder (OR 1.5 [95% CI 1.1–1.9]; P < 0.01), and years of education (OR 1.1 [95% CI 1.0–1.1]; P < 0.01) were all associated with BT3‐RA. History of cancer was associated with BT1‐RA status (OR 0.8 [95% CI 0.6–1.0]; P = 0.03).

Figure 3.

Multivariable logistic regression results for (A) baseline predictors of BT3‐RA and (B) factors associated with BT3‐RA at the time of meeting criteria. Characteristics from Table 1 identified as consequential with LASSO were included in each model. Rheumatology visits (0–2 reference), hospitalization, and concomitant medications were within the six‐month period before the observation. *Statistically significant (P < 0.05) covariates. BMI, body mass index; BT1‐RA, first biologic or targeted synthetic disease‐modifying antirheumatic drugs; BT3‐RA, third biologic or targeted synthetic disease‐modifying antirheumatic drugs; csDMARD, conventional synthetic disease‐modifying antirheumatic drug; GI, gastrointestinal; Hx, history; LASSO, Least Absolute Shrinkage and Selection Operator; NSAID, nonsteroidal anti‐inflammatory drug; PAS‐II, Patient Activity Scale II; RDCI, Rheumatic Disease Comorbidity Index; VAS, visual analog scale.

A sensitivity analysis was performed for both time points, where the search for the best models was repeated using the three components of PAS‐II individually plus the other covariates. The models remained similar in terms of estimated ORs, with the exception that variables White race and RDCI were no longer selected at baseline. At the follow‐up time point, pain scale and global were selected instead of PAS‐II with similar ORs. Finally, similar models were found to be robust when excluding health satisfaction, not indicating spurious or artificial associations (not shown).

DISCUSSION

Despite the availability of advanced therapies for the treatment of RA, a subset of patients develop reRA, and there is a need to better characterize this condition as well as its consequences on the clinical and economic burden of the disease. This study demonstrated that exposure to numerous advanced RA therapies is associated with significant disease burden and unmet health care needs, as evidenced by lower health satisfaction, higher rates of glucocorticoid use, greater comorbidity and symptom burdens, and more rheumatology visits.

The prevalence of BT3‐RA in our cohort is consistent with rates of reRA in prior work in this area. Among FORWARD participants with incident advanced therapy exposure, 11% met study BT3‐RA criteria, compared to an estimated range of 6% to 21% from prior studies on reRA. ¹ , ² , ⁴ Due to inconsistent reRA definitions, the prevalence of reRA varies widely across studies depending on the criteria used and the cohort studied.

Due to the absence of a standardized reRA definition, the limited availability of outcome data in this area, and the importance of incorporating PROs and other real‐world data into improved future definitions of reRA, ¹ , ² , ⁵ , ¹¹ we intentionally excluded any PROs or other measures of disease activity in our BT3‐RA criteria for this study. An ideal standardized definition of reRA may include PROs and/or another disease activity measure, but in this case the exclusion of a disease activity measure in our BT3‐RA cohort allowed for the assessment of differences in PROs without influencing or biasing the results. The creation of arguably our gold standard clinical RA activity measure, the DAS, was based entirely on the likelihood of changing DMARD therapy. ¹⁷ This setup facilitated the improvement of our foundational understanding of these measures in an RA population with exposure to multiple advanced therapies with varied mechanisms of action. With this improved understanding, it can become clearer how to identify the reRA subpopulation more effectively from the broader D2T RA population and how to incorporate PROs into more precise reRA definitions in the future. Additionally, the matching process ensured well‐balanced and comparable cohorts, which was essential for making accurate assessments and drawing meaningful conclusions. Prior studies in this area have identified an association between reRA and female sex as well as younger age, ⁴ , ¹⁸ which was also observed in the BT3‐RA cohort before matching. As a result of our approaches to defining BT3‐RA and to matching, the measures assessed in this study may provide a valuable baseline from which to identify key parameters and improve reRA definitions.

All PROs assessed in this study were found to be significantly worse among the BT3‐RA cohort compared to controls, at both initiation of first advanced therapy and at the point of meeting BT3‐RA criteria. This includes greater pain, fatigue, patient global, physical function, and disease activity, highlighting a substantial impact on disease burden and quality of life in this subpopulation. Due to a high degree of correlation among many of the PROs, LASSO models likely dropped correlated but less consequential measures, leaving few but perhaps the most relevant PROs for the final models. LASSO models identified pain VAS and PAS‐II as particularly consequential when comparing BT3‐RA and BT1‐RA controls, although these covariates were not statistically significant in multivariable regression models. These findings may be the direct result of BT3‐RA status but may also be impacted by adverse effects from medications, notably glucocorticoids and opioids, both of which were taken at significantly higher rates in the BT3‐RA cohort.

Several comorbid conditions were significantly associated with BT3‐RA or BT1‐RA status, including PSD or fibromyalgia, GI disorder, and cancer. Higher RDCI, a measure of comorbidity burden, was also associated with BT3‐RA status; this relationship appears to have been driven primarily by higher prevalence of GI disorders and a higher incidence of fractures. The significantly higher measures of PSD, and subsequently greater prevalence of meeting fibromyalgia criteria, among the BT3‐RA cohort suggests uncontrolled symptoms. Rates of GI disorder were significantly higher among those with BT3‐RA, and GI disorder was associated with BT3‐RA status in multivariable models. Participants with BT3‐RA may have experienced difficulty tolerating medications, including csDMARDs at baseline, as csDMARD use decreased in the BT3‐RA cohort between the two time points despite uncontrolled RA symptoms. Individuals with BT3‐RA also may have developed more GI issues from taking multiple medications, including glucocorticoids and opioids in addition to the numerous advanced RA therapies, and may also be under more severe mental and physical stress due to their BT3‐RA status. Finally, history of cancer was associated with BT1‐RA status; hesitance to increase drug exposure in the context of cancer may contribute to lower odds of reRA, ¹⁹ , ²⁰ and the potential immunosuppressive effects of cancer therapies could influence the remission of RA. ²¹

We also identified several important associations between BT3‐RA and health care interactions, which may impact clinical and economic burdens. The increased number of rheumatology visits in the BT3‐RA cohort is likely driven by uncontrolled symptoms and disease activity and subsequent change in treatments. Despite increased visits, the lower health satisfaction suggests that patients may not perceive their needs as adequately addressed. We also found that education was a predictor of future BT3‐RA status at baseline as well as associated with BT3‐RA status at follow‐up, underscoring potential socioeconomic influences on refractory status. Additionally, rural residence, although nonsignificant itself, remained in all multivariable models prompting consideration of access to care and social determinants of health. ²²

Our study contributes substantially to the understanding of reRA, providing a foundation for understanding and applying PROs and other real‐world data to reRA criteria, aiding in risk assessment, expanding our knowledge of the burden associated with exposure to numerous advanced RA therapies, and guiding future research in this area. Although this study has provided valuable insights, there are limitations to note. First, the longer disease duration at baseline may have influenced outcomes and could make the results less comparable to a more recent inception cohort closer to initiation of first advanced therapy. Second, participation bias could influence the generalizability of the results as individuals who actively engage in observational studies may exhibit distinct characteristics from the broader RA population, though we accounted for this with an internal matched control group. Third, the analysis of PROs in BT3‐RA would be strengthened by the addition of comparisons to objective clinical measures of inflammation and of serology such as the DAS28, but such measures were not available for both cohorts. As the D2T RA EULAR definition could not be used formally, we chose an analogous to DAS28 definition based only on the number of exposed biologics and allowing easier comparison within other real‐world cohorts. Finally, as in all observational studies, there is the potential for residual confounding that results from imperfect disease activity measures, and the numerous measures of health dimensions to help account for this.

In conclusion, these findings provide important information about PROs and patient experiences that can guide ongoing efforts to differentiate reRA from the broader D2T RA. This study underscores the importance of well‐defined reRA criteria and the need for further investigation into this unique RA phenotype to identify targeted treatment strategies and ultimately improve outcomes.

AUTHOR CONTRIBUTIONS

All authors contributed to at least one of the following manuscript preparation roles: conceptualization AND/OR methodology, software, investigation, formal analysis, data curation, visualization, and validation AND drafting or reviewing/editing the final draft. As corresponding author, Dr Michaud confirms that all authors have provided the final approval of the version to be published and takes responsibility for the affirmations regarding article submission (eg, not under consideration by another journal), the integrity of the data presented, and the statements regarding compliance with institutional review board/Declaration of Helsinki requirements.

Supporting information

Disclosure form.

Appendix S1: Supplementary Information.

Data availability statement

Data may be made available upon reasonable request.