Use of a molecular signal response classifier informs treatment selection and improves outcomes in Hispanic patients with rheumatoid arthritis

Objective: In this study, we evaluated the performance of a molecular signal response classifier (MSRC) predicting a signal of inadequate response (SIR) to tumor necrosis factor-α inhibitors (TNFis) in Hispanic patients with rheumatoid arthritis. We explored whether the MSRC informed treatment selection and improved outcomes over standard of care.

Methods: We compared 108 MSRC-tested patients with 206 untested controls. Outcomes included the low clinical disease activity index (CDAI-LDA, <10), minimum clinically important difference (CDAI-MCID), low Routine Assessment of Patient Index Data-3 (RAPID3-LDA), low patient global assessment of disease (PtGA-LDA), and minimal pain visual analog scale score (all ≤2 on a 0–10 scale).

Results: Seventy (64.8%) MSRC-tested patients exhibited an SIR; those received TNFis less frequently than controls [14.3% vs. 57.8%, adjusted odds ratio (aOR) 0.12 (95% confidence interval, CI, 0.06–0.25)]. Patients with an MSRC-aligned treatment selection more frequently reported CDAI-LDA [aOR 9.79 (1.70–56.58)], CDAI-MCID [aOR 6.49 (1.18–35.71)], RAPID3-LDA [aOR 11.55 (1.42–93.81)], PtGA-LDA [aOR 9.74 (1.11–85.64)], and minimal pain [aOR 6.98 (1.14–42.59)] compared with misaligned ones. Compared with controls, MSRC-aligned patients more commonly reported PtGA-LDA [aOR 1.76 (1.00–3.11)] and minimal pain [aOR 1.94 (1.08–3.51)]. Among biologic disease-modifying antirheumatic drug (bDMARD)-naïve participants, MSRC-aligned patients more frequently reported CDAI-LDA [aOR 3.67 (1.31–10.26)], CDAI-MCID [aOR 5.17 (1.05–25.47)], PtGA-LDA [aOR 3.20 (1.25–8.19)], and minimal pain [aOR 2.90 (1.07–7.85)] vs. controls. In TNFi-treated patients, the MSRC predicted SIR by CDAI-LDA with positive predictive value = 80.0% (95% CI 55.5–97.5%), sensitivity = 50% (95% CI 29.9–75.4%), specificity = 86.7% (95% CI 68.1–98.3%), and area under the curve = 0.68 (0.53–0.84).

Conclusions: Most Hispanic patients exhibited an SIR to TNFis. MSRC-aligned therapy improved outcomes vs. the MSRC-misaligned group and untested controls, predominantly among bDMARD-naïve patients.

Introduction

Rheumatoid arthritis (RA) is a chronic autoimmune disease characterized by articular and systemic inflammation, functional limitation, impaired quality of life, and increased multimorbidity. Its rate of global prevalence is estimated at 0.24%, and it imposes a major and evolving burden on modern healthcare systems through high direct medical costs and substantial indirect productivity losses despite therapeutic advances (1, 2). Attainment and maintenance of remission significantly improves outcomes, reduces healthcare costs and utilization, and decreases societal burden (3). For every year that remission is not achieved, RA patients accrue irreversible declines in function (about 0.1 Health Assessment Questionnaire unit/year) and additional chronic comorbidities (approximately 0.2 conditions/year) (2, 4). Tumor necrosis factor-α inhibitors (TNFis) are a cornerstone of RA therapy; in national registries and real-world RA practice, over 75% of inadequate responders to conventional synthetic disease-modifying antirheumatic drugs (DMARDs) receive a TNFi as an initial biologic therapy (5–11). Yet, 30%–80% of TNFi recipients fail to achieve an American College of Rheumatology 50% improvement (ACR50) after 6 months of treatment (6, 12–14). Contemporary guidelines offer no guidance for personalized treatment choices. This results in a trial-and-error prescription of a targeted synthetic DMARD (tsDMARD) or biologic DMARD (bDMARD) until an effective treatment is identified.

A molecular signature response classifier (MSRC) blood test was developed that detects gene expression patterns associated with inadequate response to TNFi therapy (15, 16). The MSRC report yielded a continuous result on a scale from 1 through 25. A numeric score of ≥10.6 had a positive predictive value (PPV) of 87.7% in identifying TNFi inadequate response by ACR50 in mostly non-Hispanic white b/tsDMARD-naïve, and TNFi-experienced patients with moderate to severe disease activity (16). In a recent study, 56.3% of patients with moderate or severe RA activity exhibited a signal of inadequate response (SIR) to TNFis; of those, 75.6% were prescribed therapy in agreement with the MSRC result (17). Patients with an SIR treated with a TNFi had nine times lower likelihood of response by ACR50 or the clinical disease activity index (CDAI) at 6 months. However, when therapies were aligned with MSRC results, responses improved; patients with an SIR to TNFi receiving an alternate b/tsDMARD exhibited up to 1.8-fold greater improvements in CDAI scores than those treated with a TNFi (15, 18, 19).

Hispanics are the fastest growing minority group in the United States (20), yet they remain underrepresented in clinical research. Compared with non-Hispanic whites, Hispanics with RA exhibit different clinical features and worse patient-reported outcomes (21–24), including global assessment of disease activity, physical function, and pain (22–25). The utility of the MSRC in RA has been evaluated predominantly in non-Hispanic whites, raising concerns about generalizability to diverse patient groups. Unique genetic, socioeconomic, and environmental factors may affect RA activity and treatment response in Hispanics, underscoring the importance of evaluating the utility of the MSRC in this group.

The goal of this study was to examine the performance of the MSRC in Hispanic patients with RA. Specific aims were to 1) determine the proportion of tested patients exhibiting an SIR (score of ≥10.6) to TNFi, 2) assess whether MSRC results informed therapeutic selection, and 3) evaluate whether MSRC-aligned treatment associated with improved physician- and patient-reported outcomes compared with MSRC-misaligned therapy or usual care. Exploratory aims were to assess whether MSRC-aligned treatment associated with improved outcomes vs. usual care in b/tsDMARD-naïve patients and evaluate the clinical validity of the MSRC in tested patients receiving TNFi therapy.

Materials and methods

Patient recruitment

The sample included 314 patients from a private rheumatology practice recruited on a first-come, first-served basis between 1 January 2021 and 1 January 2024. The two-location, six-provider practice serves the greater El Paso, Texas metropolitan area. El Paso residents are predominantly Hispanic (82.9%) and 78.2% of residents identify as Mexican-American (26). The decision to recruit patients was made during routine appointments when a rheumatology provider determined that a patient had not reached their treatment target and should initiate or change bDMARD therapy. Inclusion criteria entailed the following: age older than 18 years, fulfillment of the 2010 classification criteria for RA, presence of at least moderate disease activity based on a CDAI > 10 on current conventional synthetic (cs)- or bDMARD therapy, and plans to initiate or change bDMARD or tsDMARD therapy based on patient-provider shared decision-making. Exclusion criteria were overlapping autoimmune syndromes, acute or chronic infections, and history of malignancy.

Overall, 864 patients with RA were assessed over a period of 3 years (Figure 1); of 357 eligible patients, 334 consented to the study and 23 declined to participate. In an unmatched case–control design, 114 patients (cases) underwent MSRC testing and the result indicating the presence or absence of an SIR was shared with both the provider and the patient in advance of shared decision-making about treatment change. A total of 220 untested patients who received standard-of-care treatment change based on patient-provider shared decision-making were concurrently recruited as controls (at a case to control ratio of 1:2). Four MSRC-tested patients and nine controls did not proceed with the intended treatment change and were therefore excluded from the analysis. Two MSRC-tested patients and five controls did not have a follow-up visit. The final analysis included 108 MSRC-tested and 206 control patients (Figure 1). Participants were reevaluated within a mean [95% confidence interval (95% CI)] of 4.5 (4.2–4.7) months after treatment change. All subjects provided written informed consent. This study received approval from the local institutional review board and was carried out in accordance with the Declaration of Helsinki.

Figure 1

Figure 1. Patient enrollment into the study. CDAI, clinical disease activity index; MSRC, molecular signal response classifier.

Data collection

Data recorded at baseline were age, sex, RA duration, anticitrullinated peptide antibody (ACPA) status, and rheumatoid factor positivity. ACPA status was tested at the Scipher Medicine laboratory. Smoking status, diabetes, hypertension, hyperlipidemia, fibromyalgia, anxiety disorder, depressive disorder, and body mass index were also recorded. Using 2022 Census data, area-level socioeconomic status (SES) was estimated on the basis of ZIP code–level median household income (27) and education (proportion of adults ≥25 years who graduated high school) (28). Baseline and past treatment with corticosteroids, conventional synthetic DMARDs, and bDMARDs were documented. Data collected at both baseline and follow-up included tender and swollen joint counts, C-reactive protein, erythrocyte sedimentation rate, and physician global assessment of disease activity. All patients completed a Routine Assessment of Patient Index Data 3 (RAPID3) questionnaire, pain visual analog scale (0–10 scale), and patient global assessment (PtGA) of disease activity (0–10 scale) at baseline and follow-up. To evaluate the effect of therapy aligned with the MSRC result, patients with an SIR receiving TNFi were classified as MSRC-misaligned, while those with an SIR were prescribed a non-TNFi; those with no SIR receiving a TNFi or non-TNFi were deemed MSRC-aligned.

MSRC test description

The MSRC test is performed in peripheral blood and identifies a molecular signature predicting inadequate response to a TNFi by ACR50 at 6 months (15, 16). Details and performance characteristics of the test have been previously described (17). The MSRC combines three clinical characteristics (sex, body mass index, and PtGA), ACPA status, and RNA expression data from 19 genes (15, 16). PAXgene Blood RNA tubes were used for the analysis of gene expression, and RNA sequencing was carried out at Athena Diagnostics (Marlborough, MA, USA) and the Ambry Genetics Corporation (Aliso Viejo, CA, USA). Result processing and interpretation was carried out at Scipher Medicine. A numeric score ≥10.6 (1–25) has a PPV of 87.7% in identifying TNFi inadequate response by ACR50 in mostly non-Hispanic white b/tsDMARD-naïve and TNFi-experienced patients with moderate-to-severe disease activity (16).

Outcome variables

The CDAI (range 0–76) was the sum of tender joint count (0–28), swollen joint count (0–28), PtGA (0–10), and physician global assessment (0–10). A CDAI low disease activity (CDAI-LDA) was defined as a follow-up CDAI <10 and CDAI minimum clinically important difference (CDAI-MCID) denoted a decrease ≥12 if the baseline CDAI >22 or ≥6 if the baseline CDAI >10 to ≤22 (29). The patient-reported RAPID3 was the sum of its three subscales [overall health (0–10), pain (0–10), and physical function (0–10)] divided by 3 (RAPID3 total range 0–10). RAPID3 low disease activity (RAPID3-LDA) was defined as follow-up RAPID3 ≤2 (30). Other main patient-reported outcomes were PtGA low disease activity (PtGA-LDA) defined as follow-up PtGA ≤2 (31) and minimal pain (score ≤2 at follow-up) (31). Secondary outcomes were absolute change from baseline to follow-up in CDAI and RAPID3 subscales (overall health, pain, and physical function).

Statistical analysis

Categorical variables were reported as frequencies and percentages and continuous variables as means and standard deviations (SDs). Logistic regression compared the start of TNFi treatment between patients with an SIR vs. untested controls overall and stratified by bDMARD exposure, as well as between bDMARD exposed vs. bDMARD naïve in patients with an SIR, those without an SIR, and untested controls. MSRC-aligned and MSRC-misaligned groups were compared on main outcomes in logistic regression models adjusted for age, ACPA positivity, fibromyalgia, follow-up duration, and the baseline value outcome. Linear regressions with the same covariates compared MSRC-aligned and MSRC-misaligned groups on absolute change in the CDAI and RAPID3.

Adjusted logistic regression models compared MSRC-aligned and untested controls on main outcomes. Inverse probability weighting was used to account for an imbalance of baseline characteristics between the MSRC-aligned and the untested control groups. The individual propensities for MSRC-aligned treatment were estimated in a logistic regression including covariates of age, sex, ACPA positivity, smoking status, body mass index, methotrexate use, prior TNFi exposure, prior non-TNFi exposure, and baseline swollen and tender joint count, C-reactive protein, patient global assessment, and physician global assessment. Propensity score values were used to calculate stabilized weights. After weighting, all covariates had a standardized mean difference value <0.10, indicating balance between the groups.

Covariates in all main outcome models were age, ACPA positivity, fibromyalgia, follow-up duration, and the baseline value of the outcome (CDAI, RAPID3, pain score, or PtGA). The covariates were chosen a priori based on their hypothesized influence on the main outcomes and treatment response. The covariates differed between the outcome models and the propensity score model because the former included a limited set of prognostic factors for clinical outcomes, while the latter used a wider set of variables to enhance comparability between the MSRC-aligned and the untested control groups. Missing follow-up RAPID3 data for five MSRC-tested patients were imputed with multiple imputation by chain equations with 10 repetitions. Analyses were performed in Stata 15.0, and a two-tailed p value <0.05 was considered statistically significant.

Results

Patients were mostly middle-aged females, self-identified predominately as Hispanic, and had established, seropositive, and moderately to severely active disease (Table 1). Over 80% of control and MSRC-tested groups received methotrexate at baseline and ≥50% of each group were bDMARD-experienced. Thirteen out of 117 (11.1%) TNFi-naïve controls and 2/43 (4.7%) of TNFi-naïve MSRC-tested patients had past exposure to a non-TNFi therapy. Eighty-nine of 102 (87.3%) bDMARD-exposed controls and 65/67 (97.0%) of bDMARD-exposed MSRC-tested patients previously received a TNFi.

Table 1

Table 1. Sample characteristics at baseline.

Seventy of 108 (64.8%) MSRC-tested patients exhibited an SIR; 60/70 (85.7%) subsequently received non-TNFi therapies, whereas 10/70 (14.3%) received TNFis despite the test results (Supplementary Table S1). Among patients with an SIR, TNFi recipients had a lower area-level high school graduation rate (p = 0.04), whereas non-TNFi recipients more commonly exhibited a high baseline CDAI (>22) (p = 0.01). Of 38 patients without an SIR, 21 (55.3%) received TNFis and 17 (44.7%) received non-TNFi therapies (Supplementary Table S1). Among those without an SIR, TNFi recipients exhibited a lower area-level high school graduation rate (p = 0.03); on the other hand, non-TNFi recipients had a higher average baseline CDAI (p = 0.03), a greater prevalence of high CDAI (p = 0.02), higher physician global assessment (p = 0.006) and PtGA (p = 0.007) scores, and greater prior TNFi exposure (p = 0.02). Frequencies of TNFi and non-TNFi therapies selected at baseline across untested controls and MSRC-tested patients according to prior bDMARD exposure and presence of an SIR are summarized in Supplementary Table S2.

MSRC test results inform therapy selection

Figure 2A shows TNFi initiation compared between MSRC-tested patients with an SIR and untested controls. Tested patients with SIR were less likely to be prescribed a TNFi than untested controls overall [14.3% vs. 57.8%, adjusted odds ratio (aOR) 0.12 (95% CI 0.06–0.25), p < 0.001] and particularly among bDMARD-naïve [9.1% vs. 76.9%, aOR 0.03 (95% CI 0.01–0.13), p < 0.001] after covarying for age, ACPA positivity, and fibromyalgia.

Figure 2

Figure 2. TNFi treatment initiation comparisons between groups. (A) MSRC-tested with SIR vs. untested control groups overall and stratified by prior bDMARD exposure, and (B) prior bDMARD-exposed vs. bDMARD-naïve in untested controls, tested with SIR and tested without SIR groups. Adjusted models covary for age, fibromyalgia, and anticitrullinated peptide antibody positivity. MSRC, molecular signal response classifier; SIR, signal of inadequate response; TNFi, tumor necrosis factor-α inhibitors; bDMARD, biologic disease-modifying antirheumatic drugs; 95% CI, 95% confidence interval.

Figure 2B shows the effect of previous bDMARD use on TNFi initiation. Among untested controls and the MSRC-tested without SIR group, those with bDMARD exposure were less likely to receive a TNFi prescription than bDMARD-naïve patients (p ≤ 0.03). In contrast, there was no difference in the limited number of TNFi prescriptions between bDMARD-experienced and bDMARD-naive MSRC-tested patients with an SIR (p = 0.69). Frequencies of TNFi and non-TNFi classes and agents at baseline in untested controls and MSRC-tested patients stratified by prior bDMARD exposure and presence of SIR are shown in Supplementary Table S3.

MSRC-aligned therapy selection improves clinical outcomes

Main outcomes compared between MSRC-aligned and MSRC-misaligned groups are shown in Figure 3. Patients who received MSRC-aligned treatment were more likely to report CDAI-LDA, CDAI-MCID, RAPID3-LDA, minimal pain, and PtGA-LDA than the MSRC-misaligned group (all p ≤ 0.04) after adjusting for age, ACPA positivity, fibromyalgia, follow-up duration, and baseline value of the respective outcome. In linear regressions with the same covariates, compared with MSRC-misaligned patients, those with MSRC-aligned treatment had greater decreases in CDAI and RAPID3 overall health and pain subscales (all p ≤ 0.007) but not in the RAPID3 physical function subscale (Figure 4).

Figure 3

Figure 3. Main outcomes compared between MSRC-aligned and MSRC-misaligned (SIR-TNFi) treatment groups. (A) Unadjusted rates of outcome attainment in MSRC-aligned and -misaligned patients. (B) Adjusted odds ratios for outcome attainment in MSRC-aligned vs. -misaligned patients. Regression models adjusted for age, anticitrullinated peptide antibody positivity, fibromyalgia, follow-up duration, and the baseline value of the outcome. CDAI, clinical disease activity index; LDA, low disease activity; MCID, minimal clinically important difference; RAPID3, Routine Assessment of Patient Index Data 3; PtGA, patient global assessment of disease activity; MSRC, molecular signal response classifier; TNFi, tumor necrosis factor-α inhibitors; SIR, signal of inadequate response; 95% CI, 95% confidence interval.

Figure 4

Figure 4. Secondary outcomes compared between MSRC-aligned and MSRC-misaligned (SIR-TNFi) treatment groups. (A) Adjusted absolute change from baseline to follow-up. (B) Adjusted unstandardized regression coefficients for change in secondary outcomes in MSRC-aligned vs. -misaligned patients. Linear regression models adjusted for age, anticitrullinated peptide antibody positivity, duration of follow-up, fibromyalgia, and the baseline value of the outcome. CDAI, clinical disease activity index; RAPID3, Routine Assessment of Patient Index Data 3; MSRC, molecular signal response classifier; TNFi, tumor necrosis factor-α inhibitors; SIR, signal of inadequate response; 95% CI, 95% confidence interval.

Figure 5A shows the main outcomes compared between the MSRC-aligned group and the untested controls using adjusted logistic regression models with inverse probability weighting. Although no differences were observed on CDAI-LDA, CDAI-MCID, or RAPID3-LDA, MSRC-aligned patients were more likely to report minimal pain (p = 0.03) and PtGA-LDA (p = 0.049) than untested controls in adjusted models. In exploratory analyses limited to bDMARD-naïve patients, the likelihood of achieving CDAI-LDA, CDAI-MCID, minimal pain, and PtGA-LDA was greater in the MSRC-aligned therapy group than in the untested control group (Figure 5B).

Figure 5

Figure 5. Outcomes compared between patients with MSRC-aligned treatment and controls (A) overall and (B) among bDMARD-naïve. Inverse probability weighted logistic regression models additionally adjusted for age, anticitrullinated peptide antibody positivity, fibromyalgia, follow-up duration, and the baseline value of the outcome. MSRC, molecular signal response classifier; Tx, treatment; CDAI, clinical disease activity index; LDA, low disease activity; MCID, minimal clinically important difference; RAPID3, Routine Assessment of Patient Index Data 3; PtGA, patient global assessment of disease activity; 95% CI, 95% confidence interval.

Clinical validity of the MSRC in Hispanic patients with rheumatoid arthritis

The ability of the MSRC to predict TNFi inadequate response by CDAI-LDA was assessed in TNFi-treated patients tested with the MSRC (n = 31). Half of the 16 patients not responding to TNFi also exhibited an SIR. Eighty percent of patients with an SIR receiving a TNFi did not achieve CDAI-LDA [PPV = 80.0% (95% CI 55.5%–97.5%), sensitivity = 50.0% (95% CI 29.9%–75.4%), specificity = 86.7% (95% CI 68.1%–98.3%), and area under the curve (AUC) = 0.68 (95% CI 0.53–0.84)]. Among MSRC-tested patients treated with a TNFi, those with an SIR were less likely to report CDAI-LDA at follow-up than those without an SIR [OR 0.15 (95% CI 0.03–0.91), p = 0.04].

A summary of the results is presented in Figure 6.

Figure 6

Figure 6. Use of an MSRC informs treatment selection and improves outcomes in Hispanic patients with rheumatoid arthritis. SIR, signal of inadequate response; bDMARDs, biologic disease-modifying antirheumatic drugs; Rx, treatment; CDAI-LDA, Clinical Disease Activity Index-low disease activity; CDAI-MCID, Clinical Disease Activity Index-minimally clinically important difference; RAPID3-LDA, Routine Assessment of Patient Index Data-3-low disease activity; PtGA-LDA, Patient Global Assessment-low disease activity, 95% CI, 95% confidence interval.

Discussion

This is the first report on the performance of the MSRC as a personalized medicine tool and its impact on future therapy selection and clinical outcomes compared with standard practice in Hispanic patients with RA. Since Hispanics are underrepresented in clinical trials, there is concern about the generalizability of the potential therapeutic impact of various medications in this population (32–36). Likewise, since predictive biomarkers are generally tested and validated predominantly in non-Hispanic whites, their performance in Hispanic populations is unknown. Our study aimed to address this knowledge gap and unmet need. The majority of patients in our cohort self-identified as Mexican-American, reflective of the ethnic composition of the regional population served (78% Mexican-American) and prevalence of this specific Hispanic origin group in the continental United States (62%) (37).

We found that 64.8% of tested Hispanics with moderate or high RA activity on treatment exhibited an SIR to TNFi. This figure is aligned with the 56.3% rate reported in unselected, non-Hispanic white populations (17) and indicates that the majority of patients with significant ongoing RA activity may not be poised to respond to the preferred first-line bDMARD class. Delays in initiating effective therapy can result in disease progression, loss of physical activity, increased cost, and poor health-related quality of life (38).

Tested patients exhibiting an SIR were far less likely to receive TNFi therapy compared with untested controls, especially if they were bDMARD-naïve. Notably, 60/70 (85.6%) of all and 20/22 (90.9%) of bDMARD-naïve patients with an SIR received non-TNFis; likewise, 98/108 (90.7%) of all and 39/41 (95.1%) of bionaive tested patients received MSRC-aligned therapies. These rates are aligned with reports in a majority of white samples where 75.6% of patients received MSRC-aligned therapies (17). Patients with an SIR receiving TNFi therapy had lower area-level high school graduation rates, indicating lower socioeconomic status. The reasons underlying this finding are unclear: We observed no differences in rates of fibromyalgia and depressive or anxiety disorders that may raise concerns or influence willingness to adhere to new treatments (39). Prior studies have shown that lower educational attainment is associated with decreased health literacy, adherence, and participation in shared decision-making; however, these relationships are modifiable through targeted education and communication support (40–43). Additional considerations may include patient preference, cost/insurance coverage, or potential implicit bias in provider communication.

Among bDMARD-exposed patients without an SIR, half had previously received TNFi treatment and had an inadequate response, leading to fewer TNFi prescriptions compared with bDMARD-naïve patients. A similar pattern was observed in untested controls. Overall, healthcare providers caring for Hispanic patients with RA heavily considered MSRC results when selecting advanced therapies. More importantly, since treatment decisions were made through shared decision-making, Hispanic patients themselves appeared to heed the MSRC results when considering therapeutic options.

Patients who received MSRC-aligned treatment showed higher rates of CDAI-LDA and CDAI-MCID responses and greater decreases in the CDAI compared with those on misaligned treatments. For the first time, superior patient-reported outcomes such as RAPID3-LDA, PtGA-LDA, and minimal pain were reported to a greater extent in MSRC-aligned than in MSRC-misaligned treatment groups. PtGA and pain, which associate with impaired physical function in Hispanics with RA (25), distinguish placebo from active treatment in randomized controlled trials similarly to or better than physician-reported outcomes (44, 45). Our findings are particularly relevant as Hispanics were shown to report worse global assessments, pain, and function compared with non-Hispanic whites and African Americans (22). In the Early Rheumatoid Arthritis Treatment Evaluation Registry cohort, patient-reported outcomes in Hispanic patients were also worse than African Americans, despite no differences in physician-reported outcomes (23).

We also compared outcomes between tested patients receiving MSRC-aligned therapy and untested controls receiving standard care in the entire cohort. No differences in CDAI-LDA, CDAI-MCID, or RAPID3-LDA rates were observed. This may be explained by the rates of subsequent non-TNFi treatment choice in the untested, bDMARD-exposed group. Indeed, 59/59 (100%) of past bDMARD exposures in the MSRC-aligned group and 89/102 (87.3%) of past bDMARD exposures in the control group were TNFis; evidently patients in both these groups experienced inadequate clinical responses to TNFis up to the time of the test. Naturally, medical providers logically and intuitively opted for a non-TNFi option in 63/102 (61.8%) patients in the bDMARD-exposed control group, which was similar to the 52/67 (77.6%) rate of non-TNFi therapy choice in bDMARD-exposed MSRC-tested patients. Remarkably, despite the lack of differences in CDAI-LDA, CDAI-MCID, and RAPID3-LDA, patients with MSRC-aligned therapy still reported higher rates of PtGA-LDA and minimal pain score compared with untested controls. It is plausible that those patient-reported outcomes are better able to discriminate treatment effect (44, 45). Yet, it is also conceivable that greater confidence in an “empiric” rather than “trial and error” therapy selection based on a test poised to predict response may indicate biased patient reporting.

Among tested patients in this Hispanic cohort, the MSRC had a PPV of 80.0%, sensitivity of 50.0%, and specificity of 86.7% for identifying likely TNFi inadequate responders with CDAI-LDA achievement as outcome. These rates are comparable to the PPV, sensitivity, and specificity reported in white patients using ACR50 as an outcome (87.9%, 54.1%, and 69.9%, respectively) (18). This observation may provide additional evidence to support the broader utilization of MSRC testing to stratify patients based on a personalized likelihood of inadequate responses to TNFis (18).

Our study has several strengths. It reports on a sizeable, well-characterized sample of Hispanic RA patients with prospectively collected standardized outcomes. It is the first report on the performance of an MSRC in a sample of patients representative of the extant Hispanic population across the continental United States and confirms that the test operates equally well among Hispanic patients as it does in non-Hispanic whites. It further corroborates that providers caring for Hispanic patients seriously contemplate the test results while considering treatment options, but more importantly, that patients widely adopt them during shared decision-making with their treating provider. It also constitutes the first study with sufficient power to evaluate the impact of MSRC-driven therapy selection on patient-reported outcomes, reflective of the patient experience and health-related quality of life, beyond a series of well-accepted physician measures.

Several limitations of our study should be acknowledged. The small number of patients with MSRC-misaligned therapies resulted in wide confidence intervals in comparative analyses and may have constrained the MSRC validity subanalysis, confounding its performance characteristics such as sensitivity, specificity, and accuracy (AUC). The lack of randomization may introduce selection bias despite the use of inverse probability weighting to balance baseline group differences. The high seropositivity rates may limit the generalizability of our findings to patients with lower seropositivity rates. We lacked formal data on SES, literacy, and acculturation, which could influence disease coping and patient-reported outcomes. To address this, we considered ZIP code–level median household income and education as proxies of SES (46, 47). It should, however, be acknowledged that ZIP code–based median income and education determinations, although accessible, are imprecise SES surrogates as they are prone to ecologic bias, misclassification, and geographic distortion (48). As a single-center study, our findings require external validation in additional Hispanic cohorts. Disclosure of the test result and shared decision-making may have introduced reporting bias into patient self-reported outcomes and this should be considered in future clinical trial designs. Finally, unmeasured confounders such as variability in clinician practice and non-persistence to therapy remain the realm of possibility.

In conclusion, two-thirds of all MSRC-tested Hispanic patients with RA exhibited an SIR to TNFi, in which case such therapy should be avoided. MSRC test results informed both physician selection of and patient agreement on advanced therapy, particularly in bDMARD-naïve patients. Treatment aligned with MSRC results was linked to improved outcomes in tested patients. Patients with MSRC-aligned treatment also had improved clinical outcomes compared with untested controls, especially if they were bDMARD-naïve. Hence, a broader use of the MSRC test may optimize outcomes and alter treatment paradigms.

Data availability statement

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Ethics statement

The studies involving humans were approved by Advarra IRB registered with OHRP and FDA under IRB#00000971. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

Author contributions

GK: Conceptualization, Investigation, Methodology, Writing – original draft, Writing – review & editing. MR: Conceptualization, Investigation, Methodology, Project administration, Writing – review & editing. VB: Investigation, Writing – review & editing. VS: Investigation, Writing – review & editing. SO: Formal analysis, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing.

Funding

The author(s) declared that financial support was not received for this work and/or its publication.

Acknowledgments

We thank all patients and health personnel involved in the study.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence, and reasonable efforts have been made to ensure accuracy, including review by the authors, wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmscd.2025.1728996/full#supplementary-material

Footnote

Abbreviations ACPA, anticitrullinated peptide antibody; ACR50, American College of Rheumatology 50% improvement; AUC, area under the curve; bDMARD, biologic disease-modifying antirheumatic drug; CDAI, Clinical Disease Activity Index; CDAI-LDA, Clinical Disease Activity Index-low disease activity; CDAI-MCID, Clinical Disease Activity Index minimum clinically important difference; CI, confidence interval; MSRC, molecular signal response classifier; PPV, positive predictive value; PtGA, patient global assessment; PtGA-LDA, patient global assessment-low disease activity; RA, rheumatoid arthritis; RAPID3, Routine Assessment of Patient Index Data-3; RAPID3-LDA, Routine Assessment of Patient Index Data-3 low disease activity; SD, standard deviation; SES, socioeconomic status; SIR, signal of inadequate response; TNFi, tumor necrosis factor-α inhibitor; tsDMARD, targeted synthetic disease-modifying antirheumatic drug.

References

1. Cross M, Smith E, Hoy D, Carmona L, Wolfe F, Vos T, et al. The global burden of rheumatoid arthritis: estimates from the global burden of disease 2010 study. Ann Rheum Dis. (2014) 73:1316–22. doi: 10.1136/annrheumdis-2013-204627

PubMed Abstract | Crossref Full Text | Google Scholar

2. England BR, Roul P, Yang Y, Sayles H, Yu F, Michaud K, et al. Burden and trajectory of multimorbidity in rheumatoid arthritis: a matched cohort study from 2006 to 2015. Ann Rheum Dis. (2021) 80:286–92. doi: 10.1136/annrheumdis-2020-218282

PubMed Abstract | Crossref Full Text | Google Scholar

3. Bergman M, Zhou L, Patel P, Sawant R, Clewell J, Tundia N. Healthcare costs of not achieving remission in patients with rheumatoid arthritis in the United States: a retrospective cohort study. Adv Ther. (2021) 38:2558–70. doi: 10.1007/s12325-021-01730-w

PubMed Abstract | Crossref Full Text | Google Scholar

4. Combe B, Logeart I, Belkacemi MC, Dadoun S, Schaeverbeke T, Daurès JP, et al. Comparison of the long-term outcome for patients with rheumatoid arthritis with persistent moderate disease activity or disease remission during the first year after diagnosis: data from the ESPOIR cohort. Ann Rheum Dis. (2015) 74:724–9. doi: 10.1136/annrheumdis-2013-204178

PubMed Abstract | Crossref Full Text | Google Scholar

5. Johnson KJ, Sanchez HN, Schoenbrunner N. Defining response to TNF-inhibitors in rheumatoid arthritis: the negative impact of anti-TNF cycling and the need for a personalized medicine approach to identify primary non-responders. Clin Rheumatol. (2019) 38:2967–76. doi: 10.1007/s10067-019-04684-1

PubMed Abstract | Crossref Full Text | Google Scholar

6. Pappas DA, St John G, Etzel CJ, Fiore S, Blachley T, Kimura T, et al. Comparative effectiveness of first-line tumour necrosis factor inhibitor versus non-tumour necrosis factor inhibitor biologics and targeted synthetic agents in patients with rheumatoid arthritis: results from a large US registry study. Ann Rheum Dis. (2021) 80:96–102. doi: 10.1136/annrheumdis-2020-217209

PubMed Abstract | Crossref Full Text | Google Scholar

7. Souto A, Maneiro JR, Gómez-Reino JJ. Rate of discontinuation and drug survival of biologic therapies in rheumatoid arthritis: a systematic review and meta-analysis of drug registries and health care databases. Rheumatology. (2016) 55:523–34.26490106

PubMed Abstract | Google Scholar

8. Kearsley-Fleet L, Davies R, De Cock D, Watson KD, Lunt M, Buch MH, et al. Biologic refractory disease in rheumatoid arthritis: results from the British society for rheumatology biologics register for rheumatoid arthritis. Ann Rheum Dis. (2018) 77:1405–12. doi: 10.1136/annrheumdis-2018-213378

PubMed Abstract | Crossref Full Text | Google Scholar

9. Zink A, Manger B, Kaufmann J, Eisterhues C, Krause A, Listing J, et al. Evaluation of the RABBIT risk score for serious infections. Ann Rheum Dis. (2014) 73:1673–6. doi: 10.1136/annrheumdis-2013-203341

PubMed Abstract | Crossref Full Text | Google Scholar

10. Aaltonen KJ, Ylikylä S, Joensuu J, Isomäki P, Pirilä L, Kauppi M, et al. Effectiveness of tumor necrosis factor-inhibitors in the treatment of rheumatoid arthritis: a comparison between randomized controlled trials and routine clinical practice. Ann Rheum Dis. (2016) 75(Suppl 2):485.

Google Scholar

11. Jin Y, Desai RJ, Liu J, Choi N-K, Kim SC. Factors associated with initial or subsequent choice of biologic disease-modifying antirheumatic drugs for treatment of rheumatoid arthritis. Arthritis Res Ther. (2017) 19:159. doi: 10.1186/s13075-017-1366-1

PubMed Abstract | Crossref Full Text | Google Scholar

12. Weinblatt ME, Kremer JM, Bankhurst AD, Bulpitt KJ, Fleischmann RM, Fox RI, et al. A trial of etanercept, a recombinant tumor necrosis factor receptor:Fc fusion protein, in patients with rheumatoid arthritis receiving methotrexate. N Engl J Med. (1999) 340:253–9. doi: 10.1056/NEJM199901283400401

PubMed Abstract | Crossref Full Text | Google Scholar

13. Plenge RM, Bridges SL. Personalized medicine in rheumatoid arthritis: miles to go before we sleep. Arthritis Rheum. (2011) 63:590–3. doi: 10.1002/art.30126

PubMed Abstract | Crossref Full Text | Google Scholar

14. Curtis JR, Kremer JM, Reed G, John AK, Pappas DA. TNFi cycling versus changing mechanism of action in TNFi-experienced patients: result of the Corrona CERTAIN comparative effectiveness study. ACR Open Rheumatol. (2022) 4:65–73. doi: 10.1002/acr2.11337

PubMed Abstract | Crossref Full Text | Google Scholar

15. Cohen S, Wells AF, Curtis JR, Dhar R, Mellors T, Zhang L, et al. A molecular signature response classifier to predict inadequate response to tumor necrosis factor-α inhibitors: the NETWORK-004 prospective observational study. Rheumatol Ther. (2021) 8:1159–76. doi: 10.1007/s40744-021-00330-y

PubMed Abstract | Crossref Full Text | Google Scholar

16. Jones A, Rapisardo S, Zhang L, Mellors T, Withers JB, Gatalica Z, et al. Analytical and clinical validation of an RNA sequencing-based assay for quantitative, accurate evaluation of a molecular signature response classifier in rheumatoid arthritis. Expert Rev Mol Diagn. (2021) 21:1235–43. doi: 10.1080/14737159.2021.2000394

PubMed Abstract | Crossref Full Text | Google Scholar

17. Curtis JR, Strand V, Golombek SJ, Karpouzas GA, Zhang L, Wong A, et al. Decision impact analysis to measure the influence of molecular signature response classifier testing on treatment selection in rheumatoid arthritis. Rheumatol Ther. (2024) 11:61–77. doi: 10.1007/s40744-023-00618-1

PubMed Abstract | Crossref Full Text | Google Scholar

18. Curtis JR, Strand V, Golombek S, Zhang L, Wong A, Zielinski MC, et al. Patient outcomes improve when a molecular signature test guides treatment decision-making in rheumatoid arthritis. Expert Rev Mol Diagn. (2022):1–10. doi: 10.1080/14737159.2022.2140586

PubMed Abstract | Crossref Full Text | Google Scholar

19. Strand V, Zhang L, Arnaud A, Connolly-Strong E, Asgarian S, Withers JB. Improvement in clinical disease activity index when treatment selection is informed by the tumor necrosis factor-ɑ inhibitor molecular signature response classifier: analysis from the study to accelerate information of molecular signatures in rheumatoid arthritis. Expert Opin Biol Ther. (2022) 22:801–7. doi: 10.1080/14712598.2022.2066972

PubMed Abstract | Crossref Full Text | Google Scholar

20. Ennis S, Ríos-Vargas M, Albert N. The Hispanic Population: 2010. US Department of Commerce, Economics and Statistics Administration, US Census Bureau (2011). Available at: http://www.census.gov/prod/cen2010/briefs/c2010br-04.pdf (Accessed October 18, 2025).

Google Scholar

21. Del Rincón I, Battafarano DF, Arroyo RA, Murphy FT, Fischbach M, Escalante A. Ethnic variation in the clinical manifestations of rheumatoid arthritis: role of HLA-DRB1 alleles. Arthritis Rheum. (2003) 49:200–8. doi: 10.1002/art.11000

PubMed Abstract | Crossref Full Text | Google Scholar

22. Bruce B, Fries JF, Murtagh KN. Health status disparities in ethnic minority patients with rheumatoid arthritis: a cross-sectional study. J Rheumatol. (2007) 34:1475–9.17552045

PubMed Abstract | Google Scholar

23. Yazici Y, Kautiainen H, Sokka T. Differences in clinical status measures in different ethnic/racial groups with early rheumatoid arthritis: implications for interpretation of clinical trial data. J Rheumatol. (2007) 34:311–5.17304656

PubMed Abstract | Google Scholar

24. Ang DC, Paulus HE, Louie JS. Patient’s ethnicity does not influence utilization of effective therapies in rheumatoid arthritis. J Rheumatol. (2006) 33:870–8.16652419

PubMed Abstract | Google Scholar

25. Karpouzas GA, Dolatabadi S, Moran R, Li N, Nicassio PM, Weisman MH. Correlates and predictors of disability in vulnerable US Hispanics with rheumatoid arthritis. Arthritis Care Res. (2012) 64:1274–81. doi: 10.1002/acr.21689

PubMed Abstract | Crossref Full Text | Google Scholar

26. U.S. Census Bureau. ACS Demographic and Housing Estimates. American Community Survey, ACS 1-Year Estimates Data Profiles, Table DP05 (2022). Available online at: https://data.census.gov/table/ACSDP1Y2022.DP05?q=elpaso,texas (Accessed September 3, 2024).

27. U.S. Census Bureau. Median Income in the Past 12 Months (in 2022 Inflation-Adjusted Dollars). American Community Survey, ACS 5-Year Estimates Subject Tables, Table S1903 (2022). Available online at: https://data.census.gov/table/ACSST5Y2022.S1903?t=Income%20(Households,%20Families,%20Individuals)&g=040XX00US35$8600000,48$8600000&y=2022 (Accessed September 3, 2024).

28. U.S. Census Bureau. Educational Attainment. American Community Survey, ACS 5-Year Estimates Subject Tables, Table S1501 (2022). Available online at: https://data.census.gov/table/ACSST5Y2022.S1501?t=Educational%20Attainment&g=040XX00US35$8600000,48$8600000&y=2022&moe=false (Accessed September 3, 2024).

29. Curtis JR, Yun H, Chen L, Ford SS, van Hoogstraten H, Fiore S, et al. Real-world sarilumab use and rule testing to predict treatment response in patients with rheumatoid arthritis: findings from the RISE registry. Rheumatol Ther. (2023) 10:1055–72. doi: 10.1007/s40744-023-00568-8

PubMed Abstract | Crossref Full Text | Google Scholar

30. Uhlig T, Kvien TK, Pincus T. Test-retest reliability of disease activity core set measures and indices in rheumatoid arthritis. Ann Rheum Dis. (2009) 68:972–5. doi: 10.1136/ard.2008.097345

PubMed Abstract | Crossref Full Text | Google Scholar

31. Wells GA, Boers M, Shea B, Brooks PM, Simon LS, Strand CV, et al. Minimal disease activity for rheumatoid arthritis: a preliminary definition. J Rheumatol. (2005) 32:2016–24.16206362

PubMed Abstract | Google Scholar

32. Genovese MC, Bathon JM, Martin RW, Fleischmann RM, Tesser JR, Schiff MH, et al. Etanercept versus methotrexate in patients with early rheumatoid arthritis: two-year radiographic and clinical outcomes. Arthritis Rheum. (2002) 46:1443–50. doi: 10.1002/art.10308

PubMed Abstract | Crossref Full Text | Google Scholar

33. Furst DE, Schiff MH, Fleischmann RM, Strand V, Birbara CA, Compagnone D, et al. Adalimumab, a fully human anti tumor necrosis factor-alpha monoclonal antibody, and concomitant standard antirheumatic therapy for the treatment of rheumatoid arthritis: results of STAR (safety trial of adalimumab in rheumatoid arthritis). J Rheumatol. (2003) 30:2563–71.14719195

PubMed Abstract | Google Scholar

34. Tugwell P, Pincus T, Yocum D, Stein M, Gluck O, Kraag G, et al. Combination therapy with cyclosporine and methotrexate in severe rheumatoid arthritis. N Engl J Med. (1995) 333:137–41. doi: 10.1056/NEJM199507203330301

PubMed Abstract | Crossref Full Text | Google Scholar

35. Weinblatt ME, Kremer JM, Coblyn JS, Maier AL, Helfgott SM, Morrell M, et al. Pharmacokinetics, safety, and efficacy of combination treatment with methotrexate and leflunomide in patients with active rheumatoid arthritis. Arthritis Rheum. (1999) 42:1322–8. doi: 10.1002/1529-0131(199907)42:7%3C1322::AID-ANR4%3E3.0.CO;2-P

PubMed Abstract | Crossref Full Text | Google Scholar

36. O’Dell JR, Haire CE, Erikson N, Drymalski W, Palmer W, Eckhoff PJ, et al. Treatment of rheumatoid arthritis with methotrexate alone, sulfasalazine and hydroxychloroquine, or a combination of all three medications. N Engl J Med. (1996) 334:1287–91. doi: 10.1056/NEJM199605163342002

Crossref Full Text | Google Scholar

37. Moslimani M, Lopez MH, Noe-Bustamante L. 11 Facts About Hispanic Origin Groups in the U.S. Washington, DC: Pew Research Center (2023). Available online at: https://www.pewresearch.org/short-reads/2023/08/16/11-facts-about-hispanic-origin-groups-in-the-us/ (Accessed March 15, 2025).

Google Scholar

38. Burgers LE, Raza K, van der Helm-van Mil AH. Window of opportunity in rheumatoid arthritis—definitions and supporting evidence: from old to new perspectives. RMD Open. (2019) 5:e000870. doi: 10.1136/rmdopen-2018-000870

PubMed Abstract | Crossref Full Text | Google Scholar

39. Karpouzas G, Hernandez E, Ruiz L, Strand V, Ormseth S. Medication necessity and concerns beliefs are distinct, interactive predictors of treatment adherence in rheumatoid arthritis [abstract]. Arthritis Rheum. (2019) 71(Suppl 10).

Google Scholar

40. Iñiguez Ubiaga C, Moriano C, Garijo Bufort M, Crespo Golmar A, GonzalezFernandez I, Alvarez Castro C, et al. AB1236 therapeutic adherence and satisfaction in a rheumatology consultation. Ann Rheum Dis. (2018) 77(Suppl 2):1715. doi: 10.1136/annrheumdis-2018-eular.7399

Crossref Full Text | Google Scholar

41. Xu M, Lo SHS, Miu EYN, Choi KC. Educational programmes for improving medication adherence among older adults with coronary artery disease: a systematic review and meta-analysis. Int J Nurs Stud. (2025) 161:104924. doi: 10.1016/j.ijnurstu.2024.104924

PubMed Abstract | Crossref Full Text | Google Scholar

42. Alegria M, Nakash O, Johnson K, Ault-Brutus A, Carson N, Fillbrunn M, et al. Effectiveness of the DECIDE interventions on shared decision making and perceived quality of care in behavioral health with multicultural patients: a randomized clinical trial. JAMA Psychiatry. (2018) 75:325–35. doi: 10.1001/jamapsychiatry.2017.4585

PubMed Abstract | Crossref Full Text | Google Scholar

43. Conlin PR, Colburn J, Aron D, Pries RM, Tschanz MP, Pogach L. Synopsis of the 2017 U.S. Department of Veterans Affairs/U.S. Department of Defense clinical practice guideline: management of type 2 diabetes mellitus. Ann Intern Med. (2017) 167:655–63. doi: 10.7326/M17-1362

PubMed Abstract | Crossref Full Text | Google Scholar

44. Strand V, Cohen S, Crawford B, Smolen JS, Scott DL, Leflunomide Investigators Groups. Patient-reported outcomes better discriminate active treatment from placebo in randomized controlled trials in rheumatoid arthritis. Rheumatology. (2004) 43:640–7. doi: 10.1093/rheumatology/keh140

PubMed Abstract | Crossref Full Text | Google Scholar

45. Pincus T, Strand V, Koch G, Amara I, Crawford B, Wolfe F, et al. An index of the three core data set patient questionnaire measures distinguishes efficacy of active treatment from that of placebo as effectively as the American College of Rheumatology 20% response criteria (ACR20) or the disease activity score (DAS) in a rheumatoid arthritis clinical trial. Arthritis Rheum. (2003) 48:625–30. doi: 10.1002/art.10824

PubMed Abstract | Crossref Full Text | Google Scholar

46. Calixto O-J, Anaya J-M. Socioeconomic status. The relationship with health and autoimmune diseases. Autoimmun Rev. (2014) 13:641–54. doi: 10.1016/j.autrev.2013.12.002

PubMed Abstract | Crossref Full Text | Google Scholar

47. McCollum L, Pincus T. A biopsychosocial model to complement a biomedical model: patient questionnaire data and socioeconomic status usually are more significant than laboratory tests and imaging studies in prognosis of rheumatoid arthritis. Rheum Dis Clin North Am. (2009) 35:699–712. doi: 10.1016/j.rdc.2009.10.003

PubMed Abstract | Crossref Full Text | Google Scholar

48. Chen F, MacDonald B, Xu Y, Franco W, Campos A, Palinkas LA, et al. ZIP code and ZIP code tabulation area linkage: implications for bias in epidemiologic research. Epidemiology. (2025) 36:115–8. doi: 10.1097/EDE.0000000000001800

PubMed Abstract | Crossref Full Text | Google Scholar