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

ABSTRACT

Background

A blood-based molecular signature response classifier (MSRC) predicts non-response to tumor necrosis factor-ɑ inhibitors (TNFi) in rheumatoid arthritis (RA).

Research design and methods

This is an interim analysis of data collected in the Study to Accelerate Information of Molecular Signatures (AIMS) in RA from patients who received the MSRC test between September 2020 and November 2021. Absolute changes in clinical disease activity index (CDAI) scores from baseline were evaluated at 12 weeks (n = 470) and 24 weeks (n = 274).

Results

Predicted TNFi non-responders who received a biologic or targeted synthetic disease-modifying antirheumatic drug (b/tsDMARD) with an alternative mechanism of action (altMOA) experienced up to 1.8-fold greater improvements in CDAI scores than those treated with a TNFi (12 weeks: 12.2 vs 8.0; p-value = 0.083; 24 weeks: 14.2 vs 7.8 p-value = 0.009). In patients with a molecular signature of non-response to TNFi in high disease activity at baseline, this corresponded to 43.2% relative improvement in achieving a lower CDAI disease activity level when likely TNFi non-responders were treated with a non-TNFi therapy (38.9% vs 55.7%). Commensurate improvements in efficiency of spend are expected when TNFi are avoided in favor of altMOA.

Conclusions

RA treatment selection informed by MSRC test results improves clinical outcomes in real-world care and offers improvements in efficiency of healthcare spending.

KEYWORDS:

• Clinical utility
• molecular signature
• precision medicine
• real-world evidence
• response prediction
• rheumatoid arthritis
• TNFi therapy
• tumor necrosis factor-ɑ inhibitors
• b/tsDMARD
• MSRC

View correction statement:

Correction

1. Introduction

Rheumatoid arthritis (RA) is a systemic autoimmune disease that presents in patients as joint pain, swelling and tenderness as a symmetric polyarthritis. Without timely therapeutic intervention, chronic articular inflammation causes progressive joint damage leading to deformities, functional impairment and disability, as well as increased mortality [1–3]. Biologic and targeted synthetic disease-modifying antirheumatic drugs (b/tsDMARDs) have become integral to the treat-to-target paradigm designed to gain low disease activity or remission as promptly as possible [4–6].

In patients with RA who are inadequate responders to methotrexate (MTX-IR), tumor necrosis factor-ɑ inhibitor (TNFi) therapies result in clinically meaningful responses according to American College of Rheumatology criteria for ≥50% (ACR50) responses at 6 months in 27–38% of patients [7]. Near equivalent efficacy and safety profiles between b/tsDMARDs are evident in patients with RA; lack of clinically validated biomarkers for patient stratification, and inability to prioritize b/tsDMARD selection by clinical guidelines has led to a preponderance of trial-and-error treatment selection, paradigms by physicians driven by administrative priorities and formularies, instead of by patient biology. As a result, TNFi are the customary first-line b/tsDMARD class in RA [8,9].

The population of patients with RA exert a disproportionate impact on total healthcare expenditure and overall quality of life, including lost productivity and early mortality [10]. Treatments for RA and specifically the TNFi class (adalimumab, certolizumab pegol, etanercept, golimumab, infliximab) represent one of the leading cost drivers of pharmacy spend in the United States, estimated at approximately $11 billion in 2021 [11] with list prices for TNFi therapies continuing to increase year over year [12,13].

A blood-based molecular signature response classifier (MSRC) was analytically and clinically validated to identify patients with RA who will be non-responders to TNFi therapies [14–16]. The MSRC combines biological features such as gene expression, anti-cyclic citrullinated protein (anti-CCP) serostatus and sex into a single predictive algorithm, the output of which is a numerical value from one to 25, where patients with a result of ≥10.6 have a molecular signature of non-response to TNFi detected [15]. Compared to those without, patients with a molecular signature of non-response to TNFi are 3.4–6.7 times less likely to reach low disease activity (LDA) by 6 months if treated with a TNFi therapy [15]. In the first cross-sectional analysis of the real-world clinical Study to Accelerate Information of Molecular Signatures (AIMS) in RA, the MSRC had high clinical utility, providing patient-specific information that informed providers’ treatment selection approximately 90% of the time. This resulted in improved responses to b/tsDMARDs among patients with or without the molecular signature of non-response according to ACR50 and minimally important differences in the clinical disease activity index (CDAI) criteria [17].

This study reports the second interim analysis of the AIMS real-world study evaluating the clinical utility of the MSRC test. Change from baseline in disease activity according to CDAI scores in response to treatment selections based on MSRC test results was assessed and the resulting cost savings per patient were evaluated.

2. Patients and methods

2.1. AIMS patient cohort

AIMS is a clinical database of real-world longitudinal data from patients with RA managed across a network of 72 private and academic rheumatology practices in the contiguous United States. Patient management in AIMS focuses on utilization of the commercially available MSRC test (PrismRA®, Scipher Medicine Corporation). The AIMS study received Institutional Review Board (IRB) approval at participating sites, clinical data were managed according to the Health Insurance Portability and Accountability Act Authorization of 1996 (HIPAA) and all patients provided informed consent prior to study participation.

This study reports an interim analysis of data collected from patients who received the MSRC test between September 2020 and November 2021. Eligibility criteria for analysis were the following: age ≥18 years, a clinical diagnosis of RA treated by a rheumatologist, b/tsDMARD-naïve or TNFi-exposed at time of MSRC testing with moderate-to-severe disease at baseline according to CDAI >10, patients to have made a b/tsDMARD treatment decision and MSRC testing, and data to calculate CDAI were available at baseline and at least one follow-up visit. Patients were excluded from analysis if, treated with a b/tsDMARD other than a TNFi at the time of MSRC testing or if the b/tsDMARD was changed between the initial treatment decision and first follow-up visit. Changes in doses of concomitant medications were permitted at the discretion of the provider and were not considered exclusionary for this analysis. Patient eligibility did not consider conventional synthetic DMARD use. Patients in the study had previously received methotrexate or conventional synthetic DMARD and were ready to start a new b/tsDMARD. Baseline patient characteristics such as anti-CCP and RF serostatus and duration of disease were considered as potential covariates in statistical analyses (Section 2.4) but were not used to define the patient cohort at baseline nor were they used to determine enrollment eligibility into the AIMS in RA observational study. Analyses evaluated data collected at baseline and follow-up visits at 12 and 24 weeks after treatment decision. Data included MSRC test results, clinical assessments, RA medical history, routine laboratory testing, and treatment decisions. The study was conducted in accordance with the Declaration of Helsinki and the International Committee on Harmonization of Good Clinical Practice.

2.2. Clinical outcomes analysis

The primary and secondary endpoints of the clinical outcomes analysis were changes from baseline in absolute CDAI scores at 24-weeks and 12-weeks, respectively. Treatment responses were assessed within and between four patient subsets, defined by MSRC test results and b/tsDMARD selection: 1) molecular signature of non-response to TNFi, treated with non-TNFi b/tsDMARDs (predicted non-responder [PNR]-altMOA); 2) molecular signature of non-response to TNFi, treated with TNFi therapy (PNR-TNFi); 3) no detected molecular signature of non-response, treated with a TNFi (NPNR-TNFi); and 4) no detected molecular signature of non-response, treated with a non-TNFi b/tsDMARD (NPNR-altMOA). Changes in CDAI scores were reported in patients who completed 12-week follow-up visits (n = 470) and in those who completed 24-week visits (n = 274). Clinical outcomes were imputed using last observation carried forward if a treatment switch occurred between the first and second follow-up visit. CDAI low disease activity was defined as ≤10 and high disease activity was defined as >22. Improvement in efficiency of spend was defined as commensurate to the relative improvement in CDAI scores from baseline.

2.3. Description of the MSRC test

The MSRC consists of a feedforward artificial neural network model that evaluates 23 biological features, including gene expression features validated in Mellors et al. [14] and Cohen et al. [15] and clinical features including anti-CCP, sex, body mass index (BMI) and patient global assessment of disease activity. The raw output of the predictive model is transformed by a monotonic function to a continuous variable between one and 25. Results ≥10.6 indicate detection of a molecular signature of non-response, the higher the number the greater likelihood of TNFi non-response. Results <10.6 indicate that a signal of non-response is not detected. The MSRC has been clinically validated in b/tsDMARD-naïve and TNFi-exposed RA patients [14,15,17].

The MSRC test kit includes two PAXgene Blood RNA Tubes for gene expression feature analysis and one serum separator tube for anti-CCP testing. Kits were processed and anti-CCP testing was performed on the Roche Cobas® system according to standard operating procedures in Scipher Medicine’s Durham, North Carolina CLIA laboratory. RNA sequencing was performed under CLIA laboratory standard operating procedures at the Ambry Genetics Corporation (Aliso Viejo, CA), as previously described [16,17]. Algorithmic analysis is performed at the Scipher Medicine Laboratory.

2.4. Statistical analysis

In the real-world data setting, no baseline characteristics were controlled between patient subsets. A generalized linear model with forward selection was implemented with covariates that showed a significant difference (p-value <0.05) between patient subsets used in the analyses. To evaluate differences between baseline demographic data between patient subsets, Student’s t-test and Chi-squared tests were used for continuous and categorical variables, respectively.

3. Results

3.1. Patient cohort demographics

Demographics and baseline characteristics were evaluated in four AIMS cohort subsets (Table 1), characterized by MSRC test results and b/tsDMARD selection (see Patients and Methods for details). There were no significant differences observed at baseline in sex, race, CRP values, RF status, anti-CCP status, swollen joint counts or tender joint counts (p-value >0.05).

Table 1. Patient demographics and baseline characteristics.

Variable	Sub-categories	PNR-TNFi	PNR-altMOA	NPNR-TNFi	NPNR-altMOA	P-value (PNR-TNFi vs.PNR-altMOA)	P-value (NPNR-TNFi vs.NPNR-altMOA)
N		151	156	129	70
Age (year), mean (SD)		53.9 (13.7)	57.4 (12.6)	56.3 (14.0)	60.1 (12.5)	0.021	0.051
Female, n (%)		n = 135 (89)	n = 142 (91)	n = 89 (69)	n = 54 (77)	0.715	0.265
Duration of disease (years), median (IQR; n)		3.00 (Q1 = 0.50, Q3 = 8.75, IQR = 8.25, n = 150)	4.90 (Q1 = 2.00, Q3 = 10.50, IQR = 8.50, n = 155)	1.80 (Q1 = 0.50, Q3 = 6.90, IQR = 6.40, n = 129)	5.90 (Q1 = 2.20, Q3 = 16.65, IQR = 14.45, n = 70)	0.020	<0.001
Race, n (%)						0.463	0.657
	White	n = 118 (78)	n = 129 (83)	n = 107 (83)	n = 56 (8)	0.387	0.700
	Other	n = 20 (13)	n = 12 (8)	n = 7 (5)	n = 6 (9)	0.124	0.508
	Black or African American	n = 11 (7)	n = 10 (6)	n = 10 (8)	n = 5 (7)	0.816	1.000
	Asian	n = 1 (1)	n = 2 (1)	n = 4 (3)	n = 1 (1)	1.000	0.659
	American Indian or Alaskan Native	n = 1 (1)	n = 3 (2)	n = 1 (1)	n = 2 (3)	0.644	0.570
TNFi-naïve, n(%)		n = 115 (76)	n = 134 (86)	n = 95 (74)	n = 64 (91)	0.040	0.005
RF positive, n (%)		n = 51 (37)	n = 51 (38)	n = 55 (54)	n = 32 (58)	0.892	0.717
CCP positive, n (%)		n = 41 (27)	n = 42 (27)	n = 62 (48)	n = 38 (54)	1.000	0.457
CRP at baseline, mg/dl		10 (13.8)	9.5 (13.9)	11.9 (19.7)	10.7 (17.9)	0.725	0.665
HAQ at baseline, mean (SD)		0.8 (0.6)	0.8 (0.5)	0.5 (0.5)	0.7 (0.5)	0.824	0.089
Tender joints at baseline, mean (SD)		14.8 (9.1)	16.5 (9.4)	11.4 (8.2)	11.3 (8.2)	0.105	0.947
Swollen joints at baseline, mean (SD)		7.9 (6.7)	9 (6.8)	7.1 (6.5)	5.5 (5.8)	0.142	0.083
Methotrexate, n (%)		n = 83 (55)	n = 82 (53)	n = 76 (59)	n = 28 (4)	0.740	0.016
Prednisone, n (%)		n = 31 (21)	n = 44 (28)	n = 45 (35)	n = 22 (31)	0.139	0.632
Hydroxychloroquine, n (%)		n = 18 (12)	n = 11 (7)	n = 16 (12)	n = 9 (13)	0.173	1.000
TNFi, n (%)		n = 151 (100)		n = 129 (100)
	Etanercept	n = 20 (13)		n = 21 (16)
	Infliximab	n = 17 (11)		n = 11 (9)
	Golimumab	n = 40 (26)		n = 25 (19)
	Certolizumab	n = 21 (14)		n = 25 (19)
	Adalimumab	n = 53 (35)		n = 47 (36)
JAK, n (%)			n = 75 (48)		n = 35 (5)
	Upadacitinib		n = 45 (29)		n = 19 (27)
	Baricitinib		n = 5 (3)		n = 9 (13)
	Tofacitinib		n = 25 (16)		n = 7 (1)
T-cell, n (%)			n = 52 (34)		n = 22 (31)
	Abatacept		n = 52 (34)		n = 22 (31)
IL-6, n (%)			n = 22 (14)		n = 9 (13)
	Tocilizumab		n = 20 (13)		n = 8 (11)
	Sarilumab		n = 2 (1)		n = 1 (1)
B-cell, n (%)			6 (4)		4 (6)
	Rituximab		6 (4)		4 (6)

b/tsDMARD, biologic or targeted synthetic disease-modifying antirheumatic drug; SD, standard deviation; Q, quartile; IQR, interquartile range; TNFi, tumor necrosis factor-a inhibitor; JAK, Janus kinase inhibitor; IL-6, Interleukin- 6; PNR-TNFi, non-response to TNFi signal present therapy TNFi; PNR-altMOA, non-response to TNFi signal present, therapy alternative mechanisms of action to TNFi; NPNR-TNFi, non-response signal absent, therapy TNFi, NPNR- altMOA; non-response signal absent, therapy alternative mechanisms of action to TNFi

3.2. Utilization of MSRC test results can improve responses to b/tsDMARDs

Changes in absolute CDAI scores from baseline were evaluated in those patients with 12-week follow-ups (n = 470) and in patients with 24-week follow-ups (n = 274) (Figure 1A). When patients with a molecular signature of non-response are treated with a non-TNFi b/tsDMARD, they experienced a 1.8-fold greater improvement in CDAI scores than when patients were treated with a TNFi (Table 2). Consistent changes in CDAI from baseline were observed at 12 weeks (8.0 in PNR-TNFi vs 12.2 in PNR-altMOA; p-value = 0.083) and 24 weeks (7.8 in PNR-TNFi vs 14.2 in PNR-altMOA; p-value = 0.009) (Figure 1B).

Figure 1. Study design and outcomes. a) Inclusion and exclusion criteria for patients included in the outcome analyses of this study. All patients were enrolled in the real-world AIMS study and in keeping with the inclusion criteria of the parent study were ≥18 years of age, had a clinical diagnosis of RA and MSRC test results. b) Mean change from baseline in absolute CDAI at 12- and 24-week follow-up visits for predicted non-responders to TNFi therapy who were treated with a TNFi (PNR-TNFi) or a b/tsDMARD with an alternative mechanism of action (PNR-altMOA).

Table 2. Mean change in CDAI scores from baseline to 12-week and 24-week follow-up visits in the AIMS cohort subsets.

Timepoint	Cohort	Endpoint (CDAI)	N	PNR-TNFi	PNR-altMOA
12 weeks	All	Absolute change	470	7.964	12.199
	Baseline High CDAI	Absolute change	313	10.13	13.994
	Baseline Moderate CDAI	Absolute change	157	2.168	4.507
24 weeks	All	Absolute change	274	7.757	14.192
	Baseline High CDAI	Absolute change	230	8.853	15.09
		% achieved lower than MDA		38.9%	55.7%
		% achieved lower than LDA		16.7%	22.8%
	Baseline Moderate CDAI	Absolute change	71	3.118	9.7
		% achieved lower than LDA		47.1%	71.4%

CDAI, clinical disease activity index; LDA, low disease activity according to CDAI scores; MDA, moderate disease activity according to CDAI scores; PNR-TNFi, non-response to TNFi signal present therapy TNFi; PNR-altMOA, non-response to TNFi signal present, therapy alternative mechanisms of action to TNFi

Patients in high disease activity at baseline with a molecular signature of non-response showed a 1.7-fold greater improvement in CDAI scores when treated with a non-TNFi b/tsDMARD as compared to a TNFi therapy (8.9 in PNR-TNFi vs 15.1 in PNR-altMOA) and patients in moderate disease activity at baseline showed a 3.1-fold greater improvement (3.1 in PNR-TNFi vs 9.7 in PNR-altMOA). In patients in CDAI high disease activity at baseline, this corresponded to 38.9% of patients achieving of a lower disease activity level in response to TNFi compared to 55.7% with non-TNFi b/tsDMARDs; a 43.2% relative improvement. Furthermore, a greater proportion of predicted TNFi non-responders experienced worsening CDAI scores when treated with a TNFi compared with an altMOA therapy (high baseline disease activity: 27.8% [20/72] in PNR-TNFi vs 17.1% [12/70] in PNR-altMOA).

In patients treated with TNFi therapy, those without a molecular signature of TNFi non-response had 1.6x greater improvements in CDAI scores at 12 weeks (8.0 in PNR-TNFi vs 10.2 in NPNR-TNFi; p-value = 0.087) and 1.9x at 24 weeks (7.8 in PNR-TNFi vs 12.7 in NPNR-TNFi; p-value = 0.012).

3.3. Using MSRC results to inform treatment selection can improve efficiency of spend

In patients predicted to have non-response to TNFi, those receiving an alternative MOA b/tsDMARD had 1.6x greater mean improvements from baseline in CDAI scores from baseline at 12 weeks and 1.9x greater mean improvements at 24 weeks compared with patients treated with a TNFi therapy. Avoidance of TNFi in patients with a signal of non-response to TNFi in favor of alternative MOAs would be expected to result in a commensurate increase in the efficiency of spend between 1.6 and 1.9 times.

4. Discussion

There are many FDA-approved b/tsDMARDs for treatment of RA that offer similar reported clinical responses based on randomized controlled trials. However, matching the unique disease biology of each patient to an appropriate MOA is one of the greatest challenges in RA [7,18]. The blood-based MSRC provides patient-specific data to inform treatment selection in those who are intending to initiate, switch, or dose escalate a TNFi therapy. When a molecular signature of non-response to TNFi is detected, patients can be offered an alternative b/tsDMARD. The MSRC test has been clinically validated to predict non-response according to multiple response outcome criteria including ACR50, DAS28-CRP and CDAI [15,16]. Treatment selection in RA that is informed by MSRC test results has high clinical utility; patients experience a near two-fold greater improvement in CDAI scores. This is consistent with prior clinical utility analyses showing that when providers used MSRC results to inform treatment decisions, ACR20/50/70 responses at 24-weeks to b/tsDMARD treatment were improved two to fivefold [17].

Providing effective treatment early in the disease course results in lower levels of disease progression and structural joint damage that results in functional impairments and disability [19]. Most of the improvements in CDAI scores were evident by the 12-week follow-up visit, indicating that when MSRC test results are used to inform treatment selection, outcomes improve in a timely manner in b/tsDMARD-naïve (373/470) and TNFi-exposed (97/470) RA patients. Furthermore, in patients with a molecular signature of TNFi non-response who nonetheless received a TNFi, no further improvements in absolute CDAI scores occurred between the 12- and 24-weeks. This is consistent with the prospective clinical validation of the MSRC that such patients are approximately 9-times less likely to achieve CDAI remission at 24-weeks [15]. Treating to target is supported by RA treatment guidelines [6,18] and defines monitoring and treatment adjustments to preserve health related quality of life, mobility, function, and prevent joint damage [4]. A greater percentage of patients with a molecular signature of non-response to TNFi and treated with a non-TNFi b/tsDMARD achieved remission or low disease activity than did those nonetheless provided a TNFi. By combining gene expression and clinical features, the MSRC provides insights into the disease biology of patients with RA beyond what is available via clinical assessments to enable providers to make more informed treatment selections. MSRC test results can inform treat-to-target decisions and complement clinical judgment to help patients with RA achieve remission or low disease activity.

Computational approaches are being applied to other areas of rheumatology to identify predictors of responses to the IL6R antagonist sarilumab [20], clinical characteristics predictive of response to biologics in RA and ankylosing spondylitis [21], and predictors of disease flares in RA [22]. These biomarker studies offer encouraging advances in technology in rheumatology and future work to clinically validate these biomarkers is warranted. Many prior biomarker strategies to inform care in RA have failed to be reproduced when tested in new patient cohorts [23,24]. In contrast, the MSRC is clinically and analytically validated to predict non-response in RA to the entire class of TNFi therapies [14–17].

Until now, a trial-and-error approach to treatment posed a challenge that can necessitate an annual therapy cost of more than $83,000 per patient in RA [13]. An independent meta-analysis of randomized controlled trials (RCTs) reported ACR50 responses at 6 months with TNFi of 27.1 to 37.5% [7]. Therefore, it can be estimated that between 63 and 73 cents of every dollar spent on TNFi therapy will not result in meaningful improvements for patients. Informing treatment selection with MSRC test results improves treatment responses [17], leading to improvements in efficiency of spend and cost-effectiveness [25].

There are limitations of this study. Given the pragmatic design of the study, differences in baseline covariates between patient subgroups may have impacted treatment responses, although statistical analyses were designed to address these. MSRC test results may have influenced measurements and perceptions of treatment response. Future studies will focus on other factors perceived to influence treatment responses that were not evaluated in the current study. While a metric of value-based care was presented here, further analyses evaluating the population-level impact of MSRC informed care on both clinical outcomes and the cost of care are warranted.

5. Conclusions

There are many treatment options for RA patients, and more in development. A trial-and-error approach to treatment selection in RA has thus far been a necessity given the paucity of evidence supporting pairing of individual patient disease biology to a specific b/tsDMARD options. This study demonstrated that when MSRC test results are used to inform b/tsDMARD selections, patients with molecular signatures of TNFi non-response can, on average, experience nearly two-fold greater improvements in CDAI scores if treated with a non-TNFi b/tsDMARD. These data add to the growing evidence supporting the clinical utility of the MSRC test and show that integration of the MSRC into RA management results in improved clinical outcomes [17].

Declaration of interest

L Zhang, A Arnaud, E Connolly-Strong, S Asgarian and J Withers are full-time employees of Scipher Medicine Corporation. V Strand serves as a consultant to Scipher Medicine. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Author contributions

Conceptualization: all authors; Methodology: Lixia Zhang, Alix Arnaud; Formal analysis and investigation: Lixia Zhang, Alix Arnaud, Erin Connolly-Strong, Vibeke Strand; Writing - original draft preparation: Lixia Zhang, Alix Arnaud, Johanna Withers; Writing - review and editing: all authors.

All named authors meet the International Committee of Medical Journal Editors (ICMJE) criteria for authorship for this article, take responsibility for the integrity of the work as a whole, and have given their approval for this version to be published.

Compliance with ethics guidelines

This study was conducted in accordance with the ethical principles of the Declaration of Helsinki and are consistent with the International Committee on Harmonization of Good Clinical Practice, as well as other applicable local and federal laws, regulations, and guidelines. All participants gave informed consent.

Data availability

The algorithm underlying the MSRC is proprietary to Scipher Medicine Corporation.

Acknowledgments

The authors thank the healthcare providers and participants who made this study possible.

Additional information

Funding

This study was funded by Scipher Medicine Corporation.