Patient Groups in Rheumatoid Arthritis Identified by Deep Learning Respond Differently to Biologic or Targeted Synthetic Disease Modifying Antirheumatic Drugs

Background: In rheumatoid arthritis (RA), there is often no adequate treatment response to the first biologic or targeted synthetic disease modifying antirheumatic drug (b/tsDMARD) which prevents and delays disease control.

Objectives: To assess and validate treatment response of b/tsDMARDs among RA patient groups identified by deep learning.

Methods: In the Swiss Clinical Quality Management of Rheumatic Diseases registry (SCQM) [1998-2018], we clustered RA patients at first-time b/tsDMARD (i.e. cohort entry) through several runs of deep embedded clustering. To increase robustness of our findings, we grouped similar patient clusters together (i.e. groups).
Our outcomes were b/tsDMARD stop due to non-response, and separately a ≥20% reduction in DAS28 (RA disease activity in 28 joints) as a proxy for treatment response. Follow-up stopped at b/tsDMARD stop or a maximum of 15 months. We assessed comparative effectiveness of b/tsDMARDs (ref. adalimumab) using Cox proportional hazard regression in each patient group by estimating hazard ratios (HR) with 95% confidence intervals (CI).
We validated results obtained per group through stratified analyses according to most distinctive patient characteristics of the respective group.

Results: We obtained 24 clusters which comprised 362-1481 patients (3516 unique patients). These clusters were grouped according to most distinct characteristics at b/tsDMARD initiation: 1) ≥2 conventional synthetic (cs) DMARDs and prednisone use, 2) male sex, 3) seronegativity, female sex, and no prednisone use, 4) rather low disease burden, 5) seropositivity, female sex, and a rather high disease burden/duration.
Comparative effectiveness results among validation strata confirmed comparative effectiveness results among the groups: Patients with ≥2 csDMARDs and prednisone use, men, as well as patients with a lower disease burden responded better to tocilizumab than to adalimumab (HRs of treatment response: 5.46, 95% CI [1.76-16.94], HR 8.44 [3.43-20.74], and HR 3.64 [2.04-6.49], respectively). Furthermore, seronegative women without use of prednisone as well as seropositive women with a higher disease burden/duration had a higher risk of non-response with golimumab (HRs of b/tsDMARD stop: 2.36 [1.03-5.40] and HR 5.27 [2.10-13.21], respectively) than with adalimumab.

Conclusions: RA patient groups identified by deep clustering may respond differently to individual first-line b/tsDMARDs. Thus, our results may support the decision on the best choice of first-time b/tsDMARD for certain RA patients, which is a step forward towards personalizing treatment.