Machine learning identifies patients with RA most likely to benefit from sarilumab
Click Here to Manage Email Alerts
A model using machine learning was able to ascertain a simple selection rule to identify which patients with rheumatoid arthritis would benefit most from sarilumab, according to a speaker at the 2020 ACR Convergence Annual Meeting.
“Despite the existence of guidelines for DMARD treatment of RA, a more individualized approach to treatment is needed to maximize efficacy while minimizing risk of adverse events,” Ernest Choy, MD, FRCP, professor and head of rheumatology at Cardiff University, in Wales, told Healio Rheumatology. “At the moment, we are facing a trial and error approach with RA treatment because we do not have a way of predicting which patient will respond to which treatment. Sarilumab is one of the many treatment options.”
“In phase 3 trials, sarilumab, an IL-6 receptor inhibitor approved for treatment of moderate to severe RA, has been shown to be superior to both placebo — in the MOBILITY and TARGET clinical trials — and the TNF-alpha inhibitor adalimumab — in the MONARCH clinical trial,” he added. “However, the characteristics of patients who are most likely to benefit from sarilumab treatment remain poorly understood.”
To apply machine learning to help improve the way clinicians choose between sarilumab (Kevzara; Regeneron, Sanofi) or adalimumab (Humira, AbbVie) for RA, Choy and colleagues drew data from the sarilumab clinical development program. Using a decision-tree classification approach, and the Generalized, Unbiased, Interaction Detection and Estimation (GUIDE) algorithm, the researchers built predictive models based on the American College of Rheumatology response criteria at week 24 among participants from the MOBILITY trial.
Choy and colleagues identified 17 categorical and 25 continuous baseline variables as candidate predictors, chosen based on subject matter expertise. Predictors included protein biomarkers, disease activity scoring and demographic data. Included endpoints were the ACR20, ACR50 and ACR70 responses at week 24. The researchers validated the resulting rule using independent data sets from the MOBILITY, TARGET and MONARCH trials — which included 1,197, 546 and 369 participants, respectively — as well as on the tocilizumab-calibrator study ASCERTAIN, which included 202 total participants. The researchers’ analysis focused on the 200 mg sarilumab dose.
According to the researchers, the most successful GUIDE model was trained against the ACR20 response. Among the 42 candidate variables, the combined presence of anti-citrullinated protein antibodies and a CRP level of more than 12.3 mg/L was identified as a predictor — or rule — of better treatment outcomes with sarilumab. These rule-positive patients ranged in prevalence from 34% to 51% in sarilumab groups across the four trials and demonstrated more severe disease and poorer prognostic factors at baseline.
The researchers found that rule-positive patients experienced a better outcome than rule-negative participants for most endpoints, with the exceptions of patients with an inadequate response to TNF inhibitors. Additionally, rule-positive patients demonstrated a better response to sarilumab but an inferior response to adalimumab, with the exception of the HAQ-Disability Index minimal clinically important difference endpoint, equivalent to an approximately 5-fold difference in the chance of achieving ACR70 — 34% compared with 7%.
“For RA physicians, you want to balance and find the right treatment for your patients,” Choy said. “This study, through machine learning, looked at different predictors of response to sarilumab, to see what patients may benefit the most from treatment with sarilumab. These findings could help physicians make better informed decisions, especially when it comes to their decision making when considering sarilumab for their patients.”
“We are looking for translation of these findings into everyday practice,” he added. “The next step is to look into routine daily practice as well as this patient population and validate our findings to ensure these are the patients who would achieve a good response.”
Collapse
Choy E. Abstract 2006. Presented at: ACR Convergence 2020; November 5-9, 2020 (virtual meeting).
Read more about
Click Here to Manage Email Alerts