Matthew is an associate editor of The American Journal of Managed Care® (AJMC®). He has been working on AJMC® since 2019 after receiving his Bachelor's degree at Rutgers University–New Brunswick in journalism and economics.
Machine learning identified patients with rheumatoid arthritis who may have an increased chance of achieving clinical response with sarilumab, with those selected also showing an inferior response to adalimumab.
Machine learning was shown to identify patients with rheumatoid arthritis (RA) who present an increased chance of achieving clinical response with sarilumab, with those selected also showing an inferior response to adalimumab, according to an abstract presented at ACR Convergence, the annual meeting of the American College of Rheumatology (ACR).
In prior phase 3 trials comparing the interleukin 6 receptor (IL-6R) inhibitor sarilumab with placebo and the tumor necrosis factor α (TNF-α) inhibitor adalimumab, sarilumab appeared to provide superior efficacy for patients with moderate to severe RA. Although promising, the researchers of the abstract highlight that treatment of RA requires a more individualized approach to maximize efficacy and minimize risk of adverse events.
“The characteristics of patients who are most likely to benefit from sarilumab treatment remain poorly understood,” noted researchers.
Seeking to better identify the patients with RA who may best benefit from sarilumab treatment, the researchers applied machine learning to select from a predefined set of patient characteristics, which they hypothesized may help delineate the patients who could benefit most from either anti–IL-6R or anti–TNF-α treatment.
Following their extraction of data from the sarilumab clinical development program, the researchers utilized a decision tree classification approach to build predictive models on ACR response criteria at week 24 in patients from the phase 3 MOBILITY trial, focusing on the 200-mg dose of sarilumab. They incorporated the Generalized, Unbiased, Interaction Detection and Estimation (GUIDE) algorithm, including 17 categorical and 25 continuous baseline variables as candidate predictors. “These included protein biomarkers, disease activity scoring, and demographic data,” added the researchers.
Endpoints used were ACR20, ACR50, and ACR70 at week 24, with the resulting rule validated through application on independent data sets from the following trials:
Assessing the end points used, it was found that the most successful GUIDE model was trained against the ACR20 response. From the 42 candidate predictor variables, the combined presence of anticitrullinated protein antibodies (ACPA) and C-reactive protein >12.3 mg/L was identified as a predictor of better treatment outcomes with sarilumab, with those patients identified as rule-positive.
These rule-positive patients, which ranged from 34% to 51% in the sarilumab groups across the 4 trials, were shown to have more severe disease and poorer prognostic factors at baseline. They also exhibited better outcomes than rule-negative patients for most end points assessed, except for patients with inadequate response to TNF inhibitors.
Notably, rule-positive patients had a better response to sarilumab but an inferior response to adalimumab, except for patients of the HAQ-Disability Index minimal clinically important difference end point.
“If verified in prospective studies, this rule could facilitate treatment decision-making for patients with RA,” concluded the researchers.
Rehberg M, Giegerich C, Praestgaard A, et al. Identification of a rule to predict response to sarilumab in patients with rheumatoid arthritis using machine learning and clinical trial data. Presented at: ACR Convergence 2020; November 5-9, 2020. Accessed January 15, 2021. 021. Abstract 2006. https://acrabstracts.org/abstract/identification-of-a-rule-to-predict-response-to-sarilumab-in-patients-with-rheumatoid-arthritis-using-machine-learning-and-clinical-trial-data/