HCP Live
Contagion LiveCGT LiveNeurology LiveHCP LiveOncology LiveContemporary PediatricsContemporary OBGYNEndocrinology NetworkPractical CardiologyRheumatology Netowrk

Machine Learning Predicts TNF Inhibitor Response in RA

Machine learning algorithms identify patterns without being explicitly programmed to do so, which allows for unbiased discoveries.

Investigators developed machine learning models that were able to predict response to tumor necrosis inhibitors (TNFi) in patients with rheumatoid arthritis (RA) using data available in clinical routine, according to Rheumatic & Musculoskeletal Diseases.1

“With the increasing amount of available data in medicine, new tools are required to extract information,” investigators explained. “Machine learning algorithms learn patterns from data and assume these will reproduce in the future. These algorithms identify patterns and rules without being explicitly programmed to do so, allowing unbiased discoveries. This is especially interesting in medicine to identify markers or combination of markers unknown so far by physicians.”

Data from the RA ESPOIR, a French multicenter, longitudinal, and prospective early arthritis cohort, was utilized to train the models. Patients included in the study received 1 or more TNFi injections and fulfilled the 2019 American College of Rheumatology (ACR)/European Alliance of Associations for Rheumatology (EULAR) criteria for RA.

Endpoints included good or moderate EULAR response, assessed at 12 months after TNFi initiation, and change in the erythrocyte sedimentation rate Disease Activity Score (DAS28) at 12 months. Investigators compared performances of liner regression, random forest, CatBoost, and XGBoost models on the training set and then cross-validated them using either the mean squared error or the area under the receiver operating characteristic curve (AUROC). Of the models, the best-performing model was then analyzed in a replication cohort, ABIRISK, a prospective study used to investigate predictive factors of developing anti-drug antibodies in patients with RA treated with TNFi.

A total of 161 patients from ESPOIR (95 receiving etanercept and 96 receiving an anti-TNF monoclonal antibody) and 118 patients from ABIRISK (68 receiving etanercept and 50 receiving either adalimumab or infliximab) were included in the study. In the ESPOIR cohort, 59% reported a response to TNFi treatment and 61% responded in the validation set (ABIRISK). Key features focused on DAS28, aspartate aminotransferase (ALT), neutrophils, lymphocytes, age, weight, and smoking status.

Of the 4 tested models, CatBoost was able to predict the best EULAR response by reaching an AUROC of 0.72 (0.68–0.73) on the train set in the ESPOIR cohort. Better results were achieved when etanercept and monoclonal antibodies were analyzed separately. In the ABIRISK cohort, the models achieved an AUROC of 0.70 (0.57–0.82) and 0.71 (0.55–0.86), respectively.

Ultimately, CatBoost and random forest had the best performances. Two decision thresholds were tested, with the first prioritizing a high confidence in identifying responders and yielded a confidence up to 90% for predicting TNFi response. The second focused on a high confidence in identifying inadequate responders, which yielded a confidence up to 70% for predicting non-response to TNFi. Changes in DAS28 were predicted with an average error of 1.1 DAS28 points.

Limitations included the relatively small sample size compared with the amount of data usually used in machine learning, which limits the accuracy of algorithms and potentially prevents investigators from drawing solid conclusions. Further, the use of data between the 2 cohorts resulted in varying time between follow-ups. However, their performances were similar for both cohorts, which highlights the strengths of the study design. Lastly, modeling limitations may have resulted in immortal time bias. Recurrent Neural Networks may help to improve the modelling of longitudinal data.

“Focusing on clinical use, we developed a model and assessed its performances in two scenarios, having a high confidence in either identifying TNFi responders or identifying TNFi inadequate responders,” investigators concluded. “Both demonstrated interesting results compared with the current clinical practice and these algorithms pave the way to a personalized treatment strategy in RA.”

Reference:

Bouget V, Duquesne J, Hassler S, et al. Machine learning predicts response to TNF inhibitors in rheumatoid arthritis: results on the ESPOIR and ABIRISK cohorts. RMD Open. 2022;8(2):e002442. doi:10.1136/rmdopen-2022-002442