Predicting Drug Response in Rheumatoid Arthritis, Ankylosing Spondylitis

Although a machine-learning model proved better at predicting biologic disease-modifying antirheumatic drug (bDMARD) response in patients with rheumatoid arthritis, machine learning did not outperform a conventional statistical method for predicting bDMARD response in patients with ankylosing spondylitis. Researchers published their findings in the journal Arthritis Research & Therapy.

“Given the input clinical features, machine-learning models have no advantages compared to a logistic regression model in patients with ankylosing spondylitis,” wrote a research team from Sungkyunkwan University, Seoul, South Korea.

The team investigated machine learning to predict patient response to bDMARDs in training datasets of 625 patients with rheumatoid arthritis and 611 patients with ankylosing spondylitis. The performance of various machine-learning models was compared with that of the logistic regression model, a conventional statistical method.

For patients with rheumatoid arthritis, the random forest method predicted treatment responses more accurately than logistic regression. Researchers reported 0.726 accuracy and 0.638 area under curve (AUC) of the receiver operating characteristic curve with the random forest method, compared with 0.689 accuracy and 0.565 AUC of the receiver operating characteristic curve with logistic regression.

However, machine learning and logistic regression performed similarly in patients with ankylosing spondylitis. The specific reason why machine learning’s predictive performance was lower in patients with ankylosing spondylitis was beyond the study scope, researchers wrote.

For both patient populations, self-report scales—the patient global assessment of disease activity for rheumatoid arthritis and the Bath Ankylosing Spondylitis Functional Index for ankylosing spondylitis—were the most important input factors for machine-learning prediction, the study showed.

“It is quite surprising,” researchers observed, “because they are more important than more objective clinical features, such as laboratory results (erythrocyte sedimentation rate and C-reactive protein) and physical examination (swollen joint count and tender joint count).”

Reference:
Lee S, Kang S, Eun Y, et al. Machine learning-based prediction model for responses of bDMARDs in patients with rheumatoid arthritis and ankylosing spondylitis. Arthritis Res Ther. 2021 Oct 9;23(1):254. doi:10.1186/s13075-021-02635-3