Machine Learning Model Can Predict Anti-TNF Drug Response in RA

The model improves anti-TNF drug selection and the effect of genetic markers across multiple cohorts.

A Gaussian process regression (GPR) model can help identify patients with rheumatoid arthritis (RA) who do not respond well to anti-tumor necrosis factor (anti-TNF) treatments, according to study results published in Arthritis & Rheumatology.

The researchers used the GPR model that won the first place in the Dialogue on Reverse Engineering Assessment and Methods (DREAM): Rheumatoid Arthritis Responder Challenge. This model can combine demographic, clinical, and genetic markers, predict changes in disease activity scores 24 months after baseline assessment, and identify nonresponders to anti-TNF treatments.

Related Articles

The researchers included 1892 patients to develop and cross-validate the model, and another dataset of 680 patients to evaluate the model. The researchers assessed the efficacy of the similarity modeling and the contribution of individual features.

In the cross-validation tests, the results indicated that the model predicted changes in disease activity scores with a correlation coefficient of 0.406. The model correctly classified anti-TNF treatment response for 78% of patients.

In the independent test, the model achieved a Pearson correlation coefficient of 0.393 in predicting delta disease activity score (ΔDAS).

The researchers found that baseline disease activity scores had the highest correlation coefficient against ΔDAS (0.370). However, this alone did not explain the model’s performance; for a more accurate prediction, the model incorporated other demographic, clinical, and genetic features through its kernel function. In addition to disease activity, important features such as age, methotrexate usage, and genetic markers played a significant role in the model’s prediction.

“For the future work, various clinical markers may be potentially used for more accurate identification of nonresponding subpopulations that carry predictive biochemical traits,” the researchers wrote.


Guan Y, Zhang H, Quang D, et al. Machine learning to predict anti-TNF drug responses of rheumatoid arthritis patients by integrating clinical and genetic markers [published online July 24, 2019]. Arthritis Rheumatol. doi:10.1002/art.41056