Machine learning shows promise for predicting TNF inhibitor response
medwireNews: A machine-learning model may help to predict which rheumatoid arthritis (RA) patients are likely to respond to treatment with tumor necrosis factor (TNF) inhibitors, researchers report.
“In current practice, patients who respond inadequately to conventional therapies usually receive [TNF inhibitor] drugs,” which are “mainly chosen on a ‘trial-and-error’ basis, and about 30% of the patients respond poorly,” say Yuanfang Guan (University of Michigan, Ann Arbor, USA) and study co-authors.
They explain that “patient heterogeneity hinders identification of predictive biomarkers and accurate modeling of anti-TNF drug responses,” and therefore “effective modeling of patient heterogeneity is the key to accurate drug response prediction.”
Guan and team used a training dataset from 1892 patients of European ancestry to develop a Gaussian process regression (GPR) model that combined demographic, clinical, and genetic features, in order to predict change in DAS28 score and categorize patients into TNF inhibitor responders or nonresponders.
They say that the model was “the best-performing model in predicting anti-TNF response in the DREAM Rheumatoid Arthritis Responder Challenge,” held in 2014.
As reported in Arthritis & Rheumatology, the model predicted change in DAS28 scores among patients in the training dataset with a correlation coefficient of 0.406, and correctly classified 77.8% of individuals as TNF inhibitor responders or nonresponders, with an area under the receiver operating characteristic curve (AUC) of 0.653.
And in an independent dataset of 680 patients from the CORRONA registry, Guan and colleagues’ model predicted DAS28 changes with a correlation coefficient of 0.393, and had an AUC of 0.615 for the identification of TNF inhibitor responders.
“Compared to traditional trial-and-error practice, our model can help up to 40% of European-descent anti-TNF non-responders avoid ineffective treatments,” write the researchers.
They caution, however, that “[c]onsidering the heterogeneity of the anti-TNF responses among rheumatoid arthritis patients, we do not expect the model to achieve a similar performance on other [non-European] populations.”
Therefore, “[e]xtension of the model over other populations requires new patient data and separate feature selection,” they add.
Looking ahead, Guan et al say that they envision future research using “more diverse, and bigger population[s]” in order to develop “more robust models” for predicting TNF inhibitor response.
medwireNews is an independent medical news service provided by Springer Healthcare. © 2019 Springer Healthcare part of the Springer Nature group