Clinical predictors of response to methotrexate in patients with rheumatoid arthritis: a machine learning approach using clinical trial data

Table 2 Model performances on the training (N = 365) and test (N = 410) sets. Two models were investigated. The “DAS28-ESR” model consisted of baseline DAS28-ESR, age, sex, race, RA duration, RF status, ACPA status, glucocorticoid use, and HAQ score. The “components of DAS28-ESR” model consisted of TJC28, SJC28, ESR, PtGA, CRP, PhGA, age, sex, race, RA duration, RF status, ACPA status, glucocorticoids use, and HAQ score. High sensitivity values observed in both LASSO models using the test set implied both models performed well in identifying those who were good responders to methotrexate. For calculating sensitivity and specificity, the cutoff was set to 0.5, with predictions greater than or equal to 0.5 classified as the “good” responder group

Algorithm	Model	Training set (N = 365, 2 RCTs)						Test set (N = 410, 2 RCTs)
Algorithm	Model	AUC (95% CI)	Sensitivity	Specificity	Accuracy	PPV	NPV	AUC (95% CI)	Sensitivity	Specificity	Accuracy	PPV	NPV
LASSO	DAS28-ESR	0.76 (0.71, 0.81)	0.73	0.66	0.70	0.72	0.67	0.79 (0.74, 0.84)	0.83	0.61	0.78	0.87	0.53
LASSO	Components of DAS28-ESR	0.77 (0.72, 0.81)	0.75	0.65	0.70	0.72	0.68	0.79 (0.74, 0.84)	0.86	0.58	0.79	0.86	0.56
Random forests	DAS28-ESR	0.96 (0.94, 0.98)	0.97	0.96	0.96	0.97	0.96	0.68 (0.62, 0.73)	0.81	0.55	0.75	0.85	0.48
Random forests	Components of DAS28-ESR	1	1	1	1	1	1	0.68 (0.63, 0.74)	0.77	0.60	0.73	0.86	0.45

Abbreviations: AUC, area under the curve; DAS28-ESR, Disease Activity Score including 28-joint counts and erythrocyte sedimentation rate, LASSO, least absolute shrinkage and selection operator, NPV, negative predictive value, PPV, positive predictive value, RCTs, randomized clinical trials

ISSN: 1478-6362