Forecasting depressive relapse in Bipolar Disorder from clinical data

Bipolar disorder (BD) is a mood disorder characterized by recurrent episodes of depression and mania/hypomania. Depressive relapse in BD reach rates close to 50% in 1 year and 70% in up to 4 years of treatment. Several studies have been developed to discover more efficient treatments for BD and prevent relapses. However, most of relapse studies used only statistical methods. We aim to analyze the performance of machine learning algorithms in predicting depressive relapse using only clinical data from patients. Five well-used machine learning algorithms (Support Vector Machines, Random Forests, Naive Bayes and Multilayer Perceptron) were applied to the Systematic Treatment Enhancement Program for Bipolar Disorder (STEP-BD) dataset of a cohort of 800 patients who became euthymic during the study and were followed up for 1 year: 507 presented a depressive relapse and 293 did not. The algorithms showed reasonable performance in the prediction task, ranging from 61% to 80% in the F-measure. Random Forest algorithm had a higher average of performance (Relapse Group 68%; No Relapse Group 74%), although, the performance between classifiers showed no significant difference. Random Forest analysis demonstrated that the three most important mood symptoms observed were: interest, depression mood and energy. Results show that the machine learning algorithms could be seen as a sensible approach to better support medical decision-making in the BD treatment and prevention of future relapses.