Utilizing Machine Learning to Forecast 3-Month Remission Outcomes in Bipolar Disorder Patients Treated with Lithium.

INTRODUCTION

Lithium remains a first-line mood stabilizer for bipolar disorder; yet, only a subset of patients achieves symptomatic remission. Early prediction of treatment response could guide personalized management. In this study, we leveraged machine learning algorithms to predict 3-month remission, defined as a Montgomery-Åsberg Depression Rating Scale score≤10, in 593 patients with bipolar disorder initiating lithium.

METHODS

In this retrospective cohort, baseline sociodemographic, clinical and laboratory data as well as concomitant medication usage were collected. Montgomery Åsberg Depression Rating Scale and Mania Rating Scale were administered at baseline and 3 months. Data were preprocessed (missing imputation and normalization) and then split into 80% training and 20% test sets. We evaluated various machine learning techniques such as random forest, XGBoost, neural network and support vector machines with five-fold cross validation. Performance metrics included area under the receiver operating characteristic curve and accuracy.

RESULTS

The mean age was 44±16.9 years and 53% of participants were females. The remission rate at 3 months was 44%. The random forest model (augmented by polynomial transformations) performed best (area under the receiver operating characteristic curve=0.76 and accuracy=0.64) improving by 10% of the standard logistic model. Key predictors included the baseline Montgomery Åsberg Depression Rating Scale and Mania Rating Scale, creatinine, thyroid-stimulating hormone levels, body mass index and age.

DISCUSSION: Machine learning, particularly gradient boosted trees, can help to predict the 3-month remission in bipolar disorder patients who start lithium therapy. Incorporating clinical and laboratory features enhances the early identification of likely responders, enabling personalized treatment strategies.