This page was translated automatically from Turkish. Read the original

This page is an AI-assisted Turkish summary of the study presented at the 26th World Congress of Psychiatry. The study belongs to the presenting authors; please refer to the original source when citing it.

Machine learning to support time-dependent treatment decisions in psychotic depression: is earlier prediction possible?

Using machine learning to support time-dependent treatment decisions in psychotic depression - is it possible to make earlier predictions?

Machine learning models trained on weekly symptom data can predict treatment response and remission in psychotic depression with clinically useful accuracy as early as the third week.

The study aimed to investigate whether machine learning (ML) models could predict treatment response and remission earlier during pharmacotherapy for psychotic depression and how predictive performance changed over the course of treatment. Data were drawn from the Dutch University Depression Group randomized clinical trial, which included 119 patients with psychotic depression treated with imipramine, venlafaxine, or venlafaxine plus quetiapine. Predictors included Hamilton Depression Rating Scale (HDRS) scores over time, psychotic symptoms, demographic variables, and previous depressive episodes; the models were trained using stratified 10-fold cross-validation, and weekly models were constructed by progressively adding symptom data from preceding weeks. Outcomes were defined as treatment response (a reduction of more than 50% in HDRS score) and remission (an HDRS score of 7 or lower) at week 7; performance was assessed using the area under the receiver operating characteristic curve (AUC) and mean absolute error (MAE). For remission, the AUC increased from 0.65 at baseline to 0.80 at week 2 and exceeded 0.90 from week 3 onward; for response, the AUC rose from 0.57 at baseline to 0.74 at week 2 and reached 0.96 at week 5. Prediction error for final HDRS scores decreased progressively over the weeks (MAE declined from 8.80 at baseline to 0.49 at week 6), and remission was generally easier to predict than response. The authors note that clinically useful predictive accuracy can be achieved as early as week 3 and that ML-based decision support tools may help identify ineffective treatments earlier.

All WCP 2026 summaries