Predicting short term mood developments among depressed patients using adherence and ecological momentary assessment...

Area of Science:

• Digital mental health
• Computational psychiatry
• Machine learning in healthcare

Background:

• Technology-driven interventions generate detailed patient data, including ecological momentary assessments (EMA) and response times.
• Significant inter-patient variability in mood patterns presents challenges for predicting future mental states.
• Accurate mood prediction can enable personalized interventions and proactive mental healthcare strategies.

Purpose of the Study:

• To investigate the feasibility of short-term mood prediction using patient data.
• To evaluate the predictive performance of different modeling approaches (individual, group, single model).
• To assess the added value of adherence and usage data as predictors for mood fluctuations.

Main Methods:

• Recurrent neural networks (RNNs) were applied to handle temporal data dynamics.
• Data from 143 patients diagnosed with major depressive disorder (DSM-IV) across five countries were utilized.
• Root Mean Squared Error (RMSE) was used to measure prediction accuracy.

Main Results:

• Short-term mood change predictions achieved notable accuracy (RMSE 0.065-0.11).
• Past EMA mood ratings were the strongest predictors of future mood states.
• Adherence and usage data did not significantly enhance prediction accuracy.

Conclusions:

• Sophisticated machine learning models can achieve accurate short-term mood predictions.
• Group-level predictive models showed promise, though differences were not statistically significant.
• Further research incorporating mobile phone data may improve predictive performance for mood disorders.