Multivariate prediction of temper outbursts in youth enriched for irritability using Ecological Momentary Assessment data

Irritability and temper outbursts are among the most common reasons youth are referred for psychiatric assessment and care. Identifying clinical variables (e.g. momentary anxiety) that precede the onset of temper outbursts would provide valuable clinical utility. Here, we provide the rationale for a study to test the performance of classifiers trained to predict temper outbursts in a group of clinically-referred youth, in a home setting, enriched for symptoms of irritability and temper outbursts. Using observational data--digital based event sampling from previous Ecological Momentary Assessment data, we demonstrate promising results in our ability to predict the presence of a temper outburst based on clinical responses (e.g., whether the participant is grouchy, hungry, happy, sad, anxious, tired, etc.) prior to the emotional event, as well as external features (e.g., time of day, day of week). In exploratory analyses of existing data, consisting of n=57 subjects with a total of 1296 time points, we evaluate the feasibility of using a logistic regression-based classifier and a random-forest based classifier for predicting the temper outburst prospectively. In order to more rigorously assess these classifiers, we propose the collection of a large confirmatory set, consisting of at least an additional 20 subjects with an expected total of 400 time points, and will perform confirmatory analyses of the precision and recall of several classifiers for predicting temper outbursts. This work provides the foundation for the identification of features predictive of risk and future development of novel mobile-device-based interventions in youth affected with severe and impairing psychopathology.