Passive sensing with psyche: utilizing data from wearable technology to predict emotion states

Depression and anxiety are some of the most common mental health disorders in the world contributing to significant morbidity and mortality. Past treatments have focused primarily on treating depression and anxiety. However, there is an urgent need to detect chronic stress states and potentially intervene using just-in-time personalized interventions. Modern technology has revolutionized our ability to passively measure various biological and physiological signals. In our daily lives, we generate significant amounts of electronic data from our phones, wearable technology, watches, and even computers and cars. In this analysis, we focus on using wearable data from FitBit to passively predict daily mood states (e.g., sad/tense/anxious vs. happy). We use daily FitBit data from 38 participants and ~1200 days of data to predict mood states (e.g., sad/tense/anxious vs. happy) on a day-to-day basis using an elastic net regression machine learning algorithm. We were able to accurately predict these states using a cross-validated machine learning algorithm and identified features predictive of each of the mood states. In this proof-of-concept analysis, we show that predicting daily mood states is feasible and may help to not only detect daily mood states but also improve passive awareness and deliver just-in-time interventions.