Feed

Negative affect (NA), encompassing heightened states of sadness, anxiety, and guilt, is a key symptom across a spectrum of internalizing disorders. Recent advancements in digital phenotyping (DP) and machine learning (ML) may enable the automatic detection of short-term fluctuations in NA through digital phenotyping, a prerequisite for Just-In-Time Adaptive Interventions (JITAI). Evidence on the prediction of momentary NA with DP is sparse, but it indicates that personalized ML models are required to account for individual heterogeneities. This preregistered study is the first to analyze data from the PREACT-digital project, encompassing 242 outpatient subjects diagnosed with internalizing disorders. We examined whether passive sensor data (heart rate, steps, mobility, physical activity) could predict momentary NA, as assessed via ecologically momentary assessments (EMA). Personalized and population-based ML approaches were trained on 19,792 pairs of DP data and NA ratings. We found that personalized ML approaches substantially outperformed population-based models. The best model, however, only marginally exceeded the benchmark, predicting per-person mean NA. Our findings emphasize the need for personalized ML in DP studies. Future efforts could incorporate richer or more raw data streams or test sequential modelling approaches to help clarify whether DP and personalized ML could reliably inform just-in-time, data-driven support for individuals affected by internalizing disorders.