Smartphone-based digital phenotyping to predict emerging affective episodes in bipolar patients: a 12-month ambulatory study

Preventing affective recurrences through the monitoring of early warning signs (e.g., changes in sleep or activity) enabling timely interventions is central to the management of bipolar disorders. Traditionally, early warning sign monitoring has relied heavily on patient’s ability for introspection. However, 25–50% of patients fail to recognize prodromal symptoms before episodes. Thus, digital phenotyping—the continuous collection of smartphone-based sensor and interaction data—offers a promising bridge to this diagnostic gap.

To identify digital predictors for emerging episodes, we conducted secondary analyses on the BipoSense dataset, which comprises data of 29 BD patients over 12 months. Our “ground truth” for psychopathological status was established through bi-weekly SCID expert interviews over the whole year, paired with continuous smartphone-based digital phenotyping. We captured 20 (hypo)manic and 30 depressive episodes throughout the study period and evaluated various activity and communication parameters using diverse analytical frameworks, including the phenomenon of Critical Slowing Down (from dynamical systems theory) and Statistical Process Control (adapted from industrial quality monitoring).

This talk will summarise the results of these distinct approaches. While several parameters yielded statistically significant signals, our findings indicate that these predictors currently provide insufficient sensitivity and specificity for direct clinical application. Isolated metrics appear inadequate for capturing the complex, multi-faceted nature of BD. Consequently, future studies should involve multivariate analyses that synthesize multiple data streams and the development of digital markers more closely mapped to specific psychopathological constructs. This synthesis is essential to move from theoretical promise to a tool that reliably supports personalised, proactive care.