Navigating the paradoxes and potential of digital phenotyping for bipolar relapse prediction.

This commentary responds to the important study by Ludwig et al. on using smartphone data and critical slowing down (CSD) to predict bipolar disorder (BD) relapse. While commending the study's methodological rigor, we highlight a key paradoxical finding: a decrease in activity variance preceding manic episodes, which challenges the classic CSD model. We posit this may not be a failure of the theory but instead reveals a distinct pre-manic signature of behavioral 'rigidification' rather than instability. Furthermore, we discuss the ambiguity of the 'euthymic' baseline in a clinically complex, medicated population, suggesting that unmeasured pharmacological effects may confound the detected signals. The commentary argues that the limited individual-level predictive power underscores the need to shift from searching for universal nomothetic signals to developing personalized, idiographic (N-of-1) models. Ultimately, we conclude that Ludwig et al.'s work is pivotal in reframing the research agenda towards more nuanced, individualized, and clinically translatable digital phenotyping for BD.