Sex and menstrual cycle differences in the mood-activity association should inform cycle-aware digital phenotyping

Abstract

Passive physical activity is increasingly used as a digital-phenotyping proxy for mood, assuming a stable relationship across people and time. We tested this assumption using daily mood and activity data from 1,072 individuals (121 women, 951 men; 13,909 person-days) in the Juli app. The within-person activity-mood association was stronger in women than men (slope difference -0.194, p = 0.019). Within women, it varied across the menstrual cycle (p = 0.022): absent in the early luteal phase, significant in all other phases, and largest in the late luteal phase (+0.46). A male pseudo-cycle control showed no such modulation (p = 0.974), confirming the effect is cycle-specific rather than a general temporal pattern. Accounting for cycle phase improved out-of-sample mood prediction in women in 75% of cross-validation splits. Digital phenotyping should account for sex and menstrual cycle phase to avoid biased predictions and enable personalized recommendations for women.