Bayesian network analysis uncovers physical activity-mood dynamics: Insights from the DiAPAson study.
Individuals with schizophrenia spectrum disorders (SSDs) frequently exhibit low levels of physical activity (PA) and mood disturbances, both of which contribute to functional impairment and poorer long-term outcomes. Despite growing evidence linking PA and affective regulation, little is known about the real-time, bidirectional relationship between these domains in SSD.
In this multicenter observational study (DiAPAson project), 120 patients with SSD and 113 age- and sex-matched healthy controls (HC) underwent a 7-day ecological assessment combining smartphone-based Ecological Momentary Assessment (EMA; 8 prompts/day) and continuous wrist-worn actigraphy. We analyzed the dynamic associations between PA and mood using generalized linear mixed models (GLMMs) and Bayesian models with lagged temporal structures.
GLMMs revealed significantly lower daily mood in SSD compared to HC (estimate = -0.33, 95% CI: -0.53 to -0.12; p = .002), but no significant day-level association between PA and mood. In contrast, Bayesian models uncovered robust within-day, bidirectional associations in HC such that higher PA levels were followed by higher subsequent mood and, conversely, better mood predicted higher subsequent PA (PA → mood: posterior probability = 99.9%; mood → PA: 89.6%). In SSD, these within-day couplings were attenuated and more heterogeneous across individuals (PA → mood: 87.2%; mood → PA: 67.5%).
PA and mood are dynamically and bidirectionally linked within short temporal windows throughout the day, with stronger coupling in HC than in SSD. The substantial individual variability observed in SSD highlights the need for personalized, sensor-informed interventions, as aggregated analyses may obscure clinically relevant microtemporal dynamics.
In this multicenter observational study (DiAPAson project), 120 patients with SSD and 113 age- and sex-matched healthy controls (HC) underwent a 7-day ecological assessment combining smartphone-based Ecological Momentary Assessment (EMA; 8 prompts/day) and continuous wrist-worn actigraphy. We analyzed the dynamic associations between PA and mood using generalized linear mixed models (GLMMs) and Bayesian models with lagged temporal structures.
GLMMs revealed significantly lower daily mood in SSD compared to HC (estimate = -0.33, 95% CI: -0.53 to -0.12; p = .002), but no significant day-level association between PA and mood. In contrast, Bayesian models uncovered robust within-day, bidirectional associations in HC such that higher PA levels were followed by higher subsequent mood and, conversely, better mood predicted higher subsequent PA (PA → mood: posterior probability = 99.9%; mood → PA: 89.6%). In SSD, these within-day couplings were attenuated and more heterogeneous across individuals (PA → mood: 87.2%; mood → PA: 67.5%).
PA and mood are dynamically and bidirectionally linked within short temporal windows throughout the day, with stronger coupling in HC than in SSD. The substantial individual variability observed in SSD highlights the need for personalized, sensor-informed interventions, as aggregated analyses may obscure clinically relevant microtemporal dynamics.
Authors
Caselani Caselani, Carnevale Carnevale, Zarbo Zarbo, Martinelli Martinelli, Martella Martella, Magno Magno, de Girolamo de Girolamo, Calza Calza,
View on Pubmed