ACTIVE AND PASSIVE DIGITAL PHENOTYPING TO MEASURE OUTCOMES IN A 12-MONTH STUDY OF XANOMELINE/ TROSPIUM CHLORIDE IN SCHIZOPHRENIA: LONGITUDINAL PREDICTION OF TOTAL CHANGES WITH EARLY LEVELS OF PHYSICAL ACTIVITY

Active and passive digital phenotyping hold substantial promise for refining measurement of treatment outcomes in clinical trials. These indices include actigraphy, global positioning, facial and vocal expression, as well as device-based ecological momentary assessment (EMA) surveys. In a large-scale (n=350) treatment study, we previously found that actigraphy-based step counts increased and the severity of indicators of avolition indexed with EMA decreased, with substantial intercorrelations. Previous observational digital phenotyping studies had shown that higher step counts longitudinally predicted better cognitive performance. Further, those studies also indicated that greater within-participant baseline variance in EMA-measured productive activities longitudinally predicted better functional outcome. Following the strategies of those earlier observational studies, step-count data collected early in a 12-month open label study of xanomeline/trospium chloride were evaluated for utility to predict overall EMA-measured treatment changes, with early variability in EMA survey responses as examined as additional predictors. Participants were 312 male and female outpatients with schizophrenia answering EMA surveys 3 times per day, 7 days per week, one week per month. On EMA survey days, participants wore an actigraph. EMAs queried location, social context, and activities. EMAmeasured activities targeted behavioral indicators of avolition and were separated into recumbent, seated, standing, and moving. Participants answered 33,657 EMA surveys and provided 7957 participant-days of actigraphy. There were treatment-related reductions over 12 months in being home and alone, as well as decreases in sedentary activities and increases in standing and moving, (all X2 > 29.73, p < .002). Steps increased in the same time period, (X2 =33.10, p < .001). Increased step counts correlated with decreased sedentary activities and increases in standing and moving over the treatment period, (all X2 > 40.59, p < .001). HLM analyses indicated that step counts in the first 2 months predicted overall changes in all 4 EMA activity variables, (all X2 > 6.49, p < .01), above and beyond the effects of time in treatment. Early within-participant variance in activity variables did not predict longitudinal changes, but within-participant variance decreased for all EMA variables (all X2 > 51.57, all p < .001) and reductions in variance were concurrently correlated with decreased severity of sedentary activities and increased standing and moving over the entire treatment period (All X2 > 228.48, p < .001). Indices of physical activity measured early in a 12-month treatment study with passive digital strategies predicted study-long treatment response, suggesting that physical activity is a factor worth considering in future studies. Behavioral indices of avolition, indexed by their level of physical demands, decreased both in severity and in within-subject variance, suggesting improvements across both severity and variability indices. These two factors can index both the magnitude and momentary consistency of treatment response. Concurrent measurement of both level and variance of different treatment outcomes would not be possible without momentary digital phenotyping. The convergence of easily collected passive digital data and active EMA survey data allow for broader assessment of treatment related improvements in everyday activities and functioning than legacy clinical assessment strategies and may predict be able to measure long-term outcomes of developing treatment strategies.