Evidence-Based Study on Low-Burden Digital Phenotyping for Precision Screening of Oral Anti-Obesity Drug Efficacy

To enable scalable and accurate early screening of oral anti-obesity medications (AOMs), we propose a low-burden digital phenotyping framework based on long-term patient-submitted data, including step count, heart rate, sleep summary, short diet frequency, and patient-reported outcomes (PROs). A cohort of 400 overweight or obese individuals was monitored for 24 weeks. A "behavioral rhythm–metabolic risk" feature set, comprising evening activity ratio, sleep consistency, postprandial subjective hunger, and mood/stress entries, was constructed. The hierarchical GBT + TFT model predicted 12-week weight loss ≥5% with high accuracy, achieving an AUC of 0.88 on external time-window validation and 0.83 using only the first 14 days of data, supporting early screening and rapid feedback. This study demonstrates that low-burden digital phenotyping can serve as a core capability for precision screening of AOMs and provides a practical pathway for patient-side applications and stratified follow-up.