How Quickly Can You Know a Participant? A 3-Week Triage Point for Oura Ring Adherence in College Students

Longitudinal wearable studies lose statistical power when participants disengage, yet most protocols lack empirically derived guidance on when to flag at-risk participants for compliance support. We analyzed Oura Ring data from 584 first-year college students across two semesters (Fall 2022, ~8 weeks; Spring 2023, ~15 weeks; 442 continued into Spring, 408 with analyzable data in both early and later windows of the semester) to identify when passive wear data becomes informative for prioritizing engagement support. We built a logistic regression on six early-window wear-time features with 5-fold stratified cross-validation. At week 3, the classifier identified bottom-quartile (Fall AUC = 0.81, Spring AUC = 0.85) and top-quartile (Fall AUC = 0.82, Spring AUC = 0.86) adherence trajectories. A Fall-trained bottom-quartile classifier applied to Spring data without retraining achieved AUC = 0.84 [0.79, 0.89]. Cross-validated permutation importance identifies average early-window wear as the dominant predictor in both semesters, a ranking that holds under a gradient-boosted alternative. Three weeks of passive wear data is sufficient to prioritize engagement support in this cohort; whether intervening at that point improves retention requires prospective evaluation.