Optimizing Temporal Windows for Wearable-Augmented Post-Discharge Risk Prediction: A Methods Study

ObjectiveTo identify optimal modeling parameters for dynamically predicting hospital readmission risk using post-discharge step-count data from remote monitoring devices. MethodsWe combined data from two clinical studies that collected wearable or smartphone-based activity data for up to 6 months after hospital discharge. Analyses were limited to older adults ([&ge;]55 years). We constructed a patient-day dataset incorporating static demographic and clinical variables and dynamic activity features aggregated over retrospective windows of 3, 5, 7, or 10 days. Models predicted a composite outcome of readmission or death over prospective horizons of 3, 5, 7, or 10 days, within follow-up periods of 30-180 days. Logistic regression and LightGBM models were trained using 5-fold cross-validation on an 80:20 patient-level split. ResultsAmong 215 participants, LightGBM outperformed logistic regression across all configurations (mean AUC 0.82 vs 0.76). Performance improved with longer prospective horizons but was largely insensitive to retrospective window length. The LightGBM model was well-calibrated (Hosmer-Lemeshow {chi}2 = 2.46, p = 0.96), whereas logistic regression showed miscalibration ({chi}2 = 51.8, p < 0.001). In feature-importance analyses, LightGBM ranked static (length of stay, vitals, BMI) and dynamic (recent steps, distance) features highly, whereas logistic regression emphasized activity-based variables. DiscussionPrediction performance was impacted by horizon length and training window, with minimal effect of retrospective window. LightGBM achieved higher discrimination and better calibration, supporting flexible, non-parametric methods for post-discharge risk prediction. ConclusionPost-discharge activity data enhance readmission-risk prediction. Selecting practical temporal windows and appropriate model types can improve accuracy and calibration in wearable-augmented risk models.