Personalized AI-Driven Modeling of Longitudinal Physiological Trajectories From Wearable Devices for Early Prediction of Respiratory Disease in Initially Healthy Adults

Respiratory diseases often emerge with subtle physiological alterations before clear clinical symptoms appear, creating an important window for early detection. Wearable devices enable continuous, noninvasive monitoring of parameters such as heart rate, respiratory rate, oxygen saturation, skin temperature, sleep patterns, activity, and heart rate variability; however, many existing prediction models rely on population-level thresholds and insufficiently account for inter-individual variability in baseline physiology and temporal adaptation.This study aims to develop and evaluate a personalized artificial intelligence framework for modeling longitudinal physiological trajectories derived from wearable devices to support early prediction of respiratory disease in initially healthy adults.A longitudinal observational design is proposed using wearable-derived physiological data collected from healthy adults over extended follow-up. After signal quality control, temporal alignment, smoothing, missing-data handling, and subject-specific normalization, multivariate trajectory features will be extracted to characterize gradual deviations from each individual’s baseline. A personalized AI model will then be compared with population-level and hybrid models for prediction of respiratory disease onset within a predefined future window. Model performance will be evaluated using AUROC, sensitivity, specificity, F۱-score, calibration, and lead time to detection, consistent with STROBE and TRIPOD+AI reporting principles.The study is designed to test whether personalized trajectory-based modeling can improve discrimination and provide earlier warning than snapshot-based or non-personalized approaches, with performance summarized through AUROC, sensitivity, specificity, and median lead time before diagnosis.Personalized modeling of gradual physiological change from wearable sensors represents a promising strategy for passive, scalable, and earlier detection of respiratory illness in healthy adult populations, particularly when individual baselines and temporal dynamics are explicitly incorporated into the prediction framework.

کلیدواژه ها:

Artificial Intelligence ، Wearable Devices ، Personalized Modeling ، Longitudinal Physiological Trajectories ، Respiratory Disease Prediction ، Early Detection ، Digital Biomarkers