[Validation study of a smartwatch-based risk prediction model for respiratory tract infections]

Objective: To explore the feasibility and accuracy of predicting respiratory tract infections (RTIs) using physiological data obtained from consumer-grade smartwatches. Methods: The study used smartwatches and paired mobile applications to continuously collect physiological parameters while participants slept. A personalized baseline model was established using multi-day data, followed by the construction of RTIs risk prediction algorithm based on deviations from physiological parameter trends. This algorithm converted variations in physiological parameter into quantifiable risk trend scores. Using self-reported RTIs onset as the reference standard, we stratified participants by infection status and assessed the accuracy of the prediction algorithm. Results: A total of 472 participants were enrolled in the study, comprising 272 who developed RTIs (RTIs group) and 200 who remained healthy throughout (non-RTIs group). Key findings included: (1) Significant fluctuations in physiological parameters preceding and following RTIs onset were reliably detected by smartwatch monitoring; (2) A strong temporal correlation was observed between risk trend predictions and self-reported infection onset. The model generated no false-positive high-risk alerts for controls and correctly issued ≥1 high-risk alert within the three-day pre-onset period for 189 RTIs cases. (3) The smartwatch-based prediction model achieved sensitivity of 69.5% (95%CI: 63.7%-74.9%), specificity of 91.3% (95%CI: 86.4%-94.9%), and overall accuracy of 80.4%. Conclusion: Our findings validated the robust predictive capability of the algorithm utilizing consumer-grade smartwatch data for RTIs, supporting the potential utility of wearable technology in the early detection of RTIs.