Interpretable framework for predicting preoperative cardiorespiratory fitness using wearable data.

Area of Science:

• Cardiorespiratory fitness assessment
• Wearable technology in healthcare
• Machine learning in medicine

Background:

• Preoperative cardiorespiratory fitness (CRF) is vital for surgical risk stratification.
• Formal CRF testing (CPET, 6MWT) is often impractical for routine preoperative screening.
• Wrist-worn wearables offer a potential solution for accessible CRF estimation.

Purpose of the Study:

• To develop a clinically interpretable machine learning (ML) model using wearable data to predict CRF in older adults.
• To assess the model's ability to estimate 6-minute walk test (6MWT) distances for preoperative risk evaluation.

Main Methods:

• Collected heart rate and activity data from Fitbit devices over one week in 65 older adults undergoing major noncardiac surgery.
• Employed an ML ensemble regression model to predict CRF, using 6MWT outcomes as the index.
• Utilized Shapley feature attribution to understand wearable data's contribution to CRF prediction.

Main Results:

• Higher CRF correlated with increased moderate-to-vigorous physical activity (MVPA), maximal activity energy expenditure (aEEmax), heart rate recovery (HRR), and non-linear heart rate variability (HRV).
• Random forest and linear regression models showed strong predictive power for CRF (R²=0.91 and R²=0.81).
• Shapley analysis confirmed MVPA, aEEmax, HRR, and HRV dynamics as key indicators of enhanced CRF.

Conclusions:

• Wearable-derived activity and heart rate metrics can provide preoperative CRF assessments.
• This approach supports surgical risk stratification and personalized prehabilitation strategies.
• Integrating wearables can lead to improved patient outcomes through better preoperative evaluations.