Wearable Sleep Measures May Improve Machine Learning Prediction of Home-based Pulmonary Rehabilitation Engagement...

To evaluate whether incorporating baseline sleep measures from a wrist-worn activity monitor in machine learning (ML) models improved the prediction of 12-week engagement with home-based pulmonary rehabilitation (HBPR) in patients with chronic obstructive pulmonary disease (COPD).

Among participants with a COPD exacerbation (n=124), sleep measures were collected for 1 week before HBPR and processed (1) using a validated Tudor-Locke algorithm and (2) applying partial least squares-discriminant analysis (PLS-DA) to generate the Composite Sleep Health Score. Engagement was defined as completion of one or more recommended activities per week for the 12-week duration. Nested model comparisons for logistic regression, SVM, decision tree, and naïve bayes ML models were performed to determine if including sleep measures improved engagement prediction.

In models adjusted for age, sex, Charlson Comorbidity Index, current smoker status, modified Medical Research Council score, and forced expiratory volume in 1 second, the inclusion of the Composite Sleep Health Score significantly improved the prediction of 12-week engagement only in SVM models (AUC 0.716; p=0.010). Specificity (18.2%) and accuracy (67.7%) also improved by 20.4% and 2.5%, respectively. Including the Score in the logistic regression model yielded the highest predictive performance (AUC = 0.721).

These proof-of-concept findings support additional investigation into the use of wearable-derived sleep measures in parametric ML models to improve screening for HBPR eligibility, identifying patients who will clinically benefit from fully remote PR. Future researchers should carefully select predictors when elucidating the link between wearable sleep measures and HBPR outcomes in COPD patients.