Wearable Technology and Machine Learning for Prediction of Performance-Based and Patient-Reported Outcome Measures: A...

Area of Science:

• Biomedical Engineering
• Health Informatics
• Machine Learning

Background:

• Traditional outcome monitoring in healthcare faces limitations like recall bias and resource constraints.
• Patient-generated data from wearables presents a novel approach for continuous health assessment.
• Wearable technology offers a promising avenue for objective and frequent outcome measurement.

Purpose of the Study:

• To identify wearable-derived features linked to patient-reported and performance-based outcomes.
• To compare the predictive accuracy of various machine learning models using wearable data.
• To outline limitations and suggest future research directions for wearable-based outcome prediction.

Main Methods:

• Systematic review of 18 studies published between 2017-2024 from four major databases.
• Analysis of studies utilizing wearable devices (primarily wrist-worn) measuring accelerometry, heart rate, respiratory, and sleep metrics.
• Comparison of machine learning algorithms including random forest, support vector machines, and hidden Markov models.

Main Results:

• Wearable-derived features show potential in predicting patient outcomes, with predictive performance varying widely (AUC 0.56-0.92).
• Non-linear machine learning models generally outperformed linear models, and temporal models showed promise with longitudinal data.
• Common limitations include small sample sizes, insufficient external validation, and challenges in achieving high accuracy for non-binary predictions.

Conclusions:

• Wearable-informed machine learning holds significant potential for continuous and objective patient outcome assessment.
• Further research necessitates larger, diverse longitudinal datasets and advanced temporal modeling for clinical translation.
• Bridging the gap from proof-of-concept to clinical application requires addressing current methodological limitations.