Exploiting Unsupervised Free-Living Data for Cardiorespiratory Fitness Estimation: Systematic Review and...

Area of Science:

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

Background:

• Traditional cardiorespiratory fitness (CRF) assessments may not be suitable for frail individuals.
• CRF estimation using free-living wearable data offers a potentially more representative and accessible approach.
• Wearable data captured over time could enhance clinical usability of CRF assessments.

Purpose of the Study:

• To systematically review the evidence for estimating CRF from free-living wearable data.
• To evaluate the performance and quality of models developed for this purpose.
• To assess the potential for clinical implementation of these novel CRF estimation methods.

Main Methods:

• Systematic literature search following PRISMA guidelines across four databases (MEDLINE, Embase, Scopus, arXiv).
• Inclusion of studies developing models for CRF estimation from continuous free-living wearable data; exclusion of laboratory-based studies.
• Meta-correlation analysis using a random-effects model and Fisher Z transformation to pool performance metrics.

Main Results:

• 18 studies with 31,072 participants met eligibility criteria; mean age 46.9 years.
• Machine learning models were employed in 8 studies; pooled meta-correlation estimate was 0.83 (95% CI 0.77-0.88).
• High heterogeneity (I2=97%) and risk of bias in data analysis were observed, with concerns in data handling clarity.

Conclusions:

• A promising preliminary agreement exists between predicted and measured CRF values from wearable data.
• High heterogeneity and lack of external validation prevent definitive conclusions for clinical implementation.
• Continuous wearable data streams represent a valuable resource with the potential to significantly advance CRF measurement and monitoring.