Wearable Step Trackers Outperform Self-Reported Exercise in Predicting Health in Older Adults

As wearable devices become ubiquitous, health researchers face a practical challenge: raw device data must be carefully processed before it can support rigorous scientific conclusions. This is especially true in aging cohorts, where activity patterns, wear compliance, and data quality can vary considerably across participants.

This study from Duke University introduces a transparent, reproducible standard operating procedure (SOP) for converting epoch-level step-count data from commercial Garmin devices into participant-level analytic variables. The pipeline standardizes timestamps, reconstructs daily activity grids, infers wear time from observed step patterns, and applies a prespecified valid-day threshold of at least 10 hours of inferred wear time. It was tested in the STRRIDE-PD Reunion study, a long-term follow-up of older adults who had previously completed a supervised exercise intervention trial.

Among 67 participants (mean age 71.4 years, 65.7% women), the median valid-day count was 10 days, and median average daily steps were 5,794. Crucially, participant-level estimates were identical whether a 10-hour or 6-hour wear-time threshold was used, suggesting robustness. Wearable-derived step counts were significantly associated with 9 of 16 cardiometabolic and fitness outcomes, including cardiorespiratory fitness, body composition, and lipid profiles. By contrast, self-reported exercise—assessed via a frequency-by-duration composite ranked into deciles—was not significantly associated with any outcome.

Regression calibration analysis suggested that measurement error inherent in self-reported exercise may have substantially attenuated its apparent associations with health outcomes. This underscores how measurement methodology can distort or obscure genuine physical activity-health relationships.

Limitations include a modest analytic sample size, a relatively short monitoring duration, and the use of indirect wear-time inference rather than direct sensor-based detection. Nonetheless, the findings make a compelling case that reproducible, transparent wearable data pipelines are essential infrastructure for aging epidemiology and should be adopted as a standard practice in future longitudinal studies.