Overview
This series follows wearable data from sensor measurement through feature design, time-aware analysis, association or detection, transportability, and trustworthy delivery. It shows when evidence supports research or screening—and why diagnosis or improved outcomes require more validation.
Learning outcomes
- A wearable signal is not a diagnosis: sensor output, device estimate, feature, health relationship, and intended use are separate claims.
- Repeated measurements can reveal persistent patterns, but windows, baselines, missingness, and event timing determine what they mean.
- Observational sleep associations and diabetes classifiers support different claims; neither alone proves causality, diagnosis, or improved outcomes.
- Transportability, data integrity, quality control, privacy, and delivery must match the population and decision before clinical use.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
When does a wearable signal become a digital biomarker?
It becomes one for a defined use when its construct, measurement, features, health relationship, applicability, errors, and delivery are validated enough for that use.
Can wearable data diagnose diabetes?
The cited Oura study supports retrospective detection of physiology consistent with self-reported diabetes, not diagnosis or direct glucose measurement.
Why do wearable studies need repeated measurements?
Repeated measurements can separate persistent patterns from nightly variation, but the useful window depends on biology, completeness, comparison frame, and intended use.
Can a wearable finding be used on another device or population?
Not automatically; transport requires checking whether the measurement, errors, labels, missingness, and threshold retain their meaning in the new setting.
What does data integrity prove?
It shows that a stored record matches the value that was hashed, not that the original measurement was accurate or clinically valid.