Journey To The Right Digital Endpoint: A Framework To Derive Reliable Clinical Outcome Scores From Continuous DHT Data

This paper presents a comprehensive framework for deriving reliable clinical outcome scores from digital health technologies (DHTs) with a focus on physical activity measures relevant to metabolic diseases like diabetes. DHTs offer high-resolution, continuous, real-world data, enabling novel digital endpoints that are objective, non-invasive, and patient-centric. However, the complex nature of DHT data—characterized by large volumes, variability in aggregation methods, and patterns of missingness—necessitates careful selection and validation of endpoints.

Using data from the NHANES 2013–2014 cohort, the study compares cut point-based (e.g., MVPA) and percentile-based (e.g., 95th percentile activity) metrics. Findings reveal that while both types differentiate well between diabetic and non-diabetic individuals, percentile-based measures demonstrate greater reliability and lower susceptibility to bias from inconsistent wear time. This paper emphasizes that optimal DHT-based endpoints balance relevance, reliability, bias resistance, and feasibility.