Abstract
Wearable sensors may support scalable public health monitoring and early detection of post-COVID morbidity, but between-person heterogeneity limits group-based interpretation. We developed a privacy-preserving normative modelling framework to quantify individual deviations in digital biomarkers and applied it to acute COVID-19 and Long COVID cohorts. Using Corona Datenspende, we derived a healthy reference cohort of 16,000 adults, with 13,000 used for training and 3,000 for independent testing. Expected distributions of heart rate features, activity patterns, sleep duration, sleep timing and sleep regularity were modelled as smooth functions of age, sex and time. Each measurement was transformed into a standardized deviation from the healthy reference baseline using z-scores or quantiles. Models were then applied to 900 individuals with acute COVID-19 and 200 individuals with Long COVID.The models captured physiological variation across the adult lifespan and provided individualized reference ranges for each marker. In acute COVID-19 and Long COVID, we observed deviations from the healthy baseline across all examined domains, including sleep duration and timing, activity patterns and resting heart rate. These deviations were not limited to the acute infection phase: several markers remained altered over extended follow-up and showed slow or incomplete relaxation toward the reference baseline, indicating persistent dysregulation in a subset of individuals. Normative modelling of wearable data translates heterogeneous longitudinal sensor data into interpretable, time-resolved markers of individual deviation, recovery and persistence. Because only model parameters and standardized deviation scores need to be shared while raw data remain local, the framework supports privacy-preserving cross-cohort validation, model reuse, and scalable monitoring of Long COVID trajectories, while also enabling transfer to other infectious, chronic, and mental health conditions.
Key messages
• Wearable normative models quantify individual deviations in sleep, activity and heart rate during acute COVID-19 and Long COVID.
• The framework supports privacy-preserving validation and reuse across cohorts and may transfer to other infectious, chronic and mental health conditions.