Aggregation framework for continuous digital health measures and application to walking activity data

Abstract No standardized procedures for aggregating continuous digital health measures exist. Using a modified Delphi approach, we developed a framework to standardize the selection of aggregations that provide clinically meaningful, stable values. This was applied to aggregate seven digital mobility outcomes (DMOs:walking bout (WB) duration, step count, walking speed, stride length, cadence, stride duration, and number of WBs), collected over 7 days, in 2181 individuals with and without mobility disabilities. We defined aggregations as combinations of restrictions (data segments) and summary statistics, and developed a three-part framework. In Part 1, candidate restriction–summary statistic combinations are identified considering technical validity, interpretability, sample size, and distribution. Part 2 assesses within-participant stability with intraclass correlation coefficients (ICCs), and discriminant capacity between healthy and mobility disabilities with the area under the curve (AUC) on as many temporal levels as needed. Part 3 involves expert discussions combining subject-specific experience, guidelines, and data-driven results to select final aggregation(s). Applied to DMOs, we identified candidate aggregations (Part 1), found moderate-to-high daily ICCs (range 0.45–1.00) and varying AUCs (range 0.43–0.90) (Part 2), and finally selected 24 daily- and weekly-aggregated DMOs, classified into walking activity and gait DMOs (Part 3). In summary, reducing digital health data into clinically meaningful, statistically stable values is achievable with standardized procedures.