ASCEPT

ASCEPT identifies and selects optimal changepoints in mobile health (mHealth) datasets to distinguish changes due to lifestyle behaviors from changes caused by proprietary data-processing algorithm updates.

Key Features:

• Two-stage changepoint detection: Sequential iterations of a changepoint detection algorithm are used to identify potential changepoints in mHealth time series.
• Empirical p-values: Empirical p-values are computed to assess the statistical significance of each detected changepoint.
• Trimming mechanism: Identified changepoints are trimmed with respect to linear and seasonal trends to remove changes attributable to underlying trends rather than discrete shifts.
• Parameterization: The method operates with two parameters: a significance level and a goodness-of-fit threshold.
• Method comparison: Performance has been demonstrated against circular binary segmentation using real-world mHealth data from the Precision VISSTA study.

Scientific Applications:

• Algorithm update detection: Distinguishing changepoints caused by proprietary algorithm updates from those caused by behavioral changes in longitudinal mHealth studies.
• Downstream analysis adjustment: Informing adjustment of downstream analyses by identifying and accounting for changepoints that affect time-series consistency.

Methodology:

Iterative application of a changepoint detection algorithm to identify candidate changepoints, evaluation of each candidate using empirical p-values, and trimming of changepoints by accounting for linear and seasonal trends.