Application of Time-to-Event Models to Ecological Momentary Assessment Data

• Activity Number: 461
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Mental Health Statistics Section
• Author(s): Emily A. Blood*+ and Lin Huang and Lydia A. Shrier
• Companies: Boston Children's Hospital and Boston Children's Hospital and Boston Children's Hospital
• Keywords: Ecological momentary assessment (EMA) ; Proportional hazard ; Accelerated failure time

Abstract

Ecological momentary assessment (EMA) data consist of in-the-moment sampling several times per day aimed at capturing phenomena that vary over the course of a day. For EMA data, with multiple correlated observations per individual, the optimal way to analyze time to event data is unclear. We apply three models to data from a study of affective state and sexual behavior in depressed adolescents: a proportional hazard (PH) model with robust standard errors, a PH model with a shared frailty term, and an accelerated failure time (AFT) model estimated via least squares. We compare the results and interpretation to published models summarizing reports in time-blocks. PH models, which require a PH assumption, estimate the instantaneous risk of an event for different predictor values. AFT models estimate time to event for different predictor values. Time-block analyses, which aggregate several reports in a block of time, estimate odds of an event following time-blocks with different average predictor values. The PH assumption was not met for all predictors and significance was obtained only in the AFT and time-block models. Computational memory requirements were substantial for AFT model.