Primary Submission Category: Difference in Differences, Synthetic Control, Methods for Panel and Longitudinal Data
Estimating effects of longitudinal modified treatment policies (LMTPs) at target times in studies with irregular assessment times
Authors: Anja Shahu, Daniel Malinsky,
Presenting Author: Anja Shahu*
Longitudinal studies are often designed to assess participants at a common set of pre-specified times after baseline (e.g., annual visits over a fixed follow-up period). In practice, however, assessment times can vary considerably from these targets in both timing and frequency. Such irregular assessment times pose challenges for estimating causal effects at target times, as outcomes are not observed at those times for all participants and the subset of participants observed at or around those times may not be representative due to informative assessment timing. In Shahu et al. (2025), we introduced a framework for estimating and testing hypotheses about effects of complex interventions on rates of change in an outcome over time and demonstrated its utility for examining the effect of a longitudinal shift intervention on the trajectory of an outcome over time. The original framework was developed for balanced discrete-time longitudinal studies with fixed visit schedules. We extend this framework to accommodate a setting with irregular assessment times by introducing a novel estimator for the projection of the causal effect at a specified target time. The proposed approach enables analysis of longitudinal studies, in which participants are assessed irregularly within pre-defined visit windows and randomized to be observed at only one of two consecutive visits. Through a simulation study, we illustrate the performance of our proposed approach in this setting.