Simulating Longitudinal Data from Marginal Structural Models

• DOI: 10.48550/arxiv.2502.07991
• Publication Date: 2025-02-11
• Authors: Lin Xi; Daniel de Vassimo...; Chester A. Mathis; Jens Magelund Tarp; Robin J. Evans
• Keywords: Marginal structural model; Longitudinal data

Simulating longitudinal data from specified marginal structural models is a crucial but challenging task for evaluating causal inference methods and designing clinical trials. While generation typically proceeds in fully conditional manner using equations according to temporal ordering, require capturing effects that are over time-dependent confounders, making it difficult align distributions with target quantities. To address this, we propose flexible efficient algorithm simulating adheres exactly model. Recognizing the importance of time-to-event outcomes trials, extend method accommodate survival models. Compared existing approaches, our offers several key advantages: enables exact simulation known model rather than relying on approximations; avoids imposing restrictive assumptions data-generating process; as need only evaluate analytic functions. This last benefit contrasts use computationally intensive techniques such Monte Carlo approximations or numerical integration. Through studies replicating realistic scenarios, validate method's accuracy utility. Our will allow researchers effectively simulate structures their specific scenarios.