Online Program Home
My Program

Abstract Details

Activity Number: 309 - Advances in Causal Inference
Type: Contributed
Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
Sponsor: Section on Statistics in Epidemiology
Abstract #306965 Presentation
Title: A Simulation Study on the Performance of AIPW and TMLE in Estimating Parameters of Marginal Structural Models Based on Real-World Longitudinal Data
Author(s): Dawei Liu* and John Zhong and Carl De Moor
Companies: Biogen and Biogen and Biogen
Keywords: real world data; marginal structural model; inverse probability weighting; targeted maximum likelihood estimation

In real-world longitudinal studies, time-varying confounding often exists. It can lead to a severely biased estimate of treatment effect if not accounted for appropriately. A popular way of adjusting for time-varying confounding is through the marginal structural model (MSM). For the estimation of MSM parameters, doubly robust (DR) methods are more desirable due to their robustness against model mis-specification of either the propensity score model or the outcome regression model. Two commonly used DR estimation methods are augmented inverse probability weighting (AIPW) and targeted maximum likelihood estimation (TMLE). Although both have the DR property, their performance may not always be the same. In this research, we will conduct a simulation study to compare the performance of AIPW and TMLE. In the simulation, longitudinal data under dynamic treatment regime will be simulated and machine learning method will be employed to model both the propensity score and the outcome. The estimates from AIPW and TMLE will be compared based on their bias and mean squared error. Special attention will be paid to the impact of sample size and number of confounding variables in the model.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program