CFE 2019: Start Registration
View Submission - CMStatistics
Title: Using sequential trials to estimate treatment effects in longitudinal observational data Authors:  Ruth Keogh - London School of Hygiene and Tropical Medicine (United Kingdom) [presenting]
Shaun Seaman - University of Cambridge (United Kingdom)
Jon Michael Gran - University of Oslo (Norway)
Stijn Vansteelandt - Ghent University and London School of Hygiene and Tropical Medicine (Belgium)
Abstract: Randomized controlled trials are the gold standard for estimating causal effects of treatments on health outcomes, but can be infeasible or unethical. Longitudinal observational data offer the possibility of estimating treatment effects over long periods of follow-up and in diverse populations. However, to do this we must tackle the challenge of time-dependent confounding. Several methods have been described for estimating causal treatment effects on survival using longitudinal observational data. The focus is on the sequential trials approach, which involves creation of a sequence of artificial trials from new time origins within an observational cohort. The analysis uses pooled Cox regression and time-dependent confounding is addressed through baseline covariate adjustment, censoring people upon deviation from their baseline treatment group, and inverse probability of censoring weighting. The sequential trials approach, despite being intuitive and straightforward to implement, has not previously been compared with alternative methods, either empirically or in terms of theoretical properties. We will use the potential outcomes framework to explain what is being estimated in the sequential trials approach, and contrast this with other methods. We will also show how sequential trials can be used to compare survival probabilities if different treatment regimes were applied in the target population.