CFE 2019: Start Registration
View Submission - CMStatistics
B0809
Title: Handling time-dependent confounding using a structural nested cumulative survival time model Authors:  Shaun Seaman - University of Cambridge (United Kingdom)
Oliver Dukes - Ghent University (Belgium)
Ruth Keogh - London School of Hygiene and Tropical Medicine (United Kingdom)
Stijn Vansteelandt - Ghent University and London School of Hygiene and Tropical Medicine (Belgium)
Shaun Seaman - University of Cambridge (United Kingdom) [presenting]
Abstract: Observational studies that attempt to assess the effect of a time-varying exposure on a survival outcome typically suffer from time-varying confounding bias. Marginal structural models, fitted by inverse probability weighting, can be used, but these can be subject to the problem of highly variable weights when the confounders are strongly predictive of the exposure or when the exposure is continuous. Structural nested accelerated failure time models can be fitted by g-estimation, but this requires artificial recensoring, which causes loss of information. We have developed an alternative method using a structural nested cumulative survival time model (SNCSTM). This method avoids inverse probability weighting and artificial recensoring. It also allows investigation of effect modification by time-dependent variables. The SNCSTM assumes that intervening to set exposure at time t to zero has an additive effect on the subsequent conditional hazard given exposure and confounder histories when all subsequent exposures have already been set to zero. We show how to estimate the exposure effect using standard software for generalised linear models and describe a more efficient estimator that is available in closed form. We apply our methods to estimate the effect of delaying initiation of treatment with DNase on survival in patients with Cystic Fibrosis.