Landmark estimation of survival and treatment effects in observational studies

摘要: Clinical studies aimed at identifying effective treatments to reduce the risk of disease or death often require long term follow-up participants in order observe a sufficient number events precisely estimate treatment effect. In such studies, observing outcome interest during may be difficult and high rates censoring observed which leads reduced power when applying straightforward statistical methods developed for time-to-event data. Alternative have been proposed take advantage auxiliary information that potentially improve efficiency estimating marginal survival testing Recently, Parast et al. (J Am Stat Assoc 109(505):384-394, 2014) landmark estimation procedure effects randomized clinical trial setting demonstrated significant gains could obtained by incorporating intermediate event as well baseline covariates. However, requires assumption potential outcomes each individual under control are independent group assignment is unlikely hold an observational study setting. this paper we develop use particular, incorporate inverse probability weights (IPTW) account selection bias on (pretreatment) We demonstrate consistent estimates can using IPTW there improved information. compare our those Kaplan-Meier estimator, original procedure, estimator. illustrate resulting reduction through simulation apply AIDS dataset examine effect previous antiretroviral therapy survival.