Estimating the Causal Effect of Treatment in Observational Studies with Survival Time End Points and Unmeasured Confounding

Abstract

Summary Estimation of the effect of a treatment in the presence of unmeasured confounding is a common objective in observational studies. The Two Stage Least Squares (2SLS) Instrumental Variables (IV) procedure is frequently used but is not applicable to time-to-event data if some observations are censored. We develop a simultaneous equations model (SEM) to account for unmeasured confounding of the effect of treatment on survival time subject to censoring. The identification of the treatment effect is assisted by IVs (variables related to treatment but conditional on treatment not to the outcome) and the assumed bivariate distribution underlying the data generating process. The methodology is illustrated on data from an observational study of time to death following endovascular or open repair of ruptured abdominal aortic aneurysm. As the IV and the distributional assumptions cannot be jointly assessed from the observed data, we evaluate the sensitivity of the results to these assumptions.

Inferring causality from observational studies: the role of instrumental variable analysis

“…[2][3][4][5] The past 2 decades have seen an increase in the use of IV analysis by epidemiologists and the development of various modeling techniques that apply IV analysis to health data. [6][7][8][9][10][11][12] The greatest strength of IV analysis is that it enables causal inference from data without assuming an absence of unmeasured confounders. Thus, it has advantages over approaches such as conventional multivariable regression modeling and propensity score methods as it isolates an independent treatment effect by removing the impact of both measured and unmeasured confounders, under specific conditions.…”

Nonparametric Bayesian Instrumental Variable Analysis: Evaluating Heterogeneous Effects of Coronary Arterial Access Site Strategies

“…This assumption of homogeneous variance can be unrealistic when the underlying treatment selection might vary by different patient subgroups, such as patients treated in different locations and hospitals. Additionally, if the parameter of interest is an average treatment effect, the covariance parameter between the potential outcomes, ρ 10 , is usually assumed to be zero (Chib and Hamilton, 2000, O'Malley et al, 2011, Choi and O'Malley, 2017. However, for estimating effects that are dependent on estimating distribution of Y i (1) − Y i (0), or some functionals of it, such as identifying fractions of a population that benefit from a given treatment, these approaches are inadequate.…”

“…Assumptions 1 and 2 are required to estimate average treatment effects under various conditions in a frequentist setting (Abadie, 2002, Basu et al, 2007. Earlier work on Bayesian methodologies for estimating average treatment effects of interventions with selection bias have assumed Normal error distribution for potential outcome models, and very few are concerned with estimating heterogeneous treatment effects (Chib and Hamilton, 2000, Hirano et al, 2000, Heckman et al, 2014, Jacobi et al, 2016, Choi and O'Malley, 2017.…”

Opportunities and Challenges for Environmental Exposure Assessment in Population-Based Studies

“…Cross-sectional but representative studies, such as the National Health and Nutrition Examination Survey (NHANES) (82), which collects information on many health related factors, can be useful for characterizing the variability and co-variability of factors for multiple environmental exposure biomarkers (80, 83, 84). If there are sets of highly correlated exposures, then disentangling their individual effects will require studying them together in studies of very large sample size.…”