A Marginal Structural Model to Compare the Causal Effect of Two Possibly Misclassified Treatments on the Survival Outcome

• Activity Number: 348
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ming Geng*+ and Andrew C. Thomas and Chung-Chou Chang
• Companies: and Carnegie Mellon and University of Pittsburgh
• Keywords: marginal structural Cox proportional hazard model ; measurement error ; MCMC

Abstract

Objective: To compare the causal effect between two possibly misclassified treatments on the survival outcome. Background: Marginal structural model can give a consistent estimate of the causal effect when all variables are measured precisely. Many studies have shown that misclassified predictors could result in biased estimates. To correct this problem, a currently used approach requires using a validation data set. In our study, we develop a method to calibrate treatment misclassification without the need for a validation set. Method: We used regression calibration method. By using the information of sensitivity and specificity of self-reported treatment, we derived the relationship between the underlying and the misclassified treatment. Transformation method was then used to obtain the partial likelihood of the true treatment effect. Result: Simulations showed that if the means of the prior distribution of sensitivity and specificity are close to the true values, using our proposed method will get less bias in the estimate. Conclusions: Our estimate is close to the true treatment effect even when the sensitivity and specificity of self-reported treatment are not known exactly.