Incorporating Intermediary Information in Cox Models of Randomized Clinical Trials: The Information Balanced Intermediary Cox Model

• Activity Number: 479 - Survival Analysis II
• Type: Contributed
• Date/Time: Wednesday, August 1, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): James Troendle* and Eric Leifer and Lauren Kunz and Song Yang
• Companies: National Institutes of Health and National Heart, Lung, and Blood Institute and National Heart, Lung, and Blood Institute and NHLBI/NIH
• Keywords: Frailty; Hazard; Time-Varying-Covariate; Treatment

Abstract

In a randomized clinical trial of the effect of a treatment on a time-to-event outcome, a major problem is attenuation due to factors unknown or not apparent at baseline. Sometimes there are variables, informative of the latent factor, that can be measured post-baseline. We propose a modeling approach that allows one to capture the information from these intermediary variables without compromising the treatment effect regardless of whether these variables lie on the causal path of treatment or not. This leads us to the Information Balanced Intermediary Cox Model (IBICM). As compared to an ordinary Cox model, the IBICM is shown to substantially reduce attenuation of the estimated hazard ratio for treatment when data are simulated from a realistic Cox model with residual attenuation due to an unobserved baseline factor. Moreover, even if the intermediary variables are related to treatment, the IBICM is less attenuated than the ordinary Cox model. Simulations indicate that the remaining attenuation decreases as the informative nature of the variables increases. The proposed models are illustrated by analyzing the Systolic Blood Pressure Intervention Trial.