Estimates of Causal Effects of a Continuous Exposure Using Multiple Imputation

• Activity Number: 151
• Type: Contributed
• Date/Time: Monday, July 30, 2007 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Sundar Natarajan*+ and Stuart R. Lipsitz and Garrett Fitzmaurice and Scott Regenbogen
• Companies: Veterans Affairs New York Harbor Healthcare System/New York University and Brigham and Women's Hospital and Harvard University and Harvard School of Public Health

Abstract

Inferring causation is difficult in observational studies. Confounders may not be balanced across groups, and conclusions about 'exposure' effects are vulnerable to selection bias. For a categorical exposure, propensity score adjustments are popular. For a continuous exposure, these adjustments are difficult. Instead, we will use Rubin's multiple imputation approach to estimate the causal effects of a continuous exposure. Specifically, if there are G distinct exposure levels, for the i^th person, we create G-1 new observations with all possible confounders the same as the i^th person, but with the exposure replaced with the G-1 exposures that the subject didn't get; further, the outcome is set at missing. Multiple imputation is used to 'impute' missing outcomes. This approach is also used to estimate the mean outcome at 'doses' of the exposure of interest.