The standard measure of crude association in a case-control study is the odds ratio as derived from a 2 by 2 table, while in a cohort study interest often focuses upon the risk difference and relative risk. While missing data is generally encountered, formal attempts to deal with it are rare and a complete case analysis is the norm. In the case-control study, the probability that exposure is missing may depend on true exposure status; likewise, the chance that disease information is missing in a cohort study may vary with disease status. Therefore, the missing at random assumption is often unreasonable. We present an adjustment to the usual product binomial likelihood to properly account for missing data. Identification of model parameters without restrictive assumptions requires a supplemental data collection effort directly akin to a validation study. We provide closed-form results to facilitate valid point and confidence interval estimation of the usual measures of effect. Simulations and an example are used to assess the performance of likelihood-based estimation and inference, and to display the potential for bias in the complete-case analysis.
Learning Objectives: N/A
Keywords: Biostatistics, Epidemiology
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: None
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.