855Multiple Imputation in the Context of Case-Cohort Studies: Simulation and Case Study

Background Multiple imputation (MI) is commonly used to address missing data in epidemiological studies, but valid use requires compatibility between the imputation and analysis models. Case-cohort studies use unequal sampling probabilities for cases and controls which are often accounted for during analyses through inverse probability weighting (IPW). It is unclear how to apply MI for missing covariates while achieving compatibility in this setting. Methods A simulation study was conducted with missingness in two covariates, motivated by a case-cohort investigation within the Barwon Infant Study. MI methods considered involved including interactions between the outcome (as a proxy for weights) and analysis variables, stratification by weights, and ignoring weights, within the context of an IPW analysis. Factors such as the target estimand, proportion of incomplete observations, missing data mechanism and subcohort selection probabilities were varied to assess performance of MI methods. Results There was similar performance in terms of bias and efficiency across the MI methods, with expected improvements compared to IPW applied to the complete cases. Precision tended to decrease as the subcohort selection probability decreased. Similar results were observed irrespective of the proportion of incomplete cases. Conclusions Our results suggest that it makes little difference how weights are incorporated in the MI model in the analysis of case-cohort studies, potentially due to only two weight classes in this setting. Key messages If and how the weights are incorporated in the imputation model may have little impact in the analysis of case-cohort studies with incomplete covariates

An Overview of Population Size Estimation where Linking Registers Results in Incomplete Covariates, with an Application to Mode of Transport of Serious Road Casualties

We consider the linkage of two or more registers in the situation where the registers do not cover the whole target population, and relevant categorical auxiliary variables (unique to one of the registers; although different variables could be present on each register) are available in addition to the usual matching variable(s). The linked registers therefore do not contain full information on either the observations (often individuals) or the variables. By treating this as a missing data problem it is possible to construct a linked data set, adjusted to estimate the part of the population missed by both registers, and containing completed covariate information for all the registers. This is achieved using an Expectation-Maximization (EM)-algorithm. We elucidate the properties of this approach where the model is appropriate and in situations corresponding with real applications in official statistics, and also where the model conditions are violated. The approach is applied to data on road accidents in the Netherlands, where the cause of the accident is denoted by the police and by the hospital. Here the cause of the accident denoted by the police is considered as missing information for the statistical units only registered by the hospital, and the other way around. The method needs to be widely applied to give a better impression of the range of problems where it can be beneficial.