A0906
Title: Fractional imputation approach for Cox regression with missing covariate
Authors: Jongho Im - Yonsei University (Korea, South) [presenting]
Taesuk Park - Yonsei University (Korea, South)
Sangwook Kang - Yonsei University (Korea, South)
Abstract: In a case-cohort study, the main exposure variable is often only available for some subjects, while other covariates are available for the whole cohort. This incomplete data can be viewed as a special case of missing covariate by design. Previous works have used a popular multiple imputation approach to efficiently handle this missing data. Instead of multiple imputation, we can use fractional imputation as another repeated imputation approach. Fractional imputation is yet to be widely used in practice because it is relatively new and there is more complexity in variance estimation. However, fractional imputation has its own advantages, for example, it provides consistent variance estimation for the method-of-moment type estimators and creates a singly completed dataset rather than multiply completed datasets. A limited simulation study is implemented to confirm the performance of the proposed approach. In addition, misspecification in the imputation model is investigated to check the robustness of the proposed imputation method.
