184 – Modern Methods in Missing Data Imputation
Comparative Studies for Cox Hazards Model Based on the Suita Study
Michikazu Nakai
National Cerebral and Cardiovascular Center
Yuhlong Lio
University of South Dakota
Din Chen
University of Rochester
Kunihiro Nishimura
National Cerebral and Cardiovascular Center
Makoto Watanabe
National Cerebral and Cardiovascular Center
Yoshihiro Miyamoto
National Cerebral and Cardiovascular Center
In cohort study, there is always a time-gap between the date of occurrence and the date of diagnosis when the disease is a lifestyle-related disease such as hypertension (HT). However, we typically analyze using the date of diagnosis. In this presentation, we investigate whether these time-gaps have caused a significant difference in the analysis result using the Suita study, the population-based prospective cohort study of Japan. In this study, participants who had no HT at baseline (1,591 men and 1,973 women) aged 30-84 years were included. During median follow-up of 7.2 years, 1,325 participants (640 men and 685 women) developed HT. We created a missing value for the date of HT occurrence. Then, we compared the efficiency of median imputation (Median) and multiple imputation (MI) with original dataset (Original). The Cox proportional hazards model was used to estimate hazard ratios (HRs) and C-index of BMI by sex adjusted for age, cigarette smoking and alcohol drinking.We conclude that significant differences didn't observe among three imputations even though Median showed less accuracy.