Fully-adjusted model (i.e., final model) was adjusted for age, sex, education, geographic region, residence, marital status, financial support, economic status, smoking status, drinking status, regular exercise, social and leisure activity, personality, disability, PM2.5 and air temperature. To keep the same sample size across analyses using different ozone exposure metrics, the observations with missing lag 3 year ozone exposure were removed. Abbreviations: CI, confidence interval; O3, ozone.