The Influence of Missing Data Handling Methods on CES-D Analysis in the China Health and Retirement Longitudinal...

Most evaluations of missing data analyses rely on literature reviews or comparisons of hypothesized methods applied to a given dataset, yet the real-world impact of different strategies remains unknown. Using the widely analyzed CHARLS dataset, we reviewed missing data handling methods in 219 studies using the CES-D-10, and compared the three most prevalent procedures: complete-case analysis (CCA), impute zero (IZ), and multiple imputation (MI). Although CES-D-10 items were not missing completely at random, 90% of publications used CCA, resulting in lower mean CES-D-10 scores but higher depression prevalence, especially among adults aged 75 and older. The IZ CES-D-10 score constructed in the harmonized CHARLS dataset systematically underestimated both prevalence and mean scores, distorting age trajectories and attenuating sex differences in later life. Many CHARLS-based depression studies are therefore likely affected by systematic bias due to suboptimal treatment of missing CES-D-10 items, particularly those studies using the harmonized dataset. Moreover, because the IZ approach has similarly been used to construct CES-D scores in other major aging studies (e.g., HRS, ELSA, and KLoSA), this bias may extend beyond CHARLS and undermine the validity of depression findings across multiple international cohorts.