Complete-case analysis (CCA) is one of the most commonly used, and misused, missing data analysis methods. The general guidance for applied researchers is to use principled methods such as weighting techniques, multiple imputation, likelihood-based or Bayesian models instead of CCA to avoid biased inferences, loss of information, and reduced efficiency. Mathur, Shpitser, and VanderWeele (Am J Epidemiol. 2026;000(00):0000-0000)) argue in favor of adopting principled adjusted CCA alongside other methods as a sensitivity check to assess the validity of the missingness assumptions. While previous work has identified scenarios where CCA is a theoretically sound missing data method, including where data are not missing completely at random, Mathur et al. frame these scenarios more generally with causal inference tools already familiar to epidemiologists. In this commentary, we contextualize their arguments and proposals in the scope of incomplete data analysis in epidemiology.
Stockton et al. (Tue,) studied this question.