In Reply Dr Stevens and colleagues highlight the KNN method and further the discussion about moving away from overly simplistic, biased methods for handling missing values. Our review of missing data methods was not exhaustive. For simplicity and clarity, we focused on methods that are more common in practice and readily available in statistical software packages. As a result, we omitted certain viable techniques for handling missing values (eg, KNN, hot-deck imputation, and maximum likelihood estimation).
Craig D. Newgard, Roger J. Lewis. Accounting for Missing Data in Clinical Research—Reply. JAMA. 2016;315(5):518. doi:10.1001/jama.2015.16470