The Reviewer Asked: Explain Your Missing Data

Describe the missingness before naming a method

Start with counts. Which variables are missing, for how many participants, and at which visits? Summarize by relevant groups when that helps explain the problem. Reasons for missed measurements can be more informative than an overall percentage.

Keep different problems separate: a participant never enrolled, a missed follow-up visit, a laboratory value not measured, and an outcome that is not defined after an intercurrent event are not automatically the same missing-data problem.

Show how the analysis population was formed

A fictional report might say “all available cases were analyzed” without explaining that each model used a different set of participants. The revision should identify the sample for each analysis and the rule used to include or exclude observations. Denominators in tables should agree with that account.

Neither complete-case analysis nor imputation is assumption-free. The choice depends on the design, analysis target, observed information, and plausible reasons values are unavailable. A convenient software option is not a sufficient justification.

Build a transparent response

Document the original method. If a different approach is justified, explain the reason, the variables and procedures involved, and how the estimates change. Consider sensitivity analysis for important assumptions that the observed data cannot verify. Label any new analysis honestly as a revision-stage analysis.

Do not say that imputation “recovered the true values.” It represents uncertainty under a specified model. Do not assume that recruiting more participants eliminates bias from systematic missingness.

Download the missingness audit worksheet and use the Reviewer Comment Decoder to organize the response. A Dataset Readiness Assessment can identify data-structure issues before reanalysis.

Methodological reading: Cochrane Handbook, missing data within Chapter 10.