How to Handle Missing Data in a Systematic Review

Missing data shows up in two different places in a systematic review, and they need different solutions: missing summary statistics you need to calculate an effect size, and missing outcome data within an individual trial (participant dropout).

Missing summary statistics from a study report

• Contact the study authors first — this is the preferred approach per Cochrane guidance, and often successful, especially for more recent publications.
• Check supplementary materials and trial registries — missing statistics are sometimes reported in appendices, protocol registrations, or linked datasets rather than the main text.
• Estimate from what is reported — e.g., converting a reported median and range or IQR into an estimated mean and SD (our free median-to-mean/SD converter handles this), or deriving a missing SD from a reported confidence interval or p-value using standard formulas.
• Exclude the study from the specific analysis where data genuinely can't be obtained or reasonably estimated — report this explicitly rather than silently dropping the study.

Missing outcome data within a trial (participant dropout)

This is a risk-of-bias concern, not just a data-entry problem — both RoB 2 and ROBINS-I include a specific domain assessing bias from missing outcome data. Key questions: how much data is missing, is the amount balanced across groups, and is the reason for missingness related to the outcome itself (e.g., patients dropping out specifically because a treatment isn't working)?

Common approaches for missing outcome data

• Complete case analysis — analyzing only participants with complete data; simple, but can introduce bias if dropout isn't random.
• Last observation carried forward (LOCF) — once common, now widely criticized for assuming outcomes stay flat after dropout, which is rarely realistic.
• Multiple imputation — the current methodological standard when missingness is substantial, generating several plausible values per missing data point to reflect genuine uncertainty rather than a single guess.

What to report

State how much data was missing for each included study and outcome, what approach you used to handle it, and — where missingness was substantial — run a sensitivity analysis comparing results with and without the affected studies or imputation method. See our guide to sensitivity analysis in meta-analysis for how to structure that comparison.