How to Handle Missing Data in Clinical Research

Missing data are common in clinical research and can bias results and reduce power. Handle them by preventing them where possible, describing the amount and pattern, choosing an analysis method that suits the mechanism and checking how sensitive results are to your assumptions.

Types of missingness

• Missing completely at random (MCAR): missingness is unrelated to any data; complete-case analysis is unbiased but less efficient.
• Missing at random (MAR): missingness depends on observed data (for example older patients miss follow-up more often), which can be accounted for by modelling.
• Missing not at random (MNAR): missingness depends on the unobserved value itself (for example sicker patients drop out); this needs sensitivity analyses and cannot be fixed by standard methods.

Steps

1. Prevent: design simple forms, train collectors, follow up participants.
2. Describe: report how much is missing per variable and why, and compare those with and without missing data.
3. Choose a method (with a statistician): complete-case analysis if little is missing and MCAR is plausible; multiple imputation for MAR; model-based approaches; and sensitivity analyses for MNAR.
4. Avoid simple fixes such as replacing with the mean, or carrying the last value forward, unless justified; they can bias estimates and understate uncertainty.
5. Report the methods, the assumptions and results of sensitivity analyses (CONSORT asks for how missing data were handled in trials).

Illustrative example

Invented for teaching. In a cohort, 18% lack a follow-up blood pressure, mostly older patients. The authors compare responders and non-responders, use multiple imputation with age and baseline blood pressure as predictors, and show that the main result is similar in a complete-case analysis.

Common pitfalls

• Dropping records with any missing value without describing the loss.
• Using mean imputation and treating the data as complete.

Key takeaways

• Prevent, describe, analyse appropriately and test sensitivity.
• Report exactly how missing data were handled.

This page is general educational information about research methods. It is not medical, legal or regulatory advice; follow your institution's ethics board and your target journal's instructions.