Strata Academy

Missing data in clinical studies

Attrition, imputation, and bias in appraisal

Types of missingness

Data can be missing completely at random (unlikely), at random (related to observed variables), or not at random (related to unobserved factors, including the outcome). The last is the most dangerous for bias.

Dropout in trials often relates to side effects or lack of improvement – so complete-case analysis can bias results.

Common approaches

Complete-case analysis discards participants with any missing values – simple but can bias estimates and reduce power.

Multiple imputation creates several plausible completed datasets and combines results – generally preferred when assumptions are met and documented.

Last-observation-carried-forward and single imputation are generally discouraged for primary analyses in modern guidelines.

Risk-of-bias link

ROB 2 includes a domain for missing outcome data. ROBINS-I covers missing data in non-randomised studies. Note the proportion lost, reasons, and whether sensitivity analyses were performed.

• Report flow of participants from allocation to analysis
• Compare baseline characteristics of completers vs non-completers
• ITT with appropriate missing-data methods is preferred for RCTs

Interactive version (quizzes, walkthroughs) loads when JavaScript is enabled.