Handling of Missing Data in Clinical Trials
Objectives
Differentiate between the three mechanisms of missing data and describe their impact on study results.
Describe the main statistical techniques for dealing with missing data in epidemiological and clinical datasets.
Identify the strengths and limitations of the methods most frequently used to deal with missing data in various missing data situations.
Handle missing data in a database.
Apply and interpret a multiple imputation method using R.
Course outline
Missing data typology : MCAR, MAR and MNAR
Using DAG to understand missing data
Ad-hoc methods: Complete Case Analysis (CCA) / LOCF / Simple Imputation
MAR hypothesis: Maximum likelihood approaches
Inverse Probability Weighting (IPW)
Multiple imputation: Rubin’s law, chained equation, MCMC algorithms
Non-monotonic missing data models: MCMC algorithms
MNAR hypothesis: Model selection and Pattern-Mixture-Models (PMM)
Sensitivity analysis under the MNAR hypothesis
Prerequisites
Markov chain, Bayesian computation, Resampling methods (2A)
Mixed models (3A)