Handling Missing Data

Workshop: Handling missing outcome data in randomised trials

  • Venue: MRC Biostatistics Unit, Cambridge
  • Date: June 20, 2011
  • Participants: 37

Aims:

  • Review some popular ways to analyse trials with missing data, focussing on the assumptions underlying them
  • Discuss what assumptions may be plausible and how one should decide this in particular trials
  • Focus on mixed models: describe the theory, how they should be implemented, what are the pitfalls
  • Demonstrate how to analyse data using mixed models and multiple imputation in Stata and SAS
  • Discuss the advantages and disadvantages of the two methods
  • Discuss the different issues involved in handling missing baselines

Agenda

  • Lecture (1hr)
    Introduction to missing data: assumptions (MCAR/MAR/MNAR and others); ITT principle; common methods (CC, LOCF); handling missing baselines
  • Lecture (0.5hr)
    Mixed models
  • Practical (1hr)
    Mixed models in Stata / SAS / R
  • Lecture (0.5hr)
    Multiple imputation
  • Lecture (0.5hr)
    Sensitivity analysis
  • Practical (1.5hr)
    Multiple imputation/Sensitivity analysis in Stata / SAS / R
  • Discussion (0.5hr)
    Which methods work best in practice? Which should we use?

Outputs

To view course materials, including lecture notes, practicals, data and programs in SAS, click here.