Objectives: To understand the theory and application of methods used to account for missing outcome data in a meta-analysis.
Description: Missing outcome data in the included trials reduce the precision in the meta-analysis summary effect. If outcomes are not missing at random, the synthesis of the available data only will yield biased estimates. While the 'Risk of bias' tool can be used to evaluate the risk of attrition bias, meta-analysts often use statistical methods that impute missing outcome data in studies under various scenarios in order to increase precision. However, most statistical methods used in meta-analysis to account for missing outcome data do not propagate imputation uncertainty and treat the imputed data as if they were observed rather than assumed. As a result, uncertainty in summary estimates is underestimated. We present valid methods to estimate meta-analytic treatment effects for dichotomous and continuous outcomes when these are missing for some of the randomized individuals. The approaches make explicit assumptions about how outcomes in the unobserved data and observed data are related. The workshop will comprise a practical session in Stata using an updated version of the metamiss routine.
Participants are requested to bring their laptops with the Stata 12 or 13.