Article type
Year
Abstract
Objectives: To understand the theory of methods used to account for missing outcome data in a meta-analysis.
Description: Missing outcome data are common even in carefully designed randomized control trials. We assume at the meta-analysis level the problem with missing data is solved at the trial level and conduct an available case analysis. Case analysis will give unbiased results if data are missing at random. However, if missing data are not random, this compromises the validity of the results.
While the ‘Risk of bias’ tool can be used to evaluate the risk of incomplete outcome reporting, meta-analysts often use statistical methods to account for missing outcome data that do not propagate imputation uncertainty and treat the imputed data as if they were observed rather than imputed. This results in more precise estimates, whereas large attrition rates should have been related with less precise estimates. 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. Approaches make explicit assumptions about how outcomes in the unobserved data and observed data are related. This relationship is not known or informed by the data, but we can use expert opinion or conduct a sensitivity analysis to evaluate the robustness of the results so we do not assume data are missing at random.
This Part 1 workshop prepares for, and is continuous with Part 2.
Description: Missing outcome data are common even in carefully designed randomized control trials. We assume at the meta-analysis level the problem with missing data is solved at the trial level and conduct an available case analysis. Case analysis will give unbiased results if data are missing at random. However, if missing data are not random, this compromises the validity of the results.
While the ‘Risk of bias’ tool can be used to evaluate the risk of incomplete outcome reporting, meta-analysts often use statistical methods to account for missing outcome data that do not propagate imputation uncertainty and treat the imputed data as if they were observed rather than imputed. This results in more precise estimates, whereas large attrition rates should have been related with less precise estimates. 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. Approaches make explicit assumptions about how outcomes in the unobserved data and observed data are related. This relationship is not known or informed by the data, but we can use expert opinion or conduct a sensitivity analysis to evaluate the robustness of the results so we do not assume data are missing at random.
This Part 1 workshop prepares for, and is continuous with Part 2.