Missing data and measurement in individual patient data meta-analysis
Research Objectives:
Meta-analyses increasingly rely not just on summary statistics from each study, but on individual patient data. This allows not only investigation and adjustment for sources of heterogeneity, but also, for example, the development of prognostic models. In practice, missing data and measurement error are likely to pose significant issues. First, some variables are inevitably missing for some individuals. Second, not all studies will measure the same variables: for example in asthma some may measure forced expiratory volume and others forced vital capacity. Third, some studies may record underlying continuous variables as ordinal or binary variables. Multiple imputation offers a powerful flexible tool for overcoming these issues.
Description of work:
Building on our recent publications and software for multilevel mixed response multiple imputation, this project will develop and evaluate – using individual patient meta-analysis data available to us – a multiple imputation approach to these issues.
Host Institution: London School of Hygiene & Tropical Medicine