Meta-analysis of individual participant data (IPD-MA) is considered as the gold standard in epidemiologic research. When IPD-MA are affected by missing data, several strategies exist to obtain summary statistics.

We conducted a simulation study to compare the possible strategies for summarizing study results (through one-stage or two-stage meta-analysis) and dealing with missing values (through complete case analysis, within-study imputation, stratified imputation or hierarchical imputation). We illustrate the implementation of each strategy in an empirical example investigating the added value of C-reactive protein in diagnosing community acquired pneumonia. Finally, we provide recommendations on the implementation of imputation and meta-analysis models in an IPD-MA.

We found that stratified imputation was most problematic in terms of bias and coverage. Although complete case analysis performed fairly well in the simulation studies, it yielded unsatisfactory estimates of added value and between-study heterogeneity in the case study. Although within-study imputation performed adequately, the best results were obtained by hierarchical imputation. When summarizing the study results, one-stage and two-stage meta-analysis methods performed roughly similar. Finally, we found that recent recommendations on the order of combining imputed datasets in a two-stage IPD-MA were detrimental, and that the reverse ordering was more appropriate.

We recommended hierarchical imputation followed by one-stage meta-analysis in an IPD-MA with missing data. When statistical expertise or computational power is lacking, or when confidentiality issues exist, two-stage meta-analysis with within-study imputation may be preferred. Each of the imputed datasets should then first be meta-analyzed, and resulting estimates should subsequently be combined using Rubin's rule.