Multiple imputation for IPD meta‐analysis: allowing for heterogeneity and studies with missing covariates

Abstract

Recently, multiple imputation has been proposed as a tool for individual patient data meta‐analysis with sporadically missing observations, and it has been suggested that within‐study imputation is usually preferable. However, such within study imputation cannot handle variables that are completely missing within studies. Further, if some of the contributing studies are relatively small, it may be appropriate to share information across studies when imputing. In this paper, we develop and evaluate a joint modelling approach to multiple imputation of individual patient data in meta‐analysis, with an across‐study probability distribution for the study specific covariance matrices. This retains the flexibility to allow for between‐study heterogeneity when imputing while allowing (i) sharing information on the covariance matrix across studies when this is appropriate, and (ii) imputing variables that are wholly missing from studies. Simulation results show both equivalent performance to the within‐study imputation approach where this is valid, and good results in more general, practically relevant, scenarios with studies of very different sizes, non‐negligible between‐study heterogeneity and wholly missing variables. We illustrate our approach using data from an individual patient data meta‐analysis of hypertension trials. © 2015 The Authors. Statistics in Medicine Published by John Wiley & Sons Ltd.

Meta-analysis using individual participant data: one-stage and two-stage approaches, and why they may differ

“…The issue of missing data has also not been explicitly addressed in this paper. In particular, we did not consider differences in one‐stage and two‐stage results after the imputation of missing participant data, although recognising this is a growing area of interest 1, 93, 94, 95, 96, 97, 98. However, in our longitudinal example, we noted that missing outcome data is handled naturally in the one‐stage approach under a missing at random assumption, which allows more efficient results by accounting for correlation between time points.…”

The use of test scores from large-scale assessment surveys: psychometric and statistical considerations

“…The main concern is not that all variables have to be in the same configuration in both models, but rather that the imputation model contains all variables (in the same transformed or untransformed form and, if needed, interacted with other variables). as the substantive model (Daniels et al 2014;Bartlett von Hippel 2009;Quartagno and Carpenter 2016). However, log(wages) and wages are strictly monotone increasing functions of each other, so that a linear approximation of one by the other in a restricted interval is probably quite serviceable.…”

Meta‐analysis of test accuracy studies using imputation for partial reporting of multiple thresholds

“…There has been debate concerning the order in which meta‐analysis and Rubin's rules should be performed, as such the MIDC approach performs meta‐analysis on the imputed dataset and then applies Rubin's rules after meta‐analysis, as Rubin's theory and recent evidence suggests …”

“…There has been debate concerning the order in which meta-analysis and Rubin's rules should be performed, as such the MIDC approach performs meta-analysis on the imputed dataset and then applies Rubin's rules after meta-analysis, as Rubin's theory and recent evidence suggests. 18,26 As with common multiple imputation methods, the number of MIDC datasets (or imputations) can be increased to reduce uncertainty further; due to the large computation time of the simulations, only 5 MIDC datasets were used throughout the simulation study. However, in reality sensitivity, analyses could be conducted in individual cases to decide on an appropriate number of MIDC datasets to reduce uncertainty as required.…”