Efficient and Robust Methods for Causally Interpretable Meta-Analysis: Transporting Inferences from Multiple Randomized Trials to a Target Population

Abstract

We present methods for causally interpretable meta-analyses that combine information from multiple randomized trials to draw causal inferences for a target population of substantive interest. We consider identifiability conditions, derive implications of the conditions for the law of the observed data, and obtain identification results for transporting causal inferences from a collection of independent randomized trials to a new target population in which experimental data may not be available. We propose an estimator for the potential outcome mean in the target population under each treatment studied in the trials. The estimator uses covariate, treatment, and outcome data from the collection of trials, but only covariate data from the target population sample. We show that it is doubly robust in the sense that it is consistent and asymptotically normal when at least one of the models it relies on is correctly specified. We study the finite sample properties of the estimator in simulation studies and demonstrate its implementation using data from a multicenter randomized trial.

Systematic review and meta-analysis of randomized trials of hydroxychloroquine for the prevention of COVID-19

“…Our pooled estimates are based on the design and analytic choices made by the investigators of each trial. Ideally, these investigators could coordinate analyses of individuallevel data with standardized outcome definitions; corrections for differences in length of follow-up and treatment dose; adjustment for losses to follow-up and other deviations from protocol [31]; and adoption of a more causallyinterpretable meta-analytic approach [32]. All but two pre-exposure prophylaxis trials used laboratory-confirmed infection as the primary outcome; the remaining trials [19,23], and one of the four post-exposure prophylaxis trials [4],…”

Synthesizing studies for comparing different treatment sequences in clinical trials

“…Although these are rather strong assumptions, they are necessary in order to combine data from different trials when not all treatment sequences data are available. These assumptions are in some sense similar to the $$$ d $$$‐separation condition in the data fusion literature 49,50 . When these assumptions are violated (for instance, see References 1 and 35), one can only compare treatment sequences for which data from both the first line and the second line are available.…”

“…These assumptions are in some sense similar to the d-separation condition in the data fusion literature. 49,50 When these assumptions are violated (for instance, see References 1 and 35), one can only compare treatment sequences for which data from both the first line and the second line are available. Another possible solution is to allow for the dependence of G(a 1 , a 2 ) on T c (a 1 , a 2 ) by including T c (a 1 , a 2 ) as a covariate in the model or incorporating a frailty.…”

Challenges in Systematic Reviews and Meta-Analyses of Mediation Analyses

“…Recently, several new approaches have been proposed to overcome this challenge. These approaches aim to standardize results of different studies over the case-mix of a well-defined target population, prior to applying conventional meta-analyis techniques to summarize the findings [27,28,29]. In this section, we describe the extension of these approaches to meta-analysis of mediation studies.…”