Extending inferences from a randomized trial to a new target population

Abstract

When variables that are treatment effect modifiers also influence the decision to participate in a clinical trial, the average effect among trial participants will differ from the effect in other populations of trial-eligible individuals. In this tutorial, we consider methods for transporting inferences about a time-fixed treatment from trial participants to a new target population of trial-eligible individuals, using data from a completed randomized trial along with baseline covariate data from a sample of nonparticipants. We examine methods based on modeling the expectation of the outcome, the probability of participation, or both (doubly robust). We compare the finite-sample performance of different methods in a simulation study and provide example code to implement the methods in software. We illustrate the application of the methods to the Coronary Artery Surgery Study, a randomized trial nested within a cohort of trial-eligible patients to compare coronary artery surgery plus medical therapy versus medical therapy alone for patients with chronic coronary artery disease. Lastly, we discuss issues that arise when using the methods in applied transportability analyses.

Balancing vs modeling approaches to weighting in practice

“…One can then weight the sample appropriately toward the covariate distributions of the target population and compute weighted estimators (both linear and doubly-robust) of the TATE. 45…”

Mosaic effectiveness: measuring the impact of novel PrEP methods

“…Mosaic effectiveness requires counterfactual estimation of the HIV background incidence in the trial population and also the estimation of the HIV incidence when participants are matched to the product (TDF/FTC vs. PrEP LA) to which they would be adherent. Estimation approaches could use the transportability framework 23 on the markers of product use. This is a framework similar to one that has been proposed for HIV vaccine bridging studies.…”

The Promise, and Challenges, of Methods to Enhance the External Validity of Randomized Trial Results

“…Additional methods have also been proposed, such as those that rely on fitting a flexible outcome model in the trial sample and using that to predict outcomes in the population, and doubly robust approaches that combine the weighting and outcome model approaches. 9 The underlying causal assumptions of these approaches are generally the same as those outlined above, but have somewhat different statistical assumptions regarding the exact methods used. As shown by Kern et al, 2 the validity of effect estimates was much more dependent on the causal assumptions detailed above than on the specific estimation procedures used.…”

“…Just as there has been extensive methodological developments to account for potential internal validity biases in observational studies when randomization of treatment assignment is not ethical or logistically feasible, there has been recent methods development to account for potential external biases in both trial samples and observational studies when random sampling of the target population is not ethical or logistically feasible. 9 That said, it is important to recognize the additional assumptions required to estimate population effects and, in particular, to think carefully about the measurement of likely effect moderators in trial samples and population data, and measure them consistently across datasets, to then be used in adjustments. Stating and evaluating the assumptions under which these statistical methods will provide accurate population effect estimates for any particular application is critical for interpreting the results of the analysis.…”

“…utilize a weighting approach to weight the trial sample to look like the population on observed baseline characteristics. Additional methods have also been proposed, such as those that rely on fitting a flexible outcome model in the trial sample and using that to predict outcomes in the population, and doubly robust approaches that combine the weighting and outcome model approaches 9 . The underlying causal assumptions of these approaches are generally the same as those outlined above, but have somewhat different statistical assumptions regarding the exact methods used.…”