Generalizing Causal Inferences from Individuals in Randomized Trials to All Trial-Eligible Individuals

Abstract

We consider methods for causal inference in randomized trials nested within cohorts of trial‐eligible individuals, including those who are not randomized. We show how baseline covariate data from the entire cohort, and treatment and outcome data only from randomized individuals, can be used to identify potential (counterfactual) outcome means and average treatment effects in the target population of all eligible individuals. We review identifiability conditions, propose estimators, and assess the estimators' finite‐sample performance in simulation studies. As an illustration, we apply the estimators in a trial nested within a cohort of trial‐eligible individuals to compare coronary artery bypass grafting surgery plus medical therapy vs. medical therapy alone for chronic coronary artery disease.

Extending inferences from a randomized trial to a new target population

“…We use the term generalizability when the target population coincides or is a subset of the trial‐eligible population and transportability when the target population includes at least some individuals who are not trial‐eligible (and who, by definition, cannot be trial participants; others have proposed different definitions). In generalizability analyses, the target population often has a different distribution of effect modifiers compared with the participant population, even though both populations meet the trial eligibility criteria. In transportability analyses, such differences are even more likely.…”

“…The methods we describe in this tutorial can be used both in nested and nonnested trial designs to estimate treatment effects in the target population, using baseline covariate, treatment, and outcome data from the trial and only baseline covariate data from the target population. Treatment and outcome data from the target population, if available, can be used to evaluate assumptions, but they are not necessary for the types of analyses described in this tutorial. This is an attractive feature if treatment and outcome data from nonparticipants are unreliable (eg, due to gross measurement error), costly to obtain, or impossible to collect.…”

“…When using nested trial designs, another causal effect, the average effect in the population represented by the cohort in which the trial is nested, , may also be of interest, and we have discussed its identification and estimation in previous work . Even if the primary target of inference for nested trial designs is , investigators are typically also interested in estimating the effect among nonparticipants, .…”

“…The second approach extends inferences from trial participants to the population of nonparticipants using a model for the probability of trial participation . In essence, we are treating the randomized trial participants as a sample from the target population with sampling probabilities that depend on baseline covariates and need to be estimated, an idea that connects this approach to survey sampling .…”

“…This tutorial builds on a growing literature to show that, instead of relying exclusively on informal assessments, we can use formal statistical methods to extend causal inferences from trial participants to a new target population. We focus on the common setting in which individual‐level data on time‐fixed treatments, outcomes, and baseline covariates are available from a randomized trial, but only individual‐level data on baseline covariates are available from the target population of nonparticipants.…”