Recent Advances in Sampling Bias, Generalizability and Transportability

Recent research in causal inference has made important progress in addressing challenges to the external validity of trial findings. Such methods weight trial participant data to more closely resemble the distribution of effect-modifying covariates in a well-defined target population. In the presence of participant non-adherence to study medication, these methods effectively transport an intention-to-treat effect that averages over heterogeneous compliance behaviors. In this paper, we develop a principal stratification framework to identify causal effects conditioning on both compliance behavior and membership in the target population. We also develop non-parametric efficiency theory for and construct efficient estimators of such "transported" principal causal effects and characterize their finite-sample performance in simulation experiments. We illustrate our methods with a study of the effect of a health system intervention studied in a randomized trial in Camden, New Jersey with the aim of determining whether the intervention shows promise among high engagers with the intervention in a Midwestern health system. While this work focuses on treatment non-adherence, our framework is applicable to a broad class of estimands that target effects in clinically-relevant, possibly latent subsets of a target population defined by post-randomization events, such as effects among survivors. 

When estimating causal effects, it is important to assess external validity, i.e., determine how useful a given study is to inform a practical question for a specific target population. One challenge is that the covariate distribution in the population underlying a study may be different from that in the target population. If some covariates are effect modifiers, the average treatment effect (ATE) may not generalize to the target population. In this talk, we propose new methods to generalize or transport the ATE from a source population to a target population, in the case where the source and target populations have different sets of covariates. When the ATE in the target population is identified, we propose new doubly robust estimators and establish their rates of convergence and limiting distributions. Under regularity conditions, the doubly robust estimators provably achieve the efficiency bound and are locally asymptotic minimax optimal. A sensitivity analysis is provided when the identification assumptions fail. Simulation studies show the advantages of the proposed doubly robust estimator over simple plug-in estimators. Importantly, we also provide minimax lower bounds and higher-order estimators of the target functionals. The proposed methods are applied in transporting causal effects of dietary intake on adverse pregnancy outcomes from an observational study to the whole U.S. pregnant female population. 

Learning the Individual Treatment Effect (ITE) is essential for personalized decision-making, yet causal inference has traditionally focused on aggregated treatment effects. While integrating conformal prediction with causal inference can provide valid uncertainty quantification for ITEs, the resulting prediction intervals are often excessively wide, limiting their practical utility. To address this limitation, we introduce Surrogate-assisted Conformal Inference for Efficient iNdividual Causal Effects (SCIENCE), a framework designed to construct more efficient prediction intervals for ITEs. SCIENCE accommodates the covariate shifts between source data and target data and applies to various data configurations, including semi-supervised and surrogate-assisted semi-supervised learning. Leveraging semi-parametric efficiency theory, SCIENCE produces rate double-robust prediction intervals under mild rate convergence conditions, permitting the use of flexible non-parametric models to estimate nuisance functions. We quantify efficiency gains by comparing semi-parametric efficiency bounds with and without the surrogates. Simulation studies demonstrate that our surrogate-assisted intervals offer substantial efficiency improvements over existing methods while maintaining valid group-conditional coverage. Applied to the phase 3 Moderna COVE COVID-19 vaccine trial, SCIENCE illustrates how multiple surrogate markers can be leveraged to generate more efficient prediction intervals. 

In prevalent cohort studies with delayed entry, the time-to-event outcome is subject to left truncation when only subjects that have not experienced the event at study entry are included. This leads to selection bias, as subjects with early event times tend not to be captured. Conventional methods for handling left truncation usually rely on the random left truncation or the slightly weaker quasi-independence assumption that requires the left truncation time and the event time are independent on the observed region. This assumption can be further relaxed to conditional (quasi-)independent left truncation which assumes that the dependence-inducing covariates are measured. However, in practice, the conditional independent left truncation assumption may fail, and measured covariates may only serve as imperfect proxies for explaining the underlying mechanism that induces the dependence between the left truncation time and the event time. In this work, we propose a proximal weighting identification framework which admits that the measured covariates may only be imperfect proxies for capturing the dependence between the left truncation time and the event time. We then construct estimators based on the framework and study their asymptotic properties. We examine the finite sample performance of the proposed estimators by comprehensive simulations. We apply the proposed method to analyze cognitive impairment-free survival using data from the Honolulu Asia Aging Study.