ASA Biopharmaceutical Section Student Paper Competition Award

The analysis of contemporary longitudinal data problems involve high-dimensional measurements of time-course data collected on a small number of observations. The estimation of such models with limited sample sizes in a target population poses substantial challenges and can lead to highly variable parameter estimates, unstable predictions, and lower power. In such instances, it is natural to borrow information from additional datasets with similar covariate-outcome relations to improve inference in the target data. Here, we develop a novel Bayesian transfer learning model for longitudinal data (BTLL). BTLL leverages mixture models for the discrepancies between pivotal parameters of the outcome models of the source and target studies to enhance the accuracy of the parameter estimates and enable data-adaptive information borrowing. BTLL aims to minimize the transfer of information from source studies that would introduce large bias into posterior inference in the target study. Extensive simulation studies show that BTLL improves the precision of parameter estimates in the target study substantially, and reduces the bias, in heterogeneous settings when the outcome. 

Motivated by conflicting conclusions regarding hydrocortisone's treatment effect on ICU patients with vasopressor-dependent septic shock, we developed a novel instrumental variable (IV) estimator to assess the average treatment effect (ATE) in time-to-event data. In real-world data, IV methods are widely used for estimating causal treatment effects in the presence of unmeasured confounding, but existing approaches for time-to-event outcomes are often constrained by strong parametric assumptions and lack desired statistical properties. Based on our derived the efficient influence function (EIF), the proposed estimator possesses double robustness and achieves asymptotic efficiency. It is also flexible to accommodate machine learning models for outcome, treatment, instrument, and censoring for handling complex real-world data. Through extensive simulations, we demonstrate its double robustness, asymptotic normality, and ideal performance in complex data settings. Using electronic health records (EHR) from ICU patients, our analysis shows no significant benefit or harm of hydrocortisone on these patients. 

Personalized cancer treatment using combination therapies offers substantial therapeutic benefits over single-agent treatments in most cancers. However, unmet clinical needs and increasing market competition pressure drug developers to quickly optimize combination doses and clearly demonstrate the contribution of each component when developing and evaluating new combination treatments. We propose a Bayesian optimal phase II drug-combination (BOP2-Comb) design that optimizes the combination dose and evaluates the proof-of-concept as well as the contribution of each component in two seamless stages. Our optimal calibration scheme minimizes the total trial sample size while controlling incorrect decision rates at nominal levels. This calibration procedure is Monte Carlo simulation-free and provides a theoretical guarantee of false-positive control. We demonstrate the superior finite-sample operating characteristics of our design through extensive simulations. For illustration, we apply the proposed design to a real phase II trial evaluating the combination therapy of bevacizumab and lomustine. 

Phase II clinical trials are pivotal in experimental treatments. Despite an average cost of \$21 million per trial, only one-third of drugs succeed in Phase II. Multi-stage adaptive designs represent a significant advancement. However, current approaches face major limitations: sample-size-minimization designs are typically restricted to three stages due to computational complexity, while power-maximization designs frequently sacrifice interim cohort size optimization, resulting in sub-optimal trial configurations. To address these challenges, we propose generalized multi-stage optimal designs for Phase II trials. Our framework integrates both design classes through a unified objective function, transforming optimization into a coherent minimization task. To overcome computational bottlenecks, we introduce PSO-GO, a practical variant of particle swarm optimization (PSO) tailored for combinatorial design space optimization. Using this framework, we developthe G-BOP2 design, which incorporates multi-faceted enhancements to the Bayesian Optimal Phase II (BOP2) design. Simulations and real example demonstrate that G-BOP2 provides robust and efficient design solutions. 

Transfer learning enhances model performance in target populations with limited samples by leveraging related studies. While much work focuses on predictive performance, statistical inference remains challenging. Bayesian methods offer uncertainty quantification but often require single-source or individual-level data. We propose TRAnsfer leArning via guideD horseshoE prioR (TRADER), enabling multi-source transfer via pre-trained models in high-dimensional linear regression. TRADER shrinks target parameters toward a weighted average of source estimates, accounting for differing source scales. Theoretical results show TRADER achieves faster posterior contraction rates than standard priors when sources align well with the target while mitigating negative transfer from heterogeneous sources. Finite-sample analysis shows TRADER maintains frequentist coverage probabilities even for moderate signals, where standard priors falter. Numerical studies and a real-data application estimating blood glucose–insulin use associations in a Hispanic diabetic population demonstrate TRADER's superior estimation and inference accuracy over standard priors and state-of-the-art methods. 

A platform trial is an innovative clinical trial design that uses a master protocol (i.e., one overarching protocol) to evaluate multiple treatments, where patients are often assigned to different subsets of treatment arms based on individual characteristics, enrollment timing, and treatment availability. While offering increased flexibility, this constrained and non-uniform treatment assignment poses inferential challenges, with two fundamental ones being the precise definition of treatment effects and robust, efficient inference on these effects. Such challenges arise primarily because some commonly used analysis approaches may target estimands defined on populations inadvertently depending on randomization ratios or trial operation format, thereby undermining interpretability. This article, for the first time, presents a formal framework for constructing a clinically meaningful estimand with precise specification of the population of interest. Specifically, the proposed entire concurrently eligible (ECE) population not only preserves the integrity of randomized comparisons but also remains invariant to both the randomization ratio and trial operation format. Then, we develop weighting and post-stratification methods to estimate treatment effects under the same minimal assumptions used in traditional randomized trials. We also consider model-assisted covariate adjustment to fully unlock the efficiency potential of platform trials while maintaining robustness against model misspecification. For all proposed estimators, we derive asymptotic distributions, propose robust variance estimators, and compare them in theory and through simulations. The SIMPLIFY trial, a master protocol assessing the continuation versus discontinuation of two common therapies in cystic fibrosis, is utilized to further highlight the practical significance of this research. All analyses are conducted using the R package RobinCID. 

While trials are attractive with guaranteed treatment randomization, they often lack the power for treatment effect evaluation due to limited and unrepresentative participants. Observational studies, with their accessible large-scale data, can help increase the study power and facilitate current trial analyses. However, combining data from both sources raises concerns about general exchangeability, as the absence of treatment randomization in observational studies leads to unmeasured confounding and obscures the true effect. When targeting the survival heterogeneous treatment effect, it is crucial to address this issue and formulate an integrative inference to improve efficiency. Under the Cox model, we introduce a confounding function to quantify bias between observed and causal effects, which can be identified by integrating the two data sources. Using a linear HTE structure for interpretability, we apply sieve approximation for nuisance functions and derive an integrative estimator via penalized loss minimization. This estimator achieves a promising convergence rate, asymptotic normality, and at least trial-level efficiency, validated through simulations and an application. 

The traditional clinical trial monitoring process, which relies heavily on site visits and manual review of accumulative patient data reported through Electronic Data Capture system, is time-consuming and resource-intensive. The recently emerged risk-based monitoring (RBM) and quality tolerance limit (QTL) framework offers a more efficient alternative solution to traditional source data verification based quality assurance. These frameworks aim at proactively identifying systematic issues that impact patient safety and data integrity. In this paper, we proposed a machine learning enabled approach to facilitate real-time, automated monitoring of clinical trial QTL risk assessment. Unlike the traditional quality assurance process, where QTLs are evaluated based on single-source data and arbitrary defined fixed threshold, we utilize the QTL-ML framework to integrate information from multiple clinical domains to predict the clinical QTL of variety types at program, study, site and patient level. Moreover, our approach is assumption-free, relying not on historical expectations but on dynamically accumulating trial data to predict quality tolerance limit risks in an automated manner.