ASA BIOPHARMACEUTICAL SECTION STUDENT PAPER AWARDS
Francis Rogan (Merck)
The Biopharmaceutical Section has announced the winners of the 2025 American Statistical Association (ASA) Student Paper Awards from a total of 58 submissions. The selected students will present their papers during one of the contributed paper sessions at the 2025 Joint Statistical Meetings (JSM) in Nashville, Tennessee. The awards will be presented at the Biopharmaceutical Section's open business meeting at JSM along with a cash prize. We would like to thank all those who’ve participated in this competition and congratulate the winners! For more details on the competition, please visit: https://community.amstat.org/biop/awards/studentpapercompetition.
FIRST PRIZE: Yuhan Qian, University of Washington
TITLE: From Estimands to Robust Inference of Treatment Effects in Platform Trials
What is your paper about?
We present a clear framework for constructing a clinically meaningful estimand with a precise specification of the population of interest in a platform trial. Additionally, we develop methods for robustly estimating treatment effects with minimal assumptions. Our proposed entire concurrently eligible (ECE) population is critical for addressing key issues in future statistical research, as it provides a clear reference point for evaluating both efficiency gains and potential bias.
What are your plans after graduation?
I plan to pursue a career in academia after graduation.
SECOND PRIZE: Daoyuan Lai, University of Hong Kong
TITLE: Bayesian Transfer Learning for Enhanced Estimation and Inference
What is your paper about?
In this paper, we introduce a Bayesian transfer learning method called TRansfer leArning via guided horseshoE prioR (TRADER). This method improves estimation and inference in high-dimensional linear models by borrowing information from other datasets. TRADER has several advantages over existing frequentist approaches: (1) it requires only summary-level information from the source, (2) it can use source estimates that are close to the target estimate at a small angle, while current methods need a small Euclidean distance, which is stricter, and (3) it offers more precise credible intervals.
We conducted extensive simulations to evaluate TRADER's performance. We also investigated its posterior contraction rate and finite-sample marginal posterior behavior. Additionally, our method addresses the over-shrinkage problem often seen with standard continuous shrinkage priors when estimating coefficients with moderate signal strength. Importantly, TRADER aligns well with the principles of several established Bayesian information borrowing priors, including the meta-analytic predictive prior, commensurate prior, and unit information prior.
What are your plans after graduation?
I am currently seeking a postdoctoral position in the United States. My primary research focuses on the intersection of genetics and statistics, with the goal of developing statistical methods that offer strong theoretical foundations, robust empirical performance, and practical algorithms for real-world applications. Additionally, I am always interested in opportunities within the industry. In fact, TRADER was motivated during my internship in the Global Statistics and Data Science department at BeiGene, where I designed an R Shiny app that integrates multiple Bayesian methods to control covariate imbalance between treatment and control groups when borrowing information from historical clinical trials.
THIRD PRIZE: Xiaohan Chi, The University of Texas MD Anderson Cancer Center
TITLE: OP2-Comb Bayesian Optimal Phase II Design for Optimizing Doses and Assessing Contribution of Components in Drug Combinations
What is your paper about?
This paper describes a very easy-to-use Bayesian optimal phase II design for drug combinations (BOP2-Comb). To better align with Project Optimus, the BOP2-Comb design achieves two goals within the same trial framework: optimizing the combination dose and evaluating the contribution of each component in the drug combination. BOP2-Comb is optimized through a calibration scheme that minimizes the total trial sample size and controls incorrect decision rates. This calibration procedure is Monte Carlo simulation-free and provides a theoretical guarantee of false-positive control.
What are your plans after graduation?
After graduation, I will definitely continue delving into the field of clinical trials, further developing my expertise and applying my skills to impactful research. While I may prefer to pursue a faculty position, I am currently open to exploring both academia and industry opportunities.
HONORABLE MENTIONS
Xinying Fang, Pennsylvania State University
Title: Generalized Multi-stage Optimal Design for Phase II Studies with Particle Swarm Optimization
Summary: Our work presents a new approach to optimizing Phase II clinical trials, addressing the high costs and low success rates of drug development. We introduce a framework with a unified objective function for multi-stage designs, encompassing various optimality criteria for multi-stage designs. Additionally, we propose an advanced particle swarm optimization method, named PSO-GO, to overcome computational challenges, making multi-stage designs more feasible. The methodology, supported by theoretical foundations, simulations and a real-world case study, offers practical improvements over current trial designs, balancing scientific rigor with computational efficiency for Phase II trials.
Runjia Li, University of Pittsburgh
Title: A Doubly Robust Instrumental Variable Approach for Estimating Average Treatment Effects in Time-to-Event Data with Unmeasured Confounding: Application to Real-World Data on ICU Patients with Septic Shock
Summary: This paper proposes a novel instrumental variable (IV) estimator to assess the average treatment effect (ATE) in time-to-event data in the presence of unmeasured confounding, addressing limitations in existing methods that rely on strong parametric assumptions. The proposed estimator is doubly robust, asymptotically efficient, and adaptable to machine learning models, making it suitable for complex real-world data. Through simulations, it demonstrates strong statistical properties. Applying this method to electronic health records (EHR) of ICU patients, with physician prescribing preferences as the IV, the study finds no significant benefit or harm of hydrocortisone on mortality. This approach provides reliable causal estimates despite unmeasured confounding, aiding clinical decision-making.
Lei Yan, Florida State University, Department of Statistics
Title: Optimizing Quality Tolerance Limits Monitoring in Clinical Trials Through
Machine Learning Methods
Summary: This paper introduces a machine learning enabled approach to facilitate real-time, automated monitoring of Quality Tolerance Limits (QTLs) in clinical trials. Unlike the traditional quality assurance process, where QTLs are evaluated based on single-source data and arbitrary defined fixed threshold, our QTL-ML framework integrates 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. Embedded within ICH-E6 recommended risk-based monitoring principles, this innovative machine learning solution for QTL monitoring has the potential to transform sponsors’ ability to protect patient safety, reduce trial duration, and lower trial costs.
Jialing Liu, University of Minnesota, Division of Biostatistics and Health Data Science
Title: Variable Selection and Prediction for Longitudinal Data Using Bayesian Transfer Learning
Summary: Contemporary longitudinal data typically involve high-dimensional time-course measurements on small samples, making model estimation challenging due to variability and unstable predictions. It's natural to borrow information from additional datasets with similar covariate-outcome relations to improve inference in the target data. We propose a novel Bayesian transfer learning model for longitudinal data (BTLL) that uses mixture models to adaptively account for differences between source and target outcome parameters. This approach minimizes bias by limiting inappropriate information transfer, and simulation studies demonstrate that BTLL significantly enhances parameter precision and reduces bias in heterogeneous settings.
Siyi Liu, North Carolina State University (graduated in December, 2024)
Title: Improved inference for survival heterogeneity of treatment effect leveraging trial and observational studies
Summary:
This paper aims at leveraging observational studies to enhance the inference of the heterogeneity of treatment effect for time-to-event outcomes while addressing unmeasured confounding. To achieve this, a confounding function is introduced to quantify the discrepancy between observed and causal treatment effects based on measured covariates, and it facilitates the identification and construction of an integrative estimator by minimizing a penalized loss function. The proposed estimator demonstrates a promising convergence rate, asymptotic normality, and efficiency at least equal to that of the trial estimator, with its effectiveness further validated through simulations and real-world application.
With this year's competition coming to an end, we want to give a big shout-out to all the winners for their incredible achievements and hard work. Your success is truly inspirational! To everyone who participated, we appreciate your efforts and enthusiasm. Keep working on your skills and passions. We can't wait to see you shine in the years to come. Keep going, keep dreaming, and keep achieving!