ASA BIOP Student Paper Competition Awards Session

Leveraging external control (EC) data can improve efficiency in estimating causal effects in clinical trials, but the use of non-randomized data complicates statistical inference. Propensity score (PS)-based adjustments are commonly used to remove confounding bias under ignorability assumptions, yet the variance behavior of the resulting inferential statistics is less studied, and naive application can lead to erratic type-I error rates. This paper investigates the behavior of test statistics for commonly used procedures after PS-based EC borrowing, identifies the reasons for misbehaved type-I errors, and proposes remedies for PS matching and weighting in single-arm and hybrid randomized controlled trial (RCT) designs. Extensive simulation studies demonstrate that the proposed remedies provide adequate variance estimation and recover type-I error rates close to the nominal level. Based on these results, we offer recommendations for practitioners to ensure valid causal effect estimation when borrowing ECs and illustrate the methods using a real RCT. 

Unstructured clinical notes contain comprehensive and highly relevant clinical information regarding the conditions and treatment of patients with acute ischemic stroke. Such information is critical for identifying nuanced acute stroke patient subtypes that can inform appropriate therapeutic strategies. Although topic modeling is a powerful tool for identifying patient subtypes from clinical notes, applying it at scale across healthcare institutions remains challenging because of privacy constraints and data heterogeneity. We introduce Federated Topic-SCORE, a communication-efficient algorithm that extends a spectral topic modeling framework to multi-institutional settings without sharing patient-level data. By transmitting only summary statistics that locally estimate low-rank singular subspace of topic loadings in a single round of communication, the method accurately recovers global singular subspace and topic loadings shared across sites. In simulations, Federated Topic-SCORE closely approximates the performance of pooled analyses and substantially outperforms models trained solely on individual institutions, particularly when topic weight distributions differ across sites. When implemented on multi-institutional clinical notes to identify etiological subtype topics for ischemic stroke hospitalizations, the federated model successfully recovers stroke subtypes aligned with known categories such as cardioembolic, large-artery atherosclerosis, and small vessel disease, while also revealing clinically meaningful variations within embolic presentations. These findings highlight the utility of one-shot federated topic modeling for scalable, privacy-preserving analysis of multi-institutional unstructured clinical notes. 

The analysis of randomized controlled trials is often complicated by intercurrent events (IEs), which occur after treatment initiation and affect the interpretation or existence of outcome measurements, such as treatment discontinuation or additional medication use. In two recent clinical trials for systemic lupus erythematosus, we classify IEs into effect-informative and effect-uninformative categories. To define a clinically meaningful estimand, we adopt tailored strategies for each type. For effect-informative IEs, we use a composite strategy that assigns an outcome reflecting treatment failure. For effect-uninformative IEs, we apply a hypothetical strategy, assuming their timing is conditionally independent of the outcome given treatment and baseline covariates. We further address competing IEs, where the first event censors subsequent ones. We develop a unified framework for estimand formulation, nonparametric identification, and semiparametric estimation, and propose weighting, outcome regression, and doubly robust estimators. Applying our methods to the two trials demonstrates robustness and practical value. 

Quasi-experimental evaluations are critical for generating real-world causal evidence and complementing insights from randomized trials. The regression discontinuity design (RDD) is a quasi-experimental framework for estimating causal effects when treatment assignment depends on a running variable crossing a threshold. Such threshold-based rules are ubiquitous in healthcare, where predictive and prognostic biomarkers frequently guide treatment decisions. However, standard RDD estimators rely on complete outcome data, an assumption often violated in time-to-event analyses where censoring arises from loss to follow-up. To address this issue, we propose a nonparametric approach that leverages doubly robust censoring corrections and can be paired with existing RDD estimators. Our approach can handle multiple survival endpoints, long follow-up times, and covariate-dependent variation in survival and censoring. We discuss the relevance of our approach across multiple biomedical applications and demonstrate its usefulness through simulations and the PLCO prostate Cancer Screening Trial. We have also developed an open-source software package rdsurvival for the R language. 

Individual patient data (IPD) are essential for oncology evidence synthesis but are rarely available, motivating reconstruction from published KM curves. Existing methods are limited by digitization error, unrealistic censoring assumptions, and inability to recover subgroup IPD from aggregate summaries. We propose RESOLVE-IPD, a unified framework for high-fidelity IPD reconstruction and uncertainty-aware subgroup recovery. The reconstruction engine combines VEC-KM, which extracts precise KM coordinates and censoring marks from vectorized figures, and CEN-KM, which resolves overlapping censor symbols without assuming uniform censoring. The subgroup module, MAPLE, identifies an ensemble of data-compatible labelings that match reported statistics and enables meta-analysis with explicit propagation of subgroup uncertainty. Across four trials in advanced esophageal squamous cell carcinoma focused on the PD-L1–low population, reconstructed IPD closely matched published KM curves and summary statistics, while MAPLE recovered plausible subgroup assignments. Ensemble meta-analysis demonstrated a survival benefit of immunotherapy over chemotherapy, most pronounced between 6 and 12 months. 

A platform trial is an innovative clinical trial design that enables simultaneous and continuous evaluation of multiple treatments within a single master protocol. Existing robust methods restrict analyses to concurrently randomized participants due to concerns that including nonconcurrent data may introduce bias from temporal trends. However, this exclusion represents a missed opportunity to improve efficiency. We propose a Gaussian process framework for incorporating nonconcurrent data that exploits temporal smoothness, a key feature of platform trials. The framework includes single-task and multi-task formulations and provides data-adaptive integration of nonconcurrent data with uncertainty quantification. The connection to kernel ridge regression yields a transparent frequentist interpretation of how nonconcurrent data are integrated. We establish two theoretical guarantees: incorporating nonconcurrent controls reduces the posterior variance of the treatment effect, and the resulting bias is controlled by a non-increasing bound. We extend the framework to discrete outcomes and to covariate adjustment, illustrate it on a hypothetical platform trial constructed from SURMOUNT-1, and provide an implementation in the R package RobinCID.