| Time |
Event/Speaker/Session Title |
Session Description |
| 7:30-8:30 a.m. |
Breakfast (60 min) |
| 8:30-8:35 a.m. |
Welcome
Tomi Mori |
| 8:35-8:45 a.m. |
Opening Remarks
Charlie Roberts |
| 8:45-10:20 a.m. |
Session #1: Clinical Trials
Session Chair: Haitao Pan |
| 8:45-9:10 a.m. |
Beyond Efficacy: Case Studies in Effectiveness and Implementation Trials
Elizabeth Hill |
Over the last two decades, historic transformations in the cancer treatment landscape – including the introduction of novel targeted therapies, immunotherapies, and cellular therapies – have turned a once fatal diagnosis into a survivable and sometimes curable condition. Despite remarkable therapeutic advances, provider- and institution-level barriers to translation can induce significant gaps in the uptake of evidence-based treatment, and blunt the delivery of much-needed supportive care to cancer survivors. We present two case studies of ongoing trials at our institution, one to increase rates of timely post-operative radiation therapy in head and neck cancer patients, and the second to improve depression screening, referral and treatment uptake rates for cancer survivors. These trials fall within the broad class of hybrid effectiveness and implementation trials, studies designed to hasten the translation of efficacious interventions into clinical care by blending the dual objectives of evaluating both effectiveness and implementation outcomes within the same trial framework. |
| 9:10-9:35 a.m |
From Belief to Evidence: Lessons from Cancer Research
Ji-Hyun Lee |
How do promising medical ideas become accepted clinical practice? Behind every important medical decision lies a process of generating, evaluating, and interpreting evidence, and biostatisticians play a central role in that process. In this talk, I will share two stories from my own research that illustrate how statistical thinking guides medical decision-making. The first examines a long-standing dietary practice for patients with cancer and how a randomized clinical trial challenged conventional wisdom. The second follows the evaluation of emerging cancer therapies, from repurposed drugs to mRNA-based cancer vaccines, highlighting how observational studies, electronic health records, and randomized clinical trials each contribute to the evidence-generation process. I will conclude by discussing how AI and large-scale health data are creating new opportunities for cancer research while reinforcing the importance of rigorous study design, high-quality data, and statistical leadership. Together, these examples illustrate that the role of biostatisticians extends well beyond data analysis: we help determine what constitutes trustworthy evidence and how that evidence informs patient care. |
| 9:35-9:50 a.m. |
Advancing clinical trials in pediatric oncology and other rare disease populations through cure model-based methodology
Subodh Selukar |
Advances in diagnostics and therapeutics have succeeded in creating substantial numbers of long-term survivors for many pediatric cancers. However, these successes, when combined with the relatively low incidence of pediatric cancers, create challenges for the continued development of novel approaches. Rare populations limit the feasible numbers of participants who can be recruited for clinical trials, and increasing rates of long-term survivorship limit the observed numbers of events for traditional time-to-event endpoints. Moreover, these factors will become more pronounced as clinical research increasingly focuses on finer risk stratification, more refined disease classifications, and targeted therapies, further subdividing patient populations throughout pediatric oncology. In current practice, these challenges frequently require trialists to extend the duration of clinical trials or accept relaxed statistical operating characteristics like increased type-1 error or lowered power when employing conventional approaches to design clinical trials. Cure models are time-to-event data models that were developed to better account for long-term survivorship relative to conventional models, but they have seen limited use for the primary statistical design of clinical trials. This is because cure models make important assumptions that can meaningfully affect study conclusions when applied inappropriately. However, these assumptions have never been systematically evaluated for pediatric oncology clinical trials. In this talk, we will review recent and upcoming work to investigate the potential of cure models for pediatric oncology and to develop novel methodology that maximizes benefits while minimizing the weaknesses of cure model-based designs. This body of research aims to deliver innovative methods and practical guidance that address barriers to clinical trials in pediatric oncology and other rare disease populations. |
| 9:50-10:05 a.m. |
Randomized optimal selection design for dose optimization
Shuqi Wang |
The US Food and Drug Administration (FDA) launched Project Optimus to shift the objective of dose selection from the maximum tolerated dose to the optimal biological dose (OBD), optimizing the benefit-risk tradeoff. One approach recommended by the FDA's guidance is to conduct randomized trials comparing multiple doses. In this paper, using the selection design framework, we propose a Randomized Optimal SElection (ROSE) design, which minimizes sample size while ensuring the probability of correct selection of the OBD at pre-specified accuracy levels. The ROSE design is simple to implement, involving a straightforward comparison of the difference in response rates between two dose arms against a predetermined decision boundary. We further consider a two-stage ROSE design that allows for early selection of the OBD at the interim when there is sufficient evidence, further reducing the sample size. Simulation studies demonstrate that the ROSE design exhibits desirable operating characteristics in correctly identifying the OBD. A sample size of 15-40 patients per dosage arm typically results in a percentage of correct selection of the optimal dose ranging from 60% to 70%. |
| 10:05-10:20 a.m. |
Q & A |
| 10:20-10:40 a.m. |
Break |
| 10:40 a.m.-12:15 p.m. |
Session #2: Observational Studies
Session Chair: Kumar Srivastava |
| 10:40-11:05 a.m. |
If you build it, will diet change? Evidence from the PHRESH Natural Experiment
Bonnie Ghosh |
Diet is a key social determinant of health, yet communities with low-income and racial minorities face disproportionate burdens of chronic disease, while often having limited access to healthy foods. In response, the Healthy Food Financing Initiative (HFFI) incentivized supermarkets to open in “food deserts” to provide access to fresh food. However, can opening a supermarket really improve diet? Funded by the National Cancer Institute, the RAND PHRESH study was the first and largest natural experiment to evaluate this important question and the HFFI. Using quasi-experimental study design, we compared an intervention neighborhood with a matching “counterfactual”. Using causal methods including difference-in-differences (DID) and instrumental variables, we estimated the supermarket’s impact on diet. A community-based participatory research (CBPR) approach was central to the study’s success, supporting strong enrollment and retention. We found modest short-term improvements in diet, but these effects did not persist. In contrast, the new supermarket produced large and lasting improvements in residents’ satisfaction with their neighborhood. These findings suggest that while improving the neighborhood food environment matters, access alone may be insufficient to sustain dietary change. Meaningful improvements in diet likely require coordinated interventions at both the individual and neighborhood levels. |
| 11:05-11:30 a.m. |
An Estimand-Focused Approach for AUC Generalization and Cross-Study Benchmarking
Xiaofei Wang |
The area under the ROC curve (AUC) is the standard measure of a biomarker's discriminatory accuracy; however, AUC is rarely treated as a population-specific estimand. When validation cohorts differ from the intended target population in case mix, Naïve AUC estimates can mislead both generalization and cross-study comparison. We develop an estimand-focused framework that anchors biomarker AUC inference to a prespecified target population, aligning with the ICH E9(R1) estimand perspective adapted to discrimination rather than treatment effect. The framework supports two scientific goals: generalizing a study-specific AUC to a clinically relevant target population, and benchmarking AUCs across studies on a common population footing. Methodologically, we extend calibration weighting to the U-statistic formulation of AUC, allowing valid estimation even when the target population is characterized only by summary-level covariate information. This setting is common in biomarker validation, where individual-level target data are often unavailable, and existing transportability methods may not be applicable. When patient-level real-world data are accessible, the proposed augmented variants provide double robustness and improved efficiency. We establish asymptotic properties and study their performances through comprehensive simulations. Furthermore, we demonstrate the proposed framework on the POWER trials, evaluating baseline stair-climb power (SCP) as a prognostic marker for 6-month survival in advanced non-small-cell lung cancer (NSCLC). Unlike prior work on transporting model-based predictive accuracy, our framework targets the biomarker-level estimand directly and addresses cross-study comparability -- an issue not resolved by current methods. |
| 11:30-11:45 a.m. |
Time-to-Event Analysis Under Imperfect Recall and Informative Interval Censoring: Methods for Self-Reported Chronic Health Conditions in Cohort Studies
Sadie Mirzaei |
Large observational cohort studies often rely on self-reported chronic health conditions, where exact ages at onset are unavailable because participants cannot precisely recall when events occurred. These missing event times are commonly treated as interval-censored under the assumption of non-informative censoring. However, recall may depend on elapsed time since onset as well as treatment- or health-related factors associated with the event process, leading to informative interval censoring and potentially biased inference. Motivated by studies of childhood cancer survivors in the Childhood Cancer Survivor Study (CCSS), we develop statistical methods for time-to-event analyses under imperfect recall. First, we present a semiparametric Cox regression framework for estimating exposure-outcome associations when recall probability depends on both elapsed time and covariates related to failure risk. The proposed likelihood-based approach accounts for informative recall without specifying the underlying event-time distribution and reduces bias relative to conventional methods that ignore the recall process. We then address the more challenging setting of assessing associations between two chronic health conditions when onset times for both conditions are subject to imperfect recall and informative interval censoring. We propose a mixture modeling framework that captures the association between two events by allowing the hazard of the second event to change following the onset of the first event while explicitly accounting for recalled onset times and informative censoring. Simulation studies evaluate the finite-sample performance of the proposed methods. We illustrate the approaches using CCSS data to investigate treatment-related risk factors for growth hormone deficiency and to assess the association between diabetes and myocardial infarction among childhood cancer survivors. Together, these methods provide a flexible framework for valid time-to-event analyses of self-reported chronic health conditions in epidemiologic and clinical cohort studies. |
| 11:45 a.m.-12:00 p.m. |
Time-Dependent ROC and PRC Analysis for Left-Truncated Right-Censored Time-to-Event Data
Kendrick Li |
Time-dependent Receiver Operating Characteristics (ROC) analysis and Precision-Recall Curve (PRC) analysis are standard methods to evaluate the ability of a biomarker or predictive score to discriminate between cases and non-cases for time-to-event outcomes. Extensions of this useful method to left-truncated right-censored (LTRC) data have been understudied. In this presentation, we will introduce our recent work in developing ROC and PRC analysis under the LTRC setting. Specifically, we will introduce novel estimators for sensitivity (precision), specificity, recall, area under the ROC curve (AUROC), and area under the PRC curve (AUPRC) under different scenarios of relations between the left-truncation and right-censoring times. We will also discuss extensions to settings with competing risks. We evaluate the novel estimators of AUROC and AUPRC in comprehensive simulation studies and demonstrate the estimators by a data application using the St. Jude Lifetime Cohort Study. |
| 12:00-12:15 p.m. |
Q & A |
| 12:15-1:15 p.m. |
Lunch |
| 1:20-2:20 p.m. |
Statistics and the AI Revolution, and Hybrid Encoder-Decoder Statistical-AI Models for Complex Structured Object Data
Jeff Morris |
|
| |
Q & A |
| 2:20-3:30 p.m. |
Session #3A: Multiomics, Biomarker and AI
Session Chair: Cheng Cheng |
| 2:20-2:45 p.m. |
Rafael Irrizarry |
|
| 2:45-3:00 p.m. |
Fusion of AI and Statistics for Bulk and Spatial Omics in Pediatric High-Grade Glioma
Qian Li |
Pediatric high-grade glioma (pHGG) is a lethal pediatric brain tumor, in which recurrence is associated with poor survival regardless of therapy. H3 G34-mutant diffuse hemispheric glioma (DHG) is a newly molecularly defined subtype of pHGG, distinct from other molecular subtypes, including H3 K27-altered diffuse midline glioma (DMG), which arises in midline structures. The cell-type-aware inference in rare tumors like G34-mutant DHG is challenging because of divergent neoplastic cell states and therapy-induced immune-evasive niches. We digitally decomposed the tumor immune microenvironment (TIME) by profiling paired bulk methylome-transcriptome in a retrospective multi-institutional cohort of 31 DHG primary diagnostic samples and 13 DMG samples (as control) and constructing a single-cell RNA-seq atlas as deconvolution reference. To dissect the developmental tumor cell states and potential therapeutic vulnerabilities in DHG, we used our recently published tool MOADE, a multimodal autoencoder framework, to jointly estimate DHG tumor cell neurodevelopmental states and TIME compositions from bulk multiomes, identifying age-dependent subgroups of DHG patients and subgroup-specific prognostic cell states. Meanwhile, we applied high-resolution spatial transcriptomics (ST) profiling to patient-matched treatment-naïve and recurrent DHG tumor sections to resolve the spatial architecture of TIME. To study the TIME dynamics from diagnosis to relapse in DHG, we developed BiNVaST, an innovative boundary-preserving and synthetic data-guided generative AI tool fused with a statistical simulator, to perform reliable cell phenotyping in high-resolution ST. BiNVaST generates platform-adjusted in silico training data by integrating query spatial profiles and user-defined pre-annotated single-cell reference, combining molecular similarity and spatial proximity to refine neoplastic and non-neoplastic niches in longitudinal tumor specimens. MOADE and BiNVaST successfully distinguished disease-specific cellular subpopulations in bulk multi-omics and high-resolution ST profiles, respectively, supporting robust cellular and/or niche inferences in H3 G34-mutant DHG tumors. |
| 3:00-3:10 p.m. |
Q & A |
| 3:10-3:30 p.m. |
Break |
| 3:30-4:20 p.m. |
Session #3B: Multiomics, Biomarker and AI
Session Chair: Zachary Wooten |
| 3:30-3:55 p.m. |
Large Margin Nearest Neighbor Classification for 2D and 3D Functional Data
Todd Ogden |
Linear and generalized linear scalar-on-function modeling have been commonly used to understand the relationship between a scalar response variable (e.g. continuous, binary outcomes) and functional predictors. Such techniques are sensitive to model misspecification when the relationship between the response variable and the functional predictors is complex. The Large Margin Nearest Neighbor (LMNN) classifier is a machine learning technique that can improve on K Nearest Neighbor algorithms in complex situations. We propose a novel method to integrate functional principal component analysis with LMNN techniques to account for the continuous nature of functional data and the nonlinear relationship between the scalar response variable and the functional predictors. We demonstrate the performance of our method through simulation experiments and real data applications. |
| 3:55-4:10 p.m. |
Jacob Luber |
|
| 4:10-4:20 p.m. |
Q & A |
| 4:20-5:00 p.m. |
Fireside Chats with Former ASA Presidents
Bonnie Ghosh - 2024 ASA President
Ji-Hyun Lee - 2025 ASA President
Internal (TBA) - Moderator - Tomi Mori |
| 5:00-5:05 p.m. |
Closing Remarks
Tomi Mori |
| 5:05-6:25 p.m. |
Poster Session/Reception（50 slots) |
| 6:30-8:00 p.m. |
Dinner with Faculty, Senior Leaders and Speakers (ARC 8th Floor) |
