Junki Mizusawa is Section Chief of Biostatistics at the Center for Research Administration and Support, National Cancer Center Japan (NCCJ). With more than 15 years of experience, he has played a central role as a trial statistician in the Japan Clinical Oncology Group (JCOG), contributing to the planning, analysis, and publication of numerous multi-institutional confirmatory cancer clinical trials. His work has focused on multicenter trials in gastrointestinal, hepatobiliary and pancreatic, head and neck, and brain tumors, as well as investigator-initiated studies. Dr. Mizusawa received his PhD from Tokyo Medical and Dental University in 2024. His research interests include master protocols, quality of life, and pragmatic trials.
Kohei Uemura is an Associate Professor of Biostatistics at The University of Tokyo and a Director at Statcom, Inc. He previously worked as a statistical reviewer at Japan’s Pharmaceuticals and Medical Devices Agency (PMDA) for eight years. His research focuses on statistical methods for clinical trials, particularly adaptive designs, Bayesian methods, platform trials, and trials in rare diseases.
Oncology platform trials can improve the efficiency of drug development by evaluating multiple experimental treatments within a shared infrastructure. Borrowing non-concurrent control data may further increase efficiency, but temporal changes in patient characteristics and outcomes can bias treatment comparisons and inflate type I error. Discounting non-concurrent data alone may not adequately address these changes, whereas models assuming a common linear time trend can be sensitive to model misspecification. In this talk, we present a Bayesian approach that combines discounting of non-concurrent control data with explicit adjustment for time trends. The method extends a conditional power prior to a logistic regression model incorporating patient covariates and a time trend parameter, accounting for differences in patient characteristics while discounting the contribution of non-concurrent controls. We examine its operating characteristics through simulation studies covering linear trends, trends that differ between the non-concurrent and concurrent periods, and discontinuous changes. The findings illustrate the limitations of borrowing without explicit time trend adjustment and of relying on a correctly specified linear trend. Across the evaluated scenarios with differing or discontinuous trends, the proposed approach improved power while limiting bias and type I error inflation. We discuss the implications of these findings for the use of non-concurrent controls in oncology platform trials, focusing on the balance between efficiency gains from information borrowing and robustness to temporal changes.
J. Jack Lee is Professor of Biostatistics and the John G. & Marie Stella Kenedy Foundation Chair in Cancer Research at The University of Texas MD Anderson Cancer Center. He also directs MD Anderson's Biostatistics Resource Group. His research focuses on the design and analysis of cancer clinical trials, including Bayesian adaptive designs, statistical methods for precision medicine, drug-combination studies, and biomarker identification and validation. During more than 30 years at MD Anderson, Dr. Lee has collaborated with multidisciplinary teams across clinical, translational, and basic science research. He is a Fellow of the American Statistical Association, the Society for Clinical Trials, and the American Association for the Advancement of Science. He has published extensively in statistical and medical journals and co-authored books on Bayesian adaptive clinical-trial methods and model-assisted Bayesian dose-finding and optimization designs.
Bayesian adaptive designs, including basket and platform trials, are master-protocol clinical trials designed to evaluate one or more interventions, disease subtypes, treatment combinations, or biomarker-defined cohorts within a single infrastructure. They use prospectively planned adaptations based on accumulating data and can run perpetually. Prominent features include shared control groups, response-adaptive randomization, interim decision-making, dropping toxic or ineffective arms, adding new arms, and seamless transitions across trial stages. Compared with conventional stand-alone randomized trials, adaptive platform trials can improve efficiency, accelerate learning, reduce duplication of infrastructure, and increase the probability that patients receive promising therapies. However, they also introduce operational, regulatory, statistical, and logistical complexities, including control of type I error, non-concurrent controls, delayed outcomes, multiplicity, changing standards of care, data-sharing governance, and the need for rapid, reliable decision-making based on appropriate endpoints. Prominent examples of adaptive platform trials include NCI-MATCH, I-SPY 2, and GBM AGILE. In this talk, key statistical innovations and lessons learned will be discussed. Despite their promise, successful implementation requires careful planning, extensive simulation of operating characteristics, robust infrastructure, clear governance, timely endpoint ascertainment, and close collaboration among statisticians, clinicians, sponsors, regulators, and data monitoring committees.
Master protocol; Bayesian inference; efficient drug development; information borrowing; seamless design; simulations; operating characteristics
Toshiki Masuishi is a medical oncologist in the Department of Clinical Oncology at Aichi Cancer Center Hospital, Japan, specializing in gastrointestinal cancers. He serves as Vice Chair of the Basket Trial Committee of the West Japan Oncology Group (WJOG) and has led several investigator-initiated clinical trials. His work includes biomarker-driven basket trials evaluating brigatinib for ALK fusion-positive solid tumors and necitumumab-based therapy for EGFR-amplified solid tumors. His research focuses on developing new treatment strategies for gastrointestinal cancers and advancing precision oncology through clinical trials targeting rare molecular alterations.
Basket trials provide a framework for evaluating targeted therapies across tumor types, but biomarker selection and patient recruitment remain practical challenges. This presentation discusses two investigator-initiated phase II basket trials conducted by the West Japan Oncology Group (WJOG). WJOG15021M evaluated necitumumab in 22 patients with EGFR-amplified solid tumors identified by comprehensive genomic profiling. The confirmed objective response rate (ORR) was 22.7%, and the trial did not meet its primary endpoint. An exploratory analysis showed an ORR of 38.5% (5/13) in patients with EGFR amplification detected in circulating tumor DNA immediately before treatment, compared with 0% (0/9) in those without detectable amplification. These findings raise questions about biomarker assessment timing and the interpretation of subgroup signals after a negative primary analysis. ALLBREAK (WJOG15221M) evaluated brigatinib in ALK fusion-positive solid tumors other than non-small-cell lung cancer. Eligible patients were identified through genomic profiling or immunohistochemistry with fluorescence in situ hybridization confirmation. A decentralized trial platform enabled nine patients to participate remotely through partner hospitals. Twenty-eight patients were enrolled after the target sample size was expanded from 14 based on favorable accrual. Among 27 analyzed patients, the confirmed ORR by independent central review was 63.0%, meeting the primary endpoint. Together, these trials illustrate practical approaches to patient identification and access, while emphasizing the need for careful interpretation of evidence from small, molecularly selected populations.
Basket trials; tumor-agnostic drug development; biomarker selection; circulating tumor DNA; decentralized clinical trials
Hiroya Taniguchi is Head Physician in the Department of Clinical Oncology at Aichi Cancer Center Hospital, Japan, specializing in gastrointestinal cancers. From 2018 to 2021, he worked at the National Cancer Center Hospital East and contributed to the SCRUM-Japan, GOZILA, and CIRCULATE-Japan precision oncology projects. He served as coordinating investigator of the ctDNA-guided HERALD basket trial and has led several investigator-initiated trials in colorectal cancer. His current work focuses on decentralized and pragmatic approaches to improving clinical trial access and efficiency.
Basket trials enable efficient evaluation of molecularly targeted therapies across multiple tumor types, but the distribution of enrolled cancers is difficult to predict. Between December 2019 and March 2022, the investigator-initiated HERALD trial evaluated trastuzumab deruxtecan in patients with various advanced solid tumors other than conventionally HER2-defined breast and gastric cancers, selected on the basis of HER2 amplification detected by circulating tumor DNA. The original design used tumor-specific cohorts and a Bayesian robust exchangeability/nonexchangeability model to borrow information across cohorts while allowing for heterogeneity. However, accrual was slower and more uneven than anticipated, making the planned cohort-wise analysis impractical. The primary analysis was therefore revised to evaluate the overall response rate in the pooled population. Ultimately, 62 patients across 16 tumor types were enrolled, achieving an overall response rate of 56.5%. The HERALD trial served as key evidence in the regulatory review by the Pharmaceuticals and Medical Devices Agency and contributed to the approval of an expanded tumor-agnostic indication for trastuzumab deruxtecan in Japan in March 2026. HERALD highlights the need for trial designs that remain statistically rigorous and clinically interpretable when the basket does not fill as planned.
Basket trial; trastuzumab deruxtecan; tumor-agnostic drug development; Bayesian inference; robust EXNEX model; cohort heterogeneity