Abstract A105: The PASS-01 Challenge: Comparing Biomarkers for Treatment Selection in Advanced Pancreatic Cancer in a Randomized Controlled Trial
Wei Quan, David Henault, Syeda Mariam Hasnain, Daniela Bevacqua, Amy Zhang, Gun Ho Jang, Yangqing Deng, Julie Wilson, Anna Dodd, Andrew Aguirre, David Tuveson, Elizabeth Jaffee, Steven Gallinger, Faiyaz Notta, Jennifer J. Knox, Robert C. GrantAbstract
Background:
Predictive biomarkers for first-line treatment selection in advanced pancreatic ductal adenocarcinoma (PDAC) have been proposed, but none have been validated using data from a randomized controlled trial (RCT), which is the only design capable of unbiased assessment of differential treatment effect (DTE).
Methods:
We launched the PASS-01 Challenge, a rigorous, arms-length comparison of predictive and prognostic algorithms in PDAC using clinical, whole-genome, transcriptomic, and digitized histopathology data from PASS-01, a randomized phase II trial of FFX versus GnP in previously untreated metastatic PDAC. Participating groups receive de-identified multi-modal data, generate predictions independently, and submit locked predictions for centralized, blinded linkage to outcome data. The primary endpoint is concordance-for-benefit (C4B) for progression-free survival (PFS) in the per-protocol population; secondary endpoints include C4B and hazard ratios for overall survival (OS) and objective response (ORR), as well as the concordance index (C-index) for prognostic predictions. Participants can publish the performance of their predictions immediately after submission.
Results:
Here, we report four initial approaches under the Challenge: a multimodal machine-learning model (MULTIPL) that we developed and three previously published biomarkers (PurIST, hENT1 expression, HRDetect). MULTIPL, PurIST, and HRDetect were prognostic for OS (best C-index 0.595 [MULTIPL], P<0.001). In contrast, none of the four approaches achieved statistically significant treatment-selection performance by C4B for OS, ORR, or PFS, although exploratory subgroup analysis identified longer OS among MULTIPL-predicted GNP-favored patients treated with GnP versus FFX (HR 0.47, 95% CI 0.28–0.82, P=0.007).
Conclusions:
The PASS-01 Challenge enables rigorous, centrally-adjudicated, RCT-based validation of treatment-selection algorithms in PDAC, providing the first head-to-head benchmark of leading biomarker approaches. While the biomarkers evaluated so far are prognostic, treatment selection remains an unmet need. The Challenge remains open for submissions (hpb-research.ca) and invites investigators to test new and existing algorithms to accelerate progress in precision treatment selection for PDAC.
Citation Format:
Wei Quan, David Henault, Syeda Mariam Hasnain, Daniela Bevacqua, Amy Zhang, Gun Ho Jang, Yangqing Deng, Julie Wilson, Anna Dodd, Andrew Aguirre, David Tuveson, Elizabeth Jaffee, Steven Gallinger, Faiyaz Notta, Jennifer J. Knox, Robert C. Grant. The PASS-01 Challenge: Comparing Biomarkers for Treatment Selection in Advanced Pancreatic Cancer in a Randomized Controlled Trial [abstract]. In: Proceedings of the AACR Conference on Pancreatic Cancer: New Frontiers in Biology and Therapeutic Development; 2026 Sep 25-28; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(18_Suppl_2):Abstract nr A105.