A collaboration between BullFrog AI and H. Lee Moffitt Cancer Center, with trial data provided by Eleison Pharmaceuticals. Accepted for poster presentation at the 2026 ASCO Gastrointestinal Cancers Symposium.
Naleid N, Gosik K, Tambe A, Savkli C, Beltrán JF, Kim RD. Data-driven subtyping and differential glufosfamide benefit in pancreatic adenocarcinoma. J Clin Oncol 44, no. 2_suppl (2026): 755. doi:10.1200/JCO.2026.44.2_suppl.755
2.9× median survival in Cluster A (glufosfamide vs. best supportive care) · 4 reproducible patient clusters · ~10,000 pages of trial reports processed · 3-week analysis cycle
Background
The Phase 3 TH-CR-302 trial evaluated glufosfamide, a glucose-conjugated alkylating agent targeting GLUT1 transporters, versus best supportive care in advanced pancreatic cancer. The trial found no statistically significant overall survival benefit across the full enrolled population.
Analysis
BullFrog AI applied its bfPREP™ workflow to extract and structure approximately 10,000 pages of PDF-formatted Clinical Trial Reports into an OMOP-compatible dataset. The bfLEAP® ensemble clustering platform then identified four stable patient subgroups using baseline clinical and laboratory features, with survival outcomes evaluated within each cluster using age- and sex-adjusted Cox proportional hazards models.
Findings
In Cluster A (n=12), median overall survival was 153 days on glufosfamide versus 52 days on best supportive care (mean 149.3 vs 58.2 days, t-test p=0.065). Cluster A also had the lowest median baseline glucose of any subgroup, consistent with glufosfamide’s GLUT1-mediated uptake mechanism. Clusters B and D showed higher rates of Grade 3–5 adverse events. Cluster C showed little to no treatment effect.
Collaborators
BullFrog AI (Gaithersburg, MD) and H. Lee Moffitt Cancer Center (Tampa, FL), with trial data provided by Eleison Pharmaceuticals (Princeton, NJ).