AI-driven RNA-based homologous recombination deficiency algorithm to predict first-line platinum response in metastatic pancreatic cancer.

Abstract

4210 Background: Developing predictive diagnostics for metastatic pancreatic cancer remains an unmet clinical need. Here, we present the Tempus HRD-RNA algorithm: an AI-driven, 1,660-gene logistic regression model trained on RNA-seq (Tempus xR) from pan-cancer solid tumor tissue samples. This dynamic RNA-based signature can identify clinically relevant patient populations likely to benefit from platinum-based regimens and PARP inhibitors, including patients without BRCA1/2 or PALB2 mutations. Methods: This is a prospectively designed retrospective RWD analysis in an independent validation cohort of 1,083 patients (pts) with treatment-naive metastatic pancreatic cancer diagnosed between 2017 and 2025. Treatment arms included first-line platinum-containing regimens (Arm A; n = 779; 97.0% FOLFIRINOX, 1.4% NALFIRINOX, 1.5% Gemcitabine/Cisplatin) or a non-platinum regimen (Arm B; n = 304; 86.6% Gemcitabine/Nab-paclitaxel, 13.4% Gemcitabine/Paclitaxel). BRCA1/2 and PALB2 mutations were determined via paired Tempus xT DNA sequencing. Key prognostic criteria included age, sex, ECOG score (0 or 1), surgery status and biopsy site. Unknown ECOG was imputed using the Laboratory Imputation Framework for EHRs (LIFE; PMID: 40595435). Overall survival (OS) differences between treatment arms were evaluated as independent Cox models for HRD+ vs. HRD- pts, including covariate adjustment. Results: Overall, 9.5% of pts were classified as HRD+: 78 pts (10.0%) in the platinum arm (Arm A) and 25 pts (8.2%) in the non-platinum arm (Arm B). OS was significantly higher in platinum-treated HRD+ pts than non-platinum. Median OS (mOS) was 12.8 months (95% CI: 8.5–16.6) in HRD+ platinum pts (Arm A) vs. 8.5 months (95% CI: 5.4–12.9) in HRD+ non-platinum-treated pts (Arm B; HR: 0.57; 90% CI: 0.33–0.99; p = 0.048). In contrast, in the HRD- group, there was no statistically significant difference in OS between platinum (Arm A; mOS 8.1 months; 95% CI: 7.1–9.1) and non-platinum (mOS 7.6 months; 95% CI: 6.6–8.8) arms (Arm B; HR: 0.91; 90% CI: 0.76–1.09; p = 0.190). Additional analyses were performed within the platinum-treated arm (Arm A) to assess if the survival benefit held independent of BRCA1/2 or PALB2 mutations. HRD+ and BRCA/PALB2mut pts (4.8%; n = 37) had a mOS of 13.3 months compared to 11.5 months in HRD+ and BRCA/PALB2wt (5.3%; n = 41) pts (HR: 0.7; 90% CI: 0.4–1.1). Comparatively, HRD- and BRCA/PALB2mut pts (2.1%; n = 16) had a mOS of 10.5 months compared to 8.3 months in HRD- and BRCA/PALB2wt (80.9%; n = 623) pts (HR: 0.7; 90% CI: 0.4–1.3). Conclusions: Tempus HRD-RNA is a novel predictive clinical biomarker that identifies a higher percentage of pts who may benefit from targeted treatment than current guidelines. The HRD-RNA algorithm offers a remarkable precision oncology tool to improve outcomes in pancreatic cancer, and has important clinical value in the era of novel HRD drug development.