Development of artificial intelligence-derived histological biomarkers for first-line treatment selection in metastatic pancreatic ductal adenocarcinoma (mPDAC).

743 Background: The prognosis of metastatic pancreatic ductal adenocarcinoma (mPDAC) remains poor with a median survival time of 10-12 months. First-line treatment is largely influenced by performance status with fit patients more often receiving FOLFIRINOX (FFX) than Gemcitabine+Nab-Paclitaxel (GNP). Although the two regimens have improved outcomes over gemcitabine monotherapy, no biomarkers routinely used in clinical practice can predict which regimen is optimal to facilitate a precision medicine approach. We developed two signatures (V-FFX and V-GNP) associated with treatment outcomes for the respective first-line regimens using a retrospective cohort of mPDAC cases. Methods: We conducted a retrospective study of mPDAC patients treated at two institutions (UPMC and Cedars Sinai) from 2014 to 2021. Digitized histological H&E sections corresponding to 145 metastatic PDAC patients treated with either first-line FFX or GNP were included. Independent randomized training and test datasets were constructed for FFX-treated (train: 41, test: 25) and GNP-treated (train: 49, test: 30) patients. To construct the histological assay, a deep-learning algorithm then segmented nuclei to extract quantitative histological features. Features associated with disease-specific survival (DSS) for FFX and GNP were identified utilizing univariate Cox proportional hazards (CPH) models for the respective training sets and V-FFX and V-GNP signatures were constructed. DSS stratification of the V-FFX and V-GNP signatures were examined using Kaplan-Meier analysis and the log-rank test and DSS percentages at 12 months were calculated on the respective test sets. Results: The V-FFX and V-GNP signatures were found to be significantly associated with treatment outcomes stratified in the respective test sets (log-rank test, V-FFX: p=0.046, V-GNP: p=0.004). 29 of 55 patients tested positive for only one of either V-FFX and V-GNP signatures. Kaplan-Meier analysis demonstrated robust separation with hazard ratios for the V-FFX and V-GNP signatures of 3.01 (95% CI: 0.96, 9.45) and 4.81 (95% CI: 1.74, 13.3). DSS at 12 months for patients in V-FFX +ve vs -ve groups were 88% (8/9) vs 50% (7/14). DSS at 12 months for patients in V-GNP +ve vs -ve groups were 66% (8/12) vs 15% (2/13). Conclusions: AI derived V-FFX and V-GNP morphological signatures were strongly associated with treatment outcomes for first-line FFX and GNP and can potentially aid in the selection of first-line treatment for mPDAC patients.