Abstract 5397: Prediction of conversion surgery completion following neoadjuvant modified FOLFIRINOX in borderline resectable and locally advanced pancreatic adenocarcinoma: machine learning algorithm analysis

Abstract

Abstract Background: Neoadjuvant modified (m) FOLFIRINOX is a standard therapy for medically fit patients with borderline resectable pancreatic cancer (BRPC) and locally advanced unresectable pancreatic cancer (LAPC). Completion of conversion surgery following mFOLFIRINOX is an important prognostic factor. In this study, we evaluated multiple machine learning models to predict completion of conversion surgery following neoadjuvant mFOLFIRINOX in BRPC and LAPC. Methods: Between January 2017 and December 2020, a total of 647 patients with BRPC and LAPC treated with neoadjuvant mFOLFIRINOX at Asan Medical Center, Seoul, South Korea were enrolled. All demographic and clinicopathological data were retrospectively collected; age, ECOG performance status, BMI, resectability as per the NCCN guidelines, tumor location, tumor size, maximum standardized uptake value assessed by 18F-FDG PET/CT, number of cycles of mFOLFIRINOX, tumor response to mFOLFIRINOX, and serum CA 19-9. Several machine learning models, such as logistic regression and random forest classifier, were trained to predict curative-intent resection and were then internally validated. The relative importance of each variable was analyzed using both Gini importance and permutation importance. Results: Among 647 patients, 173 (26.7%) underwent curative-intent conversion surgery (R0 or R1) following mFOLFIRINOX. The patients who underwent surgery showed significantly better overall survival (median 42.1 vs 17.2 months, P<0.0001) than those who did not. In multivariable analysis using logistic regression, age, resectability as per the NCCN guidelines, tumor size, number of cycles of mFOLFIRINOX, tumor response to mFOLFIRINOX, and CA 19-9 change rate (percentage change in CA 19-9 after mFOLFIRINOX) were significantly correlated to tumor removal with curative-intent conversion surgery. Random forest classifier showed the best predictive capability with AUC at 0.832. The most important feature was the CA 19-9 change rate. Conclusions: In this large cohort-based analysis, completion of conversion surgery could be predicted successfully in BRPC and LAPC patients treated with neoadjuvant mFOLFIRINOX using machine learning models. Further external validation is necessary for generalization. Citation Format: Hyunseok Yoon, Kyu-Pyo Kim, Inkeun Park, Jae Ho Jeong, Heung-Moon Chang, Baek-Yeol Ryoo, Changhoon Yoo. Prediction of conversion surgery completion following neoadjuvant modified FOLFIRINOX in borderline resectable and locally advanced pancreatic adenocarcinoma: machine learning algorithm analysis. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5397.