Abstract IA-13: Machine learning for improved management of patients with pancreatic cysts

Abstract

Abstract Invasive pancreatic cancer is a deadly disease, with the lowest five year survival rate among all cancers. Cysts in the pancreas are relatively common, but while some have malignant potential, many are entirely benign, while surgical resection of the pancreas has relatively high morbidity and mortality. Clinical management of patients with pancreatic cysts must balance these risks. We developed a supervised machine learning approach to integrate clinical, imaging and molecular data, as part of a large retrospective international study of almost 900 patients with pancreatic cysts. Using the Multivariate Optimization of Combinatorial Alterations (MOCA) algorithm, patients were classified into those who required immediate surgery, those who required routine monitoring, and those who could be discharged without future surveillance, using a stepwise approach. The resulting diagnostic test, called CompCyst, was more accurate than current patient management which relies on imaging and clinical criteria. More than half of the patients sent to surgery were found to have unnecessary resection of their cysts, which could have been avoided by application of the CompCyst test. Our results demonstrate how machine learning and molecular features can be applied to improve pre-surgical diagnostic testing in the clinic. Citation Format: Rachel Karchin. Machine learning for improved management of patients with pancreatic cysts [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr IA-13.