Considering Machine Learning Application for NCD Surgical Procedure Prediction within Unified Integrated Cancer Database System

Abstract

　Since 2017, we have been developing an Integrated Cancer Treatment Database System (Cancer DB) which automatically collects confirmed and finalized treatment data from multiple information systems. Surgery data consists of several data items such as patient background, preoperative state, intraoperative remarks, postoperative diagnosis and pathological information which are organized mainly for physicians. We have implemented an NCD output conversion function for thoracic surgical oncology region, which aims to reduce NCD registration burden. To predict NCD surgical procedure code, we defined NCD surgical procedure code classification to apply XGBoost for primary lung cancer (344 cases) and metastatic lung tumor (139 cases), which came with 98％ accuracy. Adding lymph node dissection and lesion count features, NCD surgical procedure code is predictable with high accuracy.