Risk prediction model for early postoperative death in patients with hepatocellular carcinoma: a retrospective study based on random forest algorithm and logistic regression

Abstract

Background: At present, little is known about the risk factors of early postoperative death in patients with hepatocellular carcinoma (HCC).

Methods: We collected the data of patients who were diagnosed with primary liver cancer between 2010 and 2015 in the Surveillance, Epidemiology, and End Results database and further allocated them to the training set and validation set. Univariate and multivariate logistic regression analysis was used to determine the independent influencing factors of early postoperative death of HCC patients. Random forest and Least absolute shrinkage and selection operator regression analysis were used to screen out vital variables for the construction of the nomogram. It was evaluated by receiver operating characteristic curve, calibration curve and decision curve analysis.

Results: A total of 4154 patients were selected in this process, including 2647 patients with postoperative early death (outcome1) and 1507 patients with liver cancer-specific postoperative early death (outcome2). Surgery method, age category, marital status and tumor grade were the risk factors for early postoperative death. As for the liver cancer-specific early postoperative death, AJCC, surgery method, chemotherapy and tumor grade were independent prognostic factors. Early death and liver cancer-specific early death nomograms have an area under curves of 0.643 and 0.679 in the training set, respectively, and 0.617 and 0.688 in the validation set. The calibration curve and decision curve analysis shows that the nomograms have good performance.

Conclusion: This model provides an intuitive and practical tool for future studies based on large-scale cohorts by exploring the risk factors of early death in patients with HCCs undergoing surgery.