Abstract:

Objective

To construct and verify the prediction model for recurrence of young and middle-aged patients with liver cancer in different periods after radical surgery，and then provide more effective postoperative monitoring and intervention strategies for clinic.

Methods

From January 2017 to December 2021，341 cases of primary hepatocellular carcinoma （PHC） were selected as the research object，and the cases were screened according to certain inclusion and exclusion criteria，and detailed clinical data were collected. Five machine learning algorithms，including logistic regression （LR），decision tree （DT），support vector machine （SVM），random forest （RF） and extreme gradient lifting algorithm （XGBoost），were used to construct the prediction model. All patients were divided into training set and validation set，and the data of training set were trained and optimized by 50% cross validation. Receiver operating characteristic curve （ROC） was used to evaluate the prediction performance of each model，and the area under the curve （AUC），sensitivity，specificity and Jordan index of each model for predicting recent recurrence and long-term recurrence were counted.

Results

Of the 341 PHC patients，173 （50.73%） had postoperative recurrence，of which 78 （22.87%） had recent recurrence and 95 （27.86%） had long-term recurrence. The levels of neutrophil to lymphocyte ratio （NLR），platelet to lymphocyte ratio （PLR），alanine aminotransferase，aspartate transaminase，total bilirubin and alpha fetoprotein in recent and long-term relapses were significantly higher than those in patients without recent and long-term relapses. At the same time，there were significant differences between those with recent recurrence and those without recent recurrence，and between those with long-term recurrence and those without long-term recurrence in portal vein tumor thrombus，tumor envelope integrity，BCLC staging and tumor differentiation （P＜0.05）. The model prediction results showed that XGBoost algorithm had the best performance in predicting recent recurrence （AUC=0.989） and long-term recurrence （AUC=0.983），followed by RF （AUC=0.926，0.939）and SVM algorithm （AUC=0.914，0.904）.

Conclusion

LR，DT，SVM，RF and XGBoost can all predict the recurrence possibility of young and middle-aged patients with liver cancer in different periods after radical surgery. Among them，the prediction performance of RF，SVM and XGBoost model is relatively good，especially XGBoost model shows high prediction accuracy. In addition，NLR，PLR，related liver function indexes，portal vein tumor thrombus，tumor capsule integrity，BCLC stage and tumor differentiation degree may be predictive factors for recurrence.

Key words: Hepatocellular carcinoma, Tumor recurrence, Risk model, Young and middleaged people, Machine learning algorithm