Prediction model for recurrence and metastasis of cervical cancer after surgery using machine learning.

Abstract

207 Background: To analyze the relationship between the results of routine blood tests and blood biochemistry in cervical cancer patients after surgery and the recurrence and metastasis of cervical cancer, and to construct a 5-year recurrence and metastasis prediction model based on these blood test results. Methods: A retrospective analysis was conducted on the clinical information, routine blood tests, and blood biochemistry data of 165 patients with recurrent and metastatic cervical cancer and 716 patients without recurrent and metastatic cervical cancer who received treatment from March 2011 to August 2020. Chi-square tests were used to analyze the factors related to the recurrence and metastasis of cervical cancer after surgery. Ridge regression analysis, support vector machine (SVM), and random forest algorithms were used to construct the 5-year recurrence and metastasis prediction models. The models were evaluated using receiver operating characteristic (ROC) curves. The SHAP (SHapley Additive exPlanations) was used to analyze and quantify the contribution of each feature to the output of the random forest model. Results: A total of 572 cases was selected based on inclusion and exclusion criteria, with 92 cases (16.1%) experiencing recurrence and metastasis within 5 years. A total of 77 indicators were collected, of which 21 indicators were found to be associated with 5-year recurrence and metastasis of cervical cancer. The prediction models based on these indicators showed the following areas under the ROC curve (AUC) on the training set: ridge regression analysis 0.918 (95%CI 0.889–0.947); SVM 0.999 (95%CI 0.997–1); random forest 1 (95%CI 1–1). In the test set, the AUC values were: ridge regression analysis 0.928 (95%CI 0.881–0.975); SVM 0.913 (95%CI 0.850–0.976); random forest 0.957 (95%CI 0.927–0.987). The top five variables contributing most to the random forest model were indirect bilirubin, SCCA, serum creatinine, magnesium ions, and leucine aminopeptidase. Conclusions: The peripheral blood routine test results of cervical cancer patients after surgery are related to recurrence and metastasis. The machine learning models based on routine blood test results show good predictive performance for the 5-year recurrence and metastasis of cervical cancer.