Application Of Random Forest Modeling To Predict The Occurrence Of Toxic And Side Effects Of High-Dose Methotrexate Chemotherapy In Patients With Osteosarcoma

Objective: Neoadjuvant chemotherapy for osteosarcoma can significantly improve the prognosis and improve survival,and high-dose methotrexate(HDMTX)is the cornerstone of neoadjuvant chemotherapy for osteosarcoma.However,the incidence of toxic and side effects of HD-MTX chemotherapy is extremely high.Find a method to accurately predict the occurrence of toxic and side effects after HD-MTX chemotherapy in osteosarcoma patients to guide clinical intervention in time and avoid delaying subsequent treatment of patients.Methods:1.Data of patients with osteosarcoma hospitalized to a top three hospital and received HD-MTX chemotherapy from 2015 to 2018 were retrospectively collected,and the incidence of hepatotoxicity,Acute kidney injury(AKI),delayed excretion and myelosuppression were calculated.2.The patient’s basic data,blood routine,liver and kidney function indexes,and blood drug concentration at 0 and 24 hours after chemotherapy were used as variables.Introduce random forest learning methods to construct prediction models of hepatotoxicity,AKI and excretion delay,and compare with the prediction models built by classic Logistic regression analysis for accuracy and the Area Under the Curve(AUC)of the receiver Operating Characteristic(ROC)Curve.Results:1.A total of 141 cases of HD-MTX chemotherapy in 45 osteosarcoma patients.The occurrence of myelosuppression after chemotherapy: 24 cases of level 0,50 cases of level I,48 cases of level II,18 cases of level III,and 1 case of level IV.Occurrence of hepatotoxicity: 34 cases of level 0,67 cases of level I,19 cases of level II,21 cases of level III,and 0 cases of level IV.2.The model of hepatotoxicity and excretion delay were successfully modeled,except that the model of AKI was not successfully modeled due to fewer positive cases.The prediction accuracy of the random forest hepatotoxicity model was 0.8889,1.0000,1.0000 and 0.7500 for the level 0,level I,level II and level III hepatotoxicity,and the overall accuracy was 0.9444.The random forest hepatotoxicity model predicted that the AUC values of level 0,level I,level II and level III of hepatotoxicity were all 1.00,and the average AUC value was also 1.00.While,the accuracy of Logistic regression analysis of the hepatotoxicity model in predicting level 0,level I,level II and level III of hepatotoxicity was 0.5556,1.0000,0.4000 and 0.7500,respectively,with an overall accuracy of 0.7778.The Logistic hepatotoxicity model predicted that the AUC values of level 0,level I,level II and level III of hepatotoxicity were 0.96,0.90,0.97 and 1.00,respectively,with an average AUC value of 0.94.The random forest excretion delay model predicts that the positive and negative accuracy rates of excretion delay are 1.000 and 0.9615 respectively,and the overall accuracy rate is 0.9722.The random forest excretion delay model predicts that the positive and negative AUC values for excretion delay are 1.00 and 1.00,respectively,and the average AUC is also1.00.While,the Logistic excretion delay model predicts that the positive and negative accuracy rates of excretion delay are 0.7000 and 1.000,respectively,and the overall accuracy rate is 0.9617.The Logistic excretion delay model predicts that the positive and negative AUC values of excretion delay are 0.99 and 0.99,respectively,and the average AUC value is also 0.99.Conclusions:1.The incidence of toxic and side effects in patients with HD-MTX chemotherapy osteosarcoma is extremely high.2.Our study demonstrated the feasibility of using clinical data from patients with osteosarcoma to model the prediction of hepatotoxicity and excretion delay.3.Random forest modeling can accurately predict the occurrence of hepatotoxicity and excretion delay caused by HD-MTX chemotherapy,and the prediction effect is better than Logistic regression model.