Construction and validation of a predictive model for surgical procedures in primary trigeminal neuralgia
Objective To develop a predictive model for surgical treatment of primary trigeminal neuralgia(PTN)based on machine learning support vector machine algorithm(SVM)and to verify its effectiveness.Methods Two datasets of PTN patients admitted consecutively to the Neurosurgery Department of the Ninth People's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine were included.Dataset 1 retrospectively collected 167 patients from June 2022 to February 2023,and Dataset 2 prospectively collected 41 patients from March to June 2023.We randomly divided dataset 1 into a training set(125 cases)and a testing set(42 cases)in a 3∶1 ratio.Surgical methods,including microvascular decompression(MVD),radiofrequency thermocoagulation(RFTC)and percutaneous ballon compression(PBC),were used as the dependent variables,Lasso regression analysis was performed to screen feature variables in training set,which were incorporated into machine learning to construct a predictive model for surgical treatment methods.We used the testing set to validate and evaluate the model,and used dataset 2(validation set)for external validation.The receiver operating characteristic(ROC)curve of the model was drawn and the area under the curve(AUC)was calculated to evaluate its predictive performance on different datasets.The visual analog scale(VAS)was used to evaluate the pain level of PTN patients before and one month after surgery,and the difference in VAS between dataset 1 and dataset 2 patients was compared.Meanwhile,we documented the incidence of surgical related adverse reactions or complications in patients from two dataset.Results The Lasso regression analysis showed that the feature variables with non-zero regression coefficients were PBC surgery history,age,magnetic resonance tomography angiography(MRTA)results,MVD surgery history,other restrictive diseases,pain distribution,disease duration,and RFTC surgery history.The overall accuracy of the prediction model in the training set,testing set,and validation set was 73.6％,69.0％,and 73.2％,respectively.The AUC(macro/micro)under the ROC curve was 0.89/0.89,0.80/0.81,and 0.85/0.85,respectively.There was no statistically significant difference(P＞0.05)in features such as age,disease duration,or surgical methods between patients in dataset 1 and dataset 2.There was no statistically significant difference in the VAS score either from before surgery or 1 month after surgery between dataset 1 and dataset 2(both P＞0.05).Compared with preoperative data,the VAS scores of dataset 1 and dataset 2 decreased one month after surgery,and the differences were statistically significant(both P＜0.001).Two patients in dataset 1 developed subcutaneous fluid accumulation after MVD surgery,and one patient in dataset 2 had poor wound healing after MVD surgery.There was no statistically significant difference in the incidence of complications between the two datasets[1.2％(2/167)vs.2.4％(1/41),P=0.484].Conclusions The multiclass machine learning prediction model based on SVM algorithm has good performance and can guide clinicians to develop personalized surgical treatment plans.The treatment is effective and does not seem to increase the incidence of complications.
Machine learningMicrovascular decompressionPrimary trigeminal neuralgiaBalloon compressionPrediction model
NETL
NSTL
万方数据
