Objective To investigate the clinical value of a machine learning‑based risk pre-diction model for hepatic encephalopathy after transjugular intrahepatic portosystemic shunt (TIPS).
Methods The retrospective cohort study was conducted. The clinical data of 1 521 patients who underwent TIPS for liver cirrhosis at The First Affiliated Hospital of Zhengzhou University from January 2015 to December 2023 were collected. There were 1 078 males and 443 females, aged 54(46, 62) years. Based on the random number table, the 1 521 patients were divided into a training set of 1 065 cases and a validation set of 456 cases with a ratio of 7∶3. The training set was used to construct the predictive model, and the validation set was used to validate the performance of the predictive model. Based on machine learning algorithms, six predictive models for hepatic encepha-lopathy after TIPS were constructed, including Logistic regression, support vector machine, decision tree, naive Bayes, neural network, and extreme gradient boosting. Observation indicators: (1) conditions of patients with hepatic encephalopathy after TIPS; (2) clinical features screening of patients; (3) construction and performance evaluation of predictive models; (4) interpretation of predictive model and clinical feature analysis. Comparison of measurement data with normal distri-bution between groups was conducted using the independent sample t test. Comparison of measure-ment data with skewed distribution between groups was conducted using the Mann‑Whitney U test. Comparison of count data between groups was conducted using the chi‑square test. The least absolute shrinkage and selection operator (LASSO) regression was used to screen predictive factors. Performance of the predic-tive model was evaluated using the area under the receiver operating characteristic (ROC) curve, accuracy, sensitivity, specificity, precision, and F1 score. The calibration curve was used to assess the agreement between predicted and observed probabilities, and decision curve was used to evaluate clinical net benefit. The interpretability analysis of machine learning algorithm and SHapley Additive exPlanations (SHAP) was used to explain the model decision separately.
Results (1) Conditions of patients with hepatic encephalopathy after TIPS: of 1 521 patients, 451 cases developed postopera-tive hepatic encephalopathy, including 320 cases of covert hepatic encephalopathy and 131 cases of overt hepatic encephalopathy. Of the 451 patients with hepatic encephalopathy, there were 326 males and 125 females, aged 57 (49, 64) years, with the Child‑Pugh score of 9 (8, 10), the model for end-stage liver disease score of 12.3 (9.8, 15.9), the preoperative serum ammonia level of 72 (55, 96) μmol/L, and the neutrophil count of 4.20(2.10, 6.95)×10⁹/L. The etiologies of cirrhosis included hepatitis B virus infection in 240 cases, hepatitis C virus infection in 65 cases, alcoholic liver disease in 62 cases, nonalcoholic fatty liver disease in 40 cases, and autoimmune disease in 25 cases or other etiologies in 19 cases. Combined embolization was performed in 213 cases, and a 10‑mm stent was used in 308 cases. (2) Clinical features screening of patients: LASSO regression identified 12 clinical features with independent predictive value, including age, cerebrovascular disease, history of spontaneous bacterial peritonitis, Child‑Pugh score, pleural effusion, alanine aminotransferase, aspartate amino-transferase, glutamyltransferase, neutrophil count, triglyceride, serum sodium, and preoperative serum ammonia. (3) Construction and performance evaluation of predictive models: the six predictive models, including Logistic regression, support vector machine, decision tree, naive Bayes, neural network, and extreme gradient boosting, were constructed in the training set and independently evaluated in the validation set. The results showed that the extreme gradient boosting model showed the highest area under the curve in both the training set and the validation set as 0.902(0.879-0.922) and 0.884(0.852-0.913), respectively. In the validation set, the accuracy, sensitivity, specificity, precision, and F1 score of the extreme gradient boosting model were 0.725, 0.735, 0.717, 0.656, and 0.694, respectively. Decision curve analysis showed that the extreme gradient boosting model had a positive net benefit and generally outperformed the other five models, as well as the treat‑all and treat‑none strategies, within a threshold probability of 0.05-0.70. Calibration curve analysis showed that the predicted probabilities of the extreme gradient boosting model were generally consistent with the observed incidence, and the curve was close to the ideal reference line. (4) Interpretation of predictive model and clinical feature analysis: SHAP analysis based on the optimal extreme gradient boosting model showed that the top six clinical features contributing to the occurrence of hepatic encephalopathy after TIPS were preoperative serum ammonia, Child‑Pugh score, neutrophil count, glutamyltranspeptidase, alanine aminotransferase, and triglyceride.
Conclusions The extreme gradient boosting‑based predictive model for hepatic encephalopathy after TIPS show a good discrimination and clinical applicability. It is suitable for individualized preoperative risk assessment and provides a basis for early intervention in high‑risk patients.
DownLoad: