Construction of risk model for central venous catheter-related thrombosis based machine learning in patients undergoing gastrointestinal tract surgery
Objective To explore the risk factors of central venous catheter related thrombosis in the patients undergoing gastrointestinal tract surgery,and to construct a risk prediction model based on machine learning algorithms.Methods A total of 385 patients receiving gastrointestinal tract surgery and central ve-nous catheter indwelling in this hospital from May 2018 to March 2024 were selected as the study subjects and divided into the thrombus group(n=62)and non-thrombus group(n=323)based on whether or not the catheter-related thrombosis forming.The age,body mass index(BMI),comorbidities,current tumors,neutro-phil/lymphocyte ratio(NLR),surgery time,catheterization vein,systemic immune inflammation index(SII),D-dimer and catheter indwelling time of the patients were collected,and the differences in baseline data were compared between the two groups.The research subjects were randomly divided into the training set and tes-ting set by a 7∶3 ratio.Based on the training set,the logistics regression model,random forest,support vector machine,decision tree and naive Bayes risk prediction models were established.The area under the operating characteristic curve(AUC),accuracy,sensitivity,specificity and F1 value in predicting catheter-related throm-bosis were compared among different models in the testing set.The importance of the predictive factors in the best prediction model conducted the visualized ranking.Results There were statistically significant differences in the proportion of tumor patients,NLR,surgical time and D-dimer level in the baseline data between the two groups(all P＜0.05).The AUC values of the five risk prediction models from great to small were the random forest(0.773),logistics regression model(0.734),support vector machine(0.680),naive Bayes(0.666)and decision tree(0.650).Among them,the accuracy(0.853),sensitivity(0.599),specificity(0.877)and F1 val-ue(0.414)of the random forest model were the highest.D-dimer,surgery time,current tumor and NLR were the top four important predictive factors in the random forest model.Conclusion The constructed random forest model for central venous catheter-related thrombosis in the patients undergoing gastrointestinal tract surgery demonstrates good performance,and the D-dimer,surgery time,current tumor and NLR are the main predictive factors.
gastrointestinal tract surgerycentral venous cathetervenous thrombosis formationma-chine learningprediction model
万方数据
维普
