Correlation between machine learning models and the risk of venous thrombosis in pregnant and postpartum women
LI Jun-lin
CHEN Ya-xue
WANG Juan
JIE Lin-lin
Abstract：Objective To explore the application value of machine learning models in predicting the risk of ve-nous thrombosis in pregnant and postpartum women.Methods A total of 1 580 singleton pregnant women who underwent prenatal examinations at our hospital from January 2018 to December 2024 were enrolled.The participants were randomly divided into a training set and a test set at a 7∶3 ratio.In the training set,the women were grouped into a venous thrombo-sis group and a non-thrombosis group according to whether venous thrombosis occurred.Multiple machine learning mod-els were developed using the training set with cross-validation to compare model accuracy.The importance of each clini-cal indicator within the models was analyzed.Both the machine learning models and a logistic regression model were used to predict thrombosis risk in the test set,and receiver operating characteristic(ROC)curves were applied to assess clini-cal performance.Results Among the 1,580 women,395(25.00％)were diagnosed with venous thrombosis.There were statistically significant differences between the thrombosis and non-thrombosis groups in cesarean delivery,BMI,TG,LDL-C,TC,D-D,TAT,PIC,t-PAIC,and sTM levels(all P＜0.05).Correlation analysis showed that BMI,cesarean delivery,TG,LDL-C,TC,D-D,TAT,PIC,t-PAIC,and sTM were highly correlated with venous throm-bosis risk,while correlations among these indicators themselves were low.A total of 14 models from 8 machine learning al-gorithms were developed,among which the neural network model had the highest accuracy during training and validation.D-D,t-PAIC,PIC,and sTM had the highest feature importance,whereas TAT,LDL-C,age,cesarean delivery,BMI,TG,APTT,FIB,TC,platelet count,hemoglobin,and PT had lower importance.The AUC of the machine learning model(0.999)was significantly higher than that of the logistic regression model(0.752)in predicting venous thrombosis risk.Conclusion BMI,cesarean delivery,TG,LDL-C,TC,D-D,TAT,PIC,t-PAIC,and sTM are associated with the risk of venous thrombosis in pregnant and postpartum women.Machine learning models built on these indicators demonstrate superior predictive value for thrombosis risk compared with logistic regression.
Keywords：machine learning modelpregnant and postpartum womenvenous thrombosiseffectiveness
Publication Date：2025-11-15
Online Publishing Date：2026-01-06(First online date of this platform, not the publication date of the document)
Pages：6( 1614-1619 )