Construction And Validation Of Prediction Model For PICC-related Thrombosis Risk In Cancer Patients

Objective To identify the independent risk factors of PICC-related thrombosis in cancer patients,and to construct the risk prediction model of PICC-related thrombosis in cancer patients,so as to provide a theoretical basis for the prevention and risk prediction of PICC-related thrombosis in cancer patients.Methods In the early stage,the data collection table of PICC-related thrombosis in cancer patients was developed through literature review and semi-structured interview.The case control method was used to collect the data of 1505 patients who met the inclusion and exclusion criteria in a Class A tertiary hospital from January 2010 to December 2022,they were divided into case group(n=500)and control group(n=1005)according to whether PICC-related thrombosis occurred.The independent risk factors of PICC related thrombosis in cancer patients were determined by univariate and multivariate analysis.On this basis,Logistic regression,Bayes discriminant analysis and artificial neural network were used to establish the prediction model for PICC-related thrombosis risk in cancer patients,and TOPSIS comprehensive analysis was used to optimize the model.Results(1)Multivariate analysis showed that the protective factors of PICC related thrombosis in cancer patients included upper limb vein(OR=0.47),valved catheter(OR=0.34),and high pressure resistant polyurethane catheter(OR=0.24);The risk factors included blind puncture-seldinger technique(OR =3.60),complicated with diabetes(OR=2.51),complicated with infection(OR=4.72),use of vascular targeted drugs(OR=1.46),history of catheterization(OR=2.93),elevated D-D concentration(OR=4.490),history of surgery(OR=6.63),anemia(OR=2.07),age increase or less than 18 years old(OR=5.25),the type of cancer is breast cancer(OR=1.75),and double lumen catheter(OR=3.90).(2)Based on logistic regression showed that the area under the ROC curve(AUC)of the training set is 0.873,the sensitivity is 84.9%,the specificity is 74.8%,the accuracy is 78.2%,the Youden index is 59.7%,the positive predictive value is 62.9% and the negative predictive value is90.7%.The sensitivity of the test set is 58.8%,the specificity is 89.9%,the accuracy is 79.8%,the Youden index is 48.3%.,the positive predictive value is 73.7% and the negative predictive value is 81.9%.(3)Based on Bayes discriminant analysis showed that the AUC of the training set is 0.873,the sensitivity is 76.1%,the specificity is 81.5%,the accuracy is 79.7%,the Youden index is 59.4%,the positive predictive value is 67.5% and the negative predictive value is 87.1%.The sensitivity of the test set is 81.1%,the specificity is 82.7%,the accuracy is 82.2% and Youden index is 63.8%,the positive predictive value is 69.3% and the negative predictive value is 90.1%.(4)Based on artificial neural network showed that the AUC of the training set is 0.908.The sensitivity is 68.3%,the specificity is 90.2%,the accuracy is 82.8%,the Youden index is 58.5%,the positive predictive value is 78.0% and the negative predictive value is 84.9%.The sensitivity of the test set is 69.2%,the specificity is 79.8%,the accuracy is 86.3%,the Youden index is 58.5%,the positive predictive value is 79.8% and the negative predictive value is 86.3%(5)The results of TOPSIS comprehensive analysis showed that the fitting and prediction effect of logistic regression model in training set and test set is the best,and the prediction ability of artificial neural network prediction model and Bayes discriminant analysis model is relatively poor.Conclusions(1)The risk factors of PICC-related thrombosis in cancer patients included blind puncture-seldinger technique,diabetes,infection,use of vascular targeting drugs,history of catheterization,elevated D-D concentration,history of surgery,anemia,age increase or less than 18 years old,the type of cancer is breast cancer,double lumen catheter.(2)The protective factors of PICC-related thrombosis in cancer patients included the vein of catheterization is upper limb vein,the catheter has valve,and the material of catheter is high-pressure resistant polyurethane.(3)Logistic regression model,Bayes discriminant analysis model and artificial neural network were used to construct prediction model for PICC-related thrombosis in cancer patients,and the Logistic regression model had the highest predictive ability,while the Bayes discriminant analysis model and artificial neural network model had relatively poor predictive ability.