Predictive value of decision tree prediction model to left atrial thrombus or left atrial spontaneous echo contrast in patients with non-valvular atrial fibrillation
Liu Bing
Zheng Nan
Yu Kai
Li Fengpeng
Yan Hongwei
Zhang Jun
Abstract：Objective To review the predictive value of decision tree prediction model to left atrial thrombus(LADT)or left atrial spontaneous echo contrast(LADSEC)in patients with non-valvular atrial fibrillation(NVAF).Methods The clinical data were retrospectively analyzed in NVAF patients(n=334)who underwent transesophageal echocardiography(TEE)in Central Hospital of Cangzhou City in Hebei Province from Jan.2016 to June 2021.The data before Dec.2019 served as training cohort,while the data from Jan.2020 onwards served as validation cohort.The risk factors for LADT/LADSEC occurrence were screened by using Logistic regression analysis in NVAF patients,and a decision tree prediction model was established.The AUC of decision tree prediction model was calculated by using ROC curve,which was compared with AUC of CHADS and CHA2DS2-VASc scores.Results The incidence of LADT/LADSEC was 9.9%.The results of Logistic regression analysis showed that stroke/transient ischemic attack(TIA,OR=103.50,95%CI:10.26～1044.37,P<0.001),non-paroxysmal atrial fibrillation(non-PAF,OR=19.94,95%CI=3.58～110.88,P=0.001),no-anticoagulant therapy(OR=159.24,95%CI:6.79～3735.02,P=0.002),increased left atrial diameter(LAD,OR=6.10,95%CI:1.43～26.08,P=0.015),decreased left ventricular ejection fraction(LVEF,OR=105.25,95%CI:10.59～1045.85,P<0.001),chronic kidney disease(CKD,OR=3.93,95%CI:1.60～5.68,P=0.003)were independent risk factors for LADT/LADSEC occurrence.AUC of decision tree prediction model of training cohort was 0.88(95%CI:0.80～0.96),sensitivity was 82.6%and specificity was 86.5%.AUC of decision tree prediction model of validation cohort was 0.86(95%CI:0.76-0.99),sensitivity was 90.0%and specificity was 80.6%.AUC of CHADS2 score was 0.68(95%CI:0.56-0.96),sensitivity was 51.5%and specificity was 85.4%.AUC of CHA2DS2-VASc score was 0.73(95%CI:0.61～0.86),sensitivity was 60.6%and specificity was 82.4%.DeLong's method comparison showed that AUC of decision tree prediction model was higher than that of CHADS2 score(Z=2.80,P=0.005),and was similar to that of CHA2DS2-VASc score(Z=1.95,P=0.052).Conclusion The decision tree prediction model can effectively predict LADT/LADSEC occurrence in NVAF patients.
Keywords：Atrial fibrillationLeft atrial thrombusLeft atrial spontaneous echo contrastDecision tree
Publication Date：2025-03-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：6( 342-346,350 )