Establishment And Evaluation Of LASSO Regression Based Prediction Model Of Esophageal Varices In Cirrhosis

Background:Cirrhosis is a serious liver disease characterized by fibrosis and structural abnormalities in the liver tissue.When liver tissue becomes fibrotic,both the structure and function of the liver are affected,leading to a variety of clinical symptoms and complications.Among them,esophageal varices(EV)are one of the most common and serious complications of cirrhosis.Esophageal varices are caused by cirrhosis-induced portal hypertension,which leads to varices in the esophagus and stomach,making the vein walls thinner and thus prone to rupture and bleeding.When the bleeding is large,it can lead to serious and even life-threatening consequences.Therefore,it is very important to predict whether a patient is at risk for esophageal varices.If patients at high risk are identified early,measures such as endoscopic treatment or surgical intervention can be taken,thus reducing the occurrence and serious consequences of esophageal varices.Currently,endoscopy is one of the main methods to diagnose esophageal varices in liver cirrhosis.However,endoscopy is invasive and costly,making it unsuitable for mass screening.Therefore,the development of a noninvasive predictive model to predict whether patients are at risk for esophageal varices while avoiding unnecessary gastroscopy is of great clinical value.Objective:To explore non-invasive indicators associated with screening esophageal varices and to construct a non-invasive predictive model for assessing esophageal variceal severity to guide clinical practice.Methods:A total of 177 patients with cirrhosis who were hospitalized at the First Hospital of Baiqiu’en,Jilin University from May 2016 to August 2022 were studied,and patients enrolled from May 2016 to November 2019 were assigned to the training cohort(n=124)and patients enrolled from December 2019 to August 2022 were assigned to the validation cohort(n=53),using time as the node.Clinical information of patients was collected through the hospital’s electronic medical record system.Using upper gastrointestinal endoscopy findings as the gold standard,the training cohort of 124 patients was divided into 75 patients in the no/mild esophageal varices group and 49 patients in the moderate/severe esophageal varices group.Risk factors associated with the severity of esophageal varices were screened.Thec~2test was used for comparison between two groups for numerical data,the t test for comparison between two groups for normally distributed measures,and the Mann-Whitney U test for comparison between two groups for skewed measures.The glment package in R language was applied for LASSO regression to screen the independent variables,and the ten-fold cross-validation method was used to screen the risk factors for cirrhosis combined with moderate/severe esophageal varices,and the independent risk factors were screened by incorporating multi-factor binary logistic regression,and the column line graph model was drawn by the rms package.The p ROC package was applied to plot the subject work characteristic(ROC)curves,calculate the area under the ROC curve(AUC)for the prediction model,and determine the optimal threshold value based on the Jorden index.Data from the validation cohort were applied to externally validate the prediction model.the Hosmer-Lemeshow goodness-of-fit test was applied to assess the fit of the prediction model.Internal validation of the model by Bootstrap method and calculation of the corrected consistency index(C-index)to assess the discrimination of the prediction model.Calibration curves are plotted to assess the calibration degree of the prediction model.Apply the rmda package to clinical decision curve analysis(DCA)to evaluate the net benefit of the prediction model under different probability thresholds and verify the clinical validity of the model,and the clinical impact curve(CIC)to assess the loss-to-benefit ratio of the prediction model.Results:A total of 124 patients were included in the training cohort,and 3 influencing factors(PLT,PV,RLR)were screened by applying LASSO regression,and the PLT and PV continuous variables were transformed into dichotomous variables by using the value corresponding to the maximum of the Jorden index as the cut-off point,and PLT≥120×109/L,PV≥13 mm,and RLR were included in binary logistic regression,according to the Akaike information criterion(AIC),the lowest(119.5993)combination was selected,and finally three independent influencing factors(PLT≥120×10~9/L,PV≥13mm,RLR,P<0.05)were derived,and a column line graph prediction model was established.The prediction model was correlated and validated with an AUC=0.857(95%CI:0.785,0.905)in the training cohort,sensitivity of 85.7%,specificity of 74.7%,positive predictive value of 68.3%,negative predictive value of87.5%,missed diagnosis rate of 16.3%,misdiagnosis rate of 25.3%,and accuracy of78.2%;the prediction model was validated in the cohort with AUC=0.823(95%CI:0.710,0.936),sensitivity of 87.0%,specificity of 70.7%,positive predictive value of69.0%,negative predictive value of 87.5%,missed diagnosis rate of 13.0%,false diagnosis rate of 30.0%,and accuracy of 77.4%,the predictive model had good predictive efficacy in both the training and validation cohorts.The Hosmer-Lemeshow goodness-of-fit test showed no statistically significant prediction deviation between predicted and actual values in the training cohort(c~2=5.747,P=0.676),and again not statistically significant in the validation cohort(c~2=3.176,P=0.923),and the prediction model was well fitted.In both the training and validation cohorts,the calibration curves of the prediction model converged to the ideal curve,and there was good agreement between predicted and actual values.Analysis of the clinical decision curves showed that the net patient benefit was higher than the other two extreme curves when the threshold probability values were in the range of approximately 10%to 80%,and the model had good clinical clinical validity in this range.The prediction model developed in this study had a larger AUC than RLR,RPR,GP,and FIB-4,with statistically significant differences(P<0.05),and had better sensitivity and specificity.Conclusions:In this study,PLT≥120×10~9/L,PV≥13mm,and RLR were found to be independent influencing factors for combined moderate/severe esophageal varices in patients with cirrhosis.A predictive model for esophageal varices in cirrhosis was established by combining the above independent influencing factors,which has good discrimination,calibration,and clinical validity and helps clinicians to screen out patients with moderate/severe esophageal varices in patients with cirrhosis in order to make early diagnosis and preventive treatment.