A Prediction Model Development And Validation Of Nonresponse To5-aminosalicylates In Patients With Ulcerative Colitis

Objective:Ulcerative colitis(UC)is a non-specific chronic inflammatory disease with unclear pathogenesis which mainly occurs in colon and rectum.5-aminosalicylic acid(5-ASA)is often the first step in the treatment of mild-moderate UC,however,some patients with UC have primary or secondary nonresponse to 5-ASA.The aim of this study is to establish and validate a clinical prediction model of nonresponse to 5-ASA in patients with UC by collecting and analyzing clinical data,and screening the predictors.Methods:The clinical information of patients with UC who were admitted to Department of Gastroenterology in First Affiliated Hospital of China Medical University from 1.2012 to 12.2019 were collected to establish a database.The general information,clinical manifestations,laboratory indicators,endoscopic characteristics and histological features were collected.Patients with UC were divided into response group and nonresponse group according to Mayo score.All the factors were analyzed with Lasso regression to screen the predictors which were subsequently included in multivariate Logistic regression analysis and were used to establish a prediction model and made a nomogram.Using bootstrap method to perform internal validation.Receiver operating characteristic(ROC)curve,area under the curve(AUC),the Hosmer-Lemeshow goodness-of-fit test,calibration curve and decision curve analysis(DCA)curve was used to validate the prediction model.Results:284 patients with UC were included in this study,198 patients were in response group,86 patients were in nonresponse group.The average age of response group and nonresponse group were 48.3 years and 44.3 years respectively,the difference is statistically significant(P=0.036).30.3%,41.9%,3.5%and 24.2%of the patients had mild,moderate,moderate-severe and severe UC in response group,5.8%,20.9%,3.4%and 69.8%of the patients had mild,moderate,moderate-severe and severe UC in non response group,the difference is statistically significant(P<0.0001).77.8%of the patients in response group and 91.9%of the patients in nonresponse group had abdominal pain respectively,the difference is statistically significant(P=0.004).33.3%of the patients in response group and 64.0%of the patients in nonresponse group had fever,the difference is statistically significant(P<0.0001).60.6%of the patients in response group and 75.6%of the patients in nonresponse group had marasmus,the difference is statistically significant(P=0.015).The average length of patient stay in response group and nonresponse group were 12.4 days and 18.3 days respectively,the difference is statistically significant(P<0.0001).The average white blood cell(WBC)in response group and nonresponse group were 7.89×10~9/L and 9.54×10~9/L respectively,the difference is statistically significant(P<0.0001).The average hemoglobin(HGB)in response group and nonresponse group were 119.31 g/L and 106.36 g/L respectively,the difference is statistically significant(P<0.0001).The average platelet(PLT)in response group and nonresponse group were 292.94×10~9/L and 361.52×10~9/L,the difference is statistically significant(P<0.0001).The average albumin(ALB)in response group and nonresponse group were 35.60 g/L and 29.87 g/L respectively,the difference is statistically significant(P<0.0001).The average C reactive protein(CRP)in response group and nonresponse group were 32.78 mg/L and 71.30 mg/L respectively,the difference is statistically significant(P<0.0001).26.8%of the patients in response group and 48.8%of the patients in nonresponse group had mucosal fragility and bleeding respectively,the difference is statistically significant(P<0.0001).12.6%of the patients in response group and 23.3%of the patients in nonresponse group had increasing abdominal soft tissue density respectively,the difference is statistically significant(P=0.024).All the factors were included in Lasso regession,Lambda.min was 0.06023419.The severity of disease,monocytes(MONO),platelet distribution width(PDW),ALB,total cholesterol(TC)andα1-globulin were included as the predictors in the prediction model according to Lambda.min,the predictors were included in the Logistic regression,and the nomogram was made.ROC curve was drew with AUC of 0.852,sensitivity and specificity were 79.3%and 80.2%when the cut-off value was 0.322,respectively.Hosmer-Lemeshow goodness-of-fit test(χ~2=6.4374,P=0.598)and calibration curve revealed that the model have good calibration.DCA curve revealed that the model had good clinical utility when the threshold probability was from 0.0%to 96.0%,and had no clinical utility when the threshold probability was greater than 96.0%.Conclusion:Patients with UC in response group and nonresponse group had difference in the age,severity of disease,abdominal pain,fever,marasmus,length of hospital stay,WBC,HGB,PLT,ALB,CRP,mucosal fragility and bleeding and increasing abdominal soft tissue density.The predictors of the prediction model were the severity of disease,MONO,PDW,ALB,TC andα1-globulin,the model has good discrimination,calibration and clinical utility.