Study On Risk Factors And Predictive Model Of Recurrent Acute Pancreatitis By Propensity Score Matching

Objective:To explore the risk factors of recurrent acute pancreatitis(RAP),estab lish a risk prediction model and verify it.Methods:The clinical data of 1365 patients diagnosed with acute pancreatitis(AP)were analyzed retrospectively,which were treated in the affiliated Hospital of Zunyi Medical University from January 2018 to January 2020.Excluded cases:483 cases were missing clinical data,17 cases had organ failure before onset,68 cases were signed out of hospital,12 cases were complicated with malignant tumor,116 cases were treated outside the hospital,46 cases were under 18 years old or over 80 years old and 623 cases were included,followed up by telephone,outpatient follow-up and readmission from October 2020 to November 2020.37 cases were lost to follow-up.Finally,586 cases were included,of which all the recurrent patients were Readmission patient,Patients without recurrence were identified by follow-up.According to the proportion of 7:3,the clinical data of AP patients were divided into two groups by computer-generated random numbers,in which 70%(n=412)of the data were taken as the modeling cohort and the remaining 30%(n=174)were taken as the verification cohort.The modeling cohort was divided into two groups:non-recurrent group(n=234)and recurrent group(n=178).The general data of the two groups were analyzed:age,sex,body mass index(BMI),disease severity grade,length of stay,short-term complications:pancreatic necrosis,shock,multiple organ dysfunction syndrome(MODS),systemic inflammatory response syndrome(SIRS),acute kidney injury(AKI),identified the confounding factors,Nearest neighbor matching was adopted to conduct propensity score match(PSM)on the modeling cohort at 1:1 ratio to exclude the confounding factors:non-recurrent group(n=164)and recurrent group(n=164).The history of diabetes,hypertension,cholelithiasis,drinking,smoking,fatty liver and laboratory indexes within 24 hours after admission were compared between the two groups.Logistic regression model was used to further screen the independent risk factors affecting RAP recurrence to establish a risk prediction model and calculate the risk score.The prediction ability and accuracy of the prediction model were evaluated by the subject working characteristic curve(ROC).In the verification dataset,the model is evaluated by ROC curve and calibration curve.The difference was statistically significant(P<0.05).Results:A total of 586 patients with AP were included,according to the proporti on of 7:3,the clinical data of AP patients were divided into two groups by compu ter-generated random numbers,in which 412 cases were taken as the modeling coh ort and 174 cases were taken as the verification cohort.According to whether the patients had recurrence or not,the modeling cohort was divided into two groups:n on-recurrent group(n=234)and recurrent group(n=178).(1)The analysis of the gener al data of the two groups showed that there were significant differences in age(P=0.037,Table 1),sex(P=0.006,Table 1),length of stay(P<0.001,Table 1)between the non-recurrent group and the recurrent group.Let age,sex and length of stay be covariables to conduct PSM at 1:1 ratio:non-recurrent group(n=164)and recurrent group(n=164).There was no significant difference between the two groups in sex(P=0.322,Table 2),age(P=0.265,Table 2)and length of stay(P=0.134,Table 2),(2)The two sets of data are comparable.Monofactor analysis showed that there wer e statistically significant differences between non-recurrent group and recurrent group in cholelithiasis(P=0.005,Table 3)、smoking history(P<0.001,Table 3),diabetes hist ory(P=0.016,Table 3),drinking history(P=0.002,Table 3),TG/ApoB(P<0.001,Table 3),TG(P<0.001,Table 3),LDL-c(P=0.025,Table 3)and blood glucose(P=0.019,Tab 1e 3).The proportion of cholecystectomy in patients with cholelithiasis in the non-r ecurrent group was significantly higher than that in the recurrent group(37(63.8%)V S11(31.4%)),P=0.002,Table 4),the difference was statistically significant.(3)TG,bloo d glucose,LDL-c,TG/ApoB,cholelithiasis,drinking history,smoking history,and d iabetes history were included in the multivariate analysis of Logistic regression mod el:TG(OR=1.094,95%CI:1.051-1.140,P<0.001,Table 5),drinking history(OR=3.735,95%C I:1.685-8.278,P=0.001,Table 5),smoking history(OR=4.641,95%CI:2.766-7.785,P<0.001,Table5),diabetes history(OR=3.314,95%CI:1.606-6.840,P=0.001,Table 5)are indep endent risk factors for recurrence of RAP.The Youden index of TG was calculated by the receiver operating characteristic curve(ROC),and the best critical value corre sponding to its maximum is:TG-16.93mmol/L,Layer the TG according to the criti cal value:TG≥16.93mmol/L and TG<16.93mmol/L,Convert it into a classified variable(Table 6,figure 2).(4)Drinking history,smoking history,diabetes history,TG≥16.93mmol/L and TG<16.93mmol/L were re-entered into the Logistic regression model to conduct forward selection analysis:diabetes history(OR=2.242,95%CI:1.261-3.986,P=0.006,Table 7),drinking history(OR=3.542,95%CI:1.688-7.060,P=0.001,Ta ble 7),smoking history(OR=4.713,95%CI:2.780-7.989,P<0.001,Table 7),TG≥16.93m mol/L(OR=8.472,95%CI:3.991-17.982,P<0.001,Table 7)are independent risk factors for recurrence in RAP.(5)Establish a risk forecast formula:diabetes history+1.239/0.807×dri nking hi story+1.550/0.807×smoking history+2.137/0.807×TG≥16.93mmol/L,Risk scoring model sco res ranged from 0 to 8 points,including drinking history:2 points;diabetes history:1 point;smoking history:2 points;TG≥16.93mmol/L:3 points.The Area under the receiver operating characteristic curve(AUC)is 0.801(95%CI:0.753-0.850)(>0.8),The cut-off value is 2.5 points,the sensitivity is 61.6%,and the specificity is 87.2%(Ta ble 8,Figure 3).(6)The remaining 30%patient data were used as validation cohort for internal validation of the risk prediction model,and the AUC is 0.799(95%CI:0.730-0.868)(>0.7),The sensitivity is 69.7%and the specificity was 89.5%mol.(Ta ble 9,Figure 4).The calibration curve shows that the prediction curve is cl ose to the standard curve,and the calibration degree of the prediction model is better.(Figure 5).Conclusion:(1)Drinking history,smoking history,diabetes history and TG≥16.93mmol/L are in dependent risk factors for recurrence of RAP.(2)Establish a risk scoring model:①Diabetes history:Yes:+1 point;No:0 point s;②Drinking history:Yes:+2 points;No:0 points;③Smoking history:Yes:+2 p oints;No:0 points;④TG:≥16.93mmol/L:+3 points;<16.93mmol/L:0 points.A t otal score of 0 to 2.5 is low risk,a total score of 2.5 to 8 is high risk.That is,p atients who meet TG≥16.93mmol/L or have at least two of the three items of drink ing,smoking and diabetes been at high-risk groups for RAP recurrence,and have h igh predictive value and stability by internal verification.