Development And Validation Of A Clinical Model For Sepsis In ICU Patients Who With Suspected Infection

Sepsis,defined as life-threatening organ dysfunction caused by the dysregulated host response to infection,has high morbidity and mortality,and the incidence is increasing year by year.Early identification of patients with sepsis and appropriate management can effectively improve the prognosis.To identify patients with sepsis early,clinicians need a tool that is simple and accurate enough to assist in improving patient outcomes.Based on the MIMIC-IV database,this study will develop a clinical prediction model based on the patient’s baseline data and past medical history.Objective:In this study,by analyzing the basic clinical data of infected patients in the ICU who met the inclusion criteria in the MIMIC-IV database,retrospectively studied the risk factors of infected patients progressing to sepsis,developed and validated the early diagnosis and prediction model of sepsis,and further analyzed the data.The model was optimized to develop a scoring scale for the early diagnosis of sepsis.Methods:All ICU inpatients with suspected infection in the MIMIC-IV database were included in the study.The exclusion criteria were: 1)age < 18 years old,2)duration of ICU stay < 24 hours.All patients were randomly divided into development group and validation group according to 2:1.According to the SOFA score 24 hours after admission,the patients were divided into sepsis group and non-sepsis group,and descriptive statistics were analyzed.Univariate and multivariate Logistic regression analysis was used to determine the risk factors of sepsis,development a predictive model for the early diagnosis of sepsis,and calculate related results such as sensitivity and specificity.And develop a rating scale,internally validated in a validation group.Results:1.The overall situation of the research object:A total of 37325 patients were included in this study,25128 patients were randomly selected as the experimental group,and the remaining patients were the validation group.According to the SOFA score 24 hours after admission,the patients were divid ed into sepsis group(n=23198)and non-sepsis group(n=1903).2.Comparison between the sepsis group and the non-sepsis group:The proportion of males in the sepsis group is significantly higher(57.7% vs.47.9%,P<0.001);there were significant differences in median age(66 vs.59,P<0.001),median systolic blood pressure(66 vs.59,P<0.001),median diastolic blood pressure(66 vs.59,P<0.001),median mean arterial pressure(66 vs.59,P<0.001)between the two groups(p<0.01),appears to be older and have lower blood pressure.In terms of past medical history,there were significant differences in the prevalence of diabetes(32.7% vs.21.3% P<0.001),hypertension(52.9% vs.49.6% P=0.06),myocardial infarction(18.9% vs.8.9% P<0.001),chronic liver disease(33.3% vs.16.6% P<0.001),chronic kidney disease(16.1% vs.6.5% P<0.001),peripheral vascular disease(12.9% vs.6.9% P<0.001),and cerebrovascular disease(13.5% vs.9.6% P<0.001)between the two groups(p<0.01).3.Identify independent risk factors and establish predictive models:Univariate analysis showed that,Risk factors for sepsis includes age(OR=1.027,95%CI:1.024-1.030,P<0.001),male(OR=1.483,95%CI:1.351-1.627,P<0.001),systolic blood pressure(OR=0.671,95%CI:0.653-0.690,P<0.001),diastolic blood pressure/10 mm Hg(OR=0.523,95%CI:0.501-0.545,P<0.001),mean arterial pressure/10 mm Hg(OR=0.533,95%CI:0.512-0.556,P<0.001),respiratory rate(OR=1.013,95%CI:1.006-1.020,P<0.001),peripheral blood oxygen saturation(OR=0.974,95%CI: 0.965-0.983,P<0.001),GCS score(OR=0.269,95%CI:0.245-0.296,P<0.001),diabetes(OR=1.794,95%CI:1.6033-2.007,P<0.001),hypertension(OR=1.140,95%CI:1.039-1.251,P<0.001),myocardial infarction(OR=2.384,95%CI:2.032-2.798,P<0.001),chronic liver disease(OR=2.772,95%CI:2.305-3.334,P<0.001),chronic renal disease(OR=6.600,95%CI:5.353-8.137,P<0.001),congestive heart failure (OR=2.505,95%CI:2.215-2.832,P<0.001),peripheral vascular disease(OR=1.944,95%CI:1.624-2.326,P<0.001),and cerebrovascular disease(OR=1.478,95%CI: 1.265-1.728,P<0.001).For continuous variables,they were converted to dichotomous variables by calculating the maximum Youden index.The above variables were includ ed in the multivariate Logistic regression model.In order to take into account the accuracy and convenience,after calculation and adjustment,finally: male(VIF=1.014),systolic blood pressure <90mm Hg(VIF=1.006,Yuden index max: 0.29),mean arterial pressure <65mm Hg(VIF=1.30,Yuden index max: 0.33),GCS score <15(VIF=1.006,Yuden index max: 0.54),history of myocardial infarction(VIF=1.075),history of congestive heart failure(VIF=1.155),history of chronic liver disease(VIF=1.014),history of chronic kidney disease(VIF=1.103)were included in the prediction model,and there was no collinearity among the variables.The area under the ROC curve of the model was 0.906,and the goodness-of-fit test p=0.245.4.Development and validation of risk scale:There are two ways to assign values to variables: 1)assign values to each variable in combination with regression coefficients;2)assign 1 point to each variable regardless of regression coefficients.The area under the ROC curve of the two risk scores was calculated separately(0.877 vs.0.837),and the results showed that scoring variable assignments according to the regression coefficient can obtain more accurate results,and more than 6 points(Yuden index max: 0.643)are considered sepsis patients,the sensitivity was 75%,the specificity was 88%,and the positive predictive value was 98%.In the validation set,similar results were obtained(the sensitivity was 75%,the specificity was 89%,and the positive predictive value was 98%).Conclusion:1.Age,male,systolic blood pressure,diastolic blood pressure,mean arterial pressure,respiratory rate,peripheral blood oxygen saturation,GCS score,diabetes,hypertension,myocardial infarction,chronic liver disease,kidney disease,congestive heart failure,peripheral Vascular disease and cerebrovascular disease are risk factors for sepsis.2.This study developed and validated a simple and accurate sepsis early diagnosis scoring scale,which can help clinicians identify sepsis patients early,but its application and promotion still need further multi-center,large sample,prospective studies.