Model Construction For Predicting The Risk Of Impaired Fasting Glucose

Objective: Impaired fasting glucose(IFG)is an early insidious but reversible state of abnormal glucose metabolism,and most patients have no typical symptoms.In recent years,the global incidence of IFG has increased dramatically,causing a serious medical and economic burden.Studies have confirmed that IFG is not only strongly associated with type 2 diabetes and increased risk of cardiovascular disease,but also with atrial fibrillation,heart failure,septal hypertrophy,cerebral hemorrhage,all-cause mortality and cardiovascular event outcomes.This shows that IFG has become a public health problem that cannot be ignored.Therefore,early identification and intervention of IFG is of great significance,not only for capturing the critical period of disease prevention and early intervention,but especially for reducing the medical and economic burden of IFG-related diseases defensively.In clinical practice,the commonly used IFG screening methods are fasting glucose test and oral glucose tolerance test(OGTT).Fasting glucose test results are more volatile and affected by diet,sleep,blood pressure,exercise status and other factors,which often lead to it is often difficult to ensure that test results are free from these factors.OGTT,on the other hand,is time-consuming,labor-intensive,making it difficult to conduct widespread screening in a large population.The lack of effective IFG risk screening tools leads to missed optimal intervention for patients with early glucose metabolism abnormalities,which in turn leads to serious health problems such as high incidence and low diagnosis of type 2 diabetes,high prevalence of cardiovascular disease and increased mortality.Therefore,in addition to the two methods mentioned above,there is a need for an efficient and convenient tool for a more accurate and comprehensive IFG risk assessment of the general population,and risk prediction models can meet this need.Risk prediction models are effective tools for risk assessment,decision selection,and benefit evaluation in health care systems,and have an important role in the tertiary disease prevention system.Common prediction models include linear regression models,logistic regression models,and COX regression models,etc.Which can also be presented in the form of a column line graph(Nomogram),making it transform complex regression equations into visual graphs,making the results of prediction models more readable and easy to apply.At present,there are already prediction models that can predict the risk of IFG,but most of them need to further improve the prediction performance for IFG.This study aims to construct and validate a Clinical and Laboratory-based Nomogram(CLN)model for IFG risk prediction.Methods:This is a retrospective cross-sectional study in which general information,anthropometric data,laboratory test indices and results of routine ancillary examinations were collected anonymously from medical examiners at the Health Management Center of the Second Affiliated Hospital of Dalian Medical University.The medical examiners’ data from 2020 and 2016 were used as the model development dataset and the model independent validation dataset,respectively.In constructing the model development dataset,a total of 39 variables were collected from 3 561 medical examiners,screened according to the inclusion criteria,and finally 2 340 study subjects were included,which were randomly divided 7:3 into 1 638 cases in the training set and 702 cases in the validation set.In the training set of the model development dataset,variables with statistically significant differences between the IFG and non-IFG groups were screened by univariate analysis;further,risk factors independently associated with IFG were identified by LASSO regression and multifactor logistic regression.Subsequently,the CLN model for predicting the risk of IFG was constructed,and a health check-up subject was randomly selected to predict the risk of IFG using the CLN model,demonstrated an example of the application of the CLN model;furthermore,the performance of the CLN model was evaluated in the training and validation sets,respectively.By assessing the discrimination of the model by subject operating characteristic(ROC)curve,area under the ROC curve(AUC)values,calibration curve to assess the calibration of the CLN model,and the level of clinical benefit using decision curve analysis(DCA).In addition,the performance of the CLN model was validated by AUC values in the model independent validation dataset(2016 medical examiner data)to assess the discriminatory performance of the CLN model in practical applications.Result:1.Analysis of the basic characteristics of the medical examiners in the model development data set.487 medical examiners in the training set and 218 medical examiners in the validation set had IFG.In the training set,a total of 31 variables differed statistically significantly(p<0.05)between the IFG and non-IFG groups,namely sex,age,vitamin D,vitamin D3,history of increased blood pressure,fatty liver,systolic blood pressure,diastolic blood pressure,height,weight,body mass index,abdominal circumference,white blood cell count,red blood cell count,neutrophil count,lymphocyte count,hemoglobin,platelets,alanine aminotransferase,aspartate aminotransferase,total protein,albumin,glutamyl transpeptidase,alkaline phosphatase,urea nitrogen,creatinine,uric acid,total cholesterol,triglycerides,high-density lipoprotein,and low-density lipoproteins.In the validation set,a total of 32 variables differed statistically significantly(p<0.05)between the IFG and non-IFG groups,in addition to the variables with statistically significant differences between groups in the training set,globulin also had statistically significant differences between IFG and non-IFG groups in the validation set.2.LASSO regression to screen variance variables.Variables with statistically significant differences between the IFG and non-IFG groups in the training set were continued to be screened using LASSO regression,and 15 predictor variables with non-zero coefficients were screened when selected =0.008603015(lambda.min),separately,they are: sex,age,vitamin D3,history of elevated blood pressure,systolic blood pressure,diastolic blood pressure,weight index,alanine aminotransferase,total protein,albumin,glutamyl transpeptidase,urea nitrogen,triglycerides,high-density lipoprotein,and low-density lipoprotein.3.Multi-factor logistic regression further confirms the predictor variables to construct the CLN model and example of the application of the CLN model.The 15 variables screened by LASSO regression were further subjected to multifactorial logistic regression(backward stepwise regression)to identify 6 predictors independently associated with IFG risk,namely age(OR: 1.055,95% CI: 1.039-1.070,p<0.001),systolic blood pressure(OR: 1.017,95% CI.1.009-1.026,p<0.001),body mass index(OR: 1.166,95% CI: 1.120-1.214,p<0.001),albumin(OR: 1.203,95% CI: 1.1360-1.273,p<0.001),urea nitrogen(OR: 1.169,95% CI: 1.044-1.309,p=0.007)and triglycerides(OR: 1.495,95% CI: 1.224-1.826,p<0.001).Subsequently,the CLN model for predicting the risk of IFG was constructed by applying the above six variables.A subject was randomly selected,and the risk of IFG was predicted to be 83.6% by using the CLN model.4.Performance validation of the CLN model in the model development dataset.To evaluate the performance of the CLN model in predicting IFG risk,it was evaluated by the following metrics in the training and validation sets,respectively.In the first step,by calculating the AUC value of the area under the ROC curve,it was found that the AUC value in the training set was 0.783(95% CI: 0.759-0.807);the AUC value in the validation set was 0.789(95% CI: 0.753-0.825),indicating that the CLN model has good discriminatory ability.In the second step,in order to assess the calibration degree of the CLN model,calibration curves were plotted,suggesting that the predicted and actual values basically overlap,using the Hosmer-Lemeshow test to show that the p-values of the training and validation sets are 0.1181 and 0.4832 are >0.05,respectively;using the Unreliability test,the p-values of the training and validation sets are 0.983 and 0.812 respectively,and the same p-values are >0.05,which proves that the CLN model has good prediction consistency.In the third step,the decision curve analysis(DCA)shows that the optimal threshold probabilities of the CLN model are obtained as follows: 2%-72% in the training set and 7%-71% in the validation set.5.Performance validation of the CLN model in the model independent validation dataset.In the model independent validation dataset(n=1,875),the AUC of the CLN model was 0.801,further indicating that the CLN has good consistency and clinical predictive value.Conclusion: In this study,we constructed and validated a CLN model capable of predicting the risk of IFG.By applying the CLN model for risk prediction and identifying people at risk for IFG,it not only aids in the early diagnosis of IFG,but also helps to reduce the burden of IFG-related diseases.