In this study, we developed a simple, quantifiable, and clinically beneficial nomogram to predict the 14-year risk of type 2 diabetes in nondiabetic populations in East China, which is also in line with the further development of the PPPM strategy. Both The calibration curve and AUC values indicate that the nomogram had good calibration and has good predictive capabilities. In addition, our DCA shows that the nomogram has good clinical application value. The predictive model can help to predict the incidence of type 2 diabetes, and early intervention may help at-risk individuals avoid the occurrence of type 2 diabetes -related complications.
The prevalence of type 2 diabetes continues to rise, placing a huge burden on patients and society[1]. Effective prevention of type 2 diabetes is essential to reduce the impact of this disease. Identifying high-risk groups of type 2 diabetes through risk prediction methods, and early intensive lifestyle intervention for high-risk groups of type 2 diabetes can help reduce the psychological burden of patients, enhance their confidence in following a healthy lifestyle, and improve their quality of life. It can also delay disease progression and reduce the risk of life-long complications in the long run. In the past few decades, various predictive models have been developed to predict the occurrence of type 2 diabetes. Well-known examples include the Finnish Diabetes Risk Score[14], Australian Type 2 Diabetes Risk[15], QRISK[16], and Framingham Offspring (FOS) Risk[17]. Most type 2 diabetes prediction models use logistic [18-20] or Cox regression [21-23] and use carriage return and automatic direction forward selection, backward elimination, or step-by-step procedures.
In our study, the nomogram first constructs a multi-factor regression model (LASSO regression and logistic regression), and assigns each value level of each influencing factor according to the degree of contribution of each influencing factor in the model to the outcome variable (the size of the regression coefficient), assign a score to each value level of each influencing factor, and then add up the scores to obtain a total score. Finally, the predicted value of the individual outcome event is calculated through the function conversion relationship between the total score and the probability of the outcome event. Simply put, the nomogram transforms complex regression equations into a visual graph, making the results of the predictive model more readable and possibly better facilitating patient evaluation. The intuitive and easy-to-understand characteristics of the nomogram have gradually garnered it more and more attention and application in both medical research and clinical practice[24, 25]. In addition, all previous T2DM risk prediction studies conducted in China have established T2DM risk scores of integer points or segment values, and our nomograms can provide more accurate and personalized risk predictions due to the use of continuous values. This is in line with EPMA. The same point of view is that individualization should become a general social trend in the field of medicine and healthcare[26].
At present, the nomogram has been previously studied in the field of type 2 diabetes in China[27-29], but the nomogram cannot be extended from one region to another, and there is no nomogram study in East China. Compared with these studies, our study has the highest AUC value and the longest follow-up time (14 years) so far used. In greater detail, we find gender, BMI, ALT, CREA, CHOL, HDL, Glu, MCHC, WBC, and age to be independent risk factors for type 2 diabetes during our 14-year sample period, and gender, BMI, HDL, GLU, and age have been also included in other predictive models[27-29]. In addition, the studies have also shown that impaired liver function is associated with type 2 diabetes[30]. And CREA[31], WBC [32] and MCHC[33] are also related to type 2 diabetes. Therefore, the application of these parameters in our model appears to be well-founded. Finally, our calibration curve shows that the nomogram has good calibration ability, and the AUC shows the statistical accuracy of the nomogram. However, accuracy does not mean that it has clinical application value. For this reason, we also conducted a decision curve analysis, which showed that the nomogram has good clinical utility.
Although our model’s predictive results are good, a key limitation of this study is that we predicted the risk of type 2 diabetes based only on laboratory data, and did not include factors such as diet, exercise, or genetics that have been shown to be closely related to type 2 diabetes because we did not collect these data. Another limitation is that the diagnosis of type 2 diabetes can only rely on fasting blood glucose, glycosylated hemoglobin, and medical history from physical examination data because the physical examination population could not perform the standard OGTT test for the diagnosis of type 2 diabetes. In addition, we did not divide our cohort into two groups for internal verification and external verification due to our relatively small amount of type 2 diabetes data.
Expert recommendations
As suggested in the Special Session “PPPM in Diabetes Mellitus” of the 2012 EPMA White Paper[7], measures focused on prevention and early identification methods deserve due consideration, and a central component of prevention strategies is identification of individuals at risk of diabetes mellitus. We constructed a type 2 diabetes prediction model based on the R language, which is also in line with the recommendations in the Special Session "Patient-specific Modeling", research and development on the predictive and preventive potential of the new IT tools for in vitro and in vivo diagnostics and consequent evaluation and implementation of these tools in daily healthcare. The model constructed in this study can predict individuals at high risk of type 2 diabetes, and personalized medical guidance for these high-risk individuals can be targeted to primary prevent type 2 diabetes. For the subsequent promotion of the model, we will combine other data (such as OGTT, eating behavior habits) to further optimize.