Development of Various Diabetes Prediction Models Using Machine Learning Techniques

Abstract

Background: There are many models for predicting diabetes mellitus (DM), but their clinical implication remains vague. Therefore, we aimed to create various DM prediction models using easily accessible health screening test parameters. Methods: Two sets of variables were used to develop eight DM prediction models. One set comprised 62 easily accessible examination results of commonly used variables from a tertiary university hospital. The second set comprised 27 of the 62 variables included in the national routine health checkups. Gradient boosting and random forest algorithms were used to develop the models. Internal validation was performed using the stratified 10-fold cross-validation method. Results: The area under the receiver operating characteristic curve (ROC-AUC) for the 62-variable DM model making 12-month predictions for subjects without diabetes was the largest (0.928) among those of the eight DM prediction models. The ROC-AUC dropped by more than 0.04 when training with the simplified 27-variable set but still showed fairly good performance with ROC-AUCs between 0.842 and 0.880. The accuracy was up to 11.5% higher (from 0.807 to 0.714) when fasting glucose was included. Conclusion:We created easily applicable diabetes prediction models that deliver good performance using parameters commonly assessed during tertiary university hospital and national routine health checkups. We plan to perform prospective external validation, hoping that the developed DM prediction models will be widely used in clinical practice.

Validation of the Framingham Diabetes Risk Model Using Community-Based KoGES Data

“…Studies that developed diabetes prediction models using health checkup data generally did not consider 2-hour glucose when defining diabetes. 25 26 However, in line with previous studies, 27 28 we observed differences in prevalence depending on the definition of diabetes. Also, one meta-study suggested lowering the cutoff values for FBG and HbA1c in relation to diabetes diagnosis.…”

“…In addition, in our study, no significant contribution of parental history of diabetes was observed in predicting diabetes. Also, like previous studies, 5 25 the contribution of sex to diabetes prediction was not found to be significant. Instead, the modified FDRM showed a predictive power of over 0.8 for both sexes.…”

“…Another Korean study using health examination data also found that γ-GTP contributed to the prediction of diabetes. 25 The role of oxidative stress in the development of diabetes-related complications is unclear; some studies found associations with peripheral neuropathy and retinopathy. 34 35 Although the Framingham offspring study developed a simple predictive model, the use of a parental history of diabetes is challenging.…”

Retrospective cohort analysis comparing changes in blood glucose level and body composition according to changes in thyroid‐stimulating hormone level

“…However, because this study used data from medical examinations at university hospitals, it contained many variables. 37 Lastly, the medical check-up did not include postprandial blood glucose measurement, one of the main diagnostic criteria for diabetes. Therefore, patients with impaired glucose tolerance may have been missed.…”

“…Third, in this study, it was difficult to identify a clear causal relationship due to the nature of the retrospective cohort study, 35,36 and since various confounding variables could not be excluded, only correlations could be estimated. However, because this study used data from medical examinations at university hospitals, it contained many variables 37 . Lastly, the medical check‐up did not include postprandial blood glucose measurement, one of the main diagnostic criteria for diabetes.…”