Development of Various Diabetes Prediction Models Using Machine Learning Techniques

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

BACKGROUND

An 8-year prediction of the Framingham Diabetes Risk Model (FDRM) was proposed, but the predictor has a gap with current clinical standards. Therefore, we evaluated the validity of the original FDRM in Korean population data, developed a modified FDRM by redefining the predictors based on current knowledge, and evaluated the internal and external validity.

METHODS

Using data from a community-based cohort in Korea (n = 5,409), we calculated the probability of diabetes through FDRM, and developed a modified FDRM based on modified definitions of hypertension (HTN) and diabetes. We also added clinical features related to diabetes to the predictive model. Model performance was evaluated and compared by area under the curve (AUC).

RESULTS

During the 8-year follow-up, the cumulative incidence of diabetes was 8.5%. The modified FDRM consisted of age, obesity, HTN, hypo-high-density lipoprotein cholesterol, elevated triglyceride, fasting glucose, and hemoglobin A1c. The expanded clinical model added γ-glutamyl transpeptidase to the modified FDRM. The FDRM showed an estimated AUC of 0.71, and the model's performance improved to an AUC of 0.82 after applying the redefined predictor. Adding clinical features (AUC = 0.83) to the modified FDRM further improved in discrimination, but this was not maintained in the validation data set. External validation was evaluated on population-based cohort data and both modified models performed well, with AUC above 0.82.

CONCLUSION

The performance of FDRM in the Korean population was found to be acceptable for predicting diabetes, but it was improved when corrected with redefined predictors. The validity of the modified model needs to be further evaluated.

Machine learning for predicting diabetic metabolism in the Indian population using polar metabolomic and lipidomic features

AIMS

To identify metabolite and lipid biomarkers of diabetes in the Indian subpopulation in newly diagnosed diabetic and long-term diabetic individuals. To utilize the global polar metabolomic and lipidomic profiles to predict the susceptibility of an individual to diabetes using machine learning algorithms.

MATERIALS AND METHODS

87 individuals, including healthy, newly diabetic, and long-term diabetics on medication, were included in the study. Post consent, their serum was used to isolate polar metabolome and lipidome. NMR and LCMS were used to identify the polar metabolites and lipids, respectively. Statistical analysis was done to determine significantly altered molecules. NMR and LCMS comprehensive data were utilized to generate diabetic models using machine learning algorithms. 10 more individuals (pre-diabetic) were recruited, and their polar metabolomic and lipidomic profiles were generated. Pre-diabetic metabolic profiles were then utilized to predict the diabetic status of the metabolome and lipidome beyond glucose levels.

RESULTS

Mannose, Betaine, Xanthine, Triglyceride (38:1), Sphingomyelin (d63:7), and Phosphatidic acid (37:2) are some of the top key biomarkers of diabetes. The predictive model generated showed the receiver operating characteristic area under the curve (ROC-AUC) as 1 on both test and validation data indicating excellent accuracy. This model then predicted the diabetic closeness of the metabolism of pre-diabetic individuals based on probability scores.

CONCLUSION

Polar metabolic and lipid profile of diabetic individuals is very different from that of healthy individuals. Lipid profile alters before the polar metabolic profile in diabetes-susceptible individuals. Without regard to glucose, the diabetic closeness of the metabolism of any individual can be determined.

The Present and Future of Artificial Intelligence-Based Medical Image in Diabetes Mellitus: Focus on Analytical Methods and Limitations of Clinical Use

Artificial intelligence (AI)-based diagnostic technology using medical images can be used to increase examination accessibility and support clinical decision-making for screening and diagnosis. To determine a machine learning algorithm for diabetes complications, a literature review of studies using medical image-based AI technology was conducted using the National Library of Medicine PubMed, and the Excerpta Medica databases. Lists of studies using diabetes diagnostic images and AI as keywords were combined. In total, 227 appropriate studies were selected. Diabetic retinopathy studies using the AI model were the most frequent (85.0%, 193/227 cases), followed by diabetic foot (7.9%, 18/227 cases) and diabetic neuropathy (2.7%, 6/227 cases). The studies used open datasets (42.3%, 96/227 cases) or directly constructed data from fundoscopy or optical coherence tomography (57.7%, 131/227 cases). Major limitations in AI-based detection of diabetes complications using medical images were the lack of datasets (36.1%, 82/227 cases) and severity misclassification (26.4%, 60/227 cases). Although it remains difficult to use and fully trust AI-based imaging analysis technology clinically, it reduces clinicians' time and labor, and the expectations from its decision-support roles are high. Various data collection and synthesis data technology developments according to the disease severity are required to solve data imbalance.

Improving Machine Learning Diabetes Prediction Models for the Utmost Clinical Effectiveness

The early prediction of diabetes can facilitate interventions to prevent or delay it. This study proposes a diabetes prediction model based on machine learning (ML) to encourage individuals at risk of diabetes to employ healthy interventions. A total of 38,379 subjects were included. We trained the model on 80% of the subjects and verified its predictive performance on the remaining 20%. Furthermore, the performances of several algorithms were compared, including logistic regression, decision tree, random forest, eXtreme Gradient Boosting (XGBoost), Cox regression, and XGBoost Survival Embedding (XGBSE). The area under the receiver operating characteristic curve (AUROC) of the XGBoost model was the largest, followed by those of the decision tree, logistic regression, and random forest models. For the survival analysis, XGBSE yielded an AUROC exceeding 0.9 for the 2- to 9-year predictions and a C-index of 0.934, while the Cox regression achieved a C-index of 0.921. After lowering the threshold from 0.5 to 0.25, the sensitivity increased from 0.011 to 0.236 for the 2-year prediction model and from 0.607 to 0.994 for the 9-year prediction model, while the specificity showed negligible changes. We developed a high-performance diabetes prediction model that applied the XGBSE algorithm with threshold adjustment. We plan to use this prediction model in real clinical practice for diabetes prevention after simplifying and validating it externally.

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

BACKGROUND

In the euthyroid state, the risk of developing diabetes according to changes in thyroid-stimulating hormone (TSH) levels remains controversial. Additionally, the correlation of various body indices affecting blood glucose levels according to changes in TSH levels over a certain period is not well known.

METHODS

Patients who underwent health check-ups twice at a 2 year interval at a tertiary university hospital between 2009 and 2018 were included. By dividing baseline TSH levels into quartiles (TSH_Q1, Q2, Q3, and Q4), various variables were compared, and their changes after 2 years (∆TSH_Q1, Q2, Q3, and Q4) were confirmed.

RESULTS

Among 15 557 patients, the incidence of diabetes mellitus after 2 years was 2.4% (377/15 557 patients). There was no statistically significant difference in the incidence of diabetes according to TSH_Q (p = 0.243) or ∆TSH_Q (p = 0.131). However, as TSH levels increased, skeletal muscle mass decreased (p < 0.001), and body fat mass and percent body fat significantly increased (p < 0.001). As ∆TSH increased, ∆fasting blood glucose and ∆body mass index also significantly increased (all p < 0.001). The incidence of diabetes decreased significantly as skeletal muscle mass increased (odds ratio 0.734, p < 0.001).

CONCLUSIONS

Owing to the short study period, it was not possible to prove a statistical relationship between the incidence of diabetes mellitus and TSH levels in the euthyroid state. Significant decreases in skeletal muscle mass and increases in body mass index and body fat mass according to baseline TSH levels were demonstrated. Therefore, a focus on improving physical functions, such as increasing muscle mass and decreasing fat, is required in this case.