LB013 - PREDICTING GLYCEMIC IMPROVEMENT USING THE CLINICAL NUTRITION PROBLEM DIAGNOSTIC MATRIX FOR TYPE 2 DIABETES (CNP-DM)

LB013

PREDICTING GLYCEMIC IMPROVEMENT USING THE CLINICAL NUTRITION PROBLEM DIAGNOSTIC MATRIX FOR TYPE 2 DIABETES (CNP-DM)

M. Kang^1,2,*, H. Lim^1,2

¹Department of Medical Nutrition, Kyung Hee University, Yongin, ²Research Institute of Medical Nutrition, Kyung Hee University, Seoul, Korea, Republic Of

Rationale: Clinical nutrition diagnoses summarize dietary, behavioral, and metabolic problems that may influence glycemic control in individuals with type 2 diabetes. This study evaluated the utility of the Clinical Nutrition Problem Diagnostic Matrix for Type 2 Diabetes (CNP-DM) for predicting short-term glycemic outcomes.

Methods: Data were obtained from the Korean National Diabetes Program. Baseline observations collected at 0 and 36 months were paired with glycemic outcomes measured 6 months later (6 and 42 months, respectively). Clinical variables, demographic characteristics, and CNP-DM variables were used to develop XGBoost models for predicting (1) achievement of an HbA1c reduction of at least 0.5 percentage points and (2) HbA1c values at 6 months. A total of 1,505 participants were randomly divided into training (80%) and test (20%) sets. Model performance was evaluated using accuracy, AUROC, RMSE, and R². SHAP analysis assessed feature importance.

Results: The classification model predicting achievement of an HbA1c reduction of at least 0.5 percentage points demonstrated good performance, achieving an accuracy of 0.799 and an AUROC of 0.856 in the test set. SHAP analysis identified glycemic indicators as the strongest predictors, while CNP-DM variables contributed additional nutrition-related information. In contrast, the regression model predicting HbA1c values at 6 months showed limited predictive performance (RMSE = 1.335, R² = -0.080).

Conclusion: Machine learning models incorporating CNP-DM variables successfully predicted short-term glycemic improvement in adults with type 2 diabetes. These findings suggest that nutrition diagnosis information may support personalized nutrition care and patient-specific risk communication.

Funding: This research was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. RS-2022-NR069092).

Disclosure of Interest: None declared