Abstract: FR-PO0010
Artificial Intelligence-Driven Prediction of CKD Based on Dietary and Clinical Features: A Population-Based Study in Korea
Session Information
- Artificial Intelligence and Digital Health at the Bedside
November 07, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Kim, Da Woon, Pusan National University Hospital, Busan, Busan, Korea (the Republic of)
- Lee, Donghyun, Pohang University of Science and Technology, Pohang-si, Gyeongsangbuk-do, Korea (the Republic of)
- Kim, Jinmi, Pusan National University Hospital, Busan, Busan, Korea (the Republic of)
- Kim, Won Hwa, Pohang University of Science and Technology, Pohang-si, Gyeongsangbuk-do, Korea (the Republic of)
- Kim, Hyo Jin, Korea University Guro Hospital, Guro-gu, Seoul, Korea (the Republic of)
Background
This study aimed to assess the predictive value of food group features in chronic kidney disease (CKD) development using machine learning models.
Methods
We analyzed data from 57,213 participants in Korean Genome and Epidemiology Study. Dietary intake was assessed via a validated food frequency questionnaire, classifying 106 food items into 21 food groups. Baseline clinical features and food data were used to train six machine learning models including gradient boosting classifier (GBC). Model performance was evaluated using accuracy, precision, recall and area under the curve (AUC). Shapley additive explanation (SHAP) values were used to interpret feature importance.
Results
Among 57,213 participants, 525 developed CKD during follow-up. CKD groups were older, more likely male, with higher prevalence of diabetes mellitus (DM) and hypertension than non-CKD groups. GBC achieved the best predictive performance when trained on both clinical and food group features: accuracy 0.86, precision 0.84, recall 0.9, and AUC 0.91. SHAP analysis (Fig 1) showed that baseline estimated glomerular filtration rate, age, and presence of DM were the most influential predictors. Importantly, food features such as other grains and soybean paste also contributed to CKD risk prediction.
Conclusion
Our findings demonstrate that integrating food group features into machine learning models enhances CKD prediction performance, revealing that dietary elements, alongside clinical features, play a non-negligible role in CKD development. These results support personalized dietary interventions as part of CKD prevention strategies.
Figure 1. Shapley additive explanation summary plots for the Gradient Boosting Classifier trained with both food group and basic features.