Kidney Week

Abstract: TH-PO971

Age-Specific Relationship between Insulin Resistance and Obesity: A Machine Learning-Based Interpretation

Session Information

• Physical Activity and Lifestyle in Kidney Diseases
  October 24, 2024 | Location: Exhibit Hall, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Health Maintenance, Nutrition, and Metabolism

• 1500 Health Maintenance, Nutrition, and Metabolism

Authors

• Oh, Sewon, Korea University, Seongbuk-gu, Korea (the Republic of)

Sewon Oh, MD, PhD

• Jang, Yookyung, Korea University, Seongbuk-gu, Korea (the Republic of)

Yookyung Jang, MD

• Choi, Young Eun, Korea University, Seongbuk-gu, Korea (the Republic of)

Young Eun Choi, MD

• Koo, Tai yeon, Korea University, Seongbuk-gu, Korea (the Republic of)

Tai yeon Koo, MD, PhD

• Kim, Myung-Gyu, Korea University, Seongbuk-gu, Korea (the Republic of)

Myung-Gyu Kim, MD

• Jo, Sang-Kyung, Korea University, Seongbuk-gu, Korea (the Republic of)

Sang-Kyung Jo, MD, PhD

Background

There has been a threefold surge in the global prevalence of obesity over the last four decades. Obesity is related to the increased risk of cardiovascular and kidney disease. Obesity in the elderly exhibits distinct features, such as sarcopenia and an increased visceral fat mass. We investigate the risk factors for obesity according to age by developing machine learning (ML) model.

Methods

We performed ML analysis on 3768 individuals whose age was over 18 from the 2021 Korea National Health and Nutrition Examination Survey dataset. ML predicted individual body mass index (BMI) values, and the predictive values were labeled into normal (BMI<25 kg/m²) and obese (BMI>25 kg/m²). Through 5-fold cross-validation, the performance and SHapley Additive exPlanations (SHAP) values, representing the feature importance in each sample, were calculated in the test set of every fold and collected.

Results

The Light Gradient Boosting Machine demonstrated reliable prediction, yielding an area under ROC curve of 0.813 (higher than other ML algorithms) and an error of 2.199±1.654. SHAP analysis revealed high importance for Homeostatic Model Assessment for Insulin Resistance (HOMA-IR) and fasting insulin, displaying an increasing trend with higher BMI (Figure 1.a). Important indicators were followed by, systolic blood pressure, alanine-transaminase, uric acid, HDL-cholesterol, hypertension, age and occupation. Age showed the highest interaction with the impact of HOMA-IR. A dependence plot (Figure 1.b) illustrated that the impact of high HOMA-IR on BMI was higher in younger adults compared to older adults. A significant disparity in obesity ratio between HOMA-IR>5 and HOMA-IR<5 was observed, particularly pronounced in individuals under 50 years.

Conclusion

While HOMA-IR is recognized for its significance in BMI and diabetic diseases, this study highlights its prominence over other known BMI-related features. The analysis showed that younger ages with high insulin resistance is more vulnerable to obesity compared to older adults.