Improving accuracy of estimating glomerular filtration rate using artificial neural network: model development and...

Area of Science:

• Nephrology
• Biomedical Engineering
• Data Science

Background:

• Existing glomerular filtration rate (GFR) estimation equations show reduced accuracy when applied to the Chinese population.
• Developing a more precise GFR estimation model for Chinese individuals is crucial for effective kidney disease management.

Purpose of the Study:

• To develop and validate a novel, highly accurate GFR estimation model tailored for the Chinese population.
• To compare the performance of artificial neural network (ANN) models against revised CKD-EPI equations.

Main Methods:

• A cohort of 1952 participants from a Chinese hospital was used, with data split for development and internal validation.
• Three models were developed: a 4-variable revised CKD-EPI equation, a 9-variable revised CKD-EPI equation, and a 9-variable artificial neural network (ANN) model.
• Models were evaluated based on bias, accuracy (IQR of difference), and P30 (percentage of estimates within 30% of the true GFR).

Main Results:

• The 9-variable revised CKD-EPI equation did not significantly improve performance over the 4-variable version.
• The 9-variable artificial neural network (ANN) model demonstrated a significant improvement in reducing bias (mean difference) and enhancing P30 accuracy compared to the CKD-EPI equations.
• The ANN model achieved a mean difference of 2.77 and a P30 of 80.0%, indicating superior performance.

Conclusions:

• Artificial neural network (ANN) models can effectively leverage multiple independent variables for superior GFR estimation.
• The developed 9-variable ANN model offers a more accurate GFR estimation for the Chinese population.
• Non-linear modeling approaches like ANN are recommended for enhancing GFR prediction accuracy.