Improving precision of glomerular filtration rate estimating model by ensemble learning.

Area of Science:

• Nephrology
• Biostatistics
• Machine Learning

Background:

• Accurate assessment of kidney function is crucial for clinical management.
• Current regression-based estimates of glomerular filtration rate (GFR) lack precision.

Purpose of the Study:

• To investigate the potential of ensemble learning to enhance the precision of GFR estimation.
• To compare the performance of ensemble learning models against traditional regression methods.

Main Methods:

• Developed and validated GFR estimation models using artificial neural networks (ANN), support vector machines (SVM), and regression.
• Utilized a dataset of 1419 participants, with 1002 in the development and 417 in the external validation cohorts.
• Employed 99mTc-DTPA renal dynamic imaging with dual plasma sample calibration as the measured GFR reference.

Main Results:

• Ensemble learning models demonstrated superior precision in GFR estimation compared to a new regression model in the external validation cohort (IQR 13.5 vs. 14.0 ml/min/1.73 m²).
• The ensemble model integrating ANN, SVM, and regression showed the best precision among the tested models.
• Bias and accuracy were comparable across models, with median differences ranging from 2.3 to 3.7 ml/min/1.73 m² and 30% accuracy between 73.1% and 76.0%.

Conclusions:

• An ensemble learning model combining ANN, SVM, and regression provided more precise GFR estimates than the new regression model.
• Further advancements in ensemble learning strategies may lead to even greater improvements in GFR estimation accuracy.