Machine Learning Model Improves Kidney Function Estimates in Young People

A machine learning model that combines two blood biomarkers with routine demographic information has shown improved agreement with measured kidney function compared with established equations in children, adolescents and young adults, according to a study published in BMC Pediatrics. The work, led by TingTing Shao, Felipe Kenji Nakano and colleagues including Pierre Delanaye and Celine Vens, systematically compared machine learning algorithms against conventional estimating equations for predicting the glomerular filtration rate, the standard measure of how well the kidneys filter waste.

Estimated glomerular filtration rate is preferred in clinical practice, especially in more vulnerable populations, because directly measuring the filtration rate involves an invasive and time-consuming procedure. In routine care, various existing equations are used to calculate the estimate, but the authors note there is room for improvement, which has motivated interest in machine learning models. Previous evidence has shown that such models can outperform equations in younger populations, though they perform comparably in general populations.

The study included seven machine learning algorithms and three equations, evaluated across three variants: models based on serum creatinine, models based on cystatin C, and models using both biomarkers. Hyperparameter optimization was performed to support model selection, and the researchers compared within each biomarker variant and between variants. They also tested models trained on data from the full available age range against models restricted to ages 2 to 25 years, and investigated whether equation-derived estimates could be included as additional input features in the machine learning models.

Performance was evaluated using Lin’s concordance correlation coefficient, the P10 and P30 measures, and Passing-Bablok regression. In these analyses, the random forest model based on both serum creatinine and cystatin C emerged as one of the most robust and strong machine learning models. It achieved a Lin’s concordance correlation coefficient of 0.80 (0.78, 0.82), with 41 percent of estimates (0.38, 0.43) within 10 percent of measured values (P10) and 89 percent (0.87, 0.91) within 30 percent (P30).

Those figures reflected improved agreement with measured glomerular filtration rate compared with the evaluated equations and with machine learning and equation variants built on a single biomarker. The dual-biomarker approach thus outperformed models that relied on serum creatinine or cystatin C alone, whether the estimator was an equation or a machine learning algorithm.

One practical finding concerned the training data. No performance difference was detected between random forest models trained on different age cohorts, meaning that restricting the training set to the 2-to-25-year window did not improve accuracy relative to training on the full age range of data.

The clearest gain came from the hybrid strategy. Integrating equation-derived estimates as additional features into the serum creatinine and cystatin C random forest improved accuracy, raising the proportion of estimates within 10 percent of measured values to 45 percent (0.42, 0.48). According to the authors, the best-performing model combined biomarkers, demographic information and equation-derived estimates, indicating a complementary role for machine learning and the domain knowledge captured by the evaluated filtration rate equations rather than a wholesale replacement of them.

The study was published as an open-access, peer-reviewed accepted article in BMC Pediatrics on 23 October 2026, shared early with a permanent digital object identifier and subject to further edits before the final version of record. Within its evaluated scope of algorithms, equations and the 2-to-25-year population, the work positions the dual-biomarker random forest, particularly when enriched with equation outputs, as one of the most robust available approaches for estimating kidney function in this age group.

Subject of Research: Machine learning models for estimating glomerular filtration rate in children, adolescents and young adults

Article Title: Machine learning models to estimate the glomerular filtration rate in children, adolescents and young adults

Article References: Shao, T., Nakano, F. K., Van Kleunen, L., Ilídio, P., Björk, J., Derain-Dubourg, L., Hansson, M., Littmann, K., Bökenkamp, A., Berg, U., Åsling-Monemi, K., Sjöström, P., Nyman, U., Lanot, A., Pottel, H., Delanaye, P., & Vens, C. (2026). Machine learning models to estimate the glomerular filtration rate in children, adolescents and young adults. BMC Pediatrics. https://doi.org/10.1186/s12887-026-07877-1

Image Credits: AI Generated

DOI: 10.1186/s12887-026-07877-1

Keywords: glomerular filtration rate, machine learning, random forest, serum creatinine, cystatin C, eGFR, pediatrics, kidney function, biomarkers, BMC Pediatrics, Machine, learning