Photo Credit: istock.com/Marco Marca

Chronic kidney disease (CKD) is characterized by the progressive injury and loss of kidney function, which affects more than 800 million individuals. However, determining who will progress to kidney failure is challenging. Artificial intelligence, particularly machine learning (ML), shows promise as a tool to optimize clinical decision-making.

Findings from a recent review sets the foundation for the development and validation of a ML model to predict the progression of CKD to kidney failure, according to researchers from Australia, who reported their findings in Nephrology.

For the review, Zane A. Miller, MD, and colleagues aimed to identify the most common and critical variables (demographic, clinical, and biochemical) used in ML models to predict progression of CKD to kidney failure. The researchers used Medline and EMBASE databases to find studies that met the criteria, including that ML was the focus of the predictive model. After screening 595 articles, 16 were included in the review.

The included studies represented modeling on 297,185 patients with CKD. The average age ranged from 44.89-81.35 years. The estimated glomerular filtration rate ranged from 24.3 mL/min/1.73m² to 86 mL/min/1.73m². The predominant ML models used were random forest, support vector machines, and XGBoost.

The most commonly occurring variables were age, gender, measures of renal function, measures of proteinuria, and full blood examination. The researchers noted that only half of all studies included clinical variables in their models. Overall, the most important variables were measures of renal function, measures of proteinuria, age, full blood examination, and serum albumin.

Kidney Failure Risk Equation (KFRE) is a widely validated tool and the most used clinical model to predict CKD progression to kidney failure. However, the researchers identified four of six studies that showed superior predictive capacity compared with KFRE.

“Despite great variation, key variables emerged in both the frequency of usage as well as the importance to model prediction. Most of the identified variables support the current clinical understanding of CKD, however, several variables such as measures of liver function and serum albumin outperformed clinical expectations and may warrant inclusion in any future ML models,” the authors concluded.