Photo Credit: freepik

The following is a summary of “Derivation and validation of a machine learning model for the prevention of unplanned dialysis,” published in the May 2024 issue of Nephrology by Klamrowski et al.

Half of all patients with advanced chronic kidney disease (CKD) progressing towards kidney failure start dialysis unexpectedly, further leading to high morbidity, mortality, and healthcare costs. 

Researchers conducted a retrospective study to develop a novel prediction model aimed at identifying advanced patients with CKD at high risk for developing kidney failure within 6-12 months. The goal was to reduce unplanned dialysis rates and improve care transitions.

They employed machine learning (ML) random forest algorithms, incorporating age, sex, and lab trends, to create and validate 6 and 12-month kidney failure risk models. The models were tested in 3 independent cohorts in Ontario, Canada. One group had 1,849 patients with advanced CKD (mean age 66±15 years, eGFR 19 [7] mL/min/1.73m²), and validated in 2 external advanced CKD cohorts (n=1,356; age 69±14 years, eGFR 22 [7] mL/min/1.73m²)

The results showed 55% developed kidney failure, with 35% starting unplanned dialysis. The 6- and 12-month models showed excellent discrimination with Area under receiver operating characteristic curve (AUC: 0.88 [95% CI: 0.87-89) and 0.87 [95% CI: 0.86-0.87]) and high accuracy (Brier scores 0.10 [95% CI 0.09-0.10] and 0.14 [95% CI 0.13-0.14). The models were well-calibrated, providing timely alerts for many unplanned dialysis cases. Similar results were found in external validation. 

Investigators concluded that the ML models effectively predict kidney failure risk in advanced patients with CKD, offering early warnings for unplanned dialysis. However, the best ways to implement them still require further exploration.

Source: journals.lww.com/cjasn/abstract/9900/derivation_and_validation_of_a_machine_learning.393.aspx