The following is a summary of “Machine Learning Predicts Acute Kidney Injury in Hospitalized Patients with Sickle Cell Disease,” published in the October 2023 issue of Nephrology by Zahr et al.
Acute Kidney Injury (AKI) is common in hospitalized sickle cell disease (SCD) patients and can lead to poor outcomes. Early diagnosis and management are key. Researchers performed a retrospective study to use machine learning (ML) to predict AKI earlier in patients with SCD using continuous minute-by-minute physiological data.
About 6,278 adult patients with SCD were admitted to five regional hospitals. Out of these, 1,178 patients were selected based on data availability. Among the selected patients, AKI was detected in 82 patients (7%) using the 2012 Kidney Disease Improving Global Outcomes (KDIGO) criteria. The remaining 1,096 patients were used as controls. Data from various physiological parameters (heart rate, respiratory rate, and blood pressure) collected every minute from bedside monitors were utilized. The classification was performed using an XGBoost classifier.
The result demonstrated strong predictive capabilities for AKI, achieving an AUROC of 0.91 (95% CI [0.89-0.93]) up to 12 hours before AKI onset and an AUROC of 0.82 (95% CI [0.80-0.83]) up to 48 hours before AKI. Patients with AKI were characterized by a higher proportion of females (64.6%) and a higher prevalence of hypertension, pulmonary hypertension, CKD, and pneumonia compared to the control group.
The study found that XGBoost accurately predicted AKI 12 hours before onset in hospitalized SCD patients, enabling innovative prevention strategies.
