A machine learning-based approach for predicting renal function recovery in general ward patients with acute kidney injury

초록

Acute kidney injury (AKI) is a significant challenge in healthcare, imposing a significant social burden. While there are considerable researches dedicated to AKI and the recovery of AKI patients, a crucial factor in their prognosis, is often overlooked. Thus, our study aims to address this issue through the development of a machine learning-based approach to predict restoration of kidney function in patients with AKI. Our study encompassed data from 350,345 cases, derived from two hospitals. AKI was classified in accordance with the Kidney Disease: Improving Global Outcomes. Criteria for recovery were established as either a 33% decrease in serum creatinine levels at AKI onset or reduction to values lower than the baseline, which was initially employed for the diagnosis of AKI. We employed various machine learning models, selecting 43 pertinent features for analysis. Our analysis contained 7,041 and 2,929 patients' data from internal cohort and external cohort respectively. The Categorical Boosting model demonstrated significant predictive accuracy, as evidenced by an internal area under the receiver operating characteristic curve (AUROC) of 0.7860, and an external AUROC score of 0.7316, thereby confirming its robustness in predictive performance. SHapley Additive exPlanations values were employed to explain key factors impacting recovery of renal function in AKI patients, highlighting factors such as elevated urine specific gravity, body temperature, and phosphorus levels. This study presented a novel machine learning framework for predicting renal function recovery in patients with AKI, offering a deeper understanding of the key variables affecting recovery. The clinical applicability of the model was assessed across distinct hospital settings, which revealed variations in its efficacy. Although the model exhibited favorable outcomes, the necessity for further enhancements and the incorporation of more diverse datasets is imperative for its application in real-world scenarios.

키워드