精准医学杂志 (Dec 2023)

PERSONALIZED PREDICTIVE VALUE OF ARTIFICIAL INTELLIGENCE FOR ACUTE KIDNEY INJURY AND MORTALITY RISK IN HOSPITALIZED PATIENTS RECEIVING FUROSEMIDE

  • CUI Lianshun, XU Lingyu, LI Tianyang, GUAN Chen, YANG Chengyu, ZHANG Ningxin, SONG Zhuo, XU Yan

DOI
https://doi.org/10.13362/j.jpmed.202306002
Journal volume & issue
Vol. 38, no. 6
pp. 475 – 480

Abstract

Read online

Objective To analyze the association between the use of furosemide in hospitalized patients and the risk of acute kidney injury (AKI) and death based on machine learning and to construct a predictive model. Methods The study inclu-ded 18 998 hospitalized patients who had used furosemide in our hospital from October 2017 to October 2020. The predictive model for evaluating furosemide-associated AKI and mortality risks was established using eight machine learning algorithms including Light Gradient Boosting Machine (LightGBM). The receiver operating characteristic (ROC) curve and decision curve analysis (DCA) were used to assess the discrimination, calibration, and clinical utility of the model. Feature importance analysis of the predictive model with the highest area under the curve (AUC) was conducted using SHAP summary plots, while SHAP force and decision plots were used to explain the individualized decision-making process for predicting AKI and mortality. Results Among the eight machine learning models, the LightGBM model exhibited an AUC of 0.814 for predicting AKI and 0.949 for predicting mortality risk, and the ROC and DCA curves confirmed its strong performance in terms of calibration and clinical application. SHAP summary plots revealed that after using furosemide, crucial factors affecting the occurrence of AKI included serum creatinine, se-rum cystatin C, urine microalbumin/creatinine ratio, and the concurrent use of adrenaline and heparin; meanwhile, the most significant factors leading to mortality in patients using furosemide encompassed 24-hour urinary protein quantification, plasma prothrombin time and international normalized ratio, plasma platelet count, the concurrent use of adrenaline, and the development of AKI du-ring hospitalization. Conclusion A predictive model has been established to assess the risk of AKI occurrence and mortality in hospitalized patients using furosemide by employing machine learning algorithms. This model effectively identifies the influencing factors in predicting the risk of AKI and mortality in these patients.

Keywords