What Other Factors Should be Considered in Addition to Accuracy When Evaluating a Contrast-Induced Acute Kidney Injury Prediction Model?

OBJECTIVES: This study aims to systematically evaluate the risk prediction models of contrast-induced acute kidney injury (CI-AKI) in patients undergoing coronary angiography (CAG) and percutaneous coronary intervention (PCI), and to provide reference for the construction and optimization of CI-AKI risk prediction models.

METHODS: Major English databases were searched systematically and references of included literatures were retrieved to obtain studies on CI-AKI risk prediction model in PCI/CAG patients. The performance and presentation of the model were summarized and analyzed. PROBAST was used to evaluate the risk of bias and applicability of the included models.

RESULTS: A total of 16 studies were included. According to the Predictive Model Risk of Bias Assessment Tool (PROBAST), all included studies had a high risk of bias. The area under the receiver operating characteristic curve (AUC) of 7 models ranged from 0.766 to 0.995. The predictive variables included in the model ranged from 3 to 20, and the top three predictors were age, estimated glomerular filtration rate (eGFR) and left ventricular ejection fraction (LVEF). Five studies reported the degree of calibration as a calibration curve, five studies reported the degree of calibration as a Hosmer-Lemeshow goodness-of-fit test, and four studies reported both the calibration curve and the Hosmer-Lemeshow goodness-of-fit test. In addition, one study validated the model externally, 10 studies validated the model internally only, and two studies used both internal and external validation of the model.

CONCLUSIONS: The current CI-AKI risk prediction models have good discrimination and considerable overall performance. Most of the models use internal validation to investigate the performance indicators of the models. In the future, it is necessary to improve the data modeling and statistical analysis methods, develop prediction models with good performance and low risk of bias, and focus on external validation and reciliation of the models.