Research Letter
Nephrology
March 5, 2025
Predicting Contrast-Associated Acute Kidney Injury
Yunlin Feng, Min Jun, Amanda Y. Wang, et al
JAMA Netw Open. 2025;8(3):e250107. doi:10.1001/jamanetworkopen.2025.0107
Introduction
Contrast-associated acute kidney injury (CA-AKI), defined as AKI after exposure to a diagnostic or therapeutic radio-contrast agent,1 is an important clinical problem.2 CA-AKI has been reported as the third leading cause of AKI among inpatient populations and is associated with short- and long-term adverse outcomes.3 Prediction models for CA-AKI increased rapidly in number in recent years, especially in the cardiology setting, to help identify patients at high risk. We conducted this review to summarize the available evidence on risk-prediction models for CA-AKI.
Methods
This systematic review and meta-analysis was performed using the PRISMA reporting guideline (eFigure in Supplement 1). As a meta-analysis, the research is exempt under 45 CFR §46.101(b)(4) from ethical review and informed consent. Critical appraisal was conducted using the PROBAST tool version 15/5/2019 (PROBAST Delphi group). Studies that developed a prediction model for CA-AKI that included at least 2 predictive variables and were published after the review by Silver et al4 were included. Discrimination data were pooled using summary receiver operating characteristic (sROC) curve analysis. Between-study heterogeneity was explored using subgroup analysis and meta-regression. Publication bias was assessed using funnel plot analysis. Methods details and extracted variables are presented in eTables 1 and 2 in Supplement 1. Statistical significance was set at a 2-tailed P < .05. Data analysis was performed using Stata version 14 MP (StataCorp) and R version 4.0.3 (R Project for Statistical Computing). Analyses were conducted between July 2023 and February 2024.
Results
Overall, 64 studies with 64 prediction models for CA-AKI were included, with 9 models (14.1%) rated as having low risk of bias. The 5 most used predictive variables were baseline kidney function, past medical history, age, coronary artery disease, and cardiac function (Table).
A total of 45 reviewed studies had requisite data to enable sROC curve analysis, resulting in a pooled C statistic of 0.83 (95% CI, 0.82-0.84) (Figure). The 95% confidence contour for the C statistic point estimate was smaller than that in Silver et al,4whereas the 95% prediction contour was little changed. Neither subgroup analysis nor metaregression identified a discrete source of the heterogeneity observed. Asymmetry was observed based on visual inspection of funnel plots.
Discussion
The number of CA-AKI prediction models has quadrupled since the 2015 review by Silver et al.4 This systematic review and meta-analysis found that despite a burgeoning literature and narrowing CIs around the point estimate of discrimination, substantial heterogeneity remained and there has been no meaningful improvement in the summary prediction estimate of model performance. Our approach of graphical presentation of sROC curves and reporting the breadth of the prediction interval better illustrates the heterogeneity and the resultant imprecision of any estimate of model performance.
It is important to bear in mind that the CI around the C statistic narrows as the total number of studies and participants increases given that it does not account for variations in study settings, patient characteristics, or methodologies. In contrast, the prediction interval does account for such variability in underlying studies and provides a more accurate representation of the imprecision in the estimate of model performance.5,6 Separating studies into 2 periods made it clear that while the CI for the C statistic was reduced with additional recent data, the prediction interval was not, suggesting that we are no closer to a recognized model for predicting CA-AKI that may have clinical or scientific utility.
Limitations include that predictive literature is dominated by retrospective studies, which are susceptible to selection, performance, and other biases. Performance bias is particularly challenging given that participants perceived to be at high risk of CA-AKI will often have the dose of contrast minimized or may be systematically excluded from datasets by never undergoing the procedure. A further challenge in interpreting this literature is the widespread use of modest changes in kidney function in the CA-AKI outcome definition, such as 25% increases in serum creatinine, which are more common but of dubious clinical significance.
