Contrast-induced acute kidney injury (CI-AKI), the third most common cause of hospital-acquired kidney injury, is associated with poor clinical outcomes, highlighting the need for accurate prediction. In this study, we used an electronic hospital monitoring system to retrospectively analyze 3,437 patients who underwent elective angiography at a large tertiary regional referral center in eastern China between 2019 and 2024, establishing a comprehensive epidemiological database. The system detected a CI-AKI incidence of 10.53% (362 cases) and revealed a striking under-diagnosis rate of 92.27% in discharge documentation. The key risk factors identified were leukocyte count, serum albumin level, and estimated glomerular filtration rate (eGFR). These predictors were used to develop and compare nine machine learning models against the conventional Mehran score. Although logistic regression (LR) showed the best overall performance, with an AUC of 0.806 and a Brier score of 0.076, the linear support vector machine (LSVM) also exhibited top-tier discriminative capability, achieving a comparable AUC of 0.807. Its slightly higher Brier score (0.124) suggests potential for improvement in calibration. Both machine learning models significantly outperformed the conventional Mehran score. This study demonstrates that electronic monitoring systems can help reduce missed diagnoses and that machine learning offers practical tools for early screening of high-risk patients and improved prognostic outcomes.
Zheng et al. (Tue,) studied this question.