Clinical prediction models in cardiac surgery-associated acute kidney injury: a systematic review and meta-analysis.

purposeCardiac surgery-associated acute kidney injury (CSA-AKI) is a frequent postoperative complication associated with substantial morbidity and mortality. Clinical prediction models (CPM) may identify high-risk patients enabling targeted preventive interventions and improved outcomes. The objective of the study was to identify available CPMs for the risk of developing CSA-AKI and summarize their performance.

methodsA systematic search of studies published in Pubmed, Scopus, the Chocrane Library and Web of Science was conducted, including studies that developed or validated CPMs for CSA-AKI in adult patients undergoing cardiac surgery with cardiopulmonary bypass. Data were extracted by two independent reviewers. Summary measures of diagnostic performance were calculated when feasible using random-effects models. Calibration and sensitivity analyses were also performed.

resultsThe pooled incidence of CSA-AKI was 26.3% (95% CI, 26.0-26.4), whereas CSA-AKI requiring renal replacement therapy (CSA-AKI RRT) occurred in 2.2% of patients (95% CI, 2.1-2.2). Among validated CPMs, Thakar 2005 demonstrated the best predictive performance for CSA-AKI RRT: pooled AUC 0.80 (95% CI, 0.77-0.82), sensitivity of 0.87, and a specificity of 0.49 at a threshold of ≥ 3 points. Calibration analyses indicated underestimation of risk. The Mehta 2006 and Wijeysundera 2007 models also demonstrated good discrimination. Discrimination for the prediction of milder stages of CSA-AKI was poorer, but newer models such as the Demirjian 2022 model showed promising performance.

conclusionAccurate and well-validated CPMs can facilitate risk stratification for CSA-AKI which can influence clinical practice and research. Thakar 2005 demonstrated very good discrimination for predicting CSA-AKI RRT, whereas more validation studies for the prediction of milder stages of CSA-AKI are warranted.