Renal Risk Prediction in Cardiac Surgery using traditional Methods of Acute Kidney Injury prediction: A Systematic Review

Abstract

Objectives Acute Kidney Injury following Cardiac Surgery (CS-AKI) remains a significant cause of morbidity and mortality. To assist early recognition, risk prediction models have been developed over the last two decades. This review evaluates the current body of evidence for non-machine learning renal risk prediction models. Methodology A systematic review of 4 databases was undertaken according to PRISMA guidelines. Included studies were those that had derived and validated a renal risk prediction model in cardiac surgery patients. Machine learning models were excluded. Outcomes measured were pre, intra or post-operative variable use within the models and metrics for prediction. The PROBAST was used to evaluate for risk of bias. Results 44 studies were finally selected in a pooled population of 907,993. 24 developed renal risk prediction models whilst 31 externally validated these. When externally validated In Caucasian populations (N= 19), pre-operative prediction models offered reliable prediction for dialysis and severe AKI with area under the receiver operating curve (AUC) between 0.7-0.93. Models using intra-operative or post-operative variables (N=14) had acceptable prediction of severe stage AKI and dialysis with AUCs between 0.7-0.81. Pre-operative predictor models were the most externally validated. AKI prediction worsened with reducing severity of AKI. Validation of non-Caucasian populations was limited with 11 (25%) of studies undertaken in this group. Conclusion Pre-operative renal risk prediction models offer the most validated, accurate prediction for dialysis and severe CS-AKI. External validation of prediction tools for less-severe AKI and in non-Caucasian populations is required.