External validation and modification of a simple prediction model for acute kidney injury after noncardiac surgeries

The Simple Postoperative AKI Risk (SPARK) index is a practical prediction model for postoperative acute kidney injury (PO-AKI) in patients after noncardiac surgery. However, previous external validation showed unsatisfactory performance, and its generalizability requires further evaluation.

This retrospective study included adults who underwent noncardiac surgery under general anesthesia from 2018 to 2020 at Peking University First Hospital. The discrimination and calibration were examined. We modified the SPARK index by refitting the model with original continuous laboratory data and evaluated the performance in our cohort.

A total of 8,386 participants were enrolled; 158 (1.9%) developed PO-AKI, and 46 (0.5%) developed critical AKI. Compared with those in the SPARK cohort, our patients were older (median [interquartile range] age, 61 years [49-69 years] vs. 56 years [44-66 years]) and had more dipstick albuminuria (14.3% vs. 9.2%). The incidence of PO-AKI and critical AKI increased in accordance with the SPARK score. The areas under the receiver operating characteristic curves (AUROCs) for PO-AKI and critical AKI were 0.77 (95% confidence interval [CI], 0.73-0.81) and 0.81 (95% CI, 0.75-0.88), respectively. The calibration plot suggested that the predicted probabilities of both PO-AKI and critical AKI were overestimated. The AUROCs of the modified model for PO-AKI and critical AKI in the validation dataset were 0.83 (95% CI, 0.75-0.90) and 0.86 (95% CI, 0.76-0.96), respectively. However, the improvement in discrimination was not significant compared to that of the original SPARK index.

The SPARK index may have the potential to predict PO-AKI in Chinese patients. Further validation and appropriate recalibrations are warranted.