Development and validation of AKI prediction model in postoperative critically ill patients: a multicenter cohort study.

ABSTRACT

BACKGROUND: Acute kidney injury (AKI) is a common complication, especially among postoperative critically ill patients. Early identification of AKI is essential for reducing mortality. METHODS: Multicenter data were used to develop an AKI prediction model for critically ill postoperative patients. A total of 1731 patients admitted to intensive care units (ICUs) were divided into a development set (n=1196) and a validation set (n=535) according to the principle of 73 randomization. Multivariate logistic regression analysis was performed on the predictors identified by univariate analysis, and a nomogram was created based on the predictors. The area under the receiver operating characteristic curve (AUROC) was used to assess the discrimination of the model. Calibration curves were generated, and the Hosmer-Lemeshow (HL) goodness of fit test was carried out. Decision curve analysis (DCA) was performed to assess the net clinical benefit. RESULTS: The final model included 7 predictors age, emergency surgery, abnormal basal creatinine level (BCr), chronic kidney disease (CKD), use of nephrotoxic drugs, diuretic use, and the Sequential Organ Failure Assessment (SOFA) score. A nomogram was drawn based on the predictors. The AUROC of the model in the development set was 0.725 (95% confidence interval (CI) 0.696-0.754). In the validation set, the AUROC was 0.706 (95% CI 0.656-0.744). The model showed good discrimination (>70%) in both sets, and the HL test indicated that the model fit was good (P>0.05). DCA showed that our model is clinically useful. CONCLUSION: The novel prediction model can be used to identify high-risk postoperative patients and provide a scientific and effective basis for clinicians to identify AKI early with a nomogram.