External validation of predictive models for acute kidney injury following cardiac surgery: A prospective multicentre cohort study.

BACKGROUND: Four predictive models for acute kidney injury associated with cardiac surgery were developed by Demirjian in the United States in 2012. However, the usefulness of these models in clinical practice needs to be established in different populations independent of that used to develop the models.

OBJECTIVES: Our aim was to evaluate the predictive performance of these models in a Spanish population.

DESIGN: A multicentre, prospective observational study.

DATA SOURCES: Twenty-three Spanish hospitals in 2012 and 2013.

ELIGIBILITY CRITERIA: Of 1067 consecutive cardiac patients recruited for the study, 1014 patients remained suitable for the final analysis.

MAIN OUTCOME MEASURES: Dialysis therapy, and a composite outcome of either a doubling of the serum creatinine level or dialysis therapy, in the 2 weeks (or until discharge, if sooner) after cardiac surgery.

RESULTS: Of the 1014 patients analysed, 34 (3.4%) required dialysis and 95 (9.4%) had either dialysis or doubled their serum creatinine level. The areas under the receiver operating characteristic curves of the two predictive models for dialysis therapy, which include either presurgical variables only, or combined presurgical and intrasurgical variables, were 0.79 and 0.80, respectively. The model for the composite endpoint that combined presurgical and intrasurgical variables showed better discriminatory ability than the model that included only presurgical variables: the areas under the receiver operating characteristic curves were 0.76 and 0.70, respectively. All four models lacked calibration for their respective outcomes in our Spanish population.

CONCLUSION: Overall, the lack of calibration of these models and the difficulty in using the models clinically because of the large number of variables limit their applicability.