Derivation and validation of plasma endostatin for predicting renal recovery from acute kidney injury: a prospective validation study

Acute kidney injury (AKI) is associated with high morbidity and mortality in surgical patients. Nonrecovery from AKI may increase mortality and early risk stratification seems key to improving clinical outcomes. The

Plasma endostatin shows a useful value for predicting failure to recover from AKI. The predictive ability can be greatly improved when endostatin is combined with the SOFA score and AKI classification.

We conducted a prospective cohort study of 198 patients without known chronic kidney disease who underwent noncardiac major surgery and developed new-onset AKI in the first 48 h after admission to the ICU. The biomarkers of plasma endostatin, neutrophil gelatinase-associated lipocalin (NGAL) and cystatin C were detected immediately after AKI diagnosis. The primary endpoint was nonrecovery from AKI (within 7 days). Cutoff values of the biomarkers for predicting nonrecovery were determined in a derivation cohort (105 AKI patients). Predictive accuracy was then analyzed in a validation cohort (93 AKI patients).

Seventy-six of 198 (38.4%) patients failed to recover from AKI onset, with 41 in the derivation cohort and 35 in the validation cohort. Compared with NGAL and cystatin C, endostatin showed a better prediction for nonrecovery, with an area under the receiver operating characteristic curve (AUC) of 0.776 (95% confidence interval (CI) 0.654-0.892, p < 0.001) and an optimal cutoff value of 63.7 ng/ml. The predictive ability for nonrecovery was greatly improved by the prediction model combining endostatin with clinical risk factors of Sequential Organ Failure Assessment (SOFA) score and AKI classification, with an AUC of 0.887 (95% CI 0.766-0.958, p < 0.001). The value of the endostatin-clinical risk prediction model was superior to the NGAL-clinical risk and cystatin C-clinical risk prediction models in predicting failure to recover from AKI, which was supported by net reclassification improvement and integrated discrimination improvement. Further, the endostatin-clinical risk prediction model achieved sensitivity and specificity of 94.6% (76.8-99.1) and 72.7% (57.2-85.0), respectively, when validated in the validation cohort.