LAUSR

We read with interest the recent article by Yoneyama et al. [1] determining two novel urinary biomarkers for acute kidney injury (AKI) and prediction of clinical outcomes following pediatric cardiac surgery. However, we noted two important issues in this study that were not well addressed. First, when the logistic regression analysis was used to identify the risk factors for AKI, a simple logistic regression analysis was performed to evaluate the variables associated with the occurrence of AKI and then a multivariate logistic regression analysis was performed with independent variables showing statistical significance (P < 0.05) in initial univariate analysis. This statistical process is questionable. Actually, all variables identified as significant (P < 0.2) in the univariate model must be subsequently included in the multivariate model for statistical adjustment [3]. As multicollinearity among candidate independent variables was not correctly applied in this study, we were concerned that results of multivariate logistic regression analysis would have been biased. Thus, some important known risk factors associated with the occurrence of AKI following pediatric cardiac surgery, such as age, RACHS score, duration of cardiopulmonary bypass, and intraoperative blood transfusion [2], were not identified. Second, to determine performance of L-FABP and NGAL for prediction of AKI and postoperative adverse outcomes, the authors directly did the receiver-operating characteristic (ROC) curve analysis and provided their areas under the ROC curve, cutoff values, sensitivities and specificities for various postoperative adverse outcomes. To determine if a biomarker is predictive for postoperative adverse outcome, however, significant association of a biomarker with postoperative adverse outcome must be firstly obtained by multivariate model for adjustment of confounders before performing the ROC curve analysis. Other than the area under ROC curve, cutoff value, sensitivity and specificity, moreover, the ROC curve analysis can also provide positive and negative predictive values, and Youden indexes at the cutoff value of a biomarker for postoperative adverse outcome. It must be emphasized that Youden index is important as it is a direct measure of diagnostic accuracy at the cutoff value, i.e., maximal overall correct classification rate that a biomarker can achieve [4]. A biomarker with a large area under ROC curve may have an unsatisfactory overall correct classification rate at the cutoff value, and vice versa. As the associations of L-FABP and NGAL with postoperative adverse outcomes were not determined, and their positive and negative predictive values, and Youden indexes at the cutoff values were not provided in this study, we cannot determine whether they are valuable predictors for postoperative adverse outcomes after pediatric cardiac surgery.