ABC | Volume 114, Nº4, April 2020

Original Article Souza et al. NT-proBNP in acute coronary syndromes Arq Bras Cardiol. 2020; 114(4):666-672 Figure 1 – The accuracy of NT-pro-BNP in the prediction of death has a value of 0.78 (95%CI 0.65 – 0.9) in the C statistic and in the Killip classification 0.69 (95%CI 0.54 – 0.84). p = 0.29 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 Sensitivity 1 – Specificity NT-proBNP (0.78) Killip (0.69) Reference Line Table 2 – Logistic regression model containing GRACE without Killip and numerical NT-proBNP in predicting deaths Variables Beta Coefficient Odds ratio (95% IC) p Value GRACE without Killip 0.043 1.04 (1.02 – 1.06) <0.001 NT-proBNP / 500 pg/ml 0.000 1.03 (1.006 – 1.05) 0.015 Some previous studies have tested the prognostic value of brain natriuretic peptides in ACS. Although there is a disagreement between studies, a careful analysis of the results shows that they all point in the same direction. Three studies conclude positively regarding the prognostic value of this type of marker; however, these studies evaluated the independent predictive value but did not test incremental value (discrimination or reclassification). 18-22 In this context, our results are not discordant. However, our negative conclusion resides in a more comprehensive analysis that was not previously done. In concordance, the two studies that evaluated the incremental value of multivariate models presented the same conclusion as ours. 7,8 Two aspects are original in the present work: it was the first study to aggregate the analysis of reclassification proposed by Pencina and the only one to adjust for the GRACE score after removal of Killip, avoiding eventual collinearity between Killip and NT-proBNP which could induce type II error. These approaches bring more veracity to our negative outcome. Methodological limitations must be recognized here, which may have promoted a false negative result. Firstly, it is known that, ideally, a risk marker should be tested in an environment where the care team is not aware of its outcome. As this marker is already available in our clinical practice, the team became aware of the NT-proBNP result, predisposing to performance bias, which could improve the prognosis of patients with high NT-proBNP. Secondly, although this study had the planned sample size, it lacked additional power for exploratory analyses. For example, it was not possible to test the incremental value of the best NT-proBNP cutoff point. To do so, it would require a sample to identify the best cutoff point and another one to test for its incremental value. However, given our sample size, we chose not to split the sample. The value of a negative result should be contextualized. Frequently, improper evaluation of markers modifies clinical reasoning with no probabilistic basis. That is, after estimating the risk based on the GRACE score, our evaluation would become less accurate if we mentally increased the risk after observing a high NT-proBNP value. It would be an improper reclassification. Therefore, it must be considered that the GRACE score has better accuracy than NT-proBNP, which should not modify the message of the first. On the other hand, the absence of prognostic value should not discredit the value of BNP in diagnosing symptoms of dyspnea during hospitalization or in monitoring the volemic status of patients who developed acute heart failure. 669