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 Table 1 – Clinical and laboratorial characteristics of the selected sample Variable N Sample Size 352 Age (years) 63 ± 14 Male Gender 210 (60%) ACS Unstable angina 102 (29%) NSTEMI 90 (26%) STEMI 160 (45%) Triarterial disease or LMCA 170 (48%) Ischemic ECG 223 (63%) Positive Troponin 250 (71%) Creatinine 1.0 ± 0.62 Killip Classification Killip I 308 (88%) Killip II 18 (5%) Killip III 25 (7%) Killip IV 1 (0.3%) NT-proBNP (pg/ml) 340 (86 – 1212) GRACE score 104 (82 – 131) Mortality 17 (4.8%) *ACS: acute coronary syndrome; NSTEMI: Non ST elevation myocardial infarction; STEMI: ST elevation myocardial infarction; LMCA: left main coronary artery; ECG: electrocardiogram; BNP: brain natriuretic peptide. NT-proBNP and Killip: Univariate Predictor Value NT-proBNP demonstrated a moderate predictive capacity for cardiovascular death, according to C-statistic of 0.78 (95% CI = 0.65-0.90, p < 0.001), while the Killip score presented C-statistic of 0.69 (95% CI = 0.54-0.84, p = 0.008), with no statistical difference between the two curves (p = 0.29) (Figure 1). The two markers agreed in the definition of heart failure in 75% of the cases (8% with heart failure and 67% without hear failure), meaning low level of agreement according to the Kappa test ( κ = 0.26; 95% CI = 0.54-0.84; p < 0.001). Independent and Incremental NT-pro-BNP Value In the logistic regression analysis, numerical NT-proBNP did not maintain statistical significance after adjustment for the traditional GRACE score (p = 0.11). On the other hand, numerical NT-proBNP remained an independent predictor when adjusted for GRACE score without Killip (p = 0.015; for each 500 pg/ml increase in NT-proBNP, a Beta of 0.029 was observed, OR = 1.03; 95% CI = 1.006 - 1.05) (Table 2). Categorical NT-proBNP was not an independent predictor after adjustment for the GRACE score (p = 0.91) or for the GRACE score without Killip (p = 0.36). For analysis of the incremental value of NT-proBNP to GRACE, we compared the C-statistics of the logistic GRACE‑BNP, categorical GRACE-BNP and traditional GRACE score. The results of the analysis were, respectively, 0.83 (95% CI = 0.69-0.97), 0.82 (95% CI = 0.68-0.96), and 0.82 (95% CI = 0.68-0.97). Therefore, no incremental value of the new approaches was identified (Figure 2). Reclassification of GRACE Score by NT-pro-BNP Regarding the net reclassification analysis, of the 17 patients who died, 3 were correctly reclassified by logistic GRACE‑BNP from low to high risk, with no incorrect reclassification, resulting in a positive net reclassification index (+ 0.18%). Among the 335 patients who survived, 9 were erroneously reclassified from low to high risk, while there was no correct reclassification. This resulted in a negative net reclassification ratio (-0.02%). In the final analysis, considering all patients, the total net reclassification index (NRI) was - 0.15% (p = 0.14) (Table 03). Reclassification based on categorical GRACE-BNP showed similar results (NRI = 0.08; p = 0.44). (Table 3) Discussion The present study demonstrates the independent prognostic value of numeric NT-proBNP after adjustment to the GRACE score. However, the NT-proBNP did not improve discrimination of the GRACE Score, nor its reclassification ability. Its findings are in line with the notions that not every independent predictor offers incremental value to traditional models. 17 In an explanatory point of view, our findings reinforce that the status of cardiac decompensation increases the risk of patients with ACS. On the other hand, from the predictive point of view, refining the prognostic evaluation with a biomarker of heart failure that is more accurate than clinical evaluation was not enough to increase the accuracy of multivariate models. This discussion is intended to debate the potential explanations for the absence of an incremental value, to confront this paper’s results with external evidence, to recognize methodological limitations and to address the relevance of the present results. Different hypotheses may explain the absence of NT‑proBNP incremental value. Three possibilities will be pointed out, which comprises the generic properties of predictors and the specificities of the clinical context in question. First, probabilistic models are created with variables that simultaneously contribute to risk prediction, each with a predictive weight that is proportional to its independent strength of association. The improvement of a single predictor (detection of ventricular dysfunction) among many may not represent a relevant change. In the present case, the incorporation of a marker related to a new phenomenon was not proposed, but rather only the replacement of the evaluation of the phenomenon of heart failure with a theoretically better marker. Second, the predictive capacity of NT-proBNP theoretically lies in its continuous characteristic (numerical variable) and in its ability to identify subclinical ventricular dysfunction. It is possible that the prognostic value of heart failure is not at initial levels, limiting to more advanced and clinically manifested degrees. Finally, the prognostic accuracy of the traditional GRACE Score is already satisfactory, represented by C-statistic above 0.8, making more difficult to improve a marker that functions with good predictive capacity. 668