ABC | Volume 114, Nº5, May 2020

Original Article Torralba et al. HEART, TIMI, and GRACE scores for MACE prediction Arq Bras Cardiol. 2020; 114(5):795-802 The area under the curve for each test was calculated and compared using the nonparametric DeLong test (p = 0.05), and, finally, a calibration test was also made for each score to compare expected and actual major cardiovascular events in the study population, according to the calibration belt method described by Finazzi S, et al. 23 from the Italian Group for the Evaluation of Interventions in Intensive CareMedicine (GiViTi). 23 Analysis was carried out using the statistical program R, version 3.3.3 (the R Foundation for Statistical Computing, Vienna, Austria). Results Patients were recruited between August 2017 and February 2018. The present study’s patient flow is shown in Figure 1. A total of 519 patients were included in the analysis, with a follow-up period of 30 days. Baseline patient characteristics are shown in Table 1. MACE were confirmed in 224 patients within the first 30 days of follow-up, with a total of 351 events (AMI, revascularization, or death). These account for a MACE incidence of 43% and an average of 1.56MACE per patient with the primary outcome. NSTEMI was diagnosed in 194 patients. Of these patients, 108 underwent percutaneous revascularization; 46 underwent surgical revascularization, and 3 died. HEART, GRACE, and TIMI score comparison Risk stratification for each score is shown in Table 2. Based on the HEART score, patients in the low, intermediate, and high risk groups had 3.1%, 46.2%, and 93.7% incidence of MACE, respectively. The MACE rate in the low-risk group calculated according to the HEART score was lower than that of the low-risk groups calculated by the other two scores. A HEART score ≤ 3 had a sensitivity of 99.5% and a negative predictive value (NPV) of 99% to predict MACE in the low risk category (Table 3). Both parameters were higher than those obtained with the other scores for the low risk MACE group (TIMI: sensitivity 90%, NPV 89.9%; GRACE: sensitivity 70%, NPV 77.8%). The ROC curves for each score are shown in Figure 2. The C statistic for the HEART score was 0.937, which was higher than the other two scores, and a statistically significant difference was found using the nonparametric DeLong test (p < 0.0001). Finally, the GiViTi calibration belt test was used to compare expected and observed results (Figure 3), showing adequate calibration of the HEART score for patients with lowMACE risk. 20 Discussion We found that the HEART score for patients with chest pain is a reliable tool for predicting major cardiovascular outcomes based on the patients’ description of symptoms, clinical record data, electrocardiographic findings, and initial hsTnI value. It is readily applicable; it does not require computerized calculations, and it has been validated by international multicenter studies in multiple populations. 10,11,14,15 Conversely, the GRACE score is a model for predicting mortality in patients with acute coronary syndrome that has been adequately validated, but the fact that it must be calculated electronically limits its applicability. 17 Similarly, the TIMI score was designed to determine the need for aggressive therapy in patients with acute coronary syndrome, allowing the calculation of risk through the use of dichotomous variables without weighing the variables or taking patient’s clinical presentation into account. 16 The results of this study are favorable for the HEART score, with a C statistic value of 0.93, which indicates an excellent ability to predict the risk of patients with chest pain, compared to the TIMI and GRACE scores. This is consistent with what was previously reported by Six et al., 10 Sakamoto et al., 11 Backus et al., 15 and confirming that low scores on the HEART 24 scale are very accurate for ruling out the occurrence of MACE in low-risk patients with a 30-day follow up. 10,11,15,24 Our study had a MACE incidence of 43%, which is higher than the 13% and 36% reported in the literature. 11,15 This high rate of MACE might be due to the institution’s distinction as a referral center for cardiovascular disease, which leads to a higher than average number of patients with intermediate and high risks of coronary heart disease. Additionally, the exclusive use of hsTnI during this study might explain a higher rate of MACE detection than previously reported. However, despite higher rates of MACE, irrespective of risk status, the HEART score maintained its predictive precision, outperforming both the TIMI and GRACE scores. With regards to the sensitivity and the NPV of the tests, we found that the C statistic was higher for the HEART score when using a cutoff of 3 points, which is the limit for the low-risk category. Both the sensitivity and NPV are close to 100%, and they are significantly higher than the sensitivity and NPV of the other two scores. Based on these results it can be concluded that a HEART score below 3 identifies patients that can safely be managed with a conservative strategy with high certainty, given that the risk of adverse cardiovascular outcomes is low. Additionally, according to the GiViTi belt method, it is observed that there is adequate calibration between expected and observed outcomes for the low-risk group in the HEART score, as opposed to the low-risk groups in the two other scores analyzed. This supports the potential use of the HEART score as a first line score to stratify risk in patients with chest pain of suspected cardiac origin. Additionally, given its ease of application and adequate validation, it can be a valuable tool to enhance decision making and proper distribution of resources. This has been demonstrated by Mahler et al., 12 with the use of the “HEART pathway,” which combines the application of this score with troponin testing upon presentation and 3 hours later. This pathway led to a significant reduction of unnecessary tests and a shorter total hospital stay. 12,25 To the best of our knowledge, no other studies have reported the performance of risk scores conducted in the era of hsTnI in an exclusively Latin-American population. The prospective nature of the study strengthens the findings. Therefore, these results serve as a validation of previous findings regarding the HEART score, and they should motivate further multicentric projects with larger populations. Also, we believe that these results should expand the use of the HEART score as a valuable tool that aims to facilitate decision making in a challenging patient population. 797