Observational retrospective single-centre cohort study including all patients admitted to our hospital’s Acute Coronary Care Unit (ACCU) diagnosed with Myocardial Infarction between December 1, 2006 and September 30, 2011. Using collected baseline data at the time of MI diagnosis and outcome data from this cohort, we tested the prognostic performance of four different risk stratification schemes - the GRACE model, the CHADS2 and CHA2DS2-VASc scores and the ACHTUNG-Rule - in patients with AF at admission or occurring during hospitalization and compared it with their efficacy in those without AF. All prediction models were evaluated for their overall discriminative performance, accuracy and calibration in the prediction of short- to mid-term all-cause mortality. Furthermore, all three versions of the ACHTUNG-Rule¹⁰ were directly compared to their equivalent GRACE score versions, and a new score, entitled GRACE-CHA2DS2-VASc, was developed and compared with GRACE.

A total of 1852 patients (age 68.3±13.5, from 29 to 99) were consecutively admitted to our ACCU diagnosed with MI according to its Universal Definition.¹⁴ This sample included both the derivation and validation cohorts of the ACHTUNG-Rule,¹⁰ plus 401 additional and consecutively admitted patients. AF was detected in 294 patients (15.9%) either at admission or during hospitalization.Table 1 describes the study sample.

Through extensive review done by 4 co-investigators blind to the purpose of the study, the following data were collected: demographic features, cardiovascular risk factors and previous medical history (including history of AF), physical examination and analytical study at admission (including complete blood count, biochemical and clotting tests), angiographic data and results of electrocardiograms performed during hospitalization. Glomerular filtration rate (GFR - by MDRD formula), the GRACE scores for intrahospital, 6-month post-admission and 6-month post-discharge mortality, the CHADS2 and CHA2DS2-VASc scores and the ACHTUNG versions for intrahospital, post-discharge and total all-cause mortality were calculated for all patients. The presence of AF was defined as the electrocardiographic documentation of at least one episode of this arrhythmia (through 12-lead electrocardiogram, 24-hour Holter or electrocardiographic monitoring in the ACCU; in all cases, the presence of AF must have been validated by at least one Cardiologist), irrespective of its timing, duration (paroxysmal, persistent or permanent) or overall patient characteristics (valvular or nonvalvular AF) and symptomatology. To collect this information, a comprehensive review of available electrocardiographic recordings and daily medical history notes was made.

The primary endpoints of this study were: a) total all-cause mortality, comprising intrahospital and post-discharge all-cause mortality; b) intrahospital all-cause mortality and c) all-cause mortality during follow-up. The specific cause of mortality in patients discharged following hospitalization for a MI is sometimes very hard to ascertain, and thus all-cause mortality is likely to be the most robust and objective endpoint. The secondary outcome was the occurrence of a primary International Classification of Diseases diagnosis of stroke, confirmed through cerebral computed tomography (CT).

This information was collected from hospital charts and clinical records from outpatient clinic and hospital ward and emergency department admission(s), including the reports of performed cerebral CT, and through proxy interviews when appropriate.

Patients were followed for 17.4±8.7 months following their discharge. Follow-up data was obtained through review of clinical records from outpatient clinic and hospital ward and emergency department admission(s), and through phone calls by the end of a 2-year period after discharge for patients not followed at our institution.

Statistical analysis was done using SPSS, v.17.0. When needed, baseline characteristics are described with mean ± standard deviation for continuous data and counts and proportions for categorical data. The Kolmogorov-Smirnov test was used to test the normal distribution of continuous variables. The Chi-square test, Student’s t-test and non-parametric equivalent tests were used when appropriate. Regression estimation techniques were applied to replace missing values whenever the number of missing values was negligible, otherwise cases with missing values would be omitted. P values < 0.05 (two-sided) were considered statistically significant.

Univariate analysis was performed to evaluate a potential association between AF at admission or any time during hospitalization and the study endpoints.

The three versions of the ACHTUNG-Rule were calculated according to their respective coefficients obtained from its original derivation cohort and described elsewhere.¹⁰

Discrimination is usually measured in terms of the area under each receiver operating characteristic curve (AUC) and refers to the ability of a prediction model to assign a higher probability to patients reaching the study endpoint than to those not reaching it.Through the calculation of the AUC, an assessment of the discriminatory power of the GRACE score versions for intrahospital (GRACE-IH), 6-month post-admission (GRACE-6PA) and 6-month post-discharge (GRACE-6PD) all-cause mortality, the CHADS2 and CHA2DS2-VASc scores and the ACHTUNG versions for intrahospital (ACHTUNG-IH), post-discharge (ACHTUNG-R) and total (ACHTUNG-T) all-cause mortality was performed.

Intrahospital mortality: assessment of the AUC of GRACE-IH, CHADS2, CHA2DS2-VASc and ACHTUNG-IH.

Mortality during follow-up: assessment of the AUC of GRACE-6PD, CHADS2, CHA2DS2-VASc and ACHTUNG-R.

Total mortality (intrahospital mortality plus mortality during follow-up): evaluation of the AUC of GRACE-6PA, CHADS2, CHA2DS2-VASc and ACHTUNG-T.

Stroke: CHADS2 and CHA2DS2-VASc.

Accuracy and calibration of each score were also assessed through the Brier score and the Hosmer-Lemeshow test, respectively. Accuracy is a measure of the average distance (residual) between the observed outcome and its predicted probability for each individual patient. A popular accuracy measure is the Brier score, which is the squared mean of the residual values¹⁵ The Brier score is sensitive to both discrimination and calibration of the predicted probabilities and describes how well a particular model predicts the likelihood of an outcome in an individual patient [a score of 0.0 implies perfect prediction, while a Brier score of 0.25 or higher suggests lack of utility in endpoint prediction].

A comparison through ROC curve analysis and the integrated discrimination improvement index (IDI) was performed between each of the GRACE score versions and their equivalent ACHTUNG-Rule versions. The IDI, which may be seen as a continuous form of the net reclassification improvement index, assesses improvement in risk discrimination by estimating the change in the difference in the mean predicted probabilities of the outcome between those with and without the outcome in question.¹⁶ AUC comparisons were performed using MedCalc for Windows version 9.2.0.1. The overall tendency of increasing total mortality event rates with increasing GRACE-6PA or ACHTUNG-T risk scores was tested using chi-square for trend (gamma). ACHTUNG’s risk strata were defined as described in ACHTUNG-Rule’s original paper.¹⁰

Finally, binary logistic regression was performed to analyse whether the CHADS2 or CHA2DS 2-VASc scores would improve GRACE’s ability to predict total or post-discharge all-cause mortality. A new score, entitled GRACE-CHA2DS2-VASc, was evaluated for its discriminative performance, accuracy and calibration in both the overall cohort and patients with AF. Intrahospital and total mortality rates according to both GRACE and CHA2DS2-VASc risk strata were also reported aiming to identify truly low- or high-risk sub-groups of patients. GRACE’s risk strata were defined according to published criteria.¹⁷ Kaplan-Meier curves were constructed to evaluate survival during follow-up according to combined GRACE and CHA2DS2-VASc risk strata: a CHA2DS2-VASc score ≤ 1 was defined as low risk and a score ≥ 2 as high risk.

This study complies with the Declaration of Helsinki and has been approved by our institution’s review board.

In univariate analysis, the occurrence of AF at admission or during hospitalization associated with higher risk for intrahospital mortality (11.6% vs. 6.7%, p=0.004, OR 1.81, 95% CI 1.20-2.73), mortality during follow-up (28.5% vs. 12.2%, p<0.001, OR 2.88, 95% CI 2.1-4.0), total mortality (37.1% vs. 18.3%, p<0.001, OR 2.63, 95% CI 2.0-3.46) and non-fatal stroke (7.2% vs. 3.4%, p=0.005, OR 2.24, 95% CI 1.25-3.98). In multivariate analysis, the presence/absence of AF did not improve the ability of GRACE-IH in predicting intrahospital mortality, but AF predicted mortality during follow-up independently of the GRACE-6PD score. A multivariate predictive model for post-discharge mortality included variables GRACE-6PD (p<0.001, OR 1.0123, 95% CI 1.018-1.028) and AF (p=0.001, OR 1.84, 95% CI 1.30-2.59).

See Table 2.

See Table 3.

Compared to their equivalent GRACE score versions, the ACHTUNG-Rule versions have shown non significantly lower discriminative performance when predicting intrahospital mortality risk (0.741±0.044 vs. 0.788±0.042, p=0.289), a trend for higher discriminative power in the prediction of post-discharge mortality (0.790±0.032 vs. 0.685±0.038, p=0.079) and non significantly higher ability to predict total mortality (0.762±0.031 vs. 0.712±0.033, p=0.144).

The IDI index provided a more rigorous and powerful statistical approach to assess the potential improvement in risk reclassification with the ACHTUNG-Rule. For post-discharge mortality, the IDI and relative IDI were 0.166 and 83.7%, respectively, translating a very sizeable improvement in risk reclassification with the use of ACHTUNG-R instead of GRACE-6PD. The IDI index for total mortality prediction was 0.042 (relative IDI of 11.7%), suggesting an improvement in risk stratification with ACHTUNG-T in detriment of GRACE-6PA.

Table 4 reports the overall tendency of increasing mortality event rates with increasing GRACE-6PD or ACHTUNG-R risk scores, tested using chi-square for trend (gamma).

Multivariate analysis by binary logistic regression (method forward conditional) did not include the CHADS2 or CHA2DS2-VASc scores along with GRACE in a prediction model for intrahospital mortality in patients with AF. However, the post-discharge prediction model included GRACE-6PD (p<0.001, OR 1.022, 95% CI 1.015-1.030) and CHA2DS2-VASc (p<0.001, OR 1.631, 95% CI 1.46-1.822), while the multivariate model for total all-cause mortality prediction included GRACE-6PA (p<0.001, OR 1.017, 95% CI 1.012-1.023) and CHA2DS2-VASc (p<0.001, OR 1.411, 95% CI 1.261-1.578).

GRACE-CHA2DS2-VASc performance in patients without AF is reported in Table 5, while the ability of GRACE-CHA2DS2-VASc to predict all-cause mortality in patients with AF is described in Table 6. Intrahospital and total mortality rates according to GRACE and CHA2DS2VASc risk strata are reported in Table 7 and Kaplan-Meier curves in Figure 1 illustrate survival during follow-up (including intrahospital events). It must be noted that no deaths were observed in patients with AF and a CHA2DS2-VASc score <2. Additionally, high-risk patients according to GRACE and with a CHA2DS2-VASc score >4 had a very high total all-cause mortality rate (39.6%).

Figure 1. Kaplan-Meier curves illustrating cumulative survival according to GRACE and CHA2DS2-VASc risk strata.

---

---

The relatively small number of patients with AF (294) represents the main limitation of this study. Our findings should be validated in larger cohorts of patients admitted for MI and presenting AF at admission or during hospitalization, preferably involving multicentre and/or prospective registries.

The difference between de novo AF occurring during hospitalization or previously known permanent AF has not been addressed. Although this study was not performed to evaluate the prognostic impact of AF, it should be noted that some studies have shown that the development of AF during hospitalization associates with higher mortality rate, while those who are in AF at the time of admission have a mortality rate that is not significantly different from that of patients in sinus rhythm.^33,34 The latter is presumably a reflection of the difference between persistent/chronic AF documented at admission and de novo AF as a manifestation of acute haemodynamic compromise in the acute MI setting. Despite the fact that some studies have not confirmed these data,^21,35 the prognostic performance of GRACE, ACHTUNG, CHADS2 and CHA2DS2-VASc might have been different in these two clinical contexts (chronic AF vs. de novo AF). Future studies should address this subject.

As many patients were not submitted to autopsy, the cause of death has not been systematically evaluated. It would have been worthwhile assessing whether GRACE’s apparent limitation in the prognostication of these patients is due to an excess of fatal thromboembolic events. Such data could help explain our results.

Our study sample included the 1051 patients assigned to ACHTUNG-Rule’s derivation cohort, the 400 assigned to the validation sample and 401 new patients. Therefore, comparisons between the GRACE score and the ACHTUNG-Rule may have been biased towards the latter. A more reliable comparison between these two risk stratification schemes might be obtained in a larger, independent, patient sample. The potential validation of the ACHTUNG-Rule in multicentre registries may eventually confirm the attributed versatility that its original research has suggested.

Patients with AF and a CHA2DS2-VASc score ≤ 1 had a very low mortality risk irrespective of their GRACE score. However, very few AF patients had a CHA2DS2-VASc score ≤ 1. Although similar results were found in the sub-group without AF (comprising 213 [13.7%] individuals with a CHA2DS2-VASc score ≤ 1) this analysis is far from being conclusive and should be validated in bigger cohorts considering its potential clinical value.

Even if our findings are validated in future studies, the optimal way to stratify the risk of these patients and its potential clinical impact remains to be determined. A new score should have a good balance between complexity, applicability and prognostic discriminatory performance. Most importantly, it should lead to improved risk reclassification and impact on therapeutic decisions. Further studies will be necessary to evaluate whether the GRACE-CHA2DS2-VASc score or the ACHTUNG-Rule may help lower mid-term mortality of patients with both an MI and AF through a more reliable identification of individuals eligible for aggressive therapies and those who should be treated rather more conservatively.