Prognostic Significance of the Initial Electrocardiogram in Patients With Acute Myocardial Infarction FREE

THROMBOLYTIC THERAPY and direct angioplasty have significantly advanced the treatment of acute myocardial infarction. Nonetheless, 30-day mortality has ranged from 6% to 9% in recent clinical trials, and it is even higher for those not enrolled.¹ Risk assessment during the initial patient evaluation is critical to facilitate optimal treatment and the appropriate intensity of monitoring. The initial clinical variables that most influence the 30-day prognosis after acute infarction include age, systolic blood pressure, Killip class, and heart rate.²

The electrocardiogram (ECG), a critical component of the early assessment and risk stratification of patients presenting with acute infarction, contains valuable diagnostic and prognostic information. Despite the central role of the ECG in patient evaluation, large studies have provided only limited qualitative information about ECG variables most critical to the assessment of prognosis. The aim of the present study was to determine the initial ECG predictors of 30-day mortality after thrombolysis for acute myocardial infarction and to assess their incremental value when combined with known initial clinical prognostic indicators.

Te Global Utilization of Streptokinase and t-PA (alteplase) for Occluded Coronary Arteries (GUSTO-I) trial enrolled 41021 patients with acute myocardial infarction from December 1990 to February 1993 from 1081 hospitals in 15 countries. Patients presented after 20 minutes but within 6 hours of the onset of chest pain accompanied by ECG signs of 0.1 mV or greater ST-segment elevation in 2 or more limb leads or 0.2 mV or greater elevation in 2 or more contiguous precordial leads. The full design and data collection methods of GUSTO-I have been described.³ Patients were randomized by telephone, with selected initial characteristics recorded to ensure eligibility. Exclusion criteria included a history of stroke, active or recent bleeding or major coagulation abnormality, recent trauma or major surgery, noncompressible vascular punctures, and previous treatment with streptokinase or anistreplase. There were no restrictions because of age, presentation in cardiogenic shock, or prior bypass surgery or infarction.

Qualifying patients were randomly allocated to 1 of 4 treatment strategies: streptokinase, 1.5 million U over 1 hour, with subcutaneous heparin, 12500 U twice daily, beginning 4 hours after the start of thrombolytic therapy; streptokinase, 1.5 million U over 1 hour, with intravenous heparin in a bolus dose of 5000 U then 1000 U per hour, with the infusion adjusted to maintain an activated partial thromboplastin time of 60 to 85 seconds; accelerated alteplase in a bolus dose of 15 mg, then an infusion of 0.75 mg/kg (up to 50 mg) over 30 minutes and 0.5 mg/kg (up to 35 mg) over the next hour, with the same intravenous heparin regimen; or combined intravenous alteplase (1.0 mg/kg over 1 hour, up to 90 mg, with 10% given as a bolus) and streptokinase (1.0 million U over 1 hour), given concurrently but through separate catheters, with the same intravenous heparin regimen.

Initial clinical data were collected on all patients with a standard data collection form. Definitions of the clinical variables in this trial have been published.^{3 - 4} Extensive quality control checks were used during data entry, and missing or questionable answers were queried. A sample of 12% of the forms was audited by comparing them with the hospital medical record.

The initial ECG was the one that fulfilled enrollment criteria and resulted in randomization. First ECG tracings obtained after the start of thrombolytic therapy were not considered to be initial tracings and resulted in patient exclusion from this analysis. Patients who lacked an initial ECG or 30-day mortality data were also excluded from analyses, as were those who failed to meet entry criteria for ST-segment deviation and those who showed confounding factors on the initial ECG (such as paced rhythms, ventricular rhythms, or left bundle-branch block). All ECGs were analyzed by a central core laboratory (Duke Clinical Research Institute, Durham, NC), whose personnel were blinded to treatment assignment and patient information. The ECG variables included heart rate per minute; heart rhythm (atrial flutter, atrial fibrillation); conduction intervals (QRS interval in milliseconds); ST-segment measures (elevation and depression for each lead in millimeters and the sum of each variable for all leads, including the inferior leads [leads II, III, and aVF] and the precordial leads [leads V₁ to V₄]; sum of absolute deviation [elevation or depression] for all leads that measured 60 milliseconds after the J point relative to the TP segment; maximum elevation in a single lead; number of leads with ST-segment elevation ≥1 mm); conduction disturbances (first-degree or second-degree atrioventricular block, right bundle-branch block, left anterior hemiblock, left posterior hemiblock, complete heart block); location of infarction (anterior, inferior, or other); left ventricular hypertrophy (Estes-Romholt or Sokolow-Lyon criteria); and pathological Q waves (considered present if they measured ≥20 milliseconds in lead V₄; ≥30 milliseconds in leads I, II, aVL, V₅, or V₆; or were present at any width in leads V₁ to V₃).^{5 - 6}

The primary end point of the trial was death from any cause within 30 days of randomization. The study coordinator at each site collected mortality information on the main case report form for patients who died in the hospital. Survival status after discharge but within 30 days was obtained by a postcard returned by patients or their families. When no postcard was received, follow-up status was determined by telephone. Thirty-day mortality status was known for 40830 patients (99.5%) overall and for all 34166 patients in the substudy.

All categorical characteristics are described as percentages, and continuous measures are summarized with medians and interquartile ranges. The univariable relation of each factor to 30-day mortality was tested with the log-likelihood ratio χ² test. Logistic regression modeling techniques were used to evaluate the univariable relations of continuous factors to 30-day mortality and to estimate the joint relations of the ECG factors and the joint relations of initial clinical² and ECG factors to 30-day mortality.

Several continuous variables were found to deviate from the key modeling assumption of a linear relation between the variable and the logit of the probability of death (Figure 1). By comparing the model χ² values and graphical relations of the transformed and linear variables to outcome, appropriate spline transformations of the variables were made.⁷ Linear splines were determined to best fit QRS duration, number of leads with ST-segment elevation, sum of ST-segment depression, heart rate on the ECG, maximum ST-segment elevation in any lead, and sum of ST-segment elevation; linear splines were used in all modeling procedures.

A stepwise variable selection technique was used to choose the best multivariable model of initial ECG predictors of mortality. Variables remained in the model if they contributed significant (P<.05) additive or multiplicative prognostic information. The incremental value of the ECG predictors above that of known clinical predictors of mortality² was determined with a logistic regression model. Wald χ² values are reported for the importance of each variable after adjusting for all other factors in the model. The quality of the model was evaluated by calculating the area under the receiver operating characteristic (ROC) curve, also known as the C-index.⁸ The model was then validated internally by means of the bootstrap method,⁹ using 40 replications.

Of the 41021 patients in the GUSTO-I trial, 3187 patients failed to meet the entry ST-segment deviation criteria (maximum ST-segment elevation <0.1 mV) or had confounding ECG factors on the initial tracing (paced rhythm, left bundle-branch block, missing lead data). An additional 3526 were excluded for lack of an initial ECG being forwarded to the core laboratory, and 142 patients were excluded for lack of 30-day mortality data, leaving 34166 patients available for analysis. The clinical characteristics of this substudy population were similar to those of the entire GUSTO-I study population³ and of patients not in the substudy (Table 1). Compared with those excluded from analysis, the substudy population had a slightly higher proportion of patients who had anterior infarctions. Most patients were male and white, and 16% had experienced a prior infarction. Patients who were excluded from the ECG substudy had a 21% higher mortality rate at 30 days than those included in the substudy (8.2% vs 6.8%).

Of the categorical initial ECG variables tested, only left ventricular hypertrophy, first-degree or second-degree atrioventricular block, and complete heart block were not significant univariable predictors of 30-day mortality (Table 2). All the continuous ECG variables showed significant univariable relations with 30-day mortality (Figure 1). Both heart rate and QRS duration were characterized by U-shaped curves, with increased mortality at the extremes of the values recorded.

In a multivariable analysis of the ECG variables (Figure 2), faster heart rate, longer QRS duration, ECG evidence of prior infarction, and sum of the absolute ST-segment deviation were most highly associated with 30-day mortality. Other measures of ST-segment deviation retained significance, including the absolute value of ST-segment elevation in the inferior leads and the sum of ST-segment depression after adjustment for other ECG measures. Two interaction terms were found to enter this model. One indicated that increases in QRS duration were associated with a much greater increase in risk with an anterior infarction than with an infarction in another location. The other showed that patients with an inferior infarction who had had a prior infarction had a higher risk of death than patients with other locations of infarctions who had had a prior infarction, but inferior infarction carried a lower risk of death than other infarct locations in the absence of previous infarction.

In a multivariable model that combined the significant initial ECG and clinical predictors of mortality, the sum of absolute ST-segment deviation, QRS duration, and ECG evidence of prior infarction remained the strongest of the ECG predictors (Table 3 and Figure 3). Overall, the clinical variables age, systolic blood pressure, and Killip class were the strongest independent predictors of 30-day mortality.

The model provided excellent discrimination between patients who died and those who lived (C-index, 0.836). After correction for the bootstrap technique, there was only a minimal alteration in this result (C-index, 0.833). Finally, a nomogram created from the full model (Figure 3), which contained a reduced set of variables, also provided excellent discrimination (C-index, 0.830).

In this large population of patients treated with thrombolytic therapy, the initial ECG contained important prognostic information beyond that provided by the medical history and physical examination. The most powerful independent ECG predictor of 30-day mortality was heart rate, consistent with studies that have shown sinus tachycardia to be correlated with adverse outcomes.^{10 - 11} The U-shaped nature of the curve indicates that bradycardia is also associated with increased risk, perhaps representing patients with large infarctions and various degrees of sinus and atrioventricular nodal suppression. In the combined clinical and ECG model, we used 2 measures of heart rate—that recorded on the initial ECG and that recorded on the case report form by the nurse or study coordinator just before thrombolytic therapy began. To further investigate this, we plotted the difference between these 2 values against 30-day mortality and found that patients with the greatest difference had the highest mortality. Thus, a change in heart rate (increase or decrease) between the recording of the initial ECG and thrombolysis was accompanied by increased mortality.

A variety of ST-segment indexes of infarct size or "myocardium at risk" were investigated, and, not surprisingly, all showed univariable correlations with mortality. The strongest of these was the sum of the absolute ST-segment deviation in all leads (limb and precordial), which also added independent information to the clinical model. Many investigations, in both the prethrombolytic and thrombolytic eras, have shown a correlation between degree of ST-segment elevation and final infarct size for both anterior and inferior infarctions.^{12 - 14} The prognostic information contained in the degree of elevation was not supplanted by information in the other ECG variables, including infarct location, heart rate, and QRS duration, or in the clinical variables of Killip class and systolic blood pressure, all of which likely are indirect measures of infarct size.

Other investigators, as in this study, have found that mortality increases progressively with the number of leads that show ST-segment elevation.¹⁵ Although our investigation found this to be a strong univariable predictor, the number of leads with ST-segment elevation provided no independent prognostic information in either of our models.

The prognostic significance and physiological implications of associated ST-segment depression have been topics of considerable debate, particularly for patients with inferior infarction.^{2 ,12 - 13 ,16 - 19} In our study, the sum of ST-segment depression added major information to the combined model, a concept that is perhaps not as well recognized by clinicians as the importance of ST-segment elevation.

In this investigation, right bundle-branch block and both left anterior and left posterior hemiblock showed univariable, but not multivariable, correlations with mortality. This can be explained by the fact that QRS duration, which retained independent significance, has correlated with mortality in other studies of postinfarction patients. Although not in the setting of acute infarction, the Cardiac Arrhythmia Suppression Trial investigators, examining placebo-treated patients only, reported that an initial QRS duration of 100 milliseconds or longer on the surface 12-lead ECG imparted a hazard ratio of 1.4 for new or worsening congestive heart failure, for arrhythmic death or cardiac arrest, and for all-cause mortality.²⁰ The association of longer QRS duration with increased mortality in our analysis may indicate that when "late potentials" are of sufficient magnitude to affect the surface ECG, they reflect extensive infarcts and involvement of the interventricular conduction system. Prolonged QRS duration, therefore, might indicate slowed and heterogeneous conduction in a large mass of ventricular myocardium with increases in both the potential for heart failure and vulnerability to re-entrant ventricular dysrhythmias.

Although patients with inferior infarction had better survival than the rest of the study population, the subpopulation that had inferior infarction and ECG evidence of previous infarction had a higher risk of death. Previous infarction was classified on the basis of "distant Q waves" (Q waves in leads other than those related to the new infarction). Thus, patients with inferior infarction by definition must have had previous anterior or lateral events; conversely, those with acute infarction in the anterior leads must have had previous inferior or lateral infarction. Because this interaction term would identify patients with multivessel disease and poorer initial ventricular function, we would expect a higher 30-day mortality in this population. In contrast, a previous infarction in those with acute anterior wall infarction carried no increased risk for 30-day mortality compared with patients with no previous infarction.

As in all thrombolytic trials, this study is limited to patients eligible for enrollment in GUSTO-I. Because trial populations have better outcomes than do general, unselected populations with myocardial infarction, these models should be validated in other, unselected populations. This prognostic information applies only to the presenting ECG; serial ECGs were not analyzed in this investigation. Because the ST segment fluctuates dynamically throughout an infarction, and rhythm disturbances are often transient, analysis of the "worst" tracing likely would have altered our results.

From the results of the multivariable analysis, a predictive nomogram (Figure 3) was created to be used in the initial risk estimation of patients who present with acute myocardial infarction. A variety of clinical and ECG markers was selected, all of which are readily assessable at presentation. A point score is assigned for the given variable, and the points are summed to estimate the risk of mortality. To increase its clinical applicability, it was necessary to limit the number of variables in the nomogram. Despite reducing the number of variables, the risk estimate from the nomogram was nearly identical to that from the entire model.

The initial ECG of patients who present with myocardial infarction and ST-segment elevation contains valuable prognostic information mostly related to heart rate, ST-segment deviations, QRS duration, and infarct location. This information adds to that of known clinical predictors of mortality and should be valuable in early risk stratification, when mortality is greatest. It should facilitate the choice of optimal pharmacological and mechanical treatments, helping physicians to target aggressive therapies to those likely to benefit most. Such information will also enable risk-adjusted assessment of both patient- and provider-specific outcomes.