Methods

The study design and overall findings of the EPHESUS trial have been described previously. Briefly, this multicenter double-blind randomized trial, conducted from 2000 to 2001, included patients with acute myocardial infarction (AMI) complicated by clinical signs and symptoms of heart failure and an LVEF ≤40%. LVEF was determined at the study site by echocardiography, radionuclide angiography, or left ventricular angiography. Most patients had heart failure symptoms that were diagnosed clinically by the presence of pulmonary rales, a third heart sound, or pulmonary vascular congestion on chest radiography. In patients with diabetes who met the criteria for left ventricular function after AMI, the clinical signs and symptoms of heart failure did not have to be demonstrated for enrollment. The eligible patients were randomized within 3 to 14 days (median 7.3 ± 3 days) after AMI to receive either eplerenone (n = 3,319 patients, mean dose: 43 mg/day) or placebo (n = 3,313 patients) and were followed for up to 32 months (mean follow-up 16 months) in addition to standard medical therapy. Revascularization was performed using thrombolysis, percutaneous coronary angioplasty, and/or coronary artery bypass grafting.

Of the 6,632 EPHESUS participants, 6,510 (98%) had an electrocardiographic description of the ST segment at enrollment and were classified by the study investigators as STEMI and non-STEMI patients. To ensure transmural ischemia, only patients with STEMI (n = 4,703, 72%) were included in the present analysis. Data relative to sociodemographic, clinical, subclinical, and laboratory variables were collected at baseline and have been reported previously. The study was approved by an institutional review committee, and the subjects gave informed consent. Of the EPHESUS participants, 2,854 patients (53%) who underwent early revascularization within 12 hours of symptoms onset and with a persistent ST-segment elevation or new presumed left branch block were identified. For patients with the clinical presentation of STEMI within, early mechanical (percutaneous coronary intervention) or pharmacological reperfusion has been performed as early as possible. In these early revascularization STEMI patients, 1,253 patients (27%) underwent percutaneous coronary angioplasty, 1,557 (33%) thrombolysis, and 44 (1%) urgent surgical revascularization.

The primary objective of the present analysis was to determine the predictors of all-cause death and recurrent MI at 30-day and 1-year follow-up and to compare the latter to the established TIMI predictive risk score. Screening and baseline procedures were to be performed during the hospitalization for the index AMI, and follow-up visits occurred at 1 and 4 weeks, 3 months, and every 3 months thereafter until the termination of the study. Information about adverse events and concurrent medications was recorded at every visit. All patients who underwent randomization were followed for vital status and hospitalizations every 3 months until the termination of the study. Recurrent MI was confirmed by 2 of 3 usual criteria: changes on electrocardiography including ST-segment elevation, new Q waves, or both (transmural infarction) or T-wave inversion, ST-segment depression, or both (subendocardial infarction); typical symptoms; or a typical increase and decrease in biochemical markers of myocardial damage (e.g., creatine kinase [CK] or CK-MB) in which the maximal value reached is >2 times the upper limit of the hospital range for CK or in which the CK-MB value is ≥10% of the CK value. The clinical events were adjudicated by a blinded EPHESUS critical events committee. The association of these major clinical events with the validated TIMI risk score was determined for each patient at 1 month and 1 year.

All analyses were performed using SAS 9.3 software (SAS Institute, Cary, North Carolina) and R software (The R Foundation for Statistical Computing). The 2-tailed significance level was set at p <0.05. Time-to-event analyses were carried out using Cox proportional hazard regression. Univariate analyses first assessed the association of TIMI score predictors with all-cause death and myocardial reinfarction at 30 days and 1 year. Then, associations between TIMI predictors and outcomes were tested in multivariate analyses without covariable selection (full models). Finally, other nonpredefined TIMI predictors associated with the outcomes were identified among baseline characteristics (gender, age, weight, and body mass index [separately assessed], Killip >I at admission, revascularization >4 hours after index event; systolic, diastolic, and mean blood pressure (separately assessed); heart rate; LVEF; sodium, potassium, estimated glomerular filtration rate [eGFR; Chronic Kidney Disease Epidemiology Collaboration formula] ; previous AMI; hypertension (HTN), diabetes, angina) along with study drug using a backward interactive selection procedure. Only significant factors were included in the final models (reduced models). The validity assumptions of the Cox models were thoroughly checked (proportionality of hazards, functional form of continuous covariables, absence of interaction or collinearity, at least 10 events/predictor). Results are presented as hazard ratios and their 95% confidence intervals (CI). The predictive accuracy of each univariate Cox model was determined by Harrell C-index. The C-index is defined as the proportion of patients in which the predictions and outcomes are concordant. The ability of the new model to reclassify is summarized by the net reclassification index (NRI) statistic. This method assesses the ability of a new model to reclassify subjects with and without a clinical event during follow-up. The continuous NRI method (IDI) developed by Uno and implemented in the survIDINRI package of the R software was used. The IDI method does not require a previous definition of strata risk and considers the change in the estimation prediction as a continuous variable.