The development of myocardial necrosis after percutaneous coronary intervention (PCI) consistently is associated with an increased risk of death, characterized by a direct continuous relationship between the amount of myonecrosis and the degree of risk ((1),(2),3). These observations support the inclusion of PCI-related myocardial infarction (procedural MI) along with nonprocedural MI (spontaneous MI) as part of the MI component of randomized clinical trial endpoints. However, there is substantial debate regarding whether procedural and spontaneous MI have similar relationships with risk for subsequent mortality. The distinction between spontaneous and procedural MI was a driving force behind the MI classification proposed by the Universal Definition of Myocardial Infarction (4), but despite recommendations made in this consensus document, the 2 types of MI are not reported separately in most clinical trials (5) and are weighted equally as endpoints.

Procedural MI typically is defined using creatine kinase-MB (CK-MB) as the preferred cardiac biomarker, using an elevation of 3 times the upper limit of normal (ULN) after the procedure as the diagnostic threshold. However, the appropriateness of this threshold is disputed ((6),(7),(8),9). Although statistically significant in most studies, the absolute risk of death associated with procedural MI has been smaller than that of spontaneous MI (10), prompting debate regarding the use of procedural MI as a trial endpoint and whether higher enzymatic thresholds for procedural MI should be used ((7),8).

In the present study, we examined the prognoses of procedural and spontaneous MI using 2 large contemporary trials of patients with non–ST-segment elevation (NSTE) acute coronary syndrome (ACS). By defining procedural MI using increasing CK-MB thresholds, we compared prognoses after procedural MI and after spontaneous MI to identify an enzymatic threshold for procedural MI that was associated with 1-year mortality risk similar to that of spontaneous MI.

Patients enrolled in the EARLY-ACS (Early Glycoprotein IIb/IIIa Inhibition in Non–ST-Segment Elevation Acute Coronary Syndrome) trial and the SYNERGY (Superior Yield of the New strategy of Enoxaparin, Revascularization and GlYcoprotein IIb/IIIa inhibitors) trial (n = 19,433) were considered for analysis ((11),(12),(13),14). Briefly, the EARLY-ACS trial evaluated the use of early eptifibatide in high-risk NSTE ACS patients for whom an early invasive approach was intended; the SYNERGY trial compared enoxaparin with unfractionated heparin in high-risk NSTE ACS patients treated with an early invasive strategy. To minimize selection bias and to include patients at risk for both types of MI, the analysis was restricted to subjects treated with PCI (n = 9,087, 46.8%).

In both trials, spontaneous and procedural MI events occurring after randomization and for up to 30 days after randomization were adjudicated by the same clinical events classification (CEC) committee, and procedural MI was defined using similar criteria. Specifically, the enzymatic criterion for procedural MI required CK-MB levels of 3 or more times the ULN and an increase of 50% or more from preprocedural levels if they exceeded ULN. If CK-MB analysis was unavailable, the CEC committee considered total creatine kinase, troponin, or new Q waves on electrocardiograms. Spontaneous MI definitions also were similar: each trial used an elevation of cardiac markers of twice the ULN. However, although CK-MB was favored over troponin for spontaneous MI in the SYNERGY trial, troponin or CK-MB could be used in the EARLY-ACS trial. To account for these differences, we restricted criteria for spontaneous MI in the EARLY-ACS trial to CK-MB levels more than twice the ULN (referred to as EARLY-ACS sMI×2) for the combined dataset analyses. The distinction between spontaneous and procedural MI was guided by the protocol (Online Appendix). In complex situations, clinical judgment and agreement between at least 2 clinicians was required to assess MI type.

Baseline characteristics and outcomes were summarized using medians (25th, 75th percentiles) for continuous variables and percentages for discrete variables. The primary endpoint was all-cause mortality occurring from 24 h to 1 year after PCI. The 1-year risks of death associated with spontaneous and procedural MI were calculated using Cox proportional hazards regression modeling. The GRACE (Global Registry of Acute Cardiac Events) covariates (15) were used for multivariate risk adjustment. Spontaneous MI was included as a time-dependent covariate to account for the possibility that this event could occur throughout the first 30 days of follow-up. Spontaneous MI was not collected beyond 30 days in these trials. Procedural MI generally was known within 24 h of PCI (only 3.5% of all procedural MI occurred more than 24 h after PCI) and was included as a fixed baseline covariate. Adjusted associations of procedural and spontaneous MI with 1-year mortality were described using hazard ratios (HRs) with 95% confidence intervals (CIs).

Associations of spontaneous and procedural MI with 1-year mortality were compared in several ways. First, we qualitatively compared the risk of death associated with procedural MI (all CEC-adjudicated events with and without CK-MB levels more than 3 times the ULN) with that for spontaneous MI in models adjusted for GRACE covariates. Second, to identify a CK-MB threshold for procedural MI at which the adjusted HR for 1-year mortality was similar to that of spontaneous MI, procedural MI was redefined using a range of CK-MB peak/ULN thresholds, increasing from 3 times the ULN in increments of 0.1. If the ratio exceeded the given threshold value, the subject was considered to have a procedural MI. For each 0.1 increment, a new Cox model with GRACE covariates was fitted, and the resulting HR was compared with that of a spontaneous MI. The threshold at which the HR for procedural and spontaneous MI were the same was defined as the matching threshold. Results were displayed graphically by plotting the grid of thresholds for the peak-to-ULN ratio against their corresponding procedural MI HR and connecting the points. Similarly, confidence bands were plotted using the corresponding 95% CI for the HR at each grid point; these represent a plausible range for the risk associated with a procedural MI defined for the given peak-to-ULN ratio. In this analysis, the definition of spontaneous MI remained fixed, with a single HR (plotted as a horizontal line) for 1-year death. A CI for the matching threshold was estimated using an iterative method and was defined as the 0.025 and 0.975 percentiles of the matching thresholds resulting from analysis of 500 bootstrap replications of the data and represents a plausible range for the peak-to-ULN ratio that would result in a procedural MI having the same risk as a spontaneous MI.

All analyses were performed separately in the SYNERGY trial and EARLY-ACS trial datasets. Then, after assessing the consistency of the results, analysis was performed on the combined dataset, in which the EARLY-ACS trial sMI×2 was used to define spontaneous MI. We also performed sensitivity analyses using a second set of covariates for risk adjustment and models that included time to PCI and/or pre-PCI CK-MB level (Online Appendix).

Of the 19,433 total study participants enrolled in the SYNERGY trial (n = 10,027) and the EARLY-ACS trial (n = 9,406), 10,259 (52.8%) underwent PCI. Of these, 9,087 (88.6%) had complete data and were included in the analysis. Overall 1-year mortality was 4.6% (469 deaths). Between randomization and 30-day follow-up, 893 procedural MI events and 298 spontaneous MI events were adjudicated.

Among 893 patients with procedural MI, 773 (86.6%) had CK-MB levels of more than 3 times the ULN and were included in the analyses; other evidence (total creatine kinase, troponin, or new Q waves on electrocardiograms) was used in the remaining cases, which were excluded. Of 773 procedural MI defined by CK-MB level, 400 occurred in the SYNERGY trial and 373 occurred in the EARLY-ACS trial. Of 298 total spontaneous MI, 127 occurred in the SYNERGY trial and 171 occurred in the EARLY-ACS trial. The median (25th, 75th) time from PCI to spontaneous MI was 2 days (interquartile range: 1 to 5 days) in the combined dataset. Among the EARLY-ACS trial patients with spontaneous MI, 112 (65.5%) had CK-MB levels of more than twice the ULN (EARLY-ACS sMI×2). Thus, the combined patient set used for analysis included 773 procedural MI and 239 spontaneous MI. A total of 28 (0.27%) patients had both PCI-related and spontaneous MI and were included in both model analyses. Baseline characteristics are shown in (Table 1).

After risk adjustment, spontaneous MI had an HR for death at 1 year of 7.35 (95% CI: 4.88 to 11.06) in the SYNERGY trial, 4.42 (95% CI: 2.90 to 6.74) in the EARLY-ACS trial using the original definition, and 3.61 (95% CI: 2.15 to 6.08) using the EARLY-ACS trial sMI×2 definition compared with patients with no spontaneous MI. The HR for patients with procedural MI compared with those without was 1.45 (95% CI: 0.93 to 2.25) in the SYNERGY trial and 1.34 (95% CI: 0.89 to 2.01) in the EARLY-ACS trial. In the pooled dataset, the HR for 1-year mortality for CEC-adjudicated procedural MI using any criterion (all CEC-adjudicated procedural MI) or a combination of criteria but no CK-MB value of more than 3 times the ULN (other evidence) was similar to that for procedural MI adjudicated with CK-MB levels more than 3 times the ULN alone (Figure 05_gr1). However, the HR for CEC-adjudicated MI that used other evidence varied more by individual trial than did the HR of procedural MI determined with CK-MB levels of more than 3 times the ULN. In the combined analyses, the risk of death associated with spontaneous MI was 5.37 (95% CI: 3.90 to 7.38) and 1.38 (95% CI: 1.01 to 1.89) for procedural MI.

(Figure 05_gr2) presents the distribution of peak CK-MB indexed by ULN for spontaneous and procedural MI events. In general, peak CK-MB elevations for spontaneous MI were lower and were associated with a heavier weight in the lower tail of the distribution than procedural MI, suggesting that spontaneous MI were adjudicated more often with lower CK-MB peak than procedural MI. (Figure 05_gr3) presents the grid search over different CK-MB thresholds for defining procedural MI in the combined dataset. By increasing the CK-MB threshold for procedural MI, a corresponding continuous increase was observed in the HR for 1-year death, which was more evident at low thresholds. The threshold at which the HR of procedural MI with 1-year mortality first matched that of spontaneous MI was 57.9 times the ULN in the SYNERGY trial, 20.4 times the ULN in the EARLY-ACS trial with the spontaneous MI definition used by the CEC committee, and 17.7 times the ULN in the EARLY-ACS trial with the sMI×2 definition (Table 2). In the combined dataset, the matching threshold was 27.7 times the ULN (95% CI: 13.9 to 58.4). At this enzymatic threshold, there were only 49 (6%) procedural MI events (Figure 05_gr4).

After risk adjustment with SYNERGY trial covariates instead of GRACE covariates, matching thresholds for procedural MI were lower but consistently larger than 3 times the ULN (6). Similar results also were found when adjusting for the time to PCI and CK-MB level before PCI in addition to the clinical model.

In a large population of patients with NSTE ACS, procedural MI was more common but associated less strongly with death at 1 year than spontaneous MI. A higher enzymatic threshold for procedural MI was needed to match the prognosis of spontaneous MI, and although the estimate varied between the SYNERGY trial and the EARLY-ACS trial, the CI for the matching threshold showed substantial overlap between groups.

In addition, the estimates were characterized by broad CI (13.9 to 58.4 for combined data), because only a small portion of patients with procedural MI exhibited elevations of CK-MB levels this extreme. Despite the large population, precisely estimating the matching threshold was challenging and will require further refinement in larger datasets. The relevant intratrial and intertrial variability in the matching threshold calls for caution when extrapolating these results to other settings. Nevertheless, these data provide evidence that for the 1-year mortality risk for procedural MI to reach a level similar to that of spontaneous MI, an elevation of CK-MB level of more than 3 times the ULN is required. A conservative estimate based on the lower bound of the 95% CI in our combined dataset suggests a threshold of nearly 14 times the ULN may be required. However, based on results in the EARLY-ACS trial dataset, in which the hazard of death associated with spontaneous MI with subsequent mortality was less strong, the estimated threshold could be as low as 5 times the ULN.

The more favorable prognosis observed after procedural MI (CK-MB more than 3 times the ULN) compared with that after spontaneous MI has been reported (10). The concordance of the present findings with previous observations supports the Universal Definition of Myocardial Infarction Task Force's recommendation for separate reporting of the effects of intervention on each type of MI (4). This group also recommends reporting the effect of experimental treatments using higher biomarker thresholds for procedural MI (i.e., 5 times the ULN and 10 times the ULN) and, ideally, the total distribution of biomarkers values. Both recommendations have been implemented only sporadically across clinical trials (5).

The higher risk of 1-year mortality associated with increasing CK-MB levels after procedural MI also provides a solid rationale for reporting infarct size in clinical trials. Regardless of the specific definition chosen for procedural MI, the availability of the complete biomarker dataset would permit more comprehensive exploration of the effects of interventions.

Although small elevations in CK-MB levels occurring after PCI still may be clinically relevant and worth preventing, combining nonfatal endpoints with unequal associations with outcomes poses interpretative challenges when assessing the effects of interventions. Although death is the key component of a composite clinical outcome, the use of composite endpoints that associate death with nonfatal events is predicated on the assumption that nonfatal components are of similar clinical importance, and common statistical approaches use equal weights for individual components of a composite endpoint ((16),(17),18). Importantly, more than three–quarters of the endpoint MI across both trials were procedural MI. Prognostic homogeneity therefore is a key property of nonfatal end point components, but this is rarely achieved (19), and attempts have been made to refine composite outcomes by developing novel approaches that include weighted individual components (20) or attributable risk (21).

We focused on prognostic homogeneity between MI components to reassess the diagnostic threshold for procedural MI. By using increasing levels of CK-MB elevation after PCI, our interest was to emphasize the degree of mortality risk over the precision of its estimate. Because of the rapid reduction of procedural MI events with increasing CK-MB level threshold, we observed increasing HR but broader CI and, eventually, opposite effects on statistical significance. This suggests that an enzymatic definition for procedural MI that relies solely on statistical significance may not account fully for the prognostic implications of a procedural MI.

The composite of procedural and spontaneous MI is a combination of heterogeneous events with substantially different prognoses. This is of particular concern for interventions that are expected to have unequal effects on spontaneous MI or procedural MI, such as early invasive strategy compared with conservative management in patients with NSTE ACS ((22),23). In this situation, a positive effect on spontaneous MI associated with early invasive strategy could be obscured by more common but less prognostically relevant procedural MI. This suggests a need for separate analysis by differentially weighting effects of experimental treatments on these MI endpoints, or alternatively, considering the use of higher degrees of CK-MB elevation for procedural MI.

Interestingly, 2 recent analyses assessing the prognostic significance of spontaneous and procedural MI ((24),25) have demonstrated a similar discrepancy between the prognoses of procedural and spontaneous MI. This may be related to the different cardiac marker thresholds used to define procedural MI. In 1 analysis (24), in which no association was observed, procedural MI was defined by a cardiac marker threshold between once and twice the ULN, similar to that used for spontaneous MI; whereas in the TRITON–TIMI 38 (Trial to Assess Improvement in Therapeutic Outcomes by Optimizing Platelet Inhibition With Prasugrel–Thrombolysis in Myocardial Infarction 38) trial (25), in which procedural MI was associated with a significant 2-fold increase in the risk of death (adjusted HR: 2.4, 95% CI: 1.6 to 3.7, p < 0.001), the cardiac marker threshold used for procedural MI was between 3 times the ULN and 5 times the ULN. Our present analysis provides a link between these varying definitions by evaluating a full spectrum of candidate thresholds.

The main analytical advantage offered by composite outcomes is the increase in statistical efficiency provided by a greater number of end points (26). Although maximizing prognostic homogeneity is desirable, the substantial reduction of procedural MI events observed with increasing enzymatic thresholds poses logistical challenges, such as increased sample size and/or study duration. In the present analysis, this attenuation of events with increasing thresholds for CK-MB level corresponds to substantial variability in the estimated matching threshold. However, in our combined analysis, even the lower bound for the matching threshold was 13.9 times the ULN. The ideal threshold should balance the competing objectives of prognostic homogeneity, clinical relevance, and maximizing endpoints.

Finally, it is unclear why similar levels of CK-MB elevation were associated with different levels of risk in the setting of procedural versus spontaneous MI. It is possible that myonecrosis caused by PCI is truly more benign compared with spontaneous MI events and coronary revascularization, and/or that other periprocedural treatments may have other benefits that attenuate the association between amount of periprocedural myonecrosis and mortality.

The possibility of a differential detection of true peak CK-MB levels between spontaneous and procedural MI also should be considered. Although time of onset for procedural MI can be known precisely, the best surrogate we have for spontaneous MI is the time of onset of symptoms, and it is common for patients with NSTE ACS to defer seeking medical attention for hours or days after symptoms emerge. Hence, the observed difference in prognosis between the 2 types of MI may be influenced by ascertainment bias.

First, the enzymatic definition for spontaneous MI used (CK-MB more than twice the ULN) is higher than current recommendations. The Universal Definition of Myocardial Infarction, developed after both studies were designed, recommends an elevation of once the ULN and designates troponin and not CK-MB as the preferred biomarker (4). Although a higher threshold may have increased the observed strength of association of spontaneous MI with subsequent outcome, recent evidence suggests that minor troponin elevation (just higher than the ULN) is associated with worse prognosis than even higher levels (27). Interestingly, the inclusion of every spontaneous MI in the EARLY-ACS trial (including events with a CK-MB of less than twice the ULN) was associated with a higher risk of mortality; other authors who used troponin elevation above the ULN to define spontaneous MI found a higher HR for death compared with our analysis (10). Thus, it remains unclear whether (and if so, how) a different definition of spontaneous MI could have affected our results.

Second, it is impossible to define a single matching threshold; slight differences in populations and covariates induce inherent heterogeneity. In particular, the risk associated with spontaneous MI (especially among patients receiving medical management because of coronary anatomical features judged unsuitable for PCI) ((28),29) may differ from that of a procedural MI. Although residual confounding cannot be excluded, the analysis population was restricted to patients who actually underwent PCI, and analyses were adjusted to minimize selection bias and measured confounders.

Third, in patients undergoing very early PCI, the discrimination between the procedural MI endpoint and the qualifying, spontaneous MI may be challenging because of increasing biomarkers at the time of PCI. However, the median time from admission to PCI was 21 hours in the SYNERGY trial, and the EARLY-ACS trial included only patients expected to undergo an invasive strategy no sooner than the next calendar day after randomization. Thus, it seems unlikely that a significant proportion of patients had increasing markers before PCI. Also, we used procedural MI adjudicated by at least 2 experienced CEC committee members and included time from admission to PCI as a covariate in the model for risk adjustment.

Fourth, there are different MI locations and subtypes of spontaneous MI (ST-segment elevation vs. non–ST-segment elevation) and procedural MI (11). Part of the damage induced by PCI is unavoidable and is unlikely to be affected by antithrombotic or other therapies. A biomarker elevation may have a different association with mortality in the presence of an overt clinical syndrome or a clear angiographic complication. However, these distinctions are not well standardized and were not captured consistently across the 2 trials we examined. Thus, the present analysis is an appropriate comparison of procedural and spontaneous MI, as defined by trials, but does not necessarily extrapolate to clinical events that involve additional information.

Fifth, study follow-up was limited to 1 year. Longer follow-up could have provided different results.

Finally, there is no unique gold standard for defining the relationship of procedural MI with outcome or for ascertaining how that differs from spontaneous MI. Although other approaches can complement ours (e.g., maximal discrimination, maximal predictive ability), we believe that the emphasis on prognostic homogeneity can highlight an additional dimension for the definition of this endpoint.

Spontaneous MI (CK-MB levels more than twice the ULN) is associated strongly with death at 1 year. Using the enzymatic definition of CK-MB levels more than 3 times the ULN, procedural MI was associated more weakly with 1-year death. A higher CK-MB level threshold for procedural MI would be needed to reach an association of procedural MI with 1-year mortality similar to that of spontaneous MI. These observations raise implications for the design and interpretation of clinical trials that use MI as an endpoint.