Multivessel versus culprit-only PCI in STEMI patients with multivessel disease: meta-analysis of randomized controlled trials

We searched MEDLINE through PubMed (up to July 22, 2019), Cochrane Central Register of Controlled Trials (up to July 22, 2019), and Embase (up to July 25, 2019) for RCTs of potential interest. Only studies published in English language were considered. Four groups of search terms were used, of which at least one term in each group was required to match: (1) "STEMI", "ST-elevation myocardial infarction", "ST-segment elevation myocardial infarction", "ST-elevation myocardial infarction", "ST-segment elevation myocardial infarction", and “ST-elevation"; (2) “multivessel”, “multi vessel”, “multi-vessel”, “complete”, and “culprit”; (3) "PCI", “PPCI", "pPCI", "primary PCI", "percutaneous coronary intervention", "primary percutaneous coronary intervention", "revascularization", “revascularization", and "angioplasty"; (4) "randomized", "randomised", "random", and "randomly" (Online Supplement Section 1). The data of the COMPLETE trial were included in our analysis immediately after online publication [10].

We included RCTs comparing culprit vessel-only PCI versus multivessel PCI in patients with STEMI and multivessel coronary artery disease, reporting mortality or non-fatal myocardial infarction after at least 6 months following randomization. Multivessel revascularization could be performed immediately during the index procedure or staged, but within 2 months following culprit vessel PCI. RCTs including STEMI patients with cardiogenic shock as well as studies with staged procedures not within the above-mentioned timeframe were excluded. We excluded also RCTs which were not published in full text since the risk of bias and other critical features cannot be excluded. The present meta-analysis was registered at the PROSPERO international prospective register of systemic reviews (CRD42019142643).

After removing duplicates, titles and abstracts of the identified studies were screened for eligibility by two independent observers (HJF and AF). In case of uncertainty, full-text articles were reviewed. Any discrepancies were resolved by discussion after consultation of a third investigator (AJ).

The risk of bias of the included trials was assessed by two independent investigators (HJF and AF) according to the Cochrane Collaboration’s tool for assessing risk of bias in randomized trials [11]. Again, after consultation of a third investigator (AJ), any discrepancies were resolved by discussion.

Data on sample size, length of follow-up, revascularization strategy, medical history, baseline clinical characteristics, and outcome were independently extracted by two observers (HJF and AJ). After checking for discrepancies and plausibility, individual study results were merged in a single uniformly coded data sheet.

The primary endpoint was a combination of death and non-fatal myocardial reinfarction. Because of diverse definition of the combined endpoint (all-cause death vs. cardiovascular death) among the included trials, we performed additional analyses exclusively including studies reporting all-cause or cardiovascular death. Secondary endpoints included all-cause death, cardiovascular death, non-fatal myocardial reinfarction, and revascularization. In stratified analyses, we investigated the heterogeneity introduced by pooling studies investigating (a) different strategies for multivessel PCI (i.e. multivessel PCI within index procedure versus multivessel PCI as staged procedure), (b) fractional flow reserve (FFR)-guided versus angiography-guided multivessel PCI, and (c) all-cause death versus cardiovascular death as part of the composite endpoint of death or non-fatal myocardial reinfarction.

We analyzed data by the intention-to-treat principle. Extracted hazard ratios (HRs) were converted to natural logarithms of HRs lnHR. The variance of lnHR V* was calculated using the extracted 95% confidence intervals of the HRs according to Eq. (1). In case HRs with respective 95% confidence intervals were not reported, we estimated these measures. For each treatment group i the failure rate λ was calculated using the number of events in that group events_i, the number of patients in that group patients_i, and the mean follow-up duration in months for the total population according to formula (2). The estimated HR resulted by dividing λ of the multi-vessel group by λ of the single vessel group as shown in formula (3). For the estimated HR V* was estimated according to formula (4):

The lnHR published in the original articles showed good agreement with the estimated lnHR (Online Supplement Section 2). We, therefore, felt confident to use this approach to estimate the lnHR of studies not reporting HR in their original article. Study-level results (i.e. lnHR and V*) were pooled by means of a random-effects meta-analysis using the inverse variance method as primary analysis. Between-study variances τ² was calculated according to Paule–Mandel with Hartung–Knapp adjustment. In addition, a fixed effect meta-analysis was calculated. Cochran’s Q statistic and Higgins and Thompsons I² were calculated to assess heterogeneity. The presence of small-study effects was investigated visually by means of funnel plots. Since the number of included trials was below 10, we did not apply formal test of funnel plot asymmetry [12]. We evaluated interactions in stratified analyses in random-effects models combined from the final results.

Our search resulted in 1751 search items including 756 duplicates. Detailed trial selection is demonstrated in Fig. 1. Screening of titles and abstracts identified 12 RCTs, of which 9 trials were included in final analysis (Fig. 1, Table 1). Two trials compared revascularization strategies not eligible for our meta-analysis and were, therefore, excluded (complete revascularization versus stress-echocardiography-guided revascularization in Calvino Santos et al. and staged-complete revascularization versus ischemia-driven revascularization in Mashhour et al.) [13, 14]. Another study (PRAGUE-13) was excluded because of missing full-text publication [15]. Definitions of study endpoints of the included trials are summarized in the Online Supplement Section 3.

Five out of 9 trials included showed a low risk of bias, whereas in 4 trials the risk of bias was substantial (Online Supplement Section 4). Additional quality indicators mirror the findings of the established risk assessment (Online Supplement Section 5). Key features of the included trials are summarized in Table 1. Baseline and treatment characteristics are shown in Table 2. With the exception of one study (Politi et al.), which compared culprit vessel-only revascularization with two different strategies of complete revascularization, patients were randomized to two different revascularization strategies in all trials (Table 1) [16]. The two groups receiving complete revascularization in the study by Politi et al. (immediate PCI of non-infarct-related arteries vs. staged PCI of non-infarct-related arteries) were combined for primary outcome analysis. The duration of follow-up ranged between 6 and 36 months among the included trials, whereas mean or median follow-up was only reported in 5 of 8 trials (Online Supplement Section 5). The other trials reported only the planned fixed follow-up duration ignoring censoring due to experiencing an endpoint event or loss to follow-up. Four trials did not report outcome data regarding the combination of death and non-fatal myocardial reinfarction and therefore could not be included in the analysis of the primary endpoint (HELP AMI, Politi et al. Ghani et al. and Hamza et al.). Six studies were included in the analyses of all-cause and cardiovascular mortality. The rate of myocardial reinfarction was reported in 8 trials.

In total, 523 (8.3%) of 6314 patients suffered the combined endpoint of death or non-fatal myocardial reinfarction. In patients treated with multivessel PCI this occurred less often than in patients treated by culprit vessel-only PCI (HR 0.63, 95% confidence interval [CI] 0.43–0.93; p = 0.03; Fig. 2a). According to stratified analysis, the reduction of the primary endpoint was not related to the mode of death reported in the individual trial (i.e. all-cause death versus cardiovascular death combined with non-fatal myocardial reinfarction; p value for between-group difference = 0.40; Fig. 2b). Looking on single endpoints, multivessel PCI did not reduce all-cause (HR 0.77, 95% CI 0.44–1.35, p = 0.28; Fig. 3a) or cardiovascular mortality (HR 0.64, 95% CI 0.37–1.11, p = 0.09; Fig. 3b) in comparison to culprit vessel-only PCI. However, the risks of non-fatal myocardial reinfarction (HR 0.64, 95% CI 0.52–0.79; p = 0.001; Fig. 3c) and repeat revascularization (HR 0.33, 95% CI 0.22–0.50, p < 0.001; Fig. 3d) were reduced with multivessel PCI compared to culprit vessel-only PCI. In a stratified analysis, the use of FFR-guided multivessel PCI (HR 0.69, 95% CI 0.08–5.83) did not affect the risk for the combined endpoint of death or non-fatal myocardial reinfarction compared to an angiography-guided strategy (HR 0.55, 95% CI 0.20–1.53; p value for between-group difference = 0.44; Fig. 4). Moreover, heterogeneity within the overall analysis was not explained by FFR versus no FFR strategy (I² = 27% in the overall population versus I² = 0% in FFR-guided and I² = 59% in non-FFR-guided trials).

The benefit of multivessel PCI regarding the primary endpoint was formally greater in trials performing multivessel PCI predominantly within the index procedure as compared to trials of multivessel PCI in a staged procedure (HR 0.45, 95% CI 0.24–0.84 versus HR 0.75, 95% CI 0.55–1.02; p value for between-group difference < 0.001). This interaction was also present in the fixed effect model (p value for between-group difference = 0.03). This stratification abandoned heterogeneity (I² = 27% in the overall population versus I² = 0% in both strata). Visual inspection did not suggest small-study effects for any of the performed meta-analyses (Online Supplement Section 6, Figs. 2–6).

In contrast to most previous meta-analyses, our meta-analysis took differences in length of follow-up in consideration. However, it is only based on published aggregated data. Unfortunately, not all eligible trials reported HRs with respective 95% CI. We therefore had to estimate these measures. Our approach is mathematically very similar to the approach used by Bangalore et al. [23]. Since the weight of the trials with estimated HRs is rather low, the error introduced in our pooled estimate is most likely small. We did not include the trial of Politi et al. in our meta-analysis regarding death or non-fatal myocardial reinfarction. This endpoint was not reported and we believe that summing up event rates for death and non-fatal myocardial reinfarction should not be done. Such an approach is prone to error if the same patients had a non-fatal myocardial reinfarction event before they die. To assure a detailed risk of bias assessment, only trials with available full-text publication were included in our meta-analysis. Therefore, PRAGUE-13 was not considered [15]. However, PRAGUE-13 would account for only 3% of patients in our meta-analysis.