Does intravascular ultrasound provide clinical benefits for percutaneous coronary intervention with bare-metal stent implantation? A meta-analysis of randomized controlled trials

We performed an electronic search of PubMed, Embase and Web of Knowledge databases with the following terms: Myocardial Ischemia; Ischemic Heart Disease; Acute Coronary Syndrome; Angina; Coronary Disease; Coronary Artery Disease; Coronary Occlusion; Coronary Thrombosis and Myocardial Infarction, in association with the terms Interventional Ultraso*; Intravascular Ultraso*; Intracoronary Ultraso*; IVUS and ICUS.

A manual search was also performed to retrieve potential articles cited in previous meta-analyses, in review articles and those considered to be relevant by the reviewers. The electronic search, which evaluated the articles included in the databases through April 2011, was limited neither by publication date nor by language.

Only randomized controlled trials that compared IVUS-guided PCI with angiography-guided PCI, with clinical outcomes, and at least six months of clinical follow-up, were included in quantitative synthesis. The clinical outcomes considered were death, nonfatal MI and the combined endpoint of MACE (death, nonfatal MI, or repeat revascularization). For repeat revascularization, a report of any new coronary revascularization (surgical or percutaneous) was considered, regardless of the lesion and of the vessel treated. Surrogate outcomes, such as angiographic outcomes, were not taken into account because these can show a positive result with no effect (or harmful effect) on clinical outcomes [15]. These clinical outcomes (death, nonfatal MI and MACE) were considered primary endpoints in our meta-analysis.

The titles and abstracts from the articles retrieved by the search strategy had been independently evaluated by two reviewers (LCP, LLJ). All articles in which IVUS was mentioned were selected. These articles were fully read, and those that met the criteria were included. Disagreements were solved by consensus. If consensus was not achieved, a third reviewer (ALR) defined the question.

The intention-to-treat meta-analysis that followed the systematic review was performed by the random-effects model of the Comprehensive Meta-Analysis software (Borenstein M, Hedges L, Higgins J, Rothstein H. Version 2.2.048, Biostat, Englewood NJ, USA 2005), with the odds ratio (OR), 95% confidence intervals and two-sided P-values calculated for each outcome. The analysis of heterogeneity between studies was estimated by the I ² statistic.

Publication bias evaluation was performed by Duval and Tweedie’s Trim and Fill method [16]. Egger's test was also performed to analyze the impact of several factors on the size of the treatment effect [17]. The small study effect was also evaluated by cumulative analysis (from largest to smallest sample size) and by the one-study-removed method.

A total of 4,247 articles in PubMed, 869 in Embase and 4,260 in Web of Knowledge databases were identified. Eight studies were selected according to the inclusion criteria (Figure 1) [18‐26]. After a comprehensive analysis, three studies were excluded because they used a provisional stenting technique [19, 25, 26], which is no longer performed because of its higher restenosis rate [27]. Table 1 summarizes the clinical and angiographic characteristics of the patients included in the selected studies.

There were significant differences between the five studies included in the final analysis (Table 2). One of the current indications of IVUS-guided PCI is for patients with long lesions (greater than 15 or 25 mm) [8, 28, 29], who have been excluded from most studies [18, 20, 24]. Unlike the others, the TULIP study excluded those patients who had focal lesions (less than 20 mm in length). Every study but the AVID trial excluded patients with a current or past history of acute coronary syndrome (ACS). In the RESIST study, randomization was performed only after the intervention, which may have caused a selection bias. In the AVID trial, the IVUS analysis was only performed after implantation of the stent, excluding the initial assessment of the target lesion [8].

The criteria for optimal stent implantation were heterogeneous. Only the OPTICUS study used the criteria proposed by the MUSIC study [10]. The majority of patients underwent angiographic assessment after six months (angiographic follow-up) [20, 22, 24]. Another difference between the studies was in the criteria used for MACE. In the RESIST study, MI was not included. In the TULIP study, the MACE criteria included death, nonfatal MI and ischemia-driven target lesion revascularization (TLR). In the AVID trial, the composition of this outcome was not explained. In the other studies, the criteria for repeat revascularization were more comprehensive and included coronary artery bypass grafting (CABG) or a repeated PCI for any reason [18, 20, 21, 24].

The heterogeneity among the studies showed intermediate values in nonfatal MI (I ²  = 48.82%) and MACE (I ²  = 57.38%). For death, no heterogeneity was observed among the studies (I ²  = 0%).

We also evaluated the possibility of publication bias (B0) for MACE. Egger’s Test (B0 = −3.43; 95% CI − 6.40 to −0.47, one-tailed P- value 0.02) and the trim and fill test (observed OR 0.74, 95% CI 0.49 to 1.13; two studies imputed: adjusted OR 0.93, 95% CI 0.60 to 1.44) (Figure 2) were positive, suggesting the presence of small studies effects, which can be attributable to differences in design (not detected) or to publication bias.

A total of 1,754 patients were randomized in five studies. There was no statistically significant difference between the IVUS-guided group and the angiography-guided group (Table 3) for death (OR 1.86, 95% CI 0.88 to 3.95, P = 0.10) (Figure 3-A), nonfatal MI (OR 0.65, 95% CI 0.27 to 1.58, P = 0.35) (Figure 3-B) or MACE (OR 0.74, 95% CI 0.49 to 1.13, P = 0.16) (Figure 3-C).

In order to clarify the discrepancy in MACE results found in this analysis compared to the medical literature, the data of a previously published meta-analysis [13] were re-evaluated (Figure 4). Among the studies selected by that meta-analysis, only two were not included in the present selection because the provisional stenting technique was employed in both of them [19, 25].

A funnel plot analysis was performed along with Egger’s Test (B0 = −3.66, 95% CI − 5.54 to −1.78, one-tailed P-value = 0.002) and the trim and fill test (observed OR 0.70, 95% CI 0.50 to 0.98; three studies imputed: adjusted OR 0.89, 95% CI 0.62 to 1.27), which suggested the presence of publication bias [16, 17].

A cumulative meta-analysis by reverse order of sample size was performed. The results only became positive when the last and smallest study was included in the analysis (Figure 5). Moreover, the one-study-removed analysis showed that the removal of any one of the smaller studies gave a neutral result from the meta-analysis (Figure 6). This makes it plausible to assume that one small unpublished study with negative results would be enough to nullify the effect of that meta-analysis.

The five selected studies have important differences that might lead to completely different outcomes in another context. For example, the exclusion of patients presenting with ACS may have led to a reduction in post-interventional adverse events [32‐35]. In the AVID trial, pre-interventional IVUS was not performed, which excluded an important phase of the method because one of the roles of IVUS is to assess the target lesion to help in the choice of technique and devices for the PCI [8]. Only the OPTICUS study used the MUSIC study criteria for optimal stent implantation [10], which theoretically could be associated with a lower MACE rate [36]. Angiographic follow-up was performed in most studies, which may have led to an overestimated rate of repeat revascularization, due to the oculo-stenotic reflex [37], which is the predisposition to indicate a PCI for any significant luminal obstruction, despite the presence or absence of myocardial ischemia [38]. The fact that most studies have used more comprehensive criteria for repeat revascularization may have also increased MACE rates [18, 20, 24].

The data analysis suggested the presence of publication bias in both meta-analyses. This bias may have led to apparently positive results that could be easily modified by unpublished studies with small sample sizes. It may be harmful because it can maintain or amplify an apparent beneficial effect of the intervention [39].

Significance-chasing bias is an enticing term that refers to the clustering of the most common types of meta-analysis bias, including those in which the apparently negative results remain unpublished (study publication bias and selective outcome reporting bias), those in which negative results become positive (selective analysis reporting bias) and those in which no existing data are presented as positive (fabrication bias) [40, 41].

Selective reporting bias is the most common problem in meta-analyses. In selective outcome reporting bias, specific data with a negative result are omitted from publication. In selective analysis reporting bias, which is even more frequent, a negative result calculated from a pre-determined analysis plan is replaced by a positive result achieved through post hoc data analysis [41].

Another major problem is the potential presence of interests other than scientific truth. This matter becomes critical when the object of the meta-analysis is an industry product, as in the present study, because most of the researches are conducted or funded by manufacturers (Table 4) [42, 43]. There is a current trend towards opposing this practice [44].

Meta-analyses have gained prestige over time but they are still considered by some to be an ancillary method, accepted only when it corroborates the point of view of experts and of public policies [41]. They could play a fundamental role in changing (or in supporting) the evidence on relevant issues if conducted properly, with a pre-specified analysis plan and declared conflicts of interest for every study included, in addition to determining and reporting all possible biases.

The paucity of randomized controlled trials comparing IVUS-guided PCI and angio-guided PCI and the exclusion of groups with specific lesions (long lesions, small vessels, bifurcations or left main coronary artery) may have masked a possible benefit. The same point applies to the exclusion of patients presenting with ACS, whose rate of cardiovascular events is higher, and PCI, when indicated, may even reduce mortality [32‐34].

The low statistical power of the present study and of the re-evaluated meta-analysis [13] is due to the presence of heterogeneity and to the possibility of study publication bias. The presence of other biases might be possible but that is even more difficult to prove.