Primary clinical endpoints included death, stroke, target vessel revascularization (TVR) and reinfarction. The composite endpoint point of major adverse cardiac events (MACE) at 30 days was not evaluated due to differences in definitions used in the selected studies. Myocardial perfusion was assessed by using angiographic and electrocardiographic measures. ST resolution was defined as per the study definition. Most studies defined ST resolution as more than 70% resolution in ST score [11] (defined as the sum of ST segment elevation in the leads V1 to V6, I, and aVL for anterior and II, III, aVF, and V5 to V6 for non-anterior infarctions). Single lead ST segment resolution by comparing the most prominent ST segment deviation was used in the REMEDIA trial and by Antoniucci et al [12, 13]. Angiographic efficacy endpoints included post procedural rates of thrombolysis in myocardial infarction (TIMI) grade 3 flow, and myocardial blush grade (MBG).

MBG has been defined previously [14] as follows: 0, no myocardial blush or contrast density; 1, minimal myocardial blush or contrast density; 2, moderate myocardial blush or contrast density but less than that obtained during angiography of a contralateral or ipsilateral non-infarct-related coronary artery; and 3, normal myocardial blush or contrast density, comparable with that obtained during angiography of a contralateral or ipsilateral non-infarct-related coronary artery. When myocardial blush persisted ("staining"), this phenomenon suggested leakage of contrast medium into the extravascular space and is graded 0. The angiographic endpoints were independently analyzed at core labs. Of the 13 trials that recorded MBG, 9 trials [11, 12, 15‐21] used the densitometry classification described by Van't Hof et al [14]. Two studies [22, 23] utilized the dynamic method described by Gibson et al [3], while the remaining 2 studies [24, 25] did not specify the exact method used. The authors had full access to the data and take responsibility for its integrity. All authors have read and agree to the manuscript as written.

From each trial, results were organized into a two-by-two table to permit calculation of effect sizes for aspiration thrombectomy in comparison with conventional PCI in regards to each outcome. Data on the results were collected on an "intention-to-treat" basis. When the outcome did not occur in both groups, we were unable to calculate effect sizes due to the empty cells and data were excluded from that particular trial. Continuity correction was performed when an event did not occur in one group. We used fixed-effects and random-effects models to produce across-study summary odds ratios (OR) with 95% confidence intervals. Since there was significant heterogeneity for some of the endpoints, the random-effects models are preferentially reported for the entire group although fixed effect models gave similar results. To assess the effect of individual studies on the summary estimate of effect, we did an influence analysis, in which the pooled estimates were recalculated omitting one study at a time. Subgroup analysis suggested that most of the heterogeneity could be explained by disparate results with different device classes. There was significant difference in trial size within subgroups and to avoid undue influence of small trials, and given lack of heterogeneity for most endpoints, Mantel-Haenszel fixed effect models are preferentially reported for the subgroups although random effect models gave similar results [26]. All p values were 2-tailed, with statistical significance set at 0.05. Heterogeneity was assessed by means of Cochrane Q heterogeneity test and considered significant when p value was <0.10 [27]. Publication bias was assessed by plotting a funnel plot and calculating the rank order correlation [28] and Eggers test of intercept [29]. We also calculated fail-safe N, using Rosenberg's and Orwin's method [30‐32].

All analyses were performed using Comprehensive Meta-Analysis software, version 2.0 (Biostat, Englewood, New Jersey).

The limitations of our meta-analysis are those inherent to all meta-analyses, and they include publication bias (although tested non-significant in our study) and the difficulties in comparing the results because of the different study populations, study designs and reporting methods as well as the absence of individual patient data, which prohibits adjustment for confounding factors [67]. Data on clinical endpoints like 30-day mortality, stroke, TVR, reinfarction and angiographic endpoints like MBG 0-1, 3 were not available in some trials. Data on cardiac remodeling e.g. left ventricular ejection fraction (LVEF), left ventricular volumes and diameters were not available in most trials. Six-month data was available only in two trials and hence long-term outcomes could not be studied. Six of our studies included were not published; however, the results of sensitivity analysis did not show an impact of publication status on the study results. The event rates of clinical endpoints like death, myocardial infarction, and stroke are low in patients undergoing primary PCI as seen in large PCI registry [68]. Even with the approach of a meta-analysis the number of endpoints is low. Our meta-analysis may not be powered to detect significant differences with respect to these endpoints.

In addition, our results cannot supplant a large adequately powered randomized clinical trial.