Navigating the Left Main Bifurcation: Which Stenting Strategy Best Minimises Major Adverse Cardiac Events and Target Lesion Revascularisation? A Network Meta-analysis

The left main coronary artery (LMCA) serves a vital role in supplying blood to the left ventricle, providing approximately 70–80% of the heart’s oxygenated blood. Lesions located at the left main bifurcation represent a complex and high-risk subset during percutaneous coronary intervention (PCI), leading to dire prognoses if not addressed promptly and effectively. Studies indicate that untreated LMCA lesions can result in mortality rates exceeding 50% within 3 years, emphasising the importance of optimal management.^1,2 Further, the prevalence of LMCA stenosis ranges between 4–6% in patients undergoing coronary angiography, indicating that many individuals at significant risk may remain asymptomatic.³

Bifurcation lesions present a number of technical challenges related to their complex geometry, which includes risks for restenosis and side branch occlusion. Specifically, left main bifurcation lesions (LMBL) are more intricate compared to non-left main bifurcations and are often associated with elevated risks of adverse clinical outcomes.⁴ The choice of treatment, whether to proceed with PCI or opt for coronary artery bypass grafting (CABG), is further complicated by advances in drug-eluting stent technology, which have made PCI a more appealing option. However, CABG remains the benchmark treatment for patients with complex lesions and multivessel coronary artery disease.^5,6

When managing LMCA bifurcation lesions, two primary stenting strategies are used: provisional stenting (a single-stent approach) and planned two-stent strategy techniques, including double kissing (DK) crush, culotte and T-stenting techniques. Each approach claims unique advantages and potentially different patient outcomes. For example, some studies suggest that the DK crush technique may have superior outcomes in terms of lower rates of major adverse cardiac events (MACE) and target lesion revascularisation (TLR) compared to culotte techniques.^7,8 Nevertheless, discrepancies in findings present ongoing challenges in determining which strategy is superior, necessitating further investigation.

MACE and TLR are the primary outcomes of interest when evaluating these strategies. Variability in stenting techniques dramatically influences MACE and TLR rates in patients with LMBL. Specifically, the associated mortality and morbidity rates with LMCA occlusions can be exceedingly high, aligning with the need to develop risk-averse stenting strategies.^9,10

Despite the existence of numerous randomised controlled trials (RCTs) and observational studies, consistent evidence regarding the optimal stenting approach for LMCA lesions remains lacking. Existing studies often focus on subset techniques in isolation and there is a notable absence of consensus regarding best practices, leaving clinicians without definitive guidelines.^11,12 Additionally, the literature lacks comprehensive comparative analyses that consider all stenting techniques within a unified analytical framework, underscoring a substantial need for systematic exploration, such as a network meta-analysis (NMA). The objective of this network meta-analysis was to compare the impact of different stenting techniques on reducing MACE and TLR in patients with LMBL.

Methods

The writing of this NMA adheres to the preferred reporting items for systematic reviews and meta-analyses for network meta-analyses (PRISMA-NMA) guidelines. The protocol of this study was published in PROSPERO with ID CRD420251064116.

Search Strategies

The literature search was performed across six databases: Scopus, Semantic, Wiley, EBSCO, PubMed and Web of Science. The search used keywords organised through Boolean operators. The search terms include “Percutaneous coronary intervention” OR “PCI” OR “Stenting” OR “Stent” OR “DK-CRUSH” OR “Double Kissing” OR “Provisional” OR “Culotte” OR “T-stenting” OR “T and small protrusion” OR “TAP” OR “Single” OR “Double” OR “Crush”) AND (“Left main bifurcation” OR “Left main”) AND (“Random” OR “Randomized” OR “Randomised” OR “Observational” OR “Cohort” OR “Case control”). Medical subject headings were used for all terms during the data retrieval process.

Study Selection

The study selection process used Rayyan, an artificial intelligence-powered systematic review management platform, and involved eight independent reviewers to reduce bias and prevent the omission of relevant studies. Following the removal of duplicates, all reviewers conducted a screening of all collected articles based on year, title, and abstract. Subsequently, full-text articles were evaluated for eligibility. In instances of conflict, SS and HFR served as mediators. The inclusion criteria were established based on the PICO framework, which focuses on patient/population, intervention, comparison and outcome:

• Population: adult patients (≥18 years) with LMBL were eligible. Studies that included mixed bifurcation lesions were considered only if left main (LM)-specific data could be extracted separately.
• Interventions and comparators: eligible studies evaluated PCI using drug-eluting stent with one or more bifurcation stenting strategies, including provisional (single-stent), crush (standard, mini or nano), DK crush, culotte, T-stenting or T-stenting and small protrusion (T/TAP), and V-stenting.
• Outcomes: studies were required to report at least one of the following endpoints: major adverse cardiac events as defined by each study (commonly a composite of death, MI or repeat revascularisation) and/or TLR.
• Follow-up: a minimum clinical follow-up of 12 months was required for inclusion in the main analysis, while studies with shorter follow-up (6–12 months) were retained for sensitivity analyses.
• Study design: we included RCTs, prospective or retrospective comparative cohort studies and registry-based analyses.

Studies were excluded according to the following criteria:

• Publication in languages other than English;
• Protocol and irretrievable studies;
• Single-armed studies.

The study selection process is illustrated in the PRISMA flow diagram (Figure 1).

Risk of Bias Assessment

The selection of instruments for measuring research quality was determined by the study design of the included studies. All authors conducted the risk of bias assessment independently, with the majority judgement applied in the event of any conflict. The Newcastle-Ottawa Scale was chosen for observational studies. This instrument evaluates three primary domains: selection bias, comparability and outcome. The highest possible score for this instrument is 9. Studies are classified as high quality if they have a final score of ≥7, moderate quality with a score of 5–6 and low quality with a score of <5. RCTs were assessed using Risk of Bias (RoB) 2.0. This tool evaluates five domains: randomisation bias; bias due to deviation from the intervention; missing outcome data; outcome measures; and bias in reporting results. The results of the RoB assessment were aggregated into a single dataset and uploaded on the Confidence in Network Meta-analysis (CiNeMa) website to generate a visualisation of the risk of bias for each comparison.

Data Extraction

The data extraction process resulted in a table containing important information from the inclusion study. The main data included study characteristics, identity and origin, number of participants, study type, type of stenting, SYNTAX score and patient comorbidity. These data were important when considering heterogeneity among studies.

Data Analysis

The process involved selecting, categorising and extracting data from each study. Data on outputs, including TLR and overall MACE, were compiled into a spreadsheet and downloaded as a .csv file to streamline the analysis process. NMAstudio was used for the execution of the network meta-analysis. In this study, network meta-analysis will be conducted on two models. Model 1 compares DK crush with all other double-stenting techniques found. DK crush was analysed separately rather than being grouped within the double-stent arm to acknowledge its distinct procedural characteristics and established evidence base. Unlike conventional two-stent techniques, DK crush incorporates a systematic sequence of side-branch predilation, first kissing, crushing and final kissing balloon inflation, which differentiates it mechanically and haemodynamically from other double-stenting strategies. Model 1 also aims to prove whether DK crush is superior to the other two stent techniques overall. Model 2 compares DK crush with dual-stenting and single-stenting techniques in detail. In this model, if the included studies do not separate the data from each double-stenting, we generalised the data using the technique most commonly used in related studies. The data used is dichotomous, employing a random effects model (REM) analysis that incorporates indirectness and risk of bias as effect modifiers. The effect size used is the RR. The findings were then displayed through a ranking system and a forest plot. We used CINeMA to assess the quality of evidence for each comparison result regarding its heterogeneity, inconsistency and indirectness. Heterogeneity was quantified using τ² values, where τ²>0.04 indicated moderate variation. Inconsistency was measured using the design-by-treatment interaction model, with p<0.05 indicating significant disagreement between direct and indirect estimates. Indirectness was appraised based on the number and strength of indirect evidence paths in the network. Each treatment comparison was then graded as having low, moderate or high concern and summarised to determine the overall confidence level for each outcome.

Result

Study Selection

After searching across six databases, a total of 365 articles were collected. A total of 128 duplicate articles were eliminated, resulting in a remaining count of 237 articles, which underwent manual selection based on their titles and abstracts. A total of 202 articles were excluded according to the exclusion criteria or failing to meet the inclusion criteria. A total of 35 articles remained for full-text access and review. An extensive evaluation of those 35 articles was performed, resulting in 20 articles selected for inclusion. The specific rationale for the exclusion of the other articles is illustrated in a PRISMA flow diagram (Figure 1).

Studies Characteristic

This review includes 20 studies conducted across various countries, including China, Italy, South Korea, India, Egypt, Germany, and several other European nations. Seven were RCTs and the other 13 studies were retrospective observational. The mean age of patients ranged from about 55 years to over 70 years, with common comorbidities such as hypertension, diabetes, dyslipidaemia and smoking. The follow-up duration varied between studies, from as short as 6 months to as long as 10 years, providing short-term and long-term perspectives on the outcomes of different stenting strategies for LMBL.

When pooled across all included studies, the total number of patients was 8,731. This figure was obtained by adding the sample sizes from each study, which varied considerably from small-scale studies with about 65 participants to large cohorts exceeding 1,000 patients. This variation reflects the heterogeneity of the studies but also strengthens the analysis by capturing a broad patient population with diverse clinical characteristics. The overall sample size provides a robust foundation for comparing the effectiveness of different stenting techniques in relation to key clinical outcomes, namely MACE and TLR. The details are shown in Supplementary Table 1.

Network across Percutaneous Coronary Intervention Strategies

Figure 2 illustrates the network geometry of studies comparing different PCI strategies for LMBL. Two models were constructed. Model 1 (Figure 2A) provides a simplified classification, grouping the interventions into single-stent (provisional), double-stent (culotte, T/TAP, V-stenting) and DK crush strategies. Model 2 (Figure 2B) depicts the full set of individual techniques, including provisional, DK crush, culotte, T/TAP, V-stenting and standard crush.

Nodes represent each intervention in the network plots, while edges represent direct comparisons between interventions. The thickness of the connecting lines is proportional to the number of patients or trials contributing to each head-to-head comparison, indicating stronger evidence where lines appear thicker. Similarly, the size and colour of the nodes reflect the number of patients allocated to each intervention and the overall risk of bias assessment, respectively. Interventions with predominantly low-risk trials appear green, while the presence of moderate or high risk of bias shifts the node colour toward yellow.

Overall, the network demonstrates that provisional stenting was the most frequently studied comparator, serving as the central hub in both models. DK crush and culotte were also well represented, while other techniques such as T/TAP and V-stenting were less frequently studied, resulting in thinner connecting lines.

Ranking Model 1

Figure 3 illustrates the heatmap ranking according to p-scores for three main strategies: DK crush, single-stenting and double-stenting. The p-score is a numerical value ranging from 0 to 1 that reflects the relative ranking of each treatment in a network meta-analysis, with higher scores indicating greater likelihood of being the most effective strategy. For TLR, DK crush ranked the highest (p=1.0), followed by single-stenting (p=0.5), while double-stenting showed the lowest performance (p=0.0). A similar pattern was observed for MACE, where DK crush again achieved the top rank (p=1.0), followed by single-stenting (p=0.42) and double-stenting (p=0.08). These findings highlight the superiority of DK crush compared to other strategies in preventing repeat adverse events.

Forest Plot Model 1

Figure 4 displays the RR comparisons of DK crush and pooled double-stenting strategies against single-stenting. For TLR (top panel), DK crush demonstrated a significant reduction in risk compared with single-stenting (RR 0.58; 95% CI [0.38–0.88]), while double-stenting was associated with an increased risk (RR 1.29; 95% CI [1.06–1.58]). For MACE (bottom panel), DK crush again showed a protective effect (RR 0.51; 95% CI [0.36–0.72]), whereas double-stenting did not significantly differ from single-stenting (RR 1.08; 95% CI [0.93–1.26]). These results suggest that DK crush consistently outperforms both single- and double-stenting in reducing adverse outcomes.

Ranking Model 2

Figure 5 presents the heatmap ranking according to p-scores with a more detailed comparison of various double-stenting techniques versus provisional stenting. For TLR, DK crush consistently ranked the highest (p=0.99), followed by provisional (p=0.73) and V-stenting (p=0.48). Other double-stenting techniques such as crush (p=0.42), culotte (p=0.19), and T/TAP (p=0.18) demonstrated lower performance. For MACE, a similar trend was observed: DK crush again maintained the highest ranking (p=0.97), with V-stenting (p=0.58) and provisional (p=0.53) showing moderate performance. Conversely, culotte (p=0.02) and T/TAP (p=0.39) ranked lowest. These results reinforce the consistent superiority of DK crush, even when compared with other double-stenting variations.

Forest Plot Model 2

Figure 6 presents the RR comparisons of several double-stenting techniques against provisional stenting. For TLR (top panel), DK crush showed a significant reduction in risk compared with provisional stenting (RR 0.55; 95% CI [0.34–0.89]). In contrast, culotte (RR 1.44; 95% CI [1.02–2.04]) and crush (RR 1.36; 95% CI [1.01–1.83]) were associated with higher risk. V-stenting (RR 0.98; 95% CI [0.59–1.61]) and T/TAP (RR 1.11; 95% CI [0.70–1.77]) did not demonstrate significant differences compared to provisional stenting.

For MACE (bottom panel), DK crush again outperformed provisional stenting (RR 0.47; 95% CI [0.30–0.72]). Meanwhile, culotte (RR 1.28; 95% CI [1.01–1.62]) and T/TAP (RR 1.21; 95% CI [1.01–1.46]) showed increased risk, whereas crush (RR 1.09; 95% CI [0.86–1.38]) and V-stenting (RR 0.94; 95% CI [0.66–1.34]) revealed no significant difference. These findings reinforce the consistent superiority of DK crush across both outcomes, while some alternative two-stent techniques may even increase the risk of adverse events compared with the provisional strategy. The inclusion of 95% CI in the forest plots highlights the precision of these estimates and confirms the robustness of DK crush as the most effective strategy across both outcomes.

Consistency, Heterogeneity, Indirectness, Coherency

Table 1 summarises the assessment of network consistency, incoherence and heterogeneity across both models for the outcomes of TLR and MACE. In this analysis, inconsistency and incoherence are expressed as p-values, indicating whether statistically significant disagreement exists between direct and indirect evidence. A p-value below 0.05 would suggest meaningful inconsistency or incoherence, however, all reported values in this table are above 0.05, demonstrating no significant disagreement between different sources of evidence and supporting the internal validity of the network. Heterogeneity is presented using τ², reflecting the variance of treatment effects across included studies. The low τ² values (ranging from 0.03–0.04) indicate minimal heterogeneity, suggesting that the effect sizes were largely consistent across trials.

The assessment of indirectness (Supplementary Figure 1) revealed that most comparisons in the network had low concerns, particularly those involving DK crush versus provisional or culotte stenting, which were supported by several head-to-head RCTs. However, some strategies such as T/TAP and V-stenting were informed predominantly by indirect evidence through links with DK crush or provisional stenting, resulting in moderate concerns for indirectness. Overall, the certainty of evidence from indirectness was judged as low to moderate, with the most reliable comparisons observed in analyses related to DK crush.

Risk of Bias and Reporting Bias

Supplementary Figure 2A represents model 1, in which double-stenting techniques were compared with single stenting, while Supplementary Figure 2B represents model 2, in which multiple two-stent strategies were compared with provisional stenting.

In model 1 (Supplementary Figure 2A), the distribution of studies around the pooled effect size appeared relatively symmetrical, suggesting a low likelihood of significant publication bias for comparisons involving double versus single stenting and DK crush versus single stenting. In model 2 (Supplementary Figure 2B), the funnel plot also demonstrated an overall symmetrical distribution across comparisons, including crush versus provisional, DK crush versus provisional, and T/TAP versus provisional. Although a few studies were dispersed toward the periphery, no substantial asymmetry was observed, indicating minimal reporting bias.

Regarding the risk of bias (Supplementary Figure 3), the majority of included studies demonstrated a low risk, especially for comparisons involving DK crush, owing to the presence of large, well-conducted RCTs such as DKCRUSH-III and DKCRUSH-V. Nonetheless, several comparisons were informed mainly by retrospective observational studies, which introduced moderate concerns due to potential confounding and selection bias. A few comparisons, particularly those involving standard crush or T/TAP techniques, were rated at high risk of bias because of small sample sizes and methodological limitations. Taken together, the overall quality of evidence was considered low to moderate, with greater confidence in results derived from DK crush and provisional stenting comparisons.

Discussion

Superiority of DK Crush for Left Main Bifurcation Lesion

In this network meta-analysis, DK crush demonstrated a consistent trend toward improved outcomes in both TLR and MACE compared with other stenting strategies. To further clarify its role, we performed two analytic models. Model 1 grouped stenting techniques according to the overall strategy (single-, double- and DK crush stenting), while model 2 compared each technique individually. DK crush was intentionally separated from the double-stent category in model 1 as it represents a distinct procedural evolution with specific mechanical and haemodynamic properties. Unlike conventional two-stent approaches, DK crush integrates sequential side-branch preparation, systematic crushing and dual kissing balloon inflations, producing unique geometrical and physiological effects that may independently influence outcomes.

The observed superiority of DK crush over other techniques may be explained by several procedural and structural advantages, although these mechanisms remain speculative. The technique ensures complete circumferential ostial coverage, optimal stent expansion and improved final apposition through two sequential kissing balloon inflations.¹³ This sequence minimises the risk of plaque shift or side-branch occlusion and promotes symmetrical lumen expansion at the carina. Imaging-based and computational studies have demonstrated that DK crush may achieve more uniform stent geometry and favourable flow patterns, which could contribute to lower restenosis and thrombosis rates.^14,15 However, it should also be acknowledged that techniques such as culotte and T/TAP can achieve full ostial coverage when properly executed. DK crush’s procedural complexity, requirement for dual rewiring and triple metal layering at the side-branch ostium may increase local rigidity and procedural duration. Thus, its advantages are likely multifactorial reflecting not only mechanical optimisation but also operator experience and lesion selection.

Anatomically, the ostial-proximal circumflex segment presents unique challenges due to its dynamic, torsion-prone nature, which predisposes it to TLF. Techniques that introduce multiple metallic layers at this site, such as culotte and T/TAP, may increase stiffness and reduce vessel compliance, impairing physiological motion and flow.^13,16 In contrast, DK crush, V-stenting and provisional approaches tend to preserve greater ostial flexibility and maintain favourable shear stress distribution. The dual kissing balloon inflations in DK crush may further facilitate optimal alignment and apposition at the circumflex ostium, potentially mitigating localised stress concentration and restenosis risk.^17,18 These mechanical and flow-related factors may partially explain the lower TLF rates associated with DK crush and simpler strategies.

While T/TAP generally yields inferior outcomes when employed as a planned two-stent technique, it remains a valuable bailout option in complex bifurcations, particularly when provisional stenting fails due to plaque shift, dissection or slow flow. T/TAP provides complete ostial coverage with minimal procedural escalation and allows for final kissing balloon inflation, which enhances apposition and restores physiological flow.^19,20 Studies have shown that, when used selectively, T/TAP achieves acceptable long-term results and may reduce MACE compared with more extensive re-stenting techniques.^21–23 Its simplicity, shorter procedure time and technical versatility make it a pragmatic choice in urgent or anatomically challenging situations, aligning with modern hybrid or stepwise provisional strategies.^19,24 In addition, previous studies showed the integration of intravascular imaging such as intravascular ultrasound further enhances the outcomes of DK crush. Imaging guidance improves vessel visualisation, ensures optimal stent positioning and reduces complications including stent thrombosis and periprocedural MI.^25,26 These complementary tools maximise the advantages of DK crush and highlight its role as the most effective strategy for complex LMBL in contemporary PCI practice.

Clinical Implications and Future Directions

From a clinical perspective, these results suggest that DK crush should be considered the preferred strategy for complex LMBL, particularly when the side branch supplies a large myocardial territory or has significant disease. Provisional stenting remains a reasonable approach in less complex anatomies, aligning with current guideline recommendations. The consistent inferiority of culotte and T/TAP techniques underscores the need for caution when selecting these strategies, especially in high-risk patients. However, this does not mean that these techniques should be abandoned, considering that the results of the analysis show that the increase in risk after intervention is not that significant when compared to DK crush or provisional stenting. Therefore, other factors such as cost, operator experience, patient comorbidity, bifurcation angle and others can still be considered.

Looking ahead, further RCTs directly comparing DK crush with other two-stent techniques are warranted to validate these findings in broader populations and with longer follow-up. Future research should also explore the role of intracoronary imaging guidance, lesion complexity (for example, based on SYNTAX score or Medina classification) and operator expertise in tailoring stenting strategy. Economic evaluations and technical feasibility studies may help integrate DK crush into routine practice, ensuring both clinical benefit and procedural sustainability.

Strengths and Limitations

This study has several strengths. It is among the first NMA to integrate both direct and indirect comparisons of all major stenting techniques for LMBL, registered prospectively in PROSPERO and adhering to PRISMA-NMA guidelines. The analysis incorporated a broad range of study designs, employed validated tools for risk of bias assessment, and used CINeMA to evaluate confidence in the evidence. The consistency of results across ranking, forest plots and sensitivity analyses enhances the robustness of the conclusions. Collectively, these findings reinforce the reliability of the pooled estimates derived from both network models. Funnel plots suggest that the included studies are unlikely to suffer from serious publication or reporting bias, thereby reinforcing the robustness of the network meta-analysis findings. Nonetheless, it should be noted that the power of funnel plots is limited when the number of included studies per comparison is small.

Nonetheless, some limitations should be acknowledged. A considerable proportion of the included studies were observational, introducing potential residual confounding despite adjustment. The number of trials for certain techniques such as T/TAP and V-stenting was relatively small, reducing precision in these comparisons. Variation in definitions of MACE and heterogeneity in follow-up durations may also affect comparability. Furthermore, several inclusion studies for model 2 did not separate data for different stenting strategies. The inclusion of a mixture of RCTs and observational studies represents a key limitation of this analysis, as it may increase the risk of bias and residual confounding. Consequently, the results should be interpreted with caution and future NMA restricted to RCT data are warranted to validate these conclusions. These limitations highlight the need for cautious interpretation and underscore the importance of future high-quality RCTs.

Conclusion

This NMA demonstrates that DK crush consistently provides superior outcomes for patients with LMBL, significantly reducing MACE and TLR compared with provisional and other two-stent strategies. Provisional stenting remains a viable option for simpler lesions, but techniques such as culotte and T/TAP may be associated with higher risks. However, these techniques should not be completely abandoned, as the increased risk is not markedly higher compared to DK crush or provisional stenting and factors such as cost, operator experience, patient comorbidities and bifurcation anatomy should also be considered.

Click here to view Supplementary Material.