Delayed Coverage in Malapposed and Side-Branch Struts With Respect to Well-Apposed Struts in Drug-Eluting Stents: In Vivo Assessment With Optical Coherence Tomography

Data at follow-up from the OCT substudies of 2 different randomized trials were analyzed: the Limus Eluted From a Durable Versus E r odable Stent Coating (LEADERS) trial,^22,23 comparing a biolimus-eluting stent with biodegradable polymer in abluminal coating (BioMatrix Flex, Biosensors International, Morges, Switzerland) versus a sirolimus-eluting stent with durable polymer (Cypher SELECT, Cordis, Miami Lakes, FL); and the RESOLUTE–All Comers trial,²⁴ comparing a zotarolimus-eluting stent with hydrophilic polymer coating (Resolute, Medtronic Inc, Santa Rosa, CA) versus an everolimus-eluting stent with fluoropolymer (Xience V, Abbott Vascular, Santa Clara, CA). The design and results of these trials have been published elsewhere.^22–24 Both trials followed an all-comers design, with minimal exclusion criteria. In LEADERS, the OCT follow-up was scheduled at 9 months, whereas in RESOLUTE–All Comers, it was at 13 months.

OCT pullbacks were obtained at follow-up with M2, M3, or C7 systems (Lightlab Imaging, Westford, MA), according to the availability at the participating sites, with an occlusive or nonocclusive technique as appropriate.²⁵ Table 1 summarizes the technical specifications of each OCT system and optical catheters. Infusion of intracoronary nitroglycerin before the OCT pullback was strongly encouraged.

OCT pullbacks were analyzed offline in a core laboratory (Cardialysis BV, Rotterdam, Netherlands) by independent staff blinded to stent-type allocation and clinical and procedural characteristics of the patients, with the use of proprietary software (Lightlab Imaging). Cross sections at 1-mm intervals within the stented segment and 5-mm proximal and distal to the stent edges were analyzed. A metallic strut typically appears as a bright signal-intense structure with dorsal shadowing. Apposition was assessed strut by strut by measuring the distance between the strut marker and the lumen contour. The marker of each strut was placed at the endoluminal leading edge, in the midpoint of its long axis, and the distance was measured following a straight line connecting this marker with the center of gravity of the vessel.²⁶ Struts were classified as ISA if the distance between the strut marker and the lumen contour was larger than the specific strut thickness plus the axial resolution of OCT (14 μm). This resulted in ISA thresholds of >168 μm for sirolimus-eluting stents, >131 μm for biolimus-eluting stents, >99 μm for everolimus-eluting stents, 111 μm for zotarolimus-eluting stents, and 95 μm for the combination drug-coated balloon plus bare-metal stents. Struts located at the ostium of side branches, with no vessel wall behind, were labeled as nonapposed side-branch struts and excluded from the analysis of apposition.

Struts were classified as uncovered if any part of the strut was visibly exposed to the lumen or were classified as covered if a layer of tissue was visible over all of the reflecting surfaces. In covered struts, the thickness of coverage was measured from the marker of each visible strut to the endoluminal edge of the tissue coverage, following a straight line connecting the strut marker with the center of gravity of the vessel.^18–21,23

Each strut was finally categorized according to apposition as well-apposed, ISA, or nonapposed side-branch strut and according to coverage as covered or uncovered (Figures 1 and 2).

The risk ratio (RR) for noncoverage of ISA and nonapposed side-branch strut was calculated versus well-apposed struts by pooled analysis with the use of an inverse variance random-effects model, taking into account the between-cluster and within-cluster variability, with each stent used as an independent unit of clustering (see Methods in the online-only Data Supplement). For each comparison between apposition categories, stents with zero struts in any of the compared apposition categories (no exposition) or zero uncovered struts (no events) were considered to be not informative to evaluate the RR and were discarded for the analysis. A proportional continuity correction was applied to stents with zero uncovered struts (zero events) in only 1 of the compared apposition categories.²⁷ The RR at each individual stent and the pooled RR of the whole sample were graphically represented by mean of forest plots. Analysis of heterogeneity of the effect was performed with the Q, H, and I² parameters, with consideration of the presence of heterogeneity when the P value of the test was ≤0.1. Calculations were done with PASW 17.0 software (Chicago, IL) with the use of a specific macro (Domenech JM).²⁸ The RR for noncoverage of ISA versus nonapposed side-branch struts was calculated by the same method.

Table 5 shows the total count of struts in each apposition category in the whole sample and the raw proportions of coverage without clustering by patient, lesion, or stent. We found that 94.9% of well-apposed struts are covered at follow-up versus only 78.1% of nonapposed side-branch struts and 27.4% of ISA struts.

Seventy-three stents (41.0%) had nonapposed side-branch struts; 64 (36.0%) had ≥1 ISA strut, and 127 (71.3%) were incompletely covered at follow-up. The presence of nonapposed side-branch struts was significantly associated with incomplete coverage of the stent: 63 stents (86.3%) with nonapposed side-branch struts were incompletely covered compared with 64 (61.0%) without nonapposed side-branch struts (P=0.008). Likewise, the presence of ISA was significantly associated with incomplete coverage of the stent: 62 stents (96.9%) with ISA were incompletely covered compared with 65 totally well-apposed stents (57.0%) (P<0.0001).

One hundred five DES had no nonapposed side-branch struts; 11 DES across side branches had all nonapposed side-branch and well-apposed struts completely covered. Sixty-two DES (13 083 struts) were suitable for the pooled comparison of the risk of noncoverage in nonapposed side-branch (n=216) versus well-apposed struts (n=12 867). Nonapposed side-branch struts present a RR for noncoverage of 9.00 (95% confidence interval, 6.58 to 12.32) compared with well-apposed struts (Table 6 and Figure 5). There was no significant heterogeneity of the effect in any of the subgroups (H<1; I²=0).

One hundred thirteen DES had no ISA struts; 3 DES with malapposition had all ISA and well-apposed struts completely covered. Sixty-two DES (12 558 struts) were suitable for the pooled comparison of the risk of noncoverage in ISA (n=405) versus well-apposed struts (n=12 153). ISA struts present a RR for noncoverage of 9.10 (95% confidence interval, 7.34 to 11.28) compared with well-apposed struts (Table 6 and Figure 6). The magnitude of the effect varied substantially between the different types of DES, ranging from 5.43 in sirolimus-eluting stents to 12.36 in everolimus-eluting stents. There was considerable heterogeneity of the effect, even after stratification by type of stent.

Thirty-three DES (335 struts) were suitable for the pooled comparison of the risk of noncoverage in ISA (n=211) versus nonapposed side-branch struts (n=124). ISA struts present a RR for noncoverage of 1.73 (95% confidence interval, 1.34 to 2.23) compared with nonapposed side-branch struts (Table 6 and Figure 7). There was no significant heterogeneity of the effect in any of the subgroups (H<1; I²=0).

To the best of our knowledge, this is the first human in vivo study addressing the coverage of nonapposed side-branch struts. Most of the recently published OCT studies excluded nonapposed side-branch struts from the analysis^18–21 because they were methodologically cumbersome and their meaning was not fully understood. Our results clarify the impact of nonapposed side-branch struts on the coverage of the stent and may explain the apparent discrepancy between pathological and clinical studies in this respect. Although some pathological studies on patients deceased by stent thrombosis found incomplete neointimal coverage in struts at side branches,^7,9 stenting across continued to be a widespread practice, often the compromise solution for many challenging cases, and late and very late stent thrombosis still remained a rare complication.²⁹ The Nordic trial did not find any significant difference in clinical outcomes between a simple strategy of stenting across and a complex 2-stent technique.^11,30 The Nordic III trial compared specifically the strategy of stenting across (a nonapposed side branch–generating technique) versus systematic final kissing balloon (a nonapposed side branch–reducing technique) and also failed to find any clinical advantage for any of the strategies.¹² The inherent selection bias of autopsy studies has been thought to explain the discrepancy between pathology and clinical data. Our findings may reconcile these apparently contradictory observations: Most nonapposed side-branch struts (78%) appear covered at follow-up, explaining the good results reported in clinical trials; however, the neointimal healing is primarily impaired in nonapposed side-branch struts, and therefore the findings in autopsies may reflect a rare but real problem, which is not properly addressed in trials that were underpowered to detect late and very late stent thrombosis. Therefore, although stenting across is becoming the default technique for bifurcation stenting, it may pose some disadvantage for complete neointimal coverage.

Ozaki et al¹⁷ reported a descriptive sequential OCT study in a sirolimus-eluting stent series, showing that 65% of acutely malapposed struts were uncovered at 6 months, in contrast with only 9% of the well-apposed struts. This observation suggests that neointimal coverage of ISA struts may be delayed with respect to well-apposed struts. Our study confirms the hypothesis generated by the report of Ozaki et al, comparing for the first time the risk of incomplete coverage at follow-up between the different apposition categories within the same stent and therefore comparing struts subjected to identical environmental conditions. This approach provides solid evidence about the effect of apposition on coverage, even though it analyzes only the outcome at follow-up. Although it is generally accepted that sequential studies provide more solid evidence about risk estimation, this statement is not totally accurate in the case of OCT because of the impossibility of tracking individual struts between 2 time points. Sequential OCT studies are based on assumptions and approximations that represent some degree of inaccuracy and bias. Analyzing OCT only at follow-up, we found a raw proportion of uncovered ISA struts (72.6%) in the DES group similar to that reported by Ozaki in sirolimus-eluting stents.¹⁷

Intravascular ultrasound studies have described the association between ISA and late and very late stent thrombosis,^13,14 but the mechanism of this association remains unclear. The demonstration that ISA struts have a higher risk of noncoverage than well-apposed struts strongly suggests that the underlying mechanism explaining this association might be a deficient reendothelialization of ISA struts. This means that ISA is a risk factor for incomplete coverage in DES, and the effort to optimize apposition may be justified.

Inflammation is an alternative mechanism to explain the association between ISA and thrombosis described in first-generation DES.^8,15 Vasculitis of the 3 arterial layers may cause weakening of the vessel wall and aneurysmal dilatation, resulting in late-acquired ISA.³¹ Intravascular ultrasound studies have reported larger in-stent mean external elastic membrane area in cases of very late stent thrombosis than in control DES and also larger than in the corresponding proximal and distal reference segments.¹⁵ In regard to the inflammatory mechanism, ISA would be the consequence of a common primary process causing both thrombosis and ISA, and therefore the optimization of acutely malapposed struts would make little sense. However, the inflammatory mechanism cannot explain our results. First, this analysis stems from an asymptomatic cohort of patients who have not suffered thrombotic events to date, in contrast to prior case-control designs.¹³ Second, the analysis of lumen area in the stents with ISA does not fit well with the aneurysmal pattern described by intravascular ultrasound in the case of inflammatory stent thrombosis.¹³ Finally, and most importantly, the deficient coverage in ISA is also observed in nonapposed side-branch struts, which become nonapposed side-branch struts at the moment of implantation and certainly not as the consequence of a late hypersensitivity reaction.

A third theoretical mechanism to explain the association between ISA and stent thrombosis is confounding: A severely diseased coronary segment would be more prone to a suboptimal apposition result and more prone to suffer thrombotic events, resulting in a spurious association between ISA and thrombosis. In that case, ISA would be a marker of poor prognosis, unlikely to change by optimizing the apposition. Although this possibility cannot be strictly ruled out in our study, the parallelism with the coverage in nonapposed side-branch struts makes this mechanism unlikely to explain the results. When all of these factors are considered together, this study constitutes solid evidence that ISA is a primary risk factor for incomplete neointimal coverage, encouraging the procedural optimization of apposition to the greatest possible extent.

Although both nonapposed side-branch and ISA struts share a higher risk of delayed coverage, they seem to present slightly different scenarios. Acute ISA is caused by a severely diseased vessel or other underlying condition precluding correct apposition. This scenario may be associated with irregular drug release due to contact of the struts with the necrotic core,³² severely dysfunctional endothelium with hampered regeneration capabilities, or more severe structural deformations in the stent platform. These reasons may explain the heterogeneous effect observed in ISA struts. This heterogeneity is only partially explained by the type of stent and deserves further clarification in the future. Conversely, nonapposed side-branch struts are not in contact with plaque, necrotic core, or calcium. Presumably, they are enshrined in a less severely diseased segment of the vessel, and the stent has suffered less structural distortion at that level. Probably this constitutes a more benign and predictable scenario than the one described for ISA and may explain the homogeneous effect observed in nonapposed side-branch struts and the higher risk of noncoverage observed in ISA versus nonapposed side-branch struts.

Research in the field of OCT must face a cumbersome methodological challenge: The level of measurement (the strut) is different from the level of analysis (the patient or, exceptionally, the lesion/stent). The clustering of the units of measurement at different levels (patient, lesion, stent) makes the probability of the outcome not independent, thus violating one of the conditions of using conventional statistics. A mathematical model that takes into account between-cluster and within-cluster variability is required. Bayesian hierarchical²³ or multilevel regression models have therefore been proposed. In this study, we use a pooled analysis for the first time, which fulfills all of the methodological requirements for the analysis of clustered data, with the additional advantage of presenting the results to the biomedical community in a more familiar format. It also allows the representation of the pooled RR as a forest plot, in which the statistical significance of the result (95% confidence interval not crossing the reference line at RR=1), the contribution of each individual stent to the final result, and the heterogeneity of the effect can be easily and intuitively understood.

We advise caution in regard to the use of OCT coverage as a surrogate for neointimal healing. Although biologically plausible and intuitively accepted by the scientific community, this approach cannot be fully supported by current evidence. OCT coverage correlates with histological neointimal healing and endothelialization after stenting in animal models,^33–35 but its sensitivity and specificity in human atherosclerotic vessels are still unknown. OCT is not able to detect thin layers of endothelium below its 14-μm axial resolution, eventually resulting in an overestimation of the incidence of delayed coverage, and it cannot discern between neointima and other material, like fibrin or thrombus. The analysis of optical density may be useful in the future to discern between neointima and thrombus/fibrin.³⁶

This analysis included OCT studies performed at different follow-up periods, ranging from 9 to 13 months. Although this is a potential limitation, it is not likely to significantly affect the results because in each case the follow-up was scheduled after healing was estimated to be complete, and no substantial variations were expected. Moreover, the statistical analysis compared the different apposition categories within the same stent, thus minimizing the impact of uneven follow-up periods in the results. Nonetheless, the level of evidence provided by data pooled from different trials may be lower than the level obtained by a homogeneous protocol of study in a randomized fashion. Despite strict inclusion and exclusion criteria for the OCT substudies, some selection bias of the patients most suitable for imaging cannot be completely ruled out.

Only OCT pullbacks at follow-up were analyzed, under the assumption that the apposition status of each strut remains stable along time. This assumption is acceptable for nonapposed side-branch struts but entails some selection bias for ISA struts. It has been described in OCT sequential studies that the proportion of ISA struts drops from baseline to follow-up because of partial integration in the neointima. The risk of noncoverage therefore may have been slightly overestimated in ISA. The choice for risk and RR in the pooled analysis is also based on this assumption, as well as because mathematically it is a ratio of proportions, which fits better in the analysis performed. Although risk assessment would require the strict calculation of incidence between 2 time points, in OCT it is not possible to match individual struts between 2 different time points. With acknowledgment of its limitations, the assumption that apposition remains stable over time may be acceptable in some situations, and the complementary solution of sequential studies which assume in turn that struts can be tracked or their number does not change between 2 time points.