Author + information
- Received June 28, 2004
- Revision received November 1, 2004
- Accepted November 15, 2004
- Published online April 19, 2005.
- Stephen G. Ellis, MD⁎,⁎ (, )
- Jeffrey J. Popma, MD†,
- John M. Lasala, MD, PhD‡,
- Joerg J. Koglin, MD§,
- David A. Cox, MD∥,
- James Hermiller, MD¶,
- Charles O’Shaughnessy, MD#,
- James Tift Mann, MD⁎⁎,
- Mark Turco, MD††,
- Ronald Caputo, MD‡‡,
- Patrick Bergin, MD§§,
- Joel Greenberg, MD∥∥ and
- Gregg W. Stone, MD¶¶,##
- ↵⁎Reprint requests and correspondence:
Dr. Stephen G. Ellis, Department of Cardiovascular Medicine, The Cleveland Clinic Foundation, 9500 Euclid Avenue, Desk F25, Cleveland, Ohio 44195.
Objectives We sought to evaluate the relationship between angiographic late loss and clinical outcomes in the drug-eluting stent era.
Background The interrelationship between angiographic late loss, binary restenosis, and clinical recurrence (target lesion revascularization [TLR]) after coronary stent implantation has been incompletely evaluated.
Methods Using the angiographic substudy of the TAXUS-IV trial, in which 1,314 patients with de novo coronary lesions were randomized to either the paclitaxel-eluting TAXUS stent or to its bare-metal equivalent, we defined the relationship between in-stent and analysis segment late loss, the shape of the late loss histogram (variance and skewedness), and nine-month TLR.
Results Late loss by several measures was closely related to TLR (area under the receiver-operator curve >0.90). For individual vessels of the size in this study (2.8 ± 0.5 mm), the likelihood of TLR did not exceed 5% until analysis segment late loss was >0.5 mm, and did not exceed 10% until late loss was >0.65 mm. At greater late losses, the late loss TLR relationship was steep and nearly linear. For the overall patient cohort, the rate of TLR was related, however, not only to median late loss, but also to measures of its statistical distribution (TLR increased with lack of homogeneous biologic response [greater variance and greater right skewedness]). Similar relationships held for late loss measured within the confines of the stent itself.
Conclusions Coronary stents result in large lumens with “room” to accommodate up to ∼0.5 to 0.65 mm of tissue (angiographic analysis segment late loss) before the likelihood of clinical restenosis (TLR) exceeds 5% to 10%. These data have important implications toward understanding the absolute and relative efficacy of drug-eluting stents.
The aim of drug-eluting stents is to provide local vascular drug delivery to reduce neointimal responses that lead to restenosis and the need for further revascularization. Compared with bare-metal stents, polymer-based elution of sirolimus and paclitaxel significantly reduce the need for target lesion revascularization (TLR) (1,2). For bare-metal stents, the restenotic process is heterogeneous, occurring focally or diffusely within the stent(s), including the edges (3). In contrast, when restenosis occurs after placement of drug-eluting stents, the pattern is typically focal in nature (1,2). Quantitative angiographic algorithms and analysis may be utilized to identify the most narrowed lumen or the minimum lumen diameter (MLD) to reflect the severity of restenosis. The MLD is then used to derive most common angiographic indexes (binary restenosis rates, % diameter stenosis, late loss, and late loss index) reflecting the process of restenosis (4).
Late loss, defined as the difference in millimeters between the MLD at the completion of the procedure and at angiographic follow-up, is commonly used to measure the degree of accumulation of tissue (4). Late loss can be calculated based on measurements within the stent, at its edges, or across the entire analysis segment, as has been done historically. Some have speculated that in-stent late loss might serve as a useful measure of biological activity of drug-eluting stents, with the implication clearly being that “less is better” (5,6). In the era of drug-eluting stents, the significance and relationship of the traditional angiographic indexes of restenosis, especially late loss, to clinical outcomes has yet to be defined.
When comparing results from the SIRIUS and TAXUS-IV trials (1,2), both drug-eluting stent systems provide significant reductions in in-stent neointimal hyperplasia compared to bare-metal stents, translating into a greatly reduced and generally similar need for repeat intervention (nine-month TLR rates of 4.1% and 3.0%, respectively) in analogous patient and lesion populations. One major difference between the Cypher (Cordis Corp., Miami, Florida) and TAXUS (Boston Scientific Corp., Natick, Massachusetts) stent systems is the absolute amount of angiographic late loss seen at late follow-up (the same angiographic core laboratory and analysis technique was utilized in both studies). For the Cypher stent, in-stent late loss was 0.17 ± 0.44 mm compared with 1.00 ± 0.70 mm for bare-metal control stents (relative reduction of 83%). For the TAXUS stent, the late loss was 0.39 ± 0.50 mm for the slow-release, polymer-based paclitaxel-eluting stent compared to 0.92 ± 0.58 mm for the EXPRESS bare-metal control (Boston Scientific Corp.) (relative reduction of 58%). The degree of late loss over the entire analysis segment, however, was similar between the Cypher and TAXUS stents (0.24 ± 0.47 mm and 0.23 ± 0.44 mm, respectively). These contrasts have raised questions about the utility of late loss as an index of clinical restenosis in the drug-eluting stent era. The objective of this substudy was, therefore, to examine the relationship between late loss and TLR.
Study population and protocol
The TAXUS-IV trial randomized patients to receive either the polymer-based, paclitaxel-eluting TAXUS stent or to a bare-metal equivalent. Patients ≥18 years of age with stable or unstable angina or provocable ischemia undergoing stenting of a single de novo lesion in a native coronary artery were considered for enrollment. Angiographic eligibility for inclusion required a target lesion with visual reference vessel diameter ≥2.5 to ≤3.75 mm and lesion length 10 to 28 mm coverable by a single study stent. Clinical and angiographic exclusion criteria have been described (2). The institutional review board at each participating center approved the study, and consecutive, eligible patients signed informed, written consent; 559 of the 732 patients (76.4%) pre-selected for routine angiographic follow-up returned for restudy. Additionally, 150 patients had angiographic follow-up one to nine months after stent placement for clinical indications. The latter group’s data were added in some analyses to increase the number of events to be studied.
Randomization and stent implantation
Telephone randomization was performed before pre-dilatation, stratified by the presence of medically treated diabetes and vessel size (<3.0 or ≥3.0 mm). Patients were equally assigned in double-blind fashion to treatment with either the slow rate-release, polymer-based, paclitaxel-eluting TAXUS stent or a visually indistinguishable bare-metal EXPRESS stent. Unfractionated heparin was administered per standard practice, and glycoprotein IIb/IIIa inhibitor use was at operator discretion. After mandatory pre-dilatation, an appropriate-sized stent (approximately 2 to 4 mm longer than the lesion, with a stent-to-distal reference vessel diameter ratio of 1 to 1.1:1) was implanted at ≥12 atm. Stents were available in lengths of 16, 24, and 32 mm, and in diameters of 2.5, 3.0, and 3.5 mm. Additional study stents were permitted for edge dissections greater than or equal to type B or otherwise suboptimal results, and post-dilatation was at operator discretion. Clinical follow-up was scheduled at one, four, and nine months, and yearly thereafter for five years. Angiographic follow-up was pre-specified in a prospectively identified subgroup of 732 patients at nine months (2).
End points and definitions
Target lesion revascularization was defined as either repeat percutaneous or surgical revascularization for a lesion anywhere within the stent or the 5-mm borders proximal or distal to the stent. Target lesion revascularization was considered to be ischemia-driven if the target lesion diameter stenosis was ≥50% by quantitative analysis with either electrocardiographic changes at rest or a positive functional study in the distribution of the target lesion, or ≥70% with recurrent symptoms only. If an adverse event could not be conclusively attributed to a non-target lesion, then the event was considered a target-related event.
As previously described, the primary end point of the TAXUS-IV trial was TVR at nine months (2), defined as revascularization due to either restenosis in the target lesion, or to a new remote lesion elsewhere in the target vessel or its branches. For the present analysis, however, TLR was used as the principal clinical analysis end point, as angiographic restenosis after stent implantation (either within the stent or at its margins) is most directly correlated with TLR.
After administration of intracoronary nitroglycerin, standard angiographic image acquisition of the coronary stenosis was performed using at least two angiographic projections that were repeated at the end of the procedure and at the time of follow-up angiography. Cineangiograms were then forwarded to the Brigham and Women’s Hospital Angiographic Core Laboratory for review by observers blinded to the treatment assignment. Baseline, postprocedural, and follow-up qualitative morphologic characteristics were characterized using standard criteria (7–9). Lesion length was defined as the axial extent of the lesion that contained a shoulder-to-shoulder lumen reduction by ≥20% or more (10).
Using the contrast-filled injection catheter as the calibration source, quantitative coronary angiographic (QCA) analysis was performed using a validated automated edge detection algorithm (Medis CMS, Leiden, the Netherlands) (11). Projections for image analysis were identified using views that demonstrated the stenosis in an unforeshortened view, minimized the degree of vessel overlap, and displayed the stenosis in its “sharpest and tightest” view. A 5-mm segment of reference diameter proximal and distal to the stenosis was used to calculate the average reference vessel diameter; side branches and other anatomic landmarks were used to identify and maintain the consistency of the measurement length during the follow-up period. Minimum lumen diameters were measured at these same time points within the stent (in-stent analysis) and within the 5-mm proximal and distal edges of the stent. Total occlusions were assigned an MLD = 0 mm and a 100% diameter stenosis.
Angiographic follow-up was performed nine months after the index procedure, or earlier in the event of recurrent symptoms. Binary angiographic restenosis was defined as a follow-up diameter stenosis >50%. Acute gain was defined as the MLD immediately after the procedure minus the MLD before the procedure, and late loss was defined as the MLD immediately after the procedure minus the MLD at follow-up. For the purposes of this analyses, late loss was calculated in each of three ways: 1) within the stent itself; 2) within the analysis segment itself considering the MLD anywhere within the analysis segment at the conclusion of the procedure and at follow-up; and 3) within the analysis segment itself, but separately considering the stented segment, proximal and distal edges and taking the maximum change in MLD within those three segments and applying it to this segment as a whole (maximal regional late loss) to better reflect local dimensional changes.
Categorical variables were compared by the Fisher exact test. Continuous variables are presented as mean ± 1 SD or median with 25% and 75% interquartile ranges, and were compared by Student ttest. Late loss data by treatment group were displayed as histograms and analyzed for variance, or the dispersion of the distribution of data around the mean value, and skewedness or the asymmetry of the data distribution and for its relation to TLR. Cumulative frequency distribution curves for each of the three measures of late loss (in-stent, in-segment, and maximal regional late loss) was plotted against TLR, and the goodness of correlation assessed using receiver operator curve analyses and the c-statistic. Separate analyses were then performed dividing the population by reference vessel size <2.5 mm, 2.5 to 3.0 mm, and >3.0 mm. To assess the TLR implications of greater and lesser heterogeneity and greater right skewedness of late loss than was actually present in the TAXUS group, simulation exercises were performed assuming: 1) the same mean late loss with a Gaussian distribution, but with 50%, 75%, 125%, or 150% of the observed variance of late loss; and 2) skewed distributions, with the same mean late loss but kurtoses of 0.37 and 0.41 representing greater right skewedness; 1,000 simulations were performed for each estimate. Finally, because not all patients with binary restenosis had TLR, similar analyses were performed with binary restenosis as the end point.
Baseline demographic and angiographic findings
Selected patient characteristics and outcomes are shown for patients with and without angiographic follow-up in Table 1,and for patients randomized to the paclitaxel-eluting or control stents within the angiographic follow-up group in Table 2.Patients scheduled for angiographic follow-up differed from those not prescheduled only with regard to lesion length (given that angiographic follow-up was mandated in all patients with 32-mm stent implantation) (6) and with respect to binary restenosis (higher in the non-prescheduled group due to the ascertainment bias engendered by the propensity of patients with symptoms to return for angiography). Patients prescheduled for angiographic follow-up returning for study had a mean reference vessel diameter of 2.78 ± 0.48 mm and lesion length 13.9 ± 6.6 mm, with no significant differences between patients randomized to paclitaxel-eluting or control with respect to these or other key baseline variables. Compared to patients randomized to a bare-metal stent, paclitaxel-eluting randomized patients had significantly lower analysis segment binary restenosis rates (7.9% vs. 26.6%, p < 0.0001) and TLR rates (3.8% vs. 14.6%, p < 0.0001). Only two and three patients, respectively, had total stent occlusion at follow-up.
Distribution of late loss
Distributions of late loss for the paclitaxel-eluting and control groups are shown in Figures 1Ato 1C. Mean analysis segment late loss in paclitaxel-eluting group compared to the control group was significantly less (0.23 ± 0.44 vs. 0.61 ± 0.57, p < 0.0001), less heterogeneous by variance measurement (0.192 vs. 0.329), but less symmetrically distributed (right skewedness 1.54 vs. 0.71). Data for in-stent and maximal regional late loss were similar.
Cumulative frequency distribution curves
Patient counts for those with and without TLR by 0.2-mm late loss increments are shown in Figures 2Ato 2C. The resultant cumulative frequency distribution curves showing TLR rates as a function of late loss are shown for late loss calculated within the in-stent segment (Fig. 3A),analysis segment including the stent and the regions extending 5 mm from the stent margin (Fig. 3B) and maximal regional late loss (Fig. 3C). As the late loss-TLR curves for the patients treated with paclitaxel-eluting and control stents were very similar, combined data were used for subsequent analyses. C-statistics for the in-stent, analysis segment, and maximal regional late loss TLR curves were 0.918, 0.925, and 0.934, respectively (p = NS for between group comparison). Hosmer and Lemeshow TLR goodness of fit statistics were 5.033 (p = 0.75), 8.093 (p = 0.42), and 7.245 (p = 0.51), respectively. For binary restenosis goodness of fit statistics were less consistently good: 48.312 (p = 0.0001), 6.167 (p = 0.62), and 20.334 (p = 0.009), respectively. The inflection point and threshold for >5% and >10% risk of TLR varied across the methodologies, being approximately 0.75 and 1.0 mm, 0.5 and 0.65 mm, and 1.1 and 1.2 mm for the in-stent, analysis segment, and maximal regional late loss analyses, respectively.
Figure 4shows cumulative frequency distribution curves for analysis segment late loss versus TLR for patients with reference vessel <2.5 mm, 2.5 to 3.0 mm, and >3.0 mm. Curves for the smaller and larger vessels are shifted leftward ∼0.3 mm and rightward ∼0.7 mm, respectively.
TLR implications of altering the frequency distribution of late loss in the TAXUS group
As displayed in Table 3,greater homogeneity of the late loss response (e.g., 50% of observed variance) would be associated with lower rates of TLR (1.9%) than that seen with the observed variance. Conversely, greater heterogeneity of late loss response (e.g., 150% of observed variance) would be associated with higher rates of TLR (7.4%). Similarly, greater right skewedness of the data distribution would be expected to be associated with higher rates of TLR (Table 4).
The principal findings of this late loss analysis from the TAXUS-IV trial for de novo native vessel coronary lesions are: 1) late loss is strongly (c-statistic >0.90) but not linearly correlated to TLR, and this correlation is preserved for both paclitaxel-eluting and bare-metal stents, and for late loss calculated as in-stent or either of the analysis segment definitions (2). Individual patients have a very low probability of TLR (<5% to 10%) until analysis segment late loss exceeds 0.5 to 0.65 mm (or in-stent late loss exceeds 0.75 to 1.0 mm), and the probability of TLR exceeds 50% only when analysis segment and in-stent late loss exceeds 1.25 mm and 1.5 mm, respectively (3). In addition to the absolute amount of late loss, the risk of TLR is also related to the homogeneity of late loss effect (the lower the variance and skewedness, the lower the TLR rate).
Late loss clinical implications
The clinical translation of this finding is that the large lumens that result after coronary stent implantation in vessels of the size studied (mean 2.8 ± 0.5 mm) allow “room” to accommodate a mean late loss of up to approximately 0.75 to 1.0 mm within the stent, providing that there is a homogeneous effect across the major subpopulations of patients. When the entire analysis segment is considered, any degree of late loss ≤0.5 mm is sufficient to result in TLR rates below 5%. For smaller reference vessel dimensions, the amount of “room” to accommodate restenotic tissue diminishes, such that for the group with diameter <2.5 (mean 2.22 ± 0.20 mm), the probability of TLR exceeds 10% when analysis segment late loss is >0.5 mm, rather than 0.6 to 0.7 mm for the entire group. When the TLR threshold is lowered to 5% for small vessels (<2.5 mm), a late lumen loss of 0.3 to 0.4 mm or below is acceptable.
Late loss findings in paclitaxel-eluting stents
For the paclitaxel-eluting stent group as a whole, the likelihood of TLR is directly, but not linearly, related to late loss and increases with greater heterogeneity of effect (more variance) and with greater right skewedness of the late loss distribution curve compared with the bare-metal stent (2). These data have important implications for the requisite performance of drug-eluting stents. A mean analysis segment late loss of 0.5 mm (or in-stent late loss of 0.75 mm) after drug-eluting stent implantation is adequate to achieve TLR rates <5%. Greater reduction of late loss may not translate into significantly lower TLR rates, because the relatively flat portion of the TLR/late loss curve has been reached. These data provide insight as to why Cypher and TAXUS stents result in similar TLR rates despite exhibiting very different degrees of in-stent late loss. Moreover, the similar amount of late loss in the analysis segments after sirolimus-eluting and paclitaxel-eluting stent implantation (despite different degrees of in-stent hyperplasia) contribute to the near identical rates of TLR.
In addition, the homogeneity of response to the drug-eluting stent as a function of vessel size, lesion length, diabetic status, and other parameters must be considered to completely characterize drug-eluting stent performance. In this regard from the present analysis, the TAXUS stent performed slightly better in smaller compared to larger vessels, with less absolute and relative late loss. Moreover, bare-metal stents typically have a late loss of 0.8 to 1.0 mm (12–14). The present analysis suggests that changes in materials or manufacturing processes that could reduce this late loss to the 0.5- to 0.6-mm range (a reduction of only ∼0.2 to 0.4 mm) would have a marked impact on reducing restenosis even without a bioactive coating.
This study also demonstrates that when analysis segment late loss is <∼0.5 mm (or when in-stent late loss is <∼0.75 mm), that the homogeneity of response (variance and skew) may affect TLR rates more powerfully than any further reduction in median late loss.
Late loss findings in bare-metal stents
For bare-metal stents, a majority of patients have late loss on the steep and linear portion of the TLR/late loss curve, and, therefore, late loss itself is a good measure of clinical benefit. In contrast, for late losses in the ranges expected with drug-eluting stents, the exact amount of late loss is an insensitive determinant of clinical restenosis, with homogeneity of effect being a more important predictor of clinical benefit.
The principal limitations of this analysis are that much of the angiographic follow-up was per protocol rather then clinically driven, hence TLR may have been artificially exaggerated due to the “oculostenotic reflex” (despite attempts to systematically adjudicate against this) (15). Second, the results may not be generalizable to all stent platforms or other types of drug-eluting stents or other lesion types. Conversely, the strengths of the study lie in the large number of patients studied, the blinded, independent QCA process, the consistency of the TLR late loss relationship for both the paclitaxel-eluting stent and bare-metal stent groups, and of the various measures of late loss to that relationship.
In conclusion, considering the salutary scaffolding effects of stents when implanted in de novo coronary artery stenoses, median analysis segment late loss of up to 0.5 to 0.65 mm (or in-stent late loss of 0.75 to 1.0 mm) may be accommodated with probability of TLR <5% to 10% provided that there is homogeneous effect across all subpopulations.
For the slow-release paclitaxel formulation stents, the median (0.15 mm) and interquartile range (−0.04 to 0.41 mm) values of analysis segment late loss fall far below the threshold where the probability of TLR increases. These findings are important in selecting drug candidates and dose thresholds for future drug-eluting stents where the antirestenotic effect needs to be balanced against adequate stent coverage consistent with a well-healed and pacified surface. Results from future drug-eluting stent studies evaluating other stent platforms and drugs, as well as a broader range of stent diameters and lengths and other lesion types, will provide more insight to prospectively determine the optimum amount of late loss combining sufficient antirestenotic properties and adequate healing after stent implantation.
Finally, these results potentially have important clinical and regulatory implications for ongoing trials in which drug-eluting stents are compared against each other, using angiographic late loss as a surrogate of clinical benefit. The present analysis suggests that a relatively broad “delta” for noninferiority in such studies would provide reasonable assurance of clinical efficacy, as long as variance of the distribution is not excessive.
Drs. Ellis and Popma receive research support from Boston Scientific; Drs. Ellis, Stone, and Hermiller are consultants for Boston Scientific; and Dr. Koglin is an employee of Boston Scientific. The data reside with, but were also reviewed externally from, Boston Scientific. Dr. Ellis feels as an investigator in the filed, he has “conflict” with all major U.S. stent manufacturers, and hence feels “conflicted” in a “balanced” fashion.
- Abbreviations and acronyms
- minimum lumen diameter
- quantitative coronary angiography
- target lesion revascularization
- Received June 28, 2004.
- Revision received November 1, 2004.
- Accepted November 15, 2004.
- American College of Cardiology Foundation
- Hoffman R.,
- Mintz G.S.,
- Dussaillant G.R.,
- et al.
- Cutlip D.,
- Chauhan M.,
- Baim D.,
- et al.
- Babapulle M.,
- Eisenberg M.
- Ellis S.,
- Vandormael M.,
- Cowley M.,
- et al.
- Ryan T.,
- Faxon D.,
- Gunnar R.,
- et al.
- Mehran R.,
- Dangas G.,
- Abizaid A.,
- et al.
- van der Zwet P.,
- Reiber J.
- ten Berg J.,
- Kelder J.,
- Suttorp M.,
- Verheugt F.,
- Plokker H.