Interlesion Dependence of the Risk for Restenosis in Patients With Coronary Stent Placement in Multiple Lesions
Background—Little is known about the behavior with regard to restenosis of multiple lesions within the same patient treated with intracoronary stenting. Our objective was to test the hypothesis that there is an intrapatient dependence of restenosis between lesions.
Methods and Results—Quantitative analysis was carried out on angiograms obtained before, immediately after, and at 6 months after coronary stent placement in 1734 lesions in 1244 patients. We used a specialized logistic regression that not only accounts for intraclass correlation but also quantifies it in the form of odds ratio (OR) as the change in risk of a lesion to develop restenosis if another companion lesion had restenosis. The model was based on 23 patient- and lesion-related variables with binary restenosis (diameter stenosis ≥50%) as end point. The overall restenosis rate was 27.5%: 24.4% for single-lesion, 28.6% for double-lesion, and 33.8% for ≥3-lesion interventions. After adjustment for the influence of significant factors (hypercholesterolemia, systemic arterial hypertension, diabetes mellitus, previous PTCA, ostial lesion, location in left anterior descending coronary artery, number of stents placed, vessel size, stenosis severity, balloon-to-vessel ratio, and final result), the analysis found a significant intrapatient correlation, OR 2.5 (1.8 to 3.6). This means that in patients with multilesion interventions, the risk of a lesion to develop restenosis is 2.5 times higher if a companion lesion has restenosis, independently of the presence or absence of analyzed patient risk factors (eg, diabetes).
Conclusions—This study demonstrates that there is a dependence of restenosis between coronary lesions in patients who undergo a multilesion intervention. The likelihood of restenosis for a lesion is higher when another companion lesion has also developed restenosis. Other, as yet unidentified patient factors may be the source of this intrapatient correlation of restenosis.
Prediction of restenosis after coronary interventional procedures has remained remarkably difficult,1 primarily because of a complex underlying process. In addition, several methodological issues on the definition of restenosis are yet unresolved,2 the best method for assessment has not been defined,3 and it is not clear which factors should be analyzed as potential predictors and which statistical method should be used. Moreover, the increasing number of procedures with interventions in more than 1 lesion per patient raises the issue of how to handle these multiple lesions and complicates further efforts to identify predictive factors for restenosis. In several large studies, analyses were performed assuming the independence of lesions in their risk for restenosis.4 5 6 7 This assumption was based on results from a previous study, which found no correlation between lesions of the same patient with regard to late loss 6 months after PTCA, stenting, or atherectomy in a limited number of patients with multilesion interventions.8 In that study, the interrelation of lesions was tested by adoption of a modified form of the general linear model, which is appropriate if the error term is normally distributed. More recent studies have used more sophisticated quantitative angiographic and mathematical analyses. These have found a markedly nonnormal distribution for the principal angiographic measures of restenosis, including late loss.9 10 Under these conditions, the above-mentioned findings may not be applicable. Enormous efforts have been made to identify patient-related characteristics as predictors of restenosis,1 probably guided by the assumption that lesions are not independent in their risk for restenosis. Such an interdependence has been suggested by studies after PTCA in a limited number of patients.11 12 Still, if this interdependence were to be confirmed, multiple-lesion patients must be handled with methods that are able to account for it. An alternative method is to include only single-lesion patients or to choose only 1 lesion per patient in the multilesion patients. Such methods, however, do not use the data efficiently and may introduce a selection bias.13 A proven dependence between lesions would focus future research on potential patient-related predictive factors rather than lesion-specific factors, which currently attract the most attention.
The objective of this study was to test the hypothesis that there is an interlesion dependence of the risk for restenosis in the patient with multilesion coronary stent placement, even after potential patient- and lesion-related risk factors have been accounted for.
Stenting was considered successful if a residual stenosis of <30% was achieved at the end of the procedure. Of 1692 patients in whom stent placement was attempted during the period from May 1992 through June 1996, 116 (6.9%) were not eligible for a 6-month follow-up angiography because of either an unsuccessful procedure (35 patients, or 2.1%) or the occurrence of ≥1 major adverse cardiac events, such as death, postprocedural nonfatal myocardial infarction, or target lesion revascularization procedure (PTCA or/and aortocoronary bypass surgery) during the first month after the procedure (81 patients, or 4.8%). Control angiography was performed in 1244 patients (79% of the 1576 eligible patients) with 1734 lesions (80.3% of the 2160 eligible lesions), which are the subject of the present analysis.
Stent Placement and Postprocedural Follow-up
Slotted-tube stents were implanted as described in detail elsewhere.14 Balloon size and pressure were at the operator’s discretion. Multiple stents were deployed if necessary to cover the full extent of the target lesion or the dissection if one was incurred. Adequacy of the final result was based solely on the angiographic assessment.
For postprocedural therapy, 30.4% of patients received a full anticoagulation regimen composed of heparin for 5 to 10 days and phenprocoumon (Marcumar, Hoffmann–La Roche) for 4 to 6 weeks; 69.6% of patients were treated with combined antiplatelet therapy with ticlopidine 250 mg BID in addition to aspirin 100 mg BID.
A 6-month coronary angiographic control was scheduled for all eligible patients. An earlier restudy was warranted if the patient presented with symptoms of angina. The median interval between the intervention and the control angiography was 188.5 days (interquartile range, 173 to 205 days).
Data Collection and Definitions
Qualitative angiographic assessment was done by the operator during or immediately after the procedure. Angiograms were assessed for the presence of chronic vessel occlusion before PTCA and dissections15 immediately before stent placement. The left main, LAD, left circumflex, and right coronary arteries with all their branches, as well as bypass grafts, were defined as vessel systems for the localization of the lesions. For example, a lesion in the LAD and one in a diagonal branch were both considered to belong to the same vessel system, the LAD.
Quantitative angiographic analysis was performed off-line by use of an automated edge-detection system (CMS, Medis Medical Imaging Systems) by operators not involved in the interventional procedures. This system has an excellent accuracy and precision.16 The contrast-filled, nontapered catheter tip was used for calibration. MLD, RD, percent diameter stenosis, and the diameter of the maximally inflated balloon were obtained from this analysis system. Measurements were done on the angiograms taken before and at the conclusion of a procedure and on that recorded at follow-up. Balloon-to-vessel ratio was calculated as diameter of the inflated balloon divided by the coronary reference diameter. Angiographic restenosis was defined as ≥50% diameter stenosis. Restenosis rate was calculated on both a per-lesion and a per-patient basis. A patient with multilesion intervention was considered to have restenosis if ≥1 dilated lesions presented angiographic restenosis at follow-up.
Potential risk factors for restenosis as previously identified were continuously assessed. Patient characteristics collected were age, sex, cardiovascular risk factors, acute myocardial infarction, unstable angina pectoris, multivessel disease, and history of previous PTCA. We used the latter as a patient-specific rather than a lesion-specific variable, because it was frequently difficult to establish precisely the target vessel or the target lesion of a PTCA procedure performed elsewhere. In addition, information on postprocedural antithrombotic therapy and the type of the intervention (single-lesion or multilesion) was recorded. Specific lesion characteristics noted were chronic occlusion, dissection, and lesion location at a vessel ostium, the LAD, or a bypass graft. Quantitative angiographic data were added to the list of lesion-related variables: RD and MLD before stenting, balloon-to-vessel ratio, maximal balloon pressure, and MLD immediately after stenting. Because mainly stents of 2 different lengths (7 and 15 mm) were used, we adopted the measure of stent unit (1 unit=7 mm).
Dichotomous data are expressed as counts or percent. These variables were compared by χ2 test. Continuous data were initially tested for normal distribution with the Kolmogorov-Smirnov test. Because most of them deviated markedly from such a pattern, descriptive statistics are presented as median (interquartile range), and nonparametric methods were used for analysis (Mann-Whitney U test). All values of P<0.05 were considered statistically significant.
The main analysis assesses the interdependence of lesions within the same patient with respect to their tendency for restenosis, a concept generally known as intraclass or intracluster correlation. Clustermates frequently may respond similarly, because they are not statistically independent. Failure to account for a cluster effect will underestimate standard errors of parameters such as regression coefficients. We used a special adaptation of polychotomous logistic regression that allows for intraclass correlation among lesions.17 18 Because the number of response combinations for all lesions in the same patient may vary, polychotomous (multinomial) instead of ordinary logistic regression has to be applied. Such a model allows adjustments for omitted covariates that are correlated for different cluster members (different lesions), such as patient factors, by absorbing the effects of these covariates in the form of the OR among cluster members (lesions). This OR is defined as (odds in favor of 1 lesion developing restenosis/companion lesion has restenosis)/(odds in favor of 1 lesion developing restenosis/companion lesion has no restenosis). This OR therefore represents the measure of intrapatient correlation. The estimation of the regression coefficients for the covariates takes into account this intrapatient correlation. The intrapatient correlation was also assessed with the eligible patients without angiographic restudy (20% of the entire population) included in the analysis after the outcome (binary restenosis) was randomly imputed. In addition, the influence of potential interactions between factors on intrapatient correlation was evaluated after allowance in the model for interactions between age and diabetes, age and multivessel disease, age and arterial hypertension, vessel size and final MLD after stenting, and balloon-to-vessel ratio and maximal balloon pressure.
There were 891 patients with single-lesion stenting and 353 patients who underwent a multilesion intervention in a total of 843 lesions. The multilesion intervention was performed on different vessel systems in 50% of the latter patients (see “Methods”). Restenosis rate was 31.4% on a per-patient basis and 27.5% on a per-lesion basis. The restenosis rate increased with the number of lesions treated, both per lesion and per patient. In particular, it was 24.4% for single-lesion interventions, 28.6% per lesion for interventions in 2 lesions, and 33.8% for interventions in ≥3 lesions (P<0.003, test for trend). On a per-patient basis, it was 24.4%, 43.6%, and 63.1%, respectively (P<0.001). Table 1⇓ lists the results of the comparative analysis between the group with single-lesion and the group with multilesion interventions. Patients in the multilesion group had previous PTCA significantly more often and, as expected, a higher incidence of multivessel disease. Lesions of this group were more often located at an ostial position and had a lower incidence of dissections. Furthermore, they were treated with fewer stent units and had a smaller RD and larger MLD before the procedure and a smaller final MLD after stenting.
Patient and lesion characteristics of the group with and that without restenosis are illustrated in Table 2⇓. All variables listed in this table were entered into the logistic regression model for binary restenosis, which accounted simultaneously for intraclass (intrapatient) correlation. The model χ2 was 153, which is highly significant for 23 df (P<0.001). Most importantly, there was a significant intrapatient correlation (Figure 1⇓), as demonstrated by an OR of 2.5 (95% CI, 1.8 to 3.6). An OR of 1 would suggest independence of restenosis between the lesions of the same patient. In addition, the analysis identified several patient and lesion variables as significant predictors of restenosis (Figure 1⇓). The intrapatient correlation was hardly influenced by the inclusion in the analysis of patients without angiographic restudy or potential interactions between covariates (see “Methods”): the respective OR varied slightly between 2.3 and 2.6 in each case.
As observed above (Table 1⇑), the group with multilesion interventions differed significantly in several characteristics compared with the group with single-lesion intervention. The divergent characteristics were also tested in the logistic model for predictive factors (Table 2⇑ and Figure 1⇑). Compared with the single-lesion group, the multilesion group had a higher incidence of previous PTCA and ostial lesions, a smaller RD before stenting, and a smaller MLD at the end of the procedure. All these factors had also been identified as risk factors for restenosis by the multivariate model. However, the multilesion group also had fewer stent units per lesion and a larger MLD before stenting, which were both protective against restenosis. Furthermore, multilesion intervention per se did not constitute a significant independent factor for restenosis in multivariate analysis (P=0.7, Table 2⇑). Thus, the group with multilesion interventions may not be considered a high-risk group on the basis of the analyzed factors.
Assuming independence of restenosis between the lesions, the restenosis rate of the single-lesion patients should allow prediction of restenosis rate at 0, 1, 2, and 3 lesions in the patients with 2- and 3-lesion interventions by use of simple formulas of probability calculations. Assuming independence, the observed rates would not differ from those predicted. In an analysis of all patients with interventions in 2 or 3 lesions (n=328), we predicted the incidence of restenosis in 0, 1, and 2 or 3 lesions and compared the predicted with the observed values. The results of this analysis are displayed in Figure 2⇓. Although the predicted value did not differ from the observed number of patients with no restenotic lesions, the observed number of patients with restenosis in 2 or 3 lesions at the same time is more than double that predicted (P<0.001), at the cost of a significantly lower number of observed patients with restenosis in only 1 lesion with respect to the value predicted (P<0.02).
The number of multilesion catheter interventions has increased markedly during the past few years. This raises the issue of how to methodologically address multilesion patients in studies searching for predictive factors of restenosis. Analyzing data on a patient-specific basis by selecting the worse (tighter) stenosis at follow-up angiography introduces a selection bias and does not make efficient use of the available data, especially those related to lesion-specific factors. Random selection of only 1 lesion per patient attenuates the selection bias but does not resolve the inefficient use of data. An alternative approach would be to average the response of all lesions in 1 patient. However, binary responses cannot be averaged, and this method would exclude lesion-specific covariates from the analysis. A lesion-based analysis would make full use of the data if the problem of the potential correlation of restenosis between the lesions of the same patient is resolved appropriately. Ignoring this correlation would lead to too small standard errors for the estimated parameters, and significance might falsely be reported. If such an intrapatient correlation did not exist, the lesion-based analysis is well motivated. In fact, Gibson et al8 failed to find a similar correlation in 67 patients who had multilesion interventions in 146 lesions with conventional PTCA, directional coronary atherectomy, or stenting. These findings, however, are qualified by the small number of patients and lesions, by the different interventional methods, and by the lack of an automated quantitative system for coronary artery measurements. Shortly after that, Weintraub et al12 concluded that restenosis is an interdependent process between the lesions of the same patient. This conclusion was based on findings that patients clustered with either 0 or 2 restenotic PTCA sites.12 This study, however, had a very low angiographic restudy rate (≈40%), used only caliper measurements of coronary arterial dimensions instead of an automated quantitative angiographic system, and used statistical methods that did not account for intracluster correlation.
We designed this study to test the hypothesis that there is a dependence of restenosis between lesions of multilesion patients. This had been stimulated by previous studies identifying several patient characteristics as correlates of restenosis4 19 20 and by an impression from the daily practice that some patients return with restenosis in most or even all of the lesions dilated a few months previously. This analysis is based on a remarkable angiographic restudy rate of 80%, a well-validated automated quantitative angiographic system, and statistical models that account for possible cluster effects in the study population. We used binary restenosis as an outcome measure that not only inherits a solid historical basis but also is legitimized by the bimodal distribution pattern of restenosis after coronary interventions.9 10 21 Our study demonstrates that even after the clustering influence of several patient-specific variables is excluded, there is a significant correlation of restenosis between lesions of the same patient. In patients with a multiple-lesion intervention, the risk of a lesion to develop restenosis also depended on the status of the companion lesions: it was more than twice as high if another lesion had a restenosis. This led to an increase of the number of patients with multiple-lesion restenosis and to a reduction of those with single-lesion restenosis compared with what was expected if lesion-to-lesion independence of this complication was assumed. The multilesion group presented with a number of differences in patient and lesion characteristics, some of which emerged as risk factors and others as protective factors from the logistic regression model for intraclass correlation. However, the fact that a multilesion intervention had been performed did not constitute per se an independent risk for restenosis. Therefore, our data do not support a strategy avoiding interventions on multiple lesions.
There are 2 major implications of this study. First, because of the presence of an intrapatient correlation of restenosis in patients with intervention in multiple lesions, it is obligatory for every lesion-based analysis of predictive factors to account for this correlation in the future. Current bootstrapping techniques may achieve this goal.22 Second, the intrapatient correlation suggests that there are additional unknown patient-specific factors that play a role in the process of restenosis and were not included in the analysis. Future studies should pay major attention to the assessment of these factors, which may substantially improve the predictive power of the multivariate models for restenosis. More recent data suggest that genetic23 24 25 and infectious26 27 factors may increase the risk of restenosis in patients undergoing catheter-based coronary interventions.
The focus of this study was to assess the dependence of restenosis between lesions of the same patient. This kind of analysis required us to account for potential patient and lesion covariates. The major finding of this study, that a significant lesion-to-lesion dependence exists, makes difficult the comparison of the results of the predictive factor analysis with those of previous reports in which interlesion dependence of restenosis was not taken into consideration. Nevertheless, a point that deserves special comment is the present finding that hypercholesterolemia was associated with decreased risk for restenosis. Several limitations must be considered before one may draw conclusions from this result. A diagnosis of hypercholesterolemia was based on the serum level at the time of the intervention. Data on cholesterol levels at follow-up were not available for this study. Furthermore, the exact number of patients who took cholesterol-lowering drugs is not known, but patients with hypercholesterolemia are routinely treated with these agents at our institution. The hypothesis that patients who achieve reduction of the initially high cholesterol levels may benefit more in terms of restenosis prevention is unlikely, considering the results of larger studies on this subject.28 Assuming that a high proportion of patients may have been under therapy with 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors, as our usual recommended therapy, may give rise to the speculation about the direct role of some of these agents in attenuating smooth muscle cell proliferation.29 30 However, similar paradoxical findings have been reported previously23 as well, which may suggest that lipids may not act in the same manner in restenosis as in atherosclerosis.
Limitations of the Study
We included in the analysis those patient-related factors that are most commonly considered in predictive studies of restenosis after coronary interventions. Further studies will certainly extend the list of these factors and may reduce the magnitude of the intrapatient correlation found in the present analysis. In addition, more lesion-specific covariates, such as lesion length, eccentricity, and calcifications, that were unavailable for the present analysis would have made the list of the potential predictive factors more complete. However, these missing lesion characteristics do not weaken the main finding of the study: assessment of the intrapatient correlation of restenosis. It is highly improbable that these lesion-specific factors would induce a patient-based clustering effect on restenosis and be responsible for part of the intrapatient correlation found in this study.
A more sophisticated analysis, with >1 level of nesting, would have allowed the estimation of the intraclass correlation not only on a patient basis but also on a vessel basis. This requires validated statistical methods and a major number of lesions and should become the objective of future studies.
This study included a large number of patients who underwent coronary stenting as the only interventional procedure. Care should be taken before any full extrapolation of the findings to PTCA patients. Further studies focused principally on PTCA patients are awaited to corroborate the findings of this study.
This study demonstrates that there is a dependence of restenosis between coronary lesions in patients who undergo a multilesion intervention. This signifies that the likelihood of restenosis for a lesion is higher when another lesion in that patient has also developed restenosis. This increase in restenosis risk is independent of the presence or absence of the analyzed patient risk factors (eg, diabetes), suggesting that unidentified patient factors are the source of this intrapatient correlation of restenosis. This finding has important implications for future studies aimed at the identification of predictive factors for restenosis after coronary interventions. Analyses that account for the lesion-to-lesion dependence of restenosis have to be applied to yield valid results. Future studies are needed that direct their focus toward additional patient-specific characteristics to improve our predictive power for restenosis.
Selected Abbreviations and Acronyms
|LAD||=||left anterior descending coronary artery|
|MLD||=||minimal lumen diameter|
|PTCA||=||percutaneous transluminal coronary angioplasty|
Reprint requests to Dr A. Kastrati, Deutsches Herzzentrum München, Lazarettstr 36, 80636 München, Germany.
- Received October 8, 1997.
- Revision received February 4, 1998.
- Accepted February 10, 1998.
- Copyright © 1998 by American Heart Association
Faxon DP. Identifying the predictors of restenosis: do we need new glasses? Circulation. 1997;95:2244–2246.
Kuntz RE, Baim DS. Defining coronary restenosis: newer clinical and angiographic paradigms. Circulation. 1993;88:1310–1323.
Rensing BJ, Hermans WRM, Vos J, Tijssen JGP, Rutsch W, Danchin N, Heyndrickx GR, Mast EG, Wijns W, Serruys PW. Luminal narrowing after percutaneous transluminal coronary angioplasty: a study of clinical, procedural, and lesional factors related to long-term angiographic outcome. Circulation. 1993;88:975–985.
Foley DP, Melkert R, Serruys PW. Influence of coronary vessel size on renarrowing process and late angiographic outcome after successful balloon angioplasty. Circulation. 1994;90:1239–1251.
Violaris AG, Melkert R, Serruys PW. Long-term luminal renarrowing after successful elective coronary angioplasty of total occlusions: a quantitative angiographic analysis. Circulation. 1995;91:2140–2150.
Eeckhout E, van Melle G, Stauffer JC, Vogt P, Kappenberger L, Goy JJ. Can early closure and restenosis after endoluminal stenting be predicted from clinical, procedural, and angiographic variables at the time of intervention? Br Heart J. 1995;74:592–597.
Gibson CM, Kuntz RE, Nobuyoshi M, Rosner B, Baim DS. Lesion-to-lesion independence of restenosis after treatment by conventional angioplasty, stenting, or directional atherectomy: validation of lesion-based restenosis analysis. Circulation. 1993;87:1123–1129.
Lehmann KG, Melkert R, Serruys PW. Contributions of frequency distribution analysis to the understanding of coronary restenosis: a reappraisal of the gaussian curve. Circulation. 1996;93:1123–1132.
Schömig A, Kastrati A, Elezi S, Schühlen H, Dirschinger J, Dannegger F, Wilhelm M, Ulm K. Bimodal distribution of angiographic measures of restenosis six months after coronary stent placement. Circulation. 1997;96:3880–3887.
Harrell FE Jr. Predicting Outcomes: Applied Survival Analysis and Logistic Regression. Charlottesville, Va: University of Virginia; 1997.
Schömig A, Kastrati A, Mudra H, Blasini R, Schühlen H, Klauss V, Richardt G, Neumann FJ. Four-year experience with Palmaz-Schatz stenting in coronary angioplasty complicated by dissection with threatened or present vessel closure. Circulation. 1994;90:2716–2724.
Black AJR, Namay DL, Niederman AL, Lembo NJ, Roubin GS, Douglas JS Jr, King SB III. Tear or dissection after coronary angioplasty: morphologic correlates of an ischemic complication. Circulation. 1989;79:1035–1042.
King SB III, Weintraub WS, Tao X, Hearn J, Douglas JS Jr. Bimodal distribution of diameter stenosis 4 to 12 months after angioplasty: implications for definitions and interpretation of restenosis. J Am Coll Cardiol. 1991;17:345A. Abstract.
Efron B, Tibshirani RJ. An Introduction to the Bootstrap. New York, NY: Chapman & Hall; 1993.
van Bockxmeer FM, Mamotte CD, Gibbons FA, Burke V, Taylor RR. Angiotensin-converting enzyme and apolipoprotein E genotypes and restenosis after coronary angioplasty. Circulation. 1995;92:2066–2071.
Speir E, Modali R, Huang ES, Leon MB, Shawl F, Finkel T, Epstein SE. Potential role of human cytomegalovirus and p53 interaction in coronary restenosis. Science. 1994;265:391–394.
Violaris AG, Melkert R, Serruys PW. Influence of serum cholesterol subfractions on restenosis after successful coronary angioplasty. Circulation. 1994;90:2267–2279.
Corsini A, Raiteri M, Soma MR, Bernini F, Fumagalli R, Paoletti R. Pathogenesis of atherosclerosis and the role of 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors. Am J Cardiol. 1995;76:21A–28A.Quantitative angiographic analysis was carried out before, immediately after, and at 6 months after coronary stent placement in 1734 lesions in 1244 patients. We used a specialized logistic regression that not only accounts for intraclass correlation but also quantifies it in the form of odds ratio (OR) as the increase in risk of a lesion to develop restenosis if another companion lesion also has restenosis. Twenty-three patient- and lesion-related variables were entered into the model with binary restenosis (diameter stenosis ≥50%) as end point. After adjustment for the influence of significant factors, the analysis found a significant intrapatient correlation, OR 2.5 (1.8 to 3.6). Thus, the risk for a lesion to develop restenosis is more than doubled if a companion lesion also has restenosis.