REVIEW ARTICLESRichard P. Cambria, MD, Section Editor

FromHDScAtioanne

AuthAddRepmNU

Thetom

0741Cophttp

Bayesian network meta-analysis of nitinolstents, covered stents, drug-eluting stents, anddrug-coated balloons in the femoropopliteal arteryKonstantinos Katsanos, MSc, MD, PhD, EBIR,a Stavros Spiliopoulos, MD, PhD, EBIR,b

Narayan Karunanithy, MRCS, FRCR, EBIR,a Miltiadis Krokidis, MD, PhD, EBIR,c

Tarun Sabharwal, FRCR, EBIR,a and Peter Taylor, MA, MChir, FRCS,d London and Cambridge,United Kingdom; and Rion, Greece

Objective: Several randomized controlled trials (RCTs) have shown the superiority of some of these technologies overballoon angioplasty, but direct comparisons between these treatment options are lacking. The authors conducted anetwork meta-analysis of RCTs comparing bare nitinol stents, covered nitinol stents, paclitaxel- or sirolimus-elutingstents (PES or SES), and paclitaxel-coated balloons (PCB) with plain balloon angioplasty or with each other in thefemoropopliteal artery (PROSPERO registry: CRD42013004845).Methods: Sixteen RCTs comprising 2532 patients with 4227 person-years of follow-up were analyzed on an intention-to-treat basis. Bayesian random effects Poisson and binomial models were used for mixed treatment comparisons (Win-BUGS). Clinical heterogeneity was accounted for by incorporating a meta-regression model on trial-specific baseline risk.End points included technical success, vascular restenosis, target lesion revascularization, and major amputations. Pair-wise odds ratios and rate ratios (ORs and RRs) of absolute treatment effects were calculated, and the probabilities of eachtreatment being best are reported. Summary estimates are reported as the posterior median and associated credible in-tervals (CrIs) that serve the same purpose as confidence intervals in the context of the Bayesian framework. Extensivesensitivity, meta-regression, and network consistency analyses were performed to evaluate heterogeneity.Results: Technical success was highest with covered stents (pooled OR, 13.6; 95% CrI, 3.3-31.1, probability best 82%)followed by uncovered stents (pooled OR, 7.0; 95% CrI, 2.6-129, probability best 18%) when compared with balloonangioplasty (reference treatment). Vascular restenosis was lowest with PES (RR, 0.43; 95% CrI, 0.16-1.18, probabilitybest 45%) followed by PCB (RR, 0.43; 95% CrI, 0.26-0.67, probability best 42%). Target lesion revascularization waslowest with PCB (RR, 0.36; 95% CrI, 0.23-0.55, probability best 56%) followed by PES (RR, 0.42; 95% CrI, 0.16-1.06,probability best 33%). Major amputations were rare in all treatment and control groups (pooled amputation rate of 0.7events per 100 person-years).Conclusions: Immediate technical success is better with the use of covered stents, whereas paclitaxel-eluting stents andpaclitaxel-coated balloons offer the best long-term results in the femoropopliteal artery. (J Vasc Surg 2014;59:1123-33.)

the Department of Interventional Radiology, Guy’s and St. Thomas’ospitals, NHS Foundation Trust, King’s Health Partners, Londona; theepartment of Interventional Radiology, Patras University Hospital,hool of Medicine, Rionb; the Department of Interventional Radiology,ddenbrooke’s Hospital, Cambridge University Hospitals, NHS Founda-n Trust, Cambridgec; and the Department of Vascular Surgery, Guy’sd St. Thomas’ Hospitals, NHS Foundation Trust, King’s Health Part-rs, London.d

or conflict of interest: none.itional material for this article may be found online at www.jvascsurg.org.rint requests: Konstantinos Katsanos, MSc, MD, PhD, EBIR, Depart-ent of Interventional Radiology, Guy’s and St. Thomas’ Hospitals,HS Foundation Trust, King’s Health Partners, London SE1 7EH,K (e-mail: konstantinos.katsanos@gstt.nhs.uk).editors and reviewers of this article have no relevant financial relationshipsdisclose per the JVS policy that requires reviewers to decline review of anyanuscript for which they may have a conflict of interest.-5214/$36.00yright � 2014 by the Society for Vascular Surgery.://dx.doi.org/10.1016/j.jvs.2014.01.041

The femoropopliteal artery is the most common siteof involvement in patients with peripheral arterial disease1

and an endovascular-first strategy is now recommendedfor the majority of symptomatic patients.2,3 However, thefemoropopliteal arteries have high rates of restenosis afterendovascular treatment because of their unique anatomyand biomechanics. Balloon-expandable metal stents areno longer used in the femoropopliteal segment becauseof the risk of external compression and longitudinal axisdeformation under conditions of vessel torsion, flexion,and extension.4 Several new technologies such as barenitinol stents, drug-eluting stents, covered stents, anddrug-coated balloons have emerged to improve long-term patency outcomes after angioplasty of the femoraland popliteal arteries.5,6

Several randomized controlled trials (RCTs) haveshown the superiority of some of these technologies overballoon angioplasty (BA), but direct comparisons between

1123

JOURNAL OF VASCULAR SURGERY1124 Katsanos et al April 2014

these treatment options are lacking. A previous meta-analysis on the matter has been limited to bare stentsonly.5 Network meta-analysis is capable of comparingdifferent treatments with a common reference treatment.When more than two treatments are available (eg, treat-ments A, B, C) an RCT network allows an indirect compar-ison of any two treatments (eg, direct comparisons of A vsB and of B vs C imply an indirect comparison of A vs C).Whereas a simple meta-analysis pools trials with the sametwo treatments, network meta-analysis aggregates resultsover an RCT network into a single unified analysis whilefully respecting randomization.7,8 A large-scale compre-hensive mixed treatment comparison (MTC) networkmeta-analysis was performed to provide a qualitative anal-ysis and quantitative data synthesis of recent evidencefrom major RCTs investigating different endovasculartreatment options for the management of arterial occlusivedisease of the femoropopliteal artery.

METHODS

Selection process. This systematic review has beenregistered and published in the PROSPERO public data-base of meta-analyses (CRD42013004845; http://www.crd.york.ac.uk/PROSPERO). Electronic searches ofPubMed (Medline, 1950 to present), EMBASE (ExcerptaMedical Database, 1980 to present), AMED (Allied andComplementary Database, 1985 to present), Scopus (1970to present), the CENTRAL (Cochrane Central Register ofControlled Trials), archives of regulatory authorities (FDA,EMA, MHRA), proceedings of international conferences,and other relevant online material were performed toidentify RCTs that evaluated endovascular treatment op-tions for femoropopliteal arterial occlusive disease. Therewere no restrictions on publication language, date, or sta-tus. For the purpose of this network meta-analysis, thesearch extracted trials comparing primary placement ofbare nitinol stents (BNS), covered nitinol stents (CNS),paclitaxel-eluting stents (PES), sirolimus-eluting stents(SES), or paclitaxel-coated balloons (PCB) with BA or witheach other in the femoropopliteal artery (Appendix, onlineonly). The trial selection process complied with thePreferred Reporting Items for Systematic reviews andMeta-Analyses statement.9 The quality of the includedRCTs was evaluated with the Cochrane Collaboration’stool for assessing risk of bias10 (Appendix, online only).

Outcome measures. The outcome measures of thesystematic review were defined according to previouslypublished international guidelines and analyzed strictly onan intention-to-treat basis.11 For example, in the RESIL-IENT trial, the initial publication reported the outcomeson a treatment-received basis.12,13 However, outcomeswere converted and analyzed on the intention-to-treatprinciple within the framework of this meta-analysis. Forthe ZILVER-PTX trial, all patients assigned to a ZILVER-PTX were grouped together and data were analyzed onintention-to-treat after the secondary randomization.14 Adetailed description of data management and extraction ofnumber of events of all included RCTs is presented in the

Appendix (online only). End points included technicalsuccess, vascular restenosis, target lesion revascularization(TLR), and major amputations. Technical success wasdefined as successful recanalization of the target vessel withno significant residual stenosis on completion angiogram(30-50% threshold, depending on the trial design) andwithout any bail-out stenting in the case of BA trials.Cross-overs to the other treatment arm were considered astechnical failures. Vascular restenosis was evaluated on the50% threshold by quantitative vascular angiography orcolor Doppler ultrasonography (DUS) if angiography wasnot available (peak systolic velocity ratio [PSVR] $2.0-2.5,depending on the trial design). TLR included any repeatprocedures because of significant in-lesion restenosis asso-ciated with recurrent symptoms. Major amputationsincluded any limb loss that extended above the ankle(below-knee or above-knee).

Statistical methods. Network meta-analysis combinesonly the data summaries for each RCT without any of theraw data. Quantitative data synthesis of the connectednetwork of RCTs was performed with Bayesian inferencewith the use of Gibbs sampling methods that allow com-bined direct and indirect MTCs (WinBUGS 1.4.3, MRCBiostatistics Unit, Cambridge, United Kingdom). Designand Bayesian modeling of the present network meta-analysis complied with the guidelines of the NationalInstitute for Health and Excellence Decision Support Units(NICEDSU).15-17 For the purposes of this analysis, we firstfitted a Bayesian hierarchical random-effects model formultiple comparisons of different treatment options withthe use of BA as the treatment reference.7 A random-effectsmodel was chosen to allow for heterogeneity among trialson the assumption that various treatment effects originatedfrom a normal distribution. Bayesian inference with Win-BUGS uses Markov Chain Monte Carlo (MCMC) simu-lation to calculate the posterior distributions of theinterrogated nodes within the framework of the chosenmodel and likelihood function and on the basis of someprior assumptions. Minimally informative priors for effectsizes and precisions were used for all calculations of thepresent analysis to avoid bias.

Bayesian results are reported as the median and accom-panying 95% credibility intervals (95% CrIs) of the posteriordistribution. CrIs serve the same purpose as confidence in-tervals in frequentist statistics. In the case of Bayesian infer-ence, 95% CrIs that do not cross unity (for relative effects)or zero (for absolute effects) are considered significant andwould be equivalent to a significance of a < .05 in case offrequentist inference. Event counts were analyzed with anMTC random-effects binomial model for the purpose ofthe immediate end point of technical success and pooledeffect estimates were expressed as pairwise odds ratios(ORs). Vascular restenosis and TLR were recorded ascounts of events per 100 person-years (rates) to accountfor the variable follow-up period between RCTs andanalyzed with an MTC random-effects Poisson model tocalculate pairwise rate ratios (RRs) between different treat-ments. Extensive meta-regression analysis with the

Fig 1. Trial selection process according to the Preferred Reporting Items for Systematic reviews and Meta-Analysesstatement.

Fig 2. Detailed analysis of the risk of bias of each included randomized controlled trials (RCTs) according to theCochrane Collaboration Tool.

JOURNAL OF VASCULAR SURGERYVolume 59, Number 4 Katsanos et al 1125

introduction of several different baseline variables as cova-riates was also performed. The baseline risk of eventsvaried widely in the reference treatment arms. Baselinerisk may be a proxy for unmeasured but importantpatient-level characteristics that produce significant clinicalheterogeneity. This uncertainty was accounted for byincorporating a meta-regression model on trial-specificbaseline risk of the control arms, and reported resultsare centered on the mean value of covariate baseline riskas described elsewhere in detail.18,19 To further addressstatistical heterogeneity, we calculated and report 95%prediction intervals that incorporate effect variance be-tween studies calculated by the random-effects model topredict a more conservative summary treatment effect ofa future similar trial.19,20

The probability of each treatment being the best(lowest or highest rate of events) was calculated to providea more comprehensible measure of treatment efficacy andplots of cumulative ranking probabilities (rankograms) toillustrate how each treatment ranks against each other interms of being the first, second, or third, and so forth,best treatment option were constructed as proposed else-where.21 Potential publication bias was assessed by visualinspection of inverted funnel plot asymmetry. TheCochran’s Q (c2) and the I2 statistical test were calculatedto test for evidence of statistical heterogeneity. In the caseof MTC network meta-analyses, the risk of network incon-sistency is greatly reduced if between-trial heterogeneity islow.22 Consistency of pairwise direct and indirect effect es-timates generated by the MTC network were evaluated

Fig 3. Mixed treatment comparison (MTC) network of evidence.The straight lines denote direct head-to-head comparisons and thedotted lines denote indirect comparisons in which direct compari-son data are missing. The numbers next to the solid lines refer tothe number of randomized controlled trials (RCTs), with directcomparisons available for each link.

Fig 4. Condensed network of evidence in the case of technicalsuccess analysis. The straight lines denote direct head-to-headcomparisons and the numbers next to the solid lines refer to thenumber of randomized controlled trials (RCTs), with directcomparisons available for each link.

JOURNAL OF VASCULAR SURGERY1126 Katsanos et al April 2014

with back-calculation by means of the Bucher method.22,23

An extended description of the statistical methods is pre-sented in the Appendix (online only) (WinBUGS codesavailable on request).

RESULTS

RCT selection and description. In total, 16 RCTswere found to be eligible, and full-text publications orother detailed archived records were analyzed5,6,12,14,24-38

(Fig 1). All trials were of generally high quality (Fig 2).Baseline demographics and procedural variables of therandomized populations were well matched among theactive and control treatment arms in all 16 trials(Supplementary Tables I and II, online only). The majorityof enrolled subjects were treated for intermittent claudi-cation, and only a small percentage of cases had criticallimb ischemia. However, there was significant variationbetween trials in both the rate of initial chronic total oc-clusions and the lesion length treated. The shortest lesionswere generally randomized in trials comparing BNS withBA and the longest in trials comparing CNS with BA,whereas lesion length was average in other treatmentcomparisons (PES, SES, and PCB). Meta-regression tech-niques were incorporated in our Bayesian modeling toaccount for this variation, as outlined in the Online statis-tical methods in more detail.19 Dual antiplatelet therapy forat least 1 month (up to 6 months) was recommended afterthe procedure in the majority of trials (Appendix, onlineonly).

Network analysis. Sixteen two-arm RCTs including2532 patients with 4227 person-years of follow-up wereincorporated into the present MTC Bayesian model. Themean duration of follow-up was 22 months (range, 9-48 months). The applied MTC network of evidence is

shown in Fig 3. In the case of immediate technical suc-cess, the network was condensed to a triangle comparingstandard BA (including the control BA and active PCBstudy arms) vs uncovered nitinol stents (including theactive BNS, PES, and SES study arms) vs covered nitinolstents (the active CNS study arms) (Fig 4). For the pur-poses of vascular restenosis analysis, findings from vascularangiography (n ¼ 7 trials) were pooled together withDUS results (PSVR >2.4; n ¼ 7 trials and PSVR >2.0;n ¼ 2 trials) for detection of events of significant reste-nosis of the femoropopliteal artery (50% threshold).Overall, there was no significant visual asymmetry of therespective funnel plots (available on request), and theEgger test did not produce any significant results tosuggest small study effects or publication bias. Results ofquantitative analysis of heterogeneity are provided in theAppendix (online only).

Technical success. Data on immediate proceduraltechnical success were reported in the majority of RCTs(n ¼ 14). Balloon angioplasty was the treatment of refer-ence, with a calculated median technical success rate of73.9%. Both uncovered and covered nitinol stents achievedsignificantly higher success rates, approximately 95% and98%, respectively, in comparison with BA. Covered stentswere related with the highest probability (82.1%) of beingbest, and uncovered stents ranked second, with an 18%probability of being best (covered stents > uncoveredstents > BA; Supplementary Table III, online only, andFig 5). Both uncovered and covered stents exhibited lownumbers needed to treat (NNTs) (approximately 5), withrelatively narrow CrIs. The forest plot of posterior pairwiseORs is shown in Fig 6. ORs were statistically significantlydifferent in both uncovered (OR, 7.0; 95% CrI, 2.6-129)and covered stents (OR, 13.6; 95% CrI, 3.3-31.1)compared with BA, but no significant difference was notedwhen comparing uncovered with covered stents.

Vascular restenosis. Data on quantitative vascularangiography or DUS follow-up imaging at various timepoints were reported by all RTCs (n ¼ 16). BA was thereference treatment, with a calculated median rate of

Fig 5. Rankograms of the cumulative ranking probabilities for (A) technical success, (B) vascular restenosis, and (C)for target lesion revascularization (TLR). On the horizontal axis is the possible rank of each option, based on relativeefficacy (first best to last worse). On the vertical axis is the cumulative probability of each treatment per rank (prob-ability to be the first, second, third, etc, best option). The larger the area under the cumulative probability ranking curvethe better the treatment compared with the rest.21

JOURNAL OF VASCULAR SURGERYVolume 59, Number 4 Katsanos et al 1127

vascular restenosis of 45 events per 100 person-years (py).Compared with BA, all investigated treatments showedreduced restenosis, but the magnitude of the effect varied.PCB and PES were the most efficacious (both at 19events/100 py), CNS had an intermediate effect (27events/100 py), and BNS and SES were the least effica-cious (both at approximately 35 events/100 py). PES hadthe highest probability (45%) of being the best treatment(lowest rate of events), followed by PCB, with a 42%probability (PES > PCB > SES > CNS > BNS > BA;

Supplementary Table III, online only, and Fig 5). Alltherapies were associated with single-digit NNTs (posteriormedians), with the least uncertainty (narrow CrIs) beingobserved with PCB, PES, and CNS. A Forest plot ofposterior rates ratios of events of different pairwise com-parisons is shown in Fig 7. Compared with BA, PCB (RR,0.43; 95% CrI, 0.26-0.67) and CNS (RR, 0.60; 95% CrI,0.36-0.94) had significantly fewer events, whereas BNS(RR, 0.78; 95% CrI, 0.57-1.06) and PES (RR, 0.43; 95%CrI, 0.16-1.18) were marginally non-significantly different

Fig 6. Random-effects Forest plot of pooled estimates of technical success for the condensed network of evidence. Theblack lines denote 95% credible intervals (CrIs) and the gray lines denote 95% prediction intervals. The numbersrepresent odds ratios (ORs) and 95% CrIs.

Fig 7. Random-effects Forest plot of pooled estimates of vascular restenosis. The black lines denote 95% credibleintervals (CrIs) and the gray lines denote 95% prediction intervals. The numbers represent odds ratios (ORs) and 95%CrIs. BA, Balloon angioplasty; BNS, bare nitinol stent; CNS, covered nitinol stent; PCB, paclitaxel-coated balloon; PES,paclitaxel-eluting stent; SES, sirolimus-eluting stent.

JOURNAL OF VASCULAR SURGERY1128 Katsanos et al April 2014

(upper CrI slightly above unity). No significant differenceswere identified in the rest of potential pairwise compari-sons, except that PCB was significantly better than BNS(RR, 0.55; 95% CrI, 0.30-0.97; Fig 7).

TLR. Data on TLR events were reported by almost allRCTs (n ¼ 15). BA served as the reference treatment, witha calculated median TLR rate of 22 events per 100 py. Incomparison with BA, all other options showed decreased

Fig 8. Random-effects Forest plot of pooled estimates of target lesion revascularization (TLR). The black lines denote95% credible intervals (CrIs) and the gray lines denote 95% prediction intervals. The numbers represent odds ratios(ORs) and 95% CrIs. BA, Balloon angioplasty; BNS, bare nitinol stent; CNS, covered nitinol stent; PCB, paclitaxel-coated balloon; PES, paclitaxel-eluting stent; SES, sirolimus-eluting stent.

JOURNAL OF VASCULAR SURGERYVolume 59, Number 4 Katsanos et al 1129

rates of events, but the actual effect was variable. PCB andPES were the most efficacious (8 events/100 py and 9events/100 py, respectively), whereas the rest showed amore modest effect; BNS (16 events/100 py), CNS (16events/100 py), and SES (14 events/100 py). PCB wasassociated with the highest probability (56%) of being thebest treatment (lowest rate of events), followed by PES,with a 33% probability (PCB > PES > SES > CNS >BNS > BA; Supplementary Table III, online only, andFig 5). Only PES and PCB had single-digit NNTs (pos-terior medians), but the 95% CrIs were narrow and thelower interval did not cross zero only in the case of PCB.The Forest plot of posterior RRs of the various pairwisecomparisons is shown in Fig 8. Compared with BA, RRswere marginally non-significantly different, with the upperCrI slightly exceeding unity in the case of BNS (RR, 0.77;95% CrI, 0.57-1.07), CNS (RR, 0.74; 95% CrI, 0.45-1.18), and PES (RR, 0.42; 95% CrI, 0.16-1.06). PCB wasthe only treatment that showed consistently significantlylower RRs against BA (RR, 0.36; 95% CrI, 0.23-0.55),BNS (RR, 0.47; 95% CrI, 0.27-0.80), and CNS (RR, 0.49;95% CrI, 0.25-0.94). No significant difference was notedbetween PCB and different drug-eluting stents (Fig 8).

Finally, major amputations were rare in all treatment andcontrol groups (pooled amputation rate of 0.7 eventsper 100 py; 95% CrI, 0.4-1.1 events per 100 py;Supplementary Table III, online only).

Sensitivity, heterogeneity, and consistency anal-ysis. When examining all outcome measures, application ofa fixed-effectsmodel produced largely similar results in termsof treatment effects and overall treatment ranking apart fromsome numerical differences of pairwise RRs (Forest plotsavailable on request).However, random-effectsmodelswerefound to produce a considerably better model fit than fixed-effects and were therefore preferred in the present MTCBayesian meta-analysis. Bayesian random-effects computa-tions were repeated with the introduction of several othercovariates instead of baseline risk, and results did not deviatesignificantly. Meta-regression plots of effect summariesagainst the baseline risk of events are illustrated in Fig 9.There were no significant discrepancies when comparing95% CrI with 95% prediction intervals of all pairwise com-parisons shown in Forest plots of Figs 6 through 8. Theoverall network of evidence was highly consistent withoutany significant differences between direct and indirectcomputations, as shown in the Forest plots of Fig 10.

Fig 9. Meta-regression plots of effect summaries against the baseline risk of events for (A) technical success, (B)vascular restenosis, and (C) for target lesion revascularization (TLR). Outcomes in favor of the active treatment aredenoted with log(OR) >0 in the case of technical success and log(OR) <0 in the case of vascular restenosis and TLR.Pooled effect summaries in the present meta-analysis were reported on the mean covariate value of plain balloon an-gioplasty denoted by the dotted vertical lines in all cases.19

JOURNAL OF VASCULAR SURGERY1130 Katsanos et al April 2014

Analytical results of the meta-regression and a detailed nu-merical analysis of publication bias, trial heterogeneity, andnetwork consistency are provided in SupplementaryTables IV and V (online only).

DISCUSSION

To our knowledge, this is the first comprehensivenetwork meta-analysis of different endovascular options

for occlusive disease in the femoropopliteal arterycombining the evidence of several RCTs. Network meta-analysis with Bayesian statistics allows for robust MTCanalysis in the case of multiple treatment options. In addi-tion, it allows for combined direct and indirect compari-sons of competing treatments while maintaining thebenefits of randomization. The use of Bayesian modelingin an in-depth, quantitative analysis was performed that

Fig 10. Forest plot of network consistency analysis for comparisons, with n$ 3 trials. Numbers represent the posteriormeans and standard deviations (SDs) of the direct, indirect, and overall mixed treatment comparison (MTC) estimatesof the odds ratios (ORs) and rate ratios (RRs). BA, Balloon angioplasty; BNS, bare nitinol stent; CNS, covered nitinolstent; PCB, paclitaxel-coated balloon; PES, paclitaxel-eluting stent; SES, sirolimus-eluting stent.

JOURNAL OF VASCULAR SURGERYVolume 59, Number 4 Katsanos et al 1131

investigated different strategies with BA as the controlcomparator in most cases. The analysis focused on theend points of immediate technical success, future vascularrestenosis, and revascularization procedures performed forsymptomatic recurrent disease.

Traditionally, stents in the femoropopliteal artery arereserved for cases of failed or suboptimal BA. Typical indi-cations for use include post-angioplasty elastic recoil or aresidual stenosis (>30%) or a flow-limiting dissectionwith the aim to maximize vessel luminal diameter. Howev-er, this study shows that both standard (bare or drug-eluting) and covered nitinol stents achieved significantlyhigher success rates (95%-98%), whereas bail-out stentingwas necessary in one in four BA procedures. Therefore,permanent scaffolds in the form of bare or covered nitinolstents produce appear to achieve significantly better imme-diate recanalization results in the femoropopliteal artery.This is in line with the fact that stents inherently tackle dis-sections and provide better radial support. Balloons (bareor drug-coated) do not leave any permanent scaffoldsbehind and thus theoretically may be more safe and moresuitable for the distal femoral and popliteal arteries thathave a considerable amount of flexion and torsion.

However, neointimal hyperplasia leading to vascularrestenosis remains problematic for femoropopliteal interven-tions.2,4Methods to inhibit restenosis include stents coveredwith polytetrafluoroethylene that act as a barrier againstdevelopment of hyperplasia (ie, CNS), drug-eluting stents(ie, SES or PES), and drug-coated balloons (ie, PCB) thatdeliver agents such as sirolimus and paclitaxel to the vesselwall and inhibit neointimal hyperplasia. In the presentMTC network meta-analysis, a benefit of reduced restenosisand fewerTLR eventswas observed for BNS,CNS, PES, andPCB options, whereas BA alone consistently ranked as theleast effective therapy. Both PES and PCB ranked as thebest treatments, but only PCB achieved statistically signifi-cant differences in pairwise comparisons against BA, BNS,and CNS. PES showed discernible but borderline non-significant differences against most other treatments. Onthe basis of these results, the use of stents significantlyimproves immediate and long-term results, whereaspaclitaxel-coated stents and balloons appear outperformother therapeutic strategies in the femoropopliteal artery.Therefore, this evidence questions the paradigm that BAwith provisional stent placement is the accepted standardof endovascular care for the femoropopliteal arteries.

JOURNAL OF VASCULAR SURGERY1132 Katsanos et al April 2014

There are some limitations to the present analysis. First,network meta-analyses may be prone to uncertainty andpotential bias, which may compromise the accuracy ofthe network of evidence. However, sensitivity and consis-tency analysis have shown the robustness of the presentnetwork. Second, the duration of dual antiplatelet therapywas variable between different studies (range, 1-6 months),but the ideal antiplatelet regimen for infrainguinal proce-dures remains elusive. Use of statins was scarcely reported.Meta-regression techniques were used to adjust for base-line risk variability, and sensitivity analysis was also usedto account for the heterogeneity produced by the differenttypes of stents examined. Regression against baseline riskshowed stronger treatment effects as observed risk ofevents increased (Fig 9). All other examined covariateswere associated with weak regression slopes, and no corre-lation coefficient proved to be statistically significant. Thesize of study groups was nonetheless limited to supportin depth subgroup analyses, and meta-regression of a largernumber of covariates and long-term follow-up is missing(average follow-up <2 years across all studies). Third, therewere differences in stent design and geometry with BNSstudies, in polytetrafluoroethylene cover with CNS (withor without heparin coating), in paclitaxel-coating agentswith PCB, and in drug-elution kinetics with SES andPES. Furthermore, some of the trials have not beenpublished in peer-reviewed journals (in whole or theirlong-term follow-up), and data extraction was based onor supplemented by presentations and online archived ma-terial from international meetings or regulatory authorities.However, we have chosen to include those because of thelimited number of RCTs available and to counteract anyrisk of publication bias. For example, both the INTRA-COIL study (data available from FDA files) and theVIBRANT study (reported online in www.clinicaltrials.gov) have produced negative results and have not beenpublished in peer-reviewed journals. Finally, the networkof evidence included RCTs spanned more than a decade,and therefore improvements over time or changes in gen-eral medical management were not accounted for.

CONCLUSIONS

Immediate technical success is more likely with the useof either covered or uncovered stents, whereas paclitaxel-eluting stents and paclitaxel-coated balloons offer the bestlong-term results in occlusive disease of femoropopliteal ar-tery. Results from the present meta-analysis may behypothesis-generating during design of further large-scaleclinical trials. More head-to-head RCTs are necessary, andcost-utility issues must be addressed to strengthen the evi-dence on clinical and cost-effectiveness of endovasculartreatment of femoropopliteal disease.

AUTHOR CONTRIBUTIONS

Conception and design KK, SS, NK, MK, TS, PTAnalysis and interpretation: KK, NKData collection: KK, SS

Writing the article: KK, SS, NKCritical revision of the article: KK, TS, PTFinal approval of the article: KK, SS, NK, MK, TS, PTStatistical analysis: KK, SS, MKObtained funding: Not applicableOverall responsibility: KK

REFERENCES

1. Kasapis C, Gurm HS. Current approach to the diagnosis and treatmentof femoral-popliteal arterial disease: a systematic review. Curr CardiolRev 2009;5:296-311.

2. Norgren L, Hiatt WR, Dormandy JA, Nehler MR, Harris KA,Fowkes FG. Inter-Society Consensus for the Management ofPeripheral Arterial Disease (TASC II). J Vasc Surg 2007;45(Suppl S):S5-67.

3. Rooke TW, Hirsch AT, Misra S, Sidawy AN, Beckman JA, Findeiss LK,et al. 2011 ACCF/AHA focused update of the guideline for themanagement of patients with peripheral artery disease (updating the2005 guideline): a report of the American College of CardiologyFoundation/American Heart Association Task Force on PracticeGuidelines: developed in collaboration with the Society for Cardio-vascular Angiography and Interventions, Society of InterventionalRadiology, Society for Vascular Medicine, and Society for VascularSurgery. J Vasc Surg 2011;54:e32-58.

4. Tsetis D, Belli AM. Guidelines for stenting in infrainguinal arterialdisease. Cardiovasc Intervent Radiol 2004;27:198-203.

5. Kasapis C, Henke PK, Chetcuti SJ, Koenig GC, Rectenwald JE,Krishnamurthy VN, et al. Routine stent implantation vs. percutaneoustransluminal angioplasty in femoropopliteal artery disease: a meta-analysis of randomized controlled trials. Eur Heart J 2009;30:44-55.

6. Cassese S, Byrne RA, Ott I, Ndrepepa G, Nerad M, Kastrati A, et al.Paclitaxel-coated versus uncoated balloon angioplasty reduces targetlesion revascularization in patients with femoropopliteal arterial disease:a meta-analysis of randomized trials. Circ Cardiovasc Intervent 2012;5:582-9.

7. Dias S, Sutton AJ, Ades AE, Welton NJ. Evidence synthesis for decisionmaking 2: a generalized linear modeling framework for pairwise andnetwork meta-analysis of randomized controlled trials. Med DecisMaking 2013;33:607-17.

8. Palmerini T, Biondi-Zoccai G, Della Riva D, Stettler C, Sangiorgi D,D’Ascenzo F, et al. Stent thrombosis with drug-eluting and bare-metalstents: evidence from a comprehensive network meta-analysis. Lancet2012;379:1393-402.

9. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting itemsfor systematic reviews and meta-analyses: the PRISMA statement. BMJ2009;339:b2535.

10. Higgins JP, Altman DG, Gotzsche PC, Juni P, Moher D, Oxman AD,et al. The Cochrane Collaboration’s tool for assessing risk of bias inrandomised trials. BMJ 2011;343:d5928.

11. Sacks D, Marinelli DL, Martin LG, Spies JB. Reporting standards forclinical evaluation of new peripheral arterial revascularization devices.J Vasc Interv Radiol 2003;14(9 Pt 2):S395-404.

12. Laird JR, Katzen BT, Scheinert D, Lammer J, Carpenter J,Buchbinder M, et al. Nitinol stent implantation vs. balloon angioplastyfor lesions in the superficial femoral and proximal popliteal arteries ofpatients with claudication: three-year follow-up from the RESILIENTrandomized trial. J Endovasc Ther 2012;19:1-9.

13. Laird JR, Katzen BT, Scheinert D, Lammer J, Carpenter J,Buchbinder M, et al. Nitinol stent implantation versus balloon angio-plasty for lesions in the superficial femoral artery and proximal poplitealartery: twelve-month results from the RESILIENT randomized trial.Circ Cardiovasc Intervent 2010;3:267-76.

14. Dake MD, Ansel GM, Jaff MR, Ohki T, Saxon RR, Smouse HB, et al.Paclitaxel-eluting stents show superiority to balloon angioplasty andbare metal stents in femoropopliteal disease: twelve-month Zilver PTXrandomized study results. Circ Cardiovasc Intervent 2011;4:495-504.

15. Dias S, Welton NJ, Sutton AJ, Caldwell DM, Guobing L, Ades AE.NICE DSU technical support document 4: inconsistency in networks

JOURNAL OF VASCULAR SURGERYVolume 59, Number 4 Katsanos et al 1133

of evidence based on randomised controlled trials. 2011. Available at:http://www.nicedsu.org.uk.

16. Dias S, Welton NJ, Sutton AJ, Ades AE. NICE DSU technical supportdocument 2: a generalised linear modelling framework for pairwise andnetwork meta-analysis of randomized controlled trials. 2011. Availableat: http://www.nicedsu.org.uk.

17. Dias S, Sutton AJ, Welton NJ, Ades AE. NICE DSU technical supportdocument 3: heterogeneity: subgroups, meta-regression, bias and biasadjustment. 2011. Available at: http://www.nicedsu.org.uk.

18. Achana FA, Cooper NJ, Dias S, Lu G, Rice SJ, Kendrick D, et al.Extending methods for investigating the relationship between treat-ment effect and baseline risk from pairwise meta-analysis to networkmeta-analysis. Stat Med 2013;32:752-71.

19. Dias S, Sutton AJ, Welton NJ, Ades AE. Evidence synthesis for decisionmaking 3: heterogeneity: subgroups, meta-regression, bias, and bias-adjustment. Med Decis Making 2013;33:618-40.

20. Guddat C, Grouven U, Bender R, Skipka G. A note on the graphicalpresentation of prediction intervals in random-effects meta-analyses.Syst Rev 2012;1:34.

21. Salanti G, Ades AE, Ioannidis JP. Graphical methods and numericalsummaries for presenting results from multiple-treatment meta-anal-ysis: an overview and tutorial. J Clin Epidemiol 2011;64:163-71.

22. Dias S, Welton NJ, Caldwell DM, Ades AE. Checking consistency inmixed treatment comparison meta-analysis. Stat Med 2010;29:932-44.

23. Bucher HC, Guyatt GH, Griffith LE, Walter SD. The results of directand indirect treatment comparisons in meta-analysis of randomizedcontrolled trials. J Clin Epidemiol 1997;50:683-91.

24. GORE VIABAHN ENDOPROSTHESIS Peripheral Vascular DiseaseStudy. Available at: http://www.clinicaltrials.gov/ct2/show/study/NCT00228384. Accessed January 2013.

25. Intracoil Self-expanding Peripheral Stent. Available at: http://www.accessdata.fda.gov/cdrh_docs/pdf/P000033bpdf.Accessed January 2013.

26. Chalmers N, Walker PT, Belli AM, Thorpe AP, Sidhu PS, Robinson G,et al. Randomized Trial of the SMART Stent versus Balloon Angio-plasty in Long Superficial Femoral Artery Lesions: the SUPER Study.Cardiovasc Intervent Radiol 2012.

27. Dick P, Wallner H, Sabeti S, Loewe C, Mlekusch W, Lammer J, et al.Balloon angioplasty versus stenting with nitinol stents in intermediatelength superficial femoral artery lesions. Catheter Cardiovasc Interv2009;74:1090-5.

28. Duda SH, Bosiers M, Lammer J, Scheinert D, Zeller T, Tielbeek A,et al. Sirolimus-eluting versus bare nitinol stent for obstructive super-ficial femoral artery disease: the SIROCCO II trial. J Vasc Interv Radiol2005;16:331-8.

29. Fanelli F, Cannavale A, Boatta E, Corona M, Lucatelli P, Wlderk A,et al. Lower limb multilevel treatment with drug-eluting balloons:6-month results from the DEBELLUM randomized trial. J EndovascTher 2012;19:571-80.

30. Lammer J. One year results of the VIASTAR trial. Presented at IROS,Berlin, Germany. Available at: http://www.esir.org. AccessedFebruary 2013.

31. Saxon RR, Dake MD, Volgelzang RL, Katzen BT, Becker GJ. Random-ized, multicenter study comparing expanded polytetrafluoroethylene-covered endoprosthesis placement with percutaneous transluminalangioplasty in the treatment of superficial femoral artery occlusive disease.J Vasc Interv Radiol 2008;19:823-32.

32. Schillinger M, Sabeti S, Dick P, Amighi J, Mlekusch W, Schlager O,et al. Sustained benefit at 2 years of primary femoropopliteal stentingcompared with balloon angioplasty with optional stenting. Circulation2007;115:2745-9.

33. Schmidt A. Lessons learnt from LEVANT-1 first in human study: 24-month final analysis. Available at: http://www.esir.org. Presented atCIRSE 2012. Accessed January 2013.

34. Tepe G, Zeller T, Albrecht T, Heller S, Schwarzwalder U, Beregi JP,et al. Local delivery of paclitaxel to inhibit restenosis during angioplastyof the leg. N Engl J Med 2008;358:689-99.

35. Werk M, Albrecht T, Meyer DR, Ahmed MN, Behne A, Dietz U, et al.Paclitaxel-coated balloons reduce restenosis after femoro-popliteal an-gioplasty: evidence from the randomized PACIFIER trial. Circ Car-diovasc Intervent 2012;5:831-40.

36. Werk M, Langner S, Reinkensmeier B, Boettcher HF, Tepe G,Dietz U, et al. Inhibition of restenosis in femoropopliteal arteries:paclitaxel-coated versus uncoated balloon: femoral paclitaxel random-ized pilot trial. Circulation 2008;118:1358-65.

37. Dake MD, Ansel GM, Jaff MR, Ohki T, Saxon RR, Smouse HB,et al. Sustained safety and effectiveness of paclitaxel-eluting stentsfor femoropopliteal lesions: two-year follow-up from the ZilverPTX randomized and single-arm clinical studies. J Am Coll Cardiol2013;61:2417-27.

38. Lammer J, Zeller T, Hausegger KA, Schaefer PJ, Gschwendtner M,Mueller-Huelsbeck S, et al. Heparin-bonded covered stents versus baremetal stents for complex femoro-popliteal artery lesions: the random-ized VIASTAR trial. J Am Coll Cardiol 2013;62:1320-7.

Submitted Jul 25, 2013; accepted Jan 19, 2014.

Additional material for this article may be found onlineat www.jvascsurg.org.

JOURNAL OF VASCULAR SURGERY1133.e1 Katsanos et al April 2014

APPENDIX (online only).

Search process. Evaluation of the quality of the RCTsincluded in the meta-analysis was performed independentlyby two authors (K.K., M.K.) by applying the CochraneCollaboration’s tool for assessing risk of bias.1 The trialselection process complied with the Preferred ReportingItems for Systematic reviews and Meta-Analyses (PRISMA)statement.2 Three authors (K.K., S.S., and M.K.) designedthe systematic review and individually selected the trials tobe included and independently extracted all presented dataas outlined in this material in detail. Any disagreementswere resolved by consensus between the investigators.

Each study was evaluated for inclusion in the networkmeta-analysis with the following criteria: (1) Only RCTswere included, (2) different endovascular options wereeligible provided that they investigated the aforementionedtherapies, (3) the target population included patients withfemoropopliteal occlusive disease causing either intermit-tent claudication (IC) or critical limb ischemia (CLI), (4)immediate outcomes and clinical and angiographicfollow-up of at least 6 months was available, and selectedend points were reported (see outcome measures below).Only self-expanding nitinol stents were considered for in-clusion, and balloon-expandable stents were excluded.

The last literature search for eligible RCTs was per-formed in January 2013 through the use of various relevantterms and keywords along with corresponding MedicalSubjects Headings (MeSH) as follows: Cochrane, femoralartery, popliteal artery, femoropopliteal, vascular restenosis,angioplasty and stent, randomized controlled trial, balloonangioplasty, primary nitinol stents, bare nitinol stents,covered nitinol stents, viabahn-covered stents, drug-eluting stents, drug-coated stents, sirolimus-eluting stents,everolimus-eluting stents, paclitaxel-eluting stents, drug-eluting balloons, drug-coated balloons, paclitaxel-coatedballoons, and paclitaxel-eluting balloons.

Data extraction. Descriptive variables extracted fromeach trial included a number of baseline demographics,procedural parameters, follow-up imaging modality, anti-platelet therapy, and primary and secondary end points foreach treatment group. Data were extracted from the text,survival curves, and tables of published manuscripts or frommeeting abstracts and presentations indexed online afterinternational conferences and symposia were held. In thecase of missing or conflicting data, the trial investigatorswere contacted. After the PRISMA process, the title andthe abstract of 169 scientific records were screened forpotential inclusion in this MTC analysis. Of those, 149citations were characterized as irrelevant or incomplete ornot meeting the predefined inclusion criteria and wereexcluded from further analysis. Another four items werediscarded because they were duplicate studies reporting aninterim follow-up.

Cochrane Collaboration’s tool for assessing risk ofbias. This method evaluates seven different methodolog-ical items of an RCT: (1) random sequence generation,(2) allocation concealment, (3) blinding of participants

and personnel, (4) blinding of outcome assessment, (5)incomplete outcome data, (6) selective outcome reporting,and (7) other sources of bias (eg, reporting of sample sizecalculation). Each of the above-mentioned domains wasevaluated as high, low, or unclear risk of bias, and dis-agreements were resolved by consensus.

Statistical methods. Quantitative data synthesis of theconnected network of RCTs was performed by means ofBayesian inference with the use of Gibbs sampling methodsthat allow combined direct and indirect mixed treatmentcomparisons (WinBUGS 1.4.3; MRC Biostatistics Unit,Cambridge, United Kingdom). In addition, direct frequent-ist inference methods (Fixed Mantel-Haenszel [M-H] orrandom DerSimonian and Laird [D-L] effects, RevMansoftware [Review Manager Version 5.1; The CochraneCollaboration, Copenhagen, Denmark] and the Statsdirectstatistical package [Version 2.7.9, Statsdirect Ltd, Chesh-ire, United Kingdom]) were applied as necessary. In thecase of zero cells in 2 � 2 contingency tables, continuitycorrection was applied with addition of 0.5 to eachobserved frequency to avoid statistical artifacts.

Design and Bayesian modeling of the present networkmeta-analysis complied with the guidelines of the NationalInstitute for Health and Excellence Decision SupportUnits (NICEDSU).3-5 For the purposes of this analysis,we first fitted a Bayesian hierarchical random-effectsmodel for multiple comparisons of different treatmentoptions with the use of BA as the treatment reference.6

A random-effects (RE) model was chosen to allow for het-erogeneity among trials on the assumption that varioustreatment effects originated from a normal distribution.Bayesian inference with WinBUGS uses Markov ChainMonte Carlo (MCMC) simulation to calculate the posteriordistributions of the interrogated nodes within the frameworkof the chosen model and likelihood function and on thebasis of some prior assumptions. Bayesian methods requirespecification of prior distributions and extensive sensitivityanalysis to check assumptions about the model. Minimallyinformative priors for effect sizes and precisions were usedfor all calculations of the present analysis to avoid bias.

Summary statistics of relative treatment effects are re-ported as the median and accompanying 95% credibility in-tervals (95% CrIs) of the posterior distribution. CrIs servethe same purpose as confidence intervals in frequentist sta-tistics. In the case of Bayesian inference, 95% CrIs that donot cross unity (for relative effects) or zero (for absolute ef-fects) are considered significant and would be equivalent toa significance of a ¼ .05 in the case of frequentist inference.Absolute treatment effects for each individual treatmentoption were then calculated on the natural scale and onthe centered covariate value after an exploratory analysisof the balloon angioplasty control arms. Event countswere analyzed with an MTC random-effects binomialmodel for the purpose of the immediate end point of tech-nical success, and pooled effect estimates were expressed aspairwise odds ratios (ORs). Vascular restenosis and TLRwere recorded as counts of events per 100 person-years(rates) to account for the variable follow-up period

JOURNAL OF VASCULAR SURGERYVolume 59, Number 4 Katsanos et al 1133.e2

between RCTs and analyzed with the use of an MTCrandom-effects Poisson model to calculate pairwise rate ra-tios (RRs) between different treatments.

Extensive meta-regression analysis with the introduc-tion of several different baseline variables as covariateswas also performed in search of potential predictors of out-comes. The observed baseline risk of events varied widely inthe reference treatment arms. Treatment effects may varyaccording to patient-level variables or trial-specific charac-teristics. Baseline risk is a proxy for unmeasured but impor-tant patient-level characteristics that produce significantclinical heterogeneity. Naive approaches that adjust theobserved risk of events in the control groups or the averagerisk of events in the control and treatment groups areflawed and may produce seriously misleading results. How-ever, a Bayesian approach with Gibbs sampling (Win-BUGS) has been recommended as a more appropriatemethod for investigating the relationship between treat-ment effect and underlying baseline risk across trials inmeta-analyses and adjusting for baseline confounders.7,8

For a detailed description of Bayesian modeling methodsto account for baseline risk in network meta-analysis,when it is of interest to adjust for baseline imbalances toreduce both heterogeneity and inconsistency, the readermay refer to other dedicated technical resources.5,7,8

Hence, in the present study, this uncertainty wasaccounted for by incorporating a meta-regression modelon trial-specific baseline risk of the control arms, and re-ported results are centered on the mean value of covariatebaseline risk as described elsewhere in detail.8,9 To furtheraddress statistical heterogeneity, we calculated and report95% prediction intervals that incorporate effect variance be-tween studies calculated by the RE model to predict thesummary treatment effect of a future similar trial.9,10 Pre-diction intervals may imply the degree of heterogeneityacross trials and will depend on the between-trial variationof observed treatment effects.

The probability of each treatment being the best(lowest or highest rate of events) and the numbers-needed-to-treat (NNTs; 95% CrIs) were also calculatedto provide more comprehensible measures of treatment ef-ficacy. We constructed rankograms of cumulative rankprobabilities of how each treatment ranks against eachother in terms of being the first, second, third, and so forth,best treatment option as proposed elsewhere.11 Plot cumu-lative ranking probabilities (rankograms) offer an intuitiveway to illustrate how each treatment effect compares withthe rest of the examined treatment options.

WinBUGS and MCMC. WinBUGS code was writtenand adapted according to recommendations of theNICEDSU (http://www.nicedsu.org.uk).6 Bayesianinference with WinBUGS uses Markov Chain Monte Carlo(MCMC) simulation to calculate the posterior distribu-tions of the interrogated nodes within the framework of thechosen model and likelihood function and on the basis ofsome prior assumptions. Minimally informative priors foreffect sizes and precisions were used for all calculations ofthe present analysis to avoid bias. Three Markov chains

were compiled and run, whereas convergence wasconfirmed with the Brooks-Gelman-Rubin diagnostictool12 and by inspection of history plots of monitorednodes. An initial burn-in simulation of 50,000 iterationswas discarded, and inference of final summary statistics wasbased on further simulation of an additional 100,000iterations.

Heterogeneity, consistency, and sensitivity ana-lyses. The validity of a network meta-analysis depends onthe distribution of effect modifiers (covariates) not onlybetween studies with a particular comparison (ie, hetero-geneity in the case of standard pairwise meta-analysis) butalso between different sets of comparisons (ie, inconsis-tency between direct and indirect study estimates).13 Thestraightforward Cochran’s Q (c2) and the I2 statistical testwere calculated to test for statistical evidence of heteroge-neity between RCTs with the same control and active arms.Briefly, I2 values <25% indicate low, 25% to 50% moderate,and >50% high heterogeneity.14 Funnel plots are plots ofthe trials’ effect estimates against respective sample sizesand in the presence of publication or other bias they mayappear to be skewed and asymmetrical.15 In the case ofMTC network meta-analyses, the risk of network incon-sistency is greatly reduced if between-trial heterogeneity islow.16 Consistency of pairwise direct and indirect effectestimates generated by the MTC network were evaluatedwith back-calculation by means of the Bucher method.16,17

The random- vs fixed-effects model were examined inthe sensitivity analysis. Random-effects models wereinitially used to account for between-trial heterogeneityin calculating the pooled effect estimates. Bayesian compu-tations were then repeated with a fixed-effects model to ac-count for the small number of studies included in some ofthe pairwise comparisons of the present MTC. Goodness ofmodel fit was compared with the posterior mean of the to-tal residual deviance and the Deviance Information Crite-rion (DIC) criterion.6 For a good model fit, the residualdeviance must be close to the total number of study armsanalysed, and the model with the lowest DIC is preferred.The level of statistical significance was set at a ¼ .05 for fre-quentist inference, whereas results associated with 95% CrIthat do not cross unity (for relative effects) or zero (for ab-solute effects) were considered significant in the case ofBayesian inference.

SUPPLEMENTARY REFERENCES (online only).

1. Higgins JP, Altman DG, Gotzsche PC, Juni P, Moher D, Oxman AD,et al. The Cochrane Collaboration’s tool for assessing risk of bias inrandomised trials. BMJ 2011;343:d5928.

2. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting itemsfor systematic reviews and meta-analyses: the PRISMA statement. BMJ2009;339:b2535.

3. Dias S, Welton NJ, Sutton AJ, Caldwell DM, Guobing L, Ades AE.NICE DSU technical support document 4: inconsistency in networksof evidence based on randomised controlled trials. 2011. Available at:http://www.nicedsu.org.uk.

4. Dias S, Welton NJ, Sutton AJ, Ades AE. NICE DSU technical supportdocument 2: a generalised linear modelling framework for pairwise andnetwork meta-analysis of randomized controlled trials. 2011. Availableat: http://www.nicedsu.org.uk.

JOURNAL OF VASCULAR SURGERY1133.e3 Katsanos et al April 2014

5. Dias S, Sutton AJ, Welton NJ, Ades AE. NICE DSU technical supportdocument 3: heterogeneity: subgroups, meta-regression, bias and biasadjustment. 2011. Available at: http://www.nicedsu.org.uk.

6. Dias S, Sutton AJ, Ades AE, Welton NJ. Evidence synthesis for decisionmaking 2: a generalized linear modeling framework for pairwise andnetwork meta-analysis of randomized controlled trials. Med DecisMaking 2013;33:607-17.

7. Thompson SG, Smith TC, Sharp SJ. Investigating underlying risk as asource of heterogeneity in meta-analysis. Stat Med 1997;16:2741-58.

8. Achana FA, Cooper NJ, Dias S, Lu G, Rice SJ, Kendrick D, et al.Extending methods for investigating the relationship between treat-ment effect and baseline risk from pairwise meta-analysis to networkmeta-analysis. Stat Med 2013;32:752-71.

9. Dias S, Sutton AJ, Welton NJ, Ades AE. Evidence synthesis for decisionmaking 3: heterogeneity: subgroups, meta-regression, bias, and bias-adjustment. Med Decis Making 2013;33:618-40.

10. Guddat C, Grouven U, Bender R, Skipka G. A note on the graphicalpresentation of prediction intervals in random-effects meta-analyses.Syst Rev 2012;1:34.

11. Salanti G, Ades AE, Ioannidis JP. Graphical methods and numericalsummaries for presenting results from multiple-treatment meta-anal-ysis: an overview and tutorial. J Clin Epidemiol 2011;64:163-71.

12. Brooks SP, Gelman A. General methods for monitoring convergence ofiterative simulations. J Comput Graphic Stat 1998;7:434-55.

13. Jansen JP, Naci H. Is network meta-analysis as valid as standard pair-wise meta-analysis? It all depends on the distribution of effect modifiers.BMC Med 2013;11:159.

14. Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring incon-sistency in meta-analyses. BMJ 2003;327:557-60.

15. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997;315:629-34.

16. Dias S, Welton NJ, Caldwell DM, Ades AE. Checking consistency inmixed treatment comparison meta-analysis. Stat Med 2010;29:932-44.

17. Bucher HC, Guyatt GH, Griffith LE, Walter SD. The results of directand indirect treatment comparisons in meta-analysis of randomizedcontrolled trials. J Clin Epidemiol 1997;50:683-91.

18. Intracoil Self-expanding Peripheral Stent. Available at: wwwaccessdatafdagov/cdrh_docs/pdf/P000033bpdf. 2013.

19. Schillinger M, Sabeti S, Loewe C, Dick P, Amighi J, Mlekusch W, et al.Balloon angioplasty versus implantation of nitinol stents in the super-ficial femoral artery. N Engl J Med 2006;354:1879-88.

20. Schillinger M, Sabeti S, Dick P, Amighi J, Mlekusch W, Schlager O,et al. Sustained benefit at 2 years of primary femoropopliteal stentingcompared with balloon angioplasty with optional stenting. Circulation2007;115:2745-9.

21. Krankenberg H, Schluter M, Steinkamp HJ, Burgelin K, Scheinert D,Schulte KL, et al. Nitinol stent implantation versus percutaneoustransluminal angioplasty in superficial femoral artery lesions up to 10cm in length: the femoral artery stenting trial (FAST). Circulation2007;116:285-92.

22. Dick P, Wallner H, Sabeti S, Loewe C, Mlekusch W, Lammer J, et al.Balloon angioplasty versus stenting with nitinol stents in intermediatelength superficial femoral artery lesions. Catheter Cardiovasc Interv2009;74:1090-5.

23. Laird JR, Katzen BT, Scheinert D, Lammer J, Carpenter J,Buchbinder M, et al. Nitinol stent implantation vs balloon angioplastyfor lesions in the superficial femoral and proximal popliteal arteries ofpatients with claudication: three-year follow-up from the RESILIENTrandomized trial. J Endovasc Ther 2012;19:1-9.

24. Laird JR, Katzen BT, Scheinert D, Lammer J, Carpenter J,Buchbinder M, et al. Nitinol stent implantation versus balloon angio-plasty for lesions in the superficial femoral artery and proximal poplitealartery: twelve-month results from the RESILIENT randomized trial.Circ Cardiovasc Intervent 2010;3:267-76.

25. Mustapha J. The RESILIENT Randomized trial: three-year results.Presented at LINC 2012; 2013.

26. Chalmers N, Walker PT, Belli AM, Thorpe AP, Sidhu PS, Robinson G,et al. Randomized trial of the SMART Stent versus Balloon Angioplastyin Long Superficial Femoral Artery Lesions: the SUPER Study. Car-diovasc Intervent Radiol 2013;36:353-61.

27. Saxon RR, Dake MD, Volgelzang RL, Katzen BT, Becker GJ. Random-ized, multicenter study comparing expanded polytetrafluoroethylene-covered endoprosthesis placement with percutaneous transluminalangioplasty in the treatment of superficial femoral artery occlusive disease.J Vasc Interv Radiol 2008;19:823-32.

28. GORE VIABAHN ENDOPROSTHESIS Peripheral Vascular DiseaseStudy. Available at: http://www.clinicaltrials.gov/ct2/show/study/NCT00228384. 2013.

29. GORE VIABAHN� endoprosthesis with bioactive propaten surfaceversus bare nitinol stent in the treatment of TASC B, C and D lesionsin superficial femoral artery occlusive disease. Available at: http://www.clinical-trials.com/ISRCTN48164244. 2013.

30. Lammer J. One year results of the VIASTAR trial. Presented at IROS,Berlin, Germany. Available at: http://www.esir.org. 2013.

31. Lammer J, Zeller T, Hausegger KA, Schaefer PJ, Gschwendtner M,Mueller-Huelsbeck S, et al. Heparin-bonded covered stents versus baremetal stents for complex femoro-popliteal artery lesions: the random-ized VIASTAR trial. J Am Coll Cardiol 2013;62:1320-7.

32. Duda SH, Bosiers M, Lammer J, Scheinert D, Zeller T, Tielbeek A,et al. Sirolimus-eluting versus bare nitinol stent for obstructive super-ficial femoral artery disease: the SIROCCO II trial. J Vasc Interv Radiol2005;16:331-8.

33. Duda SH, Pusich B, Richter G, Landwehr P, Oliva VL, Tielbeek A, et al.Sirolimus-eluting stents for the treatment of obstructive superficialfemoral artery disease: six-month results. Circulation 2002;106:1505-9.

34. Scheinert D. Long-term outcomes of endovascular treatment in the SFA.Presented at LINC 2011. 2013.

35. Dake MD, Ansel GM, Jaff MR, Ohki T, Saxon RR, Smouse HB,et al. Paclitaxel-eluting stents show superiority to balloon angioplastyand bare metal stents in femoropopliteal disease: twelve-month Zil-ver PTX randomized study results. Circ CardiovascIntervent 2011;4:495-504.

36. Dake MD. The ZILVER-PTX randomized trial of paclitaxel-elutingstents for femoropopliteal disease: 3-year results. Vasc Intervent AdvPresented at VIVA 2012. 2012.

37. Dake MD, Ansel GM, Jaff MR, Ohki T, Saxon RR, Smouse HB,et al. Sustained safety and effectiveness of paclitaxel-eluting stentsfor femoropopliteal lesions: two-year follow-up from the ZilverPTX randomized and single-arm clinical studies. J Am Coll Car-diol 2013;61:2417-27.

38. Tepe G, Zeller T, Albrecht T, Heller S, Schwarzwalder U, Beregi JP,et al. Local delivery of paclitaxel to inhibit restenosis during angioplastyof the leg. N Engl J Med 2008;358:689-99.

39. Werk M, Langner S, Reinkensmeier B, Boettcher HF, Tepe G,Dietz U, et al. Inhibition of restenosis in femoropopliteal arteries:paclitaxel-coated versus uncoated balloon: femoral paclitaxel random-ized pilot trial. Circulation 2008;118:1358-65.

40. Werk M, Albrecht T, Meyer DR, Ahmed MN, Behne A, Dietz U, et al.Paclitaxel-coated balloons reduce restenosis after femoro-popliteal an-gioplasty: evidence from the randomized PACIFIER trial. Circ Car-diovasc Intervent 2012;5:831-40.

41. Fanelli F, Cannavale A, Boatta E, Corona M, Lucatelli P, Wlderk A,et al. Lower limb multilevel treatment with drug-eluting balloons: 6-month results from the DEBELLUM randomized trial. J EndovascTher 2012;19:571-80.

42. Schmidt A. Lessons learnt from LEVANT-1 first in human study: 24-month final analysis. Available at: http://www.esir.org. Presented atCIRSE 2012. 2013.

Supplementary Table I (online only). Design and characteristics of randomized controlled trials (RCTs) included inthe mixed treatment comparison (MTC) analysis

RCT Registration Comparisons Year Subjects Follow-up, months

INTRACOIL18 n/a BNS vs BA 2001 266 9ABSOLUTE19,20 NCT00281060 BNS vs BA 2007 104 24FAST21 n/a BNS vs BA 2007 244 12ASTRON22 NCT00715416 BNS vs BA 2009 73 12RESILIENT23-25 NCT00673985 BNS vs BA 2010 206 36SUPER26 NCT00232843 BNS vs BA 2012 150 12VIABAHN27 n/a CNS vs BA 2008 197 12VIBRANT28 NCT00228384 CNS vs BNS 2012 148 36VIASTAR29-31 ISRCTN48164244 CNS vs BNS 2013 141 12SIROCCO32-34 NCT00232869 SES vs BNS 2006 93 48ZILVER-PTX35-37 NCT00120406 PES vs BA 2012 479 36THUNDER38 NCT00156624 PCB vs BA 2008 102 24FEMPAC39 NCT00472472 PCB vs BA 2008 87 24PACIFIER40 NCT01083030 PCB vs BA 2012 91 12DEBELLUM41 n/a PCB vs BA 2012 50 12LEVANT 142 NCT00930813 PCB vs BA 2012 101 24

ITT, Intention-to-treat.For the RESILIENT trial, the initial publication reported the outcomes on a treatment-received basis. However, outcomes were analyzed on the ITT principlewithin the framework of this meta-analysis. For the ZILVER-PTX trial, all patients assigned to a ZILVER-PTX were grouped together and data were analyzedon ITT after the secondary randomization. For the SIROCCO trial, pooled data of the SIROCCO I and II sub-studies were analyzed with the use of QVAavailable at 6 months and Duplex follow-up up to 48 months. For the THUNDER trial (three-arm trial), the arm of paclitaxel dissolved in the contrastmedium was excluded from the present network of evidence. For the DEBELLUM trial, only subjects with femoropopliteal lesions were analyzed. For theLEVANT-1 trial, cases with balloon deployment malfunction (n ¼ 8) were excluded from further analysis. In the latter study randomization between plain BAand PCB was performed after provisional stent placement in a quarter of the cases (26/101). Some of the trials have not been published yet in peer-reviewedjournals either in whole (INTRACOIL, VIBRANT, VIASTAR, and LEVANT-1) or their long-term follow-up (SIROCCO and ZILVER-PTX). Dataextraction was based on or supplemented by presentations and online archived material from international meetings (VIASTAR, LEVANT-1, SIROCCO, andZILVER-PTX) or regulatory authorities (INTRACOIL and VIBRANT).

JOURNAL OF VASCULAR SURGERYVolume 59, Number 4 Katsanos et al 1133.e4

Supplementary Table II (online only). Baseline variables of randomized controlled trials (RCTs) included in thenetwork meta-analysis

Trial

INTRACOIL ABSOLUTE FAST ASTRON RESILIENT SUPER VIABAHN

Active Control Active Control Active Control Active Control Active Control Active Control Active Control

Study arms n ¼ 135 n ¼ 131 n ¼ 51 n ¼ 53 n ¼ 123 n ¼ 121 n ¼ 34 n ¼ 39 n ¼ 134 n ¼ 72 n ¼ 74 n ¼ 76 n ¼ 97 n ¼ 100Age, years 67 6 11 68 6 10 65 6 10 68 6 10 67 6 9 66 6 10 69 6 9 69 6 10 68 6 10 66 6 9 66 6 9 70 6 9 67 6 10 67 6 10Male sex 67% 63% 59% 47% 63% 75% 74% 64% 71% 67% 78% 86% 70% 82%Smoking 82% 80% 53% 36% 68% 73% 35% 44% 72% 83% 24% 26% 51% 46%Hypertension n/a n/a 92% 87% 83% 83% 82% 80% 84% 94% 66% 67% 68% 65%Hyperlipidemia n/a n/a 92% 87% 60% 61% 91% 92% 80% 76% n/a n/a 59% 60%Diabetes 38% 37% 43% 32% 36% 31% 29% 31% 38% 39% 31% 38% 34% 37%Renal

insufficiencyn/a n/a n/a n/a 15% 6% n/a n/a n/a n/a 15% 8% 10% 6%

Claudication n/a n/a 88% 87% 97% 96% 91% 97% 100% 100% 85% 79% 88% 91%Critical leg

ischemian/a n/a 12% 13% 3% 4% 9% 3% n/a n/a 15% 21% 12% 9%

Lesions treated n ¼ 177 n ¼ 175 n ¼ 51 n ¼ 53 n ¼ 123 n ¼ 121 n ¼ 34 n ¼ 39 n ¼ 134 n ¼ 72 n ¼ 74 n ¼ 76 n ¼ 97 n ¼ 100Occlusions

(CTOs)23% 17% 37% 32% 37% 25% 38% 39% n/a n/a 96% 91% 20% 30%

Lesion length(cm)

3.5 63.0

3.3 63.0

10.1 67.5

9.2 66.4

4.5 62.8

4.5 62.8

8.2 66.7

6.5 64.6

7.1 64.4

6.4 64.1

12.3 65.4

11.7 65.2

7.0 64.0

7.0 64.0

Recommendedantiplatelettherapy

Aspirin andticlopidinefor 1 month

Aspirin þclopidogrelfor 3 months

Aspirin þclopidogrelfor 1 month

Aspirin þclopidogrelfor 3 months

Aspirin þclopidogrelfor 3 months

Aspirin þclopidogrelfor 1 month(or LMWH)

Aspirin(or occasionallyticlopidine)

Trial

VIBRANT VIASTAR SIROCCO ZILVER-PTX THUNDER FEMPAC PACIFIER

Active Control Active Control Active Control Active Control Active Control Active Control Active Control

Study arms n ¼ 72 n ¼ 76 n ¼ 72 n ¼ 69 n ¼ 47 n ¼ 46 n ¼ 238 n ¼ 241 n ¼ 48 n ¼ 54 n ¼ 45 n ¼ 42 n ¼ 44 n ¼ 47Age, years 69 63 68 69 66 6 9 66 6 11 68 6 10 68 6 11 69 6 8 68 6 9 67 70 71 6 7 71 6 9Male sex 63% 64% 67% 75% 66% 78% 66% 64% 65% 63% 60% 60% 59% 64%Smoking n/a n/a 69% 70% 47% 30% 86% 84% 23% 22% 47% 36% 49% 60%Hypertension n/a n/a 83% 84% 68% 70% 89% 82% 79% 83% 78% 81% 66% 66%Hyperlipidemia n/a n/a 68% 68% 64% 63% 76% 70% 69% 63% 58% 59% 50% 47%Diabetes n/a n/a 35% 36% 43% 35% 49% 42% 50% 46% 40% 55% 43% 28%Renal

insufficiencyn/a n/a n/a n/a n/a n/a 10% 11% n/a n/a n/a n/a n/a n/a

Claudication n/a n/a 86% 91% n/a n/a 90% 91% n/a n/a 96% 93% 95% 96%Critical leg

ischemian/a n/a 14% 19% n/a n/a 10% 9% n/a n/a 4% 7% 5% 4%

Lesionstreated (n)

n ¼ 72 n ¼ 76 n ¼ 72 n ¼ 69 n ¼ 47 n ¼ 46 n ¼ 247 n ¼ 251 n ¼ 86 n ¼ 86 n ¼ 100 n ¼ 101 n ¼ 86 n ¼ 47

Occlusions(CTOs)

n/a n/a 79% 70% 69% 57% 30% 25% 27% 26% 13% 19% 23% 38%

Lesionlength (cm)

n/a n/a 17.3 18.9 8.5 64.4

8.1 65.2

5.4 64.1

5.3 64.0

7.5 66.2

4.4 66.7

4.0 4.7 7.0 65.3

6.6 65.5

Recommendedantiplatelettherapy

n/a Aspirin þclopidogrelfor 6 months

Aspirin þclopidogrel/ticlopidinefor 1 month

Aspirin þclopidogrelfor 2 months

Aspirin þclopidogrelfor 1 month

Aspirin þclopidogrellong-term

Aspirin þclopidogrelfor >2months

DEBELLUM LEVANT 1

Trial Active Control Active Control

Study arms n ¼ 25 n ¼ 25 n ¼ 49 n ¼ 52Age, years 66 6 6 67 6 6 n/a n/aMale sex 76% 72% n/a n/aSmoking 68% 56% n/a n/aHypertension 76% 60% 96% 87%Hyperlipidemia 48% 68% 59% 69%Diabetes 52% 36% 45% 50%Renal insufficiency n/a n/a n/a n/aClaudication 64% 60% 94% 93%Critical leg ischemia 36% 40% 6% 7%Lesions treated n ¼ 44 n ¼ 48 n ¼ 49 n ¼ 52Occlusions (CTOs) 21% 22% 41% 42%Lesion length (mm) 7.6 6 0.6 7.8 6 0.7 8.1 6 3.7 8.0 6 3.8Recommended

antiplatelet therapyAspirin þclopidogrelfor 1 month

n/a

JOURNAL OF VASCULAR SURGERY1133.e5 Katsanos et al April 2014

Supplementary Table III (online only). Outcome measures with 95% credible interval (CrI) and probabilities of eachone being best

Median ratea (95% CrI) Probability best, % NNT (95% CrI)

Technical successBalloon angioplasty 73.9 (52.1-88.0) 0.03 ControlUncovered stents 95.4 (85.5-99.1) 17.9 4.7 (2.6-10.4)Covered stents 97.5 (85.1-99.8) 82.1 4.4 (2.3-10.3)

Vascular restenosisBA 44.9 (14.2-141.4) 0.0 ControlBNS 34.8 (10.6-114.0) 0.04 9.8 ([�33.1]-77.7)CNS 26.8 (7.7-92.0) 5.0 5.8 (1.3-28.9)PES 19.4 (4.2-88.1) 44.5 4.0 ([�8.9]-24.7)SES 35.9 (7.2-178.4) 8.0 3.7 ([�66.5]-72.6)PCB 19.1 (5.5-65.4) 42.3 4.0 (1.2-13.9)

TLRBA 21.5 (9.4-49.3) 0.0 ControlBNS 16.8 (6.9-40.4) 0.02 20.1 ([�87.0]-142.2)CNS 15.9 (6.1-41.2) 0.4 15.9 ([�134.9]-196.0)PES 9.1 (2.6-31.2) 32.7 8.2 ([�7.3]-42.5)SES 14.3 (3.5-59.5) 11.2 8.6 ([�94.3]-108.8)PCB 7.8 (3.0-19.7) 55.5 7.4 (3.1-17.9)

Major amputationsb

BA 1.2 (0.4-2.3) n/a ControlBNS 0.5 (0.1-1.1) n/a >1000 ([�90.9]-90.9)CNS 0.4 (0.01-1.2) n/a 15.9 ([�10.3]-9.8)PES 0.0 (0.0-1.9) n/a n/aSES 0.0 (0.0-4.3) n/a n/aPCB 1.1 (0.2-2.6) n/a �500 ([�45.5]-58.8)

BA, Balloon angioplasty; BNS, bare nitinol stent; CNS, covered nitinol stent; CrI, credibility interval; ITT, intention-to-treat; NNT, number needed to treat;PCB, paclitaxel-coated balloon; PES, paclitaxel-eluting stent; SES, sirolimus-eluting stent; TLR, target lesion revascularization.All end points were analyzed on an ITT principle as described in the Appendix (online only).aOutcomes expressed as median (95% CrIs) of the posterior distribution of event rates per 100 person-years of follow-up, except of technical success expressedas the median percentage (%) of enrolled subjects. Negative NNT values indicate a number needed to harm because of a negative treatment benefit.bThe numbers of events of major amputations were too low in control and active treatment groups (zero cells in a lot of groups), and the Bayesian approachfailed to produce any meaningful results (MCMC chains did not converge). Results refer to event rates per 100 person-years of follow-up as calculated bymeans of a standard DerSimonian and Laird random-effects model.

JOURNAL OF VASCULAR SURGERYVolume 59, Number 4 Katsanos et al 1133.e6

Supplementary Table IV (online only). Between-trial heterogeneity and network consistency analysis

ComparisonsPublication

bias Heterogeneity MTC Direct Indirect Inconsistency

PActive Control Trials Egger P I2 (%) P Mean (SD) Mean (SD) Mean (SD) Mean (SD)

Technical successBalloon angioplasty Nitinol stents 7 0.49 .84 93.9 <.001 9.9 (25.1) 14.4 (95.6) 9.6 (26.0) 1.5 (99.1) .99

Vascular restenosisBA BNS 6 �7.94 .08 55.2 .04 0.78 (0.12) 0.84 (0.44) 0.77 (0.12) 1.08 (0.46) .86BA PCB 5 �2.42 .11 0 .71 0.44 (0.11) 0.44 (0.22) 0.44 (0.13) 1.0 (0.25) 1.00

TLRBA BNS 5 �2.44 .17 0 .89 0.79 (0.13) 0.80 (1.0) 0.79 (1.0) 1.0 (1.0) .98BA PCB 5 �3.12 .22 0 .49 0.37 (0.08) 0.41 (0.44) 0.37 (0.08) 1.1 (0.4) .81

Major amputations

Comparisons Publication bias Heterogeneity

MTC Direct Indirect InconsistencyActive Control Trials Egger P I2 (%) P

BA BNS 6 �0.08 .92 0 .89 n/a n/a n/a n/aBA PCB 5 �0.05 .97 0 .70 n/a n/a n/a n/a

BA, Balloon angioplasty; BNS, bare nitinol stent; MTC, mixed treatment comparison; PCB, paclitaxel-coated balloon; OR, odds ratio; RR, rate ratio; SD,standard deviation; TLR, target lesion revascularization.Heterogeneity and inconsistency statistics reported for comparisons with n $ 3 trials. Posterior means and SDs of the MTC, and the direct analysis estimate ofthe ORs and RRs were computed from the Bayesian model. Inconsistency was computed with back-calculation by subtraction of the indirect from the directestimate with the Bucher method. In the case of major amputations, the numbers of events were too low, with several zero cells in both control and activetreatment groups, and therefore the Bayesian approach was numerically unstable and failed to produce any meaningful results (MCMC chains did notconverge). Hence, MTC analyses were not feasible, and only direct comparisons with a standard DerSimonian and Laird random-effects model are presented.

JOURNAL OF VASCULAR SURGERY1133.e7 Katsanos et al April 2014

Supplementary Table V (online only). Meta-regression analysis of network covariates

Covariate Correlation coefficient 95% CrI

Technical successSmoking �2.46 [�12.4]-4.82Diabetes �0.24 [�46.8]-34.4Lesion length 0.22 [�0.26]-0.81Chronic occlusion �0.01 [�9.69]-9.84Bail-out/cross-over n/a n/aAverage follow-up 0.78 [�0.60]-1.95Publication year 0.19 [�0.15]-0.50Baseline risk �1.17 [�2.70]-0.53

Vascular restenosisSmoking 0.004 [�1.50]-1.43Diabetes 2.96 [�4.21]-10.6Lesion length �0.01 [�0.12]-0.10Chronic occlusion 0.14 [�1.13]-1.50Bail-out/Cross-over �2.24 [�8.50]-1.73Average follow-up �0.19 [�0.53]-0.18Publication year �0.04 [�0.12]-0.05Baseline risk �0.30 [�1.06]-0.46

TLRSmoking 0.47 [�1.00]-1.96Diabetes 2.43 [�5.78]-11.9Lesion length �0.02 [�0.12]-0.09Chronic occlusion �0.34 [�1.99]-1.25Bail-out/Cross-over �0.63 [�4.13]-2.93Average follow-up �0.04 [�0.37]-0.29Publication year �0.01 [�0.09]-0.07Baseline risk �0.31 [�1.26]-0.72

BA, Balloon angioplasty; CrI, credibility interval; RCT, randomizedcontrolled trial; TLR, target lesion revascularization.This table shows the results of the Bayesian random-effects meta-regressionmodeling of the effect of certain covariates on treatment outcomes.Regression slopes (correlation coefficients) with respective 95% CrIs werecalculated. For the evaluation of the effect of cross-over/bail-out, use of thecomparator device only RCTs randomized against BA (n ¼ 8) wereincluded in the meta-regression analysis. All examined covariates wereassociated with weak regression slopes, and no correlation coefficient provedto be statistically significant. In the case of technical success, a positiveregression slope indicates better outcomes with the active treatment,whereas a negative slope indicates worse outcomes as the covariate effectincreases. On the other hand, in the case of vascular restenosis and TLR, anegative regression slope would indicate better outcomes with the activetreatment, whereas a positive slope relates to worse outcomes as the co-variate effect increases. Bayesian meta-regression could not be performed formajor amputations because the model was numerically unstable. Results areconsidered statistically significant if the CrIs do not cross zero. Weak cor-relation coefficients with wide CrIs were noted in all cases. Baseline risk inthe control reference group may be a proxy for unmeasured but importantpatient-level characteristics that produce significant heterogeneity. There-fore, the correlation coefficients of baseline risk were included in the presentrandom-effects Bayesian models to adjust for and reduce clinical hetero-geneity while calculating absolute and relative treatment effects.8

JOURNAL OF VASCULAR SURGERYVolume 59, Number 4 Katsanos et al 1133.e8