Recruitment

Figure 1 shows the CONSORT diagram of the flow of patients through the trial. Between May 2001 and October 2008, 1713 patients from 50 centres in the UK, mainland Europe, Australia, New Zealand and Canada were enrolled and randomised. The trial centres, together with the members of the trial committees, location of recruiting centres, number of patients recruited at each centre and the names of the investigators at each centre are detailed in Appendix 10. Three patients (two in the stenting group and one in the endarterectomy group) withdrew consent immediately after randomisation and were excluded from the intention-to-treat (ITT) analysis. In total, 1511 patients were enrolled at experienced centres and 202 at supervised probationary centres: 751 (88%) of 853 patients assigned to carotid stenting and 760 (89%) of 857 patients assigned to endarterectomy were randomised at centres classified as experienced. Most patients had their allocated treatment initiated (stenting group, n = 828; endarterectomy group, n = 821). Nine patients allocated to stenting crossed over to surgery without an attempt at the procedure and a further 16 had no attempted ipsilateral endarterectomy or stenting procedure (Figure 1). Fifteen patients allocated to endarterectomy crossed over to stenting without an attempt at endarterectomy and 21 had no attempted ipsilateral procedure.

FIGURE 1

Consolidated Standards of Reporting Trials diagram for ICSS.

Monitoring of adverse events led to concern about the stenting results of two investigators at supervised centres. These investigators were stopped from treating further patients within the trial and their centres were suspended from randomisation. All the patients allocated to stenting (n = 11, five with disabling stroke or death) or endarterectomy during the same time period (n = 9, one with fatal stroke) at these centres were included in the analyses. One of the two centres subsequently restarted randomisation with a different investigator performing stenting.

Baseline characteristics

Table 1 shows baseline characteristics of study participants.

TABLE 1

Baseline patient characteristics at randomisation by allocated treatment

Patient baseline characteristics (Table 1) and drug treatment during the trial (Table 2) were similar between the two treatment groups. At 1 year after randomisation, 97% of patients in both the stenting group and the endarterectomy group took any antiplatelet or anticoagulant; at 5 years, the percentages were 94% and 95% (Table 2). There were slightly more patients taking antihypertensive medications in the endarterectomy group at 1 year (71% vs. 75%; p = 0.088), but at 5 years the difference had reversed (83% vs. 76%; p = 0.017). However, this did not lead to any significant difference in systolic blood pressure or diastolic blood pressure (DBP) between groups at either time point. The majority of patients were treated with lipid-lowering medications with no significant difference between the groups (82% of patients in the CAS group vs. 84% in the CEA group at 1 year, and 87% of patients in the CAS group vs. 86% in the CEA group at 5 years).

TABLE 2

Drug treatment and blood pressure readings during follow-up (ITT population)

Success of procedures and cross-overs

Figure 2 shows the delay from randomisation to first initiated ipsilateral treatment in the per-protocol analysis within the first 120 days after randomisation.

FIGURE 2

Time between randomisation and treatment. Cumulative number of patients in whom allocated treatment was initiated per protocol plotted as a proportion (%) of the total number randomised in each treatment group (y-axis), against the delay in days between (more...)

Median delay from randomisation to treatment was shorter in the stenting group than in the endarterectomy group, as was the delay from most recent ipsilateral event to treatment (Table 3).

TABLE 3

Time from randomisation and from most recent ipsilateral event to allocated treatment

Of the 828 patients in whom stenting was initiated as allocated, 64 (8%) had their procedure aborted before the insertion of a stent (38 procedures were aborted because of difficulty gaining access to the stenosis; 15 were aborted because of the finding of an occluded artery, one patient had a fatal stroke, one patient had a fatal MI before completion of treatment, two had other medical complications, and further investigation in seven patients showed the artery to be < 50% stenosed). Of the 62 patients whose stenting procedure was aborted after initiation and who did not have a fatal event, 37 went on to have an ipsilateral endarterectomy, whereas 25 continued with best medical care only. Only two of the 821 patients whose allocated endarterectomy was initiated had their procedure aborted (one patient had an allergic reaction during general anaesthesia; the other became distressed and the endarterectomy had to be abandoned). Both patients subsequently had ipsilateral stenting.

The following stents were each used in 10% or more of the 764 patients in whom stents were inserted: Carotid WALLSTENT^® (Boston Scientific), Precision (Cordis^®, Freemont, CA, USA), and Protege™ (EV3^®, Dublin, Ireland). The following were each used in < 10% of patients: Acculink (Guidant™, Santa Clara, CA, USA), XACT^® (Abbott™, Santa Clara, CA, USA), S.M.A.R.T.^® (Cordis^®, Miami Lakes, FL, USA), Cristallo Ideale (Invatec, Roncadelle, Brescia, Italy), Exponent (Medtronic^®, Minneapolis, MN, USA), Next Stent (Boston Scientific). Protection devices were known to have been used in 593 (72%) of 828 patients. The following protection devices were each used in 10% or more of the patients in whom stenting was attempted: FilterWire EZ™ (Boston Scientific), ANGIOGUARD^® (Cordis), SpiderFX™ (EV3) and Emboshield^® (Abbott). A range of other protection devices were used in < 5% of patients. In 27 patients, it was not clear whether or not a protection device was used.

Short-term outcomes

In the ITT analysis, between randomisation and 120 days, there was no significant difference in the rate of disabling stroke or death between groups (stenting group, 4.0% vs. endarterectomy group, 3.2%; Table 4). The risk of stroke, death or procedural MI 120 days after randomisation was significantly higher in patients in the stenting group than in patients in the endarterectomy group (8.5% vs. 5.1%), representing an estimated 120-day absolute risk difference of 3.3% (95% CI 0.9% to 5.7%) with a HR in favour of surgery of 1.69 (1.16 to 2.45, log-rank p-value = 0.006) (Figure 3 and Table 4).

TABLE 4

Intention-to-treat analyses: outcome measures within 120 days of randomisation

FIGURE 3

Kaplan–Meier estimates of cumulative incidence of main short-term outcome measures. Data are analysed by ITT. The numbers above the end of the lines are the incidence estimates at 120 days after randomisation. (a) Stroke, death or procedural MI (more...)

FIGURE 4

Comparison of the short-term rate of stroke, death, or procedural myocardial infarction. Subgroups are defined according to baseline characteristics and analysed by intention to treat up to 120 days after randomisation, apart from time from event to treatment, (more...)

Most outcome events, within 120 days of randomisation in the stent and endarterectomy groups occurred within 30 days of the first ipsilateral procedure (61 of 72 events vs. 31 of 44 events). A few events occurred after randomisation but before the date of treatment (two patients vs. one patient) in patients who had no attempted ipsilateral procedure (three patients vs. six patients), or more than 30 days after treatment but within 120 days of randomisation (six patients vs. six patients). Compared with endarterectomy, allocation to stenting had a greater 120-day risk of the outcome measures of any stroke, any stroke or death, any stroke or procedural death, and all-cause death (Table 4).

Most strokes within 120 days of randomisation were ipsilateral to the treated carotid artery and most were ischaemic (Table 5). There were very few haemorrhagic strokes, with only two patients in whom the cause of the stroke was uncertain.

TABLE 5

Number of outcome events recorded between randomisation and 120 days in the ITT analysis, and between initiation of treatment and 30 days after treatment in the per-protocol analysis

The observed treatment effect was largely driven by the higher number of non-disabling strokes in the stenting group, most of which had symptoms lasting for more than 7 days. There was an excess of fatal strokes in the stenting group compared with the surgery group, but little difference in the number of patients with disabling stroke within 120 days of randomisation.

The per-protocol analysis included 1649 patients (stenting group, n = 828; endarterectomy group, n = 821). Results for 30-day procedural risk mirrored the results of the ITT analysis. Risk of stroke, death or procedural MI was higher in the stenting group than in the endarterectomy group (30-day risk 7.4% vs. 4.0%) [risk difference (RD) 3.3%, 95% CI 1.1% to 5.6%; risk ratio (RR) 1.83, 95% CI 1.21 to 2.77; χ² p = 0.003] (Table 6). Risk of any stroke or death up to 30 days after treatment remained significantly higher in patients in whom stenting was initiated than in patients with surgery initiated, but there was no significant difference in the risk of disabling stroke or death between treatment groups. There were more fatal strokes in the stenting group than in the endarterectomy group (eight vs. three), but difference in the risk of death alone was no longer significant (see Table 5). Forty-three (74%) of 58 strokes in the stenting group and 12 (44%) of 27 strokes in the endarterectomy group occurred on the day of the procedure. There was no difference in the numbers of strokes occurring between day 2 and day 30 between the two treatments (15 vs. 15).

TABLE 6

Per-protocol analysis of procedural risk: outcome measures between initiation of treatment and 30 days after treatment

Few procedural MIs were recorded (three in the stenting group, all of which were fatal, compared with five in the endarterectomy group). Cranial nerve palsies were almost completely avoided by stenting (RR 0.02, 95% CI 0.00 to 0.16; p < 0.0001) (see Table 5). The one cranial nerve palsy recorded in the stenting group occurred as a complication of an endarterectomy performed within 30 days of stenting. This patient and one additional patient in the endarterectomy group required percutaneous endoscopic gastrostomy feeding as a result of the cranial nerve palsies, which was classified as disabling. There were also fewer haematomas of any severity in the stenting group than in the endarterectomy group (RR 0.59, 95% CI 0.38 to 0.93; p = 0.0197), and fewer severe haematomas requiring surgical intervention, blood transfusion or extended hospital stay (RR 0.28, 95% CI 0.13 to 0.62; p = 0.0007) (see Table 5). A post-hoc sensitivity analysis was undertaken to examine if the results of the per-protocol analysis were affected by inclusion of patients in whom the allocated procedure was initiated but not completed. Exclusion of the 64 patients allocated to stenting and two patients allocated to endarterectomy in whom the procedures were aborted after initiation (i.e. including only patients in whom the allocated procedure was completed as planned) made little difference to the results (30-day risk of stroke, death or procedural MI of 7.6% in the stenting group vs. 4.0% in the endarterectomy group) (RD 3.6%, 95% CI 1.3% to 5.9%; RR 1.88, 95% CI 1.24 to 2.86; p = 0.002).

We undertook exploratory analyses of the composite outcome of stroke, death or procedural MI for pre-defined subgroups (Figure 4). These analyses suggested that carotid stenting might have a similar risk to endarterectomy in women, but that the intervention was more hazardous than endarterectomy in men. The difference was mainly caused by a higher risk of stroke, death or procedural MI in women assigned to endarterectomy than in men (7.6% vs. 4.2%). However, the difference between the HRs comparing the risk of stenting with endarterectomy in men and women only reached borderline significance (interaction p = 0.071). Stenting was more hazardous, and endarterectomy less hazardous, in patients not taking medication for hypertension at baseline than in patients taking medication for hypertension (see Figure 4). There was also a suggestion that patients who presented with multiple ipsilateral symptoms had a similar risk of stroke death, or procedural MI with stenting and endarterectomy. However, when compared with patients with only one event before randomisation, the difference in the HRs only reached borderline significance (interaction p = 0.055). There was no evidence that the relative increase in the hazard of an event in the stenting group compared with the endarterectomy group differed significantly across any other subgroups.

Duration of follow-up in the International Carotid Stenting Study

Figure 5 shows the number of patients remaining in follow-up in ICSS plotted against time from randomisation. Patients were followed up for a maximum of 10 years after randomisation with a median of 4.2 years and an interquartile range of 3.0–5.2 years. This amounted to 7355 patient-years of follow-up, without any difference between the two arms.

FIGURE 5

Patients remaining in each arm of the study (per protocol) are plotted against year of follow-up. In total, there are 7354.45 patient-years of follow-up until time of last follow-up or death. CAS (n = 853): median follow-up = 4.2 (more...)

Long-term primary and secondary outcomes

In the ITT analysis, the primary outcome event, fatal or disabling stroke between randomisation and the end of follow-up, occurred in 52 patients in the stenting group, corresponding to a cumulative 5-year risk of 6.4%, and in 49 patients in the endarterectomy group (5-year risk n, 6.5%), without any evidence for a difference in time to first occurrence of an event (HR 1.06, 95% CI 0.72 to 1.57; p = 0.76) (Table 7 and Figure 6).

TABLE 7

Intention-to-treat analysis of cumulative risks and HRs of main outcome events during the entire study period

FIGURE 6

Kaplan–Meier estimates of cumulative incidence of major long-term outcome measures. (a) Fatal or disabling stroke; (b) any stroke; (c) procedural stroke or death, or ipsilateral stroke during follow up; (d) any stroke > 30 days (more...)

The following secondary outcome events occurred significantly more often in the stenting group in the ITT analysis between randomisation and the end of follow-up: any stroke (5-year risks 15.2% vs. 9.4%) (HR 1.71, 95% CI 1.28 to 2.30; p = 0.0003); any stroke or death (27.5% vs. 22.6%) (HR 1.34, 95% CI 1.11 to 1.63; p = 0.003); the combination of any procedural stroke, procedural death or ipsilateral stroke during follow-up (11.8% vs. 7.2%) (HR 1.72, 95% CI 1.24 to 2.39; p = 0.001). There was no difference in all-cause mortality between treatment groups (17.4% vs. 17.2%) (HR 1.17, 95% CI 0.92 to 1.48; p = 0.19).

A total of 752 patients in the stenting group (88.2% of the ITT population) and 811 patients in the endarterectomy group (94.6%) were included in the per-protocol analysis of clinical outcome events. In the per-protocol analysis of events occurring more than 30 days after completed treatment up to the end of follow-up, there was no significant difference in the long-term risks of fatal and disabling stroke after stenting compared with endarterectomy (5-year risk 3.4% vs. 4.3%) (HR 0.93, 95% CI 0.53 to 1.60; p = 0.78) (Table 8). There was also no significant difference in the rates of ipsilateral stroke in the territory of the treated carotid artery (4.7% vs. 3.4%) (HR 1.29, 95% CI 0.74 to 2.24; p = 0.36). However, stroke of any severity occurred more often after stenting (8.9% vs. 5.8%) (HR 1.53, 95% CI 1.02 to 2.31; p = 0.039) (Figure 6 and Table 8). This difference was largely attributable to stroke occurring in the territory of the contralateral carotid artery or the vertebrobasilar circulation among patients treated with stents (5-year risks 4.6% vs. 2.5%) (HR 1.92, 95% CI 1.04 to 3.53; p = 0.033) and the majority were non-disabling.

TABLE 8

Per-protocol analysis of cumulative risks and HRs of major non-procedural outcome events occurring > 30 days after treatment

A total of 737 (98.0%) patients in the stenting group and 793 (97.8%) in the endarterectomy group were followed up with carotid ultrasound for a median of 4.0 years (interquartile range, 2.3–5.0 years) after treatment. There was no significant difference in long-term rates of severe carotid restenosis (≥ 70%) or occlusion, which occurred in 72 patients in the stenting group (5-year risk 10.8%) and in 62 patients in the endarterectomy group (5-year risk 8.6%) (HR 1.25, 95% CI 0.89 to 1.75; p = 0.20; see Table 8 and Figure 6).

Exploratory subgroup analyses showed no significant modification of the HR of the primary outcome event (Figure 7), nor of the combined outcome of procedural death or stroke, or non-procedural ipsilateral stroke by any of the evaluated variables (Figure 8).

FIGURE 7

Hazard ratios of fatal or disabling stroke between randomisation and end of follow-up in various patient subgroups. Subgroups are defined according to baseline characteristics and analysed by ITT for all available follow-up, apart from time from event (more...)

FIGURE 8

Hazard ratios of procedural stroke, death or ipsilateral stroke between randomisation and end of follow-up in various patient subgroups. Subgroups are defined according to baseline characteristics and analysed by ITT for all available follow-up, apart (more...)

Long-term functional outcome

There was no difference in distribution of functional disability as measured by the mRS scores at the end of follow-up, nor was there any significant difference 1 or 5 years after randomisation (Figure 9).

FIGURE 9

The distribution of scores on the mRS: (a) after 1-year and 5-year follow-up in patients allocated CAS or CEA using the Rankin scores in patients still surviving and in follow-up or who had died before at the indicated time points [permutation test comparing (more...)

Findings of the magnetic resonance imaging substudy

The MRI substudy has been previously published in detail.⁴⁴ A total of 231 patients (124 in the stenting group and 107 in the endarterectomy group) had MRI before and after treatment. Sixty-two (50%) of 124 patients in the stenting group and 18 (17%) of 107 patients in the endarterectomy group had at least one new DWI lesion detected on post-treatment scans done a median of 1 day after treatment [adjusted odds ratio (OR) 5.21, 95% CI 2.78 to 9.79; p < 0.0001]. At 1 month, there were changes on fluid-attenuated inversion recovery sequences in 28 (33%) of 86 patients in the stenting group and six (8%) of 75 in the endarterectomy group (adjusted OR 5.93, 95% CI 2.25 to 15.62; p = 0.0003). In patients treated at a centre with a policy of using cerebral protection devices, 37 (73%) of 51 in the stenting group and eight (17%) of 46 in the endarterectomy group had at least one new DWI lesion on post-treatment scans (adjusted OR 12.20, 95% CI 4.53 to 32.84), whereas in those treated at a centre with a policy of unprotected stenting, 25 of 73 patients (34%) in the stenting group and 10 of 61 (16%) in the endarterectomy group had new lesions on DWI (adjusted OR 2.70, 95% CI 1.16 to 6.24; interaction p = 0.019).

Studies on the predictors of risk of individual procedures

Findings of the cost–utility analysis

Resource use and costs

The cost–utility economics analysis has been published elsewhere.⁴⁸ Mean index procedure duration was 107 minutes [standard deviation (SD) 47 minutes] in the endarterectomy group (n = 700) and 68 minutes (SD 33 minutes) in the stenting group (n = 691; see Appendix 9). Eighty-two per cent of endarterectomy patients (n = 794) had general anaesthetic compared with 0% of stenting patients (n = 853). Eighteen per cent of endarterectomy patients had local anesthetic compared with 100% of stenting patients. In the endarterectomy group a shunt was used in 40% of patients (n = 818) and a patch in 66% (n = 693). In the stenting group a stent was deployed in 92% of patients (n = 816) and a cerebral protection device was used in 71% (n = 824). Sixty-four per cent of endarterectomy patients (n = 813) were admitted to the ICU post-operatively versus 52% in the stenting group (n = 808). Length of stay on the ward was 5.7 days (SD 9.4 days) for endarterectomy (n = 803) and 5.1 days (SD 10.8 days) for stenting (n = 789). In patients randomised to endarterectomy the mean number of additional endarterectomies during follow-up was 0.039 (SD 0.193) and the mean number of stents was 0.023 (SD 0.159). In patients randomised to stenting, the figures were 0.066 (SD 0.257) and 0.028 (SD 0.172), respectively. No patients in the endarterectomy group had a fatal MI during the first 30 days after treatment, compared with three patients in the stenting group; five patients had a non-fatal MI in the endarterectomy group compared with none in the stenting group; 28 patients in the endarterectomy group had severe haematoma compared with eight patients in the stenting group; one patient in each group had disabling cranial nerve palsy. Patients in both groups underwent a range of imaging tests; ultrasound was the most common in the endarterectomy group (234 tests; n = 857) compared with intra-arterial angiography in the stenting group (352 tests; n = 853). Drug usage 1 month after treatment was similar for both endarterectomy (n = 785) and stenting (n = 781) groups. Seventy-one patients (8%) in the endarterectomy group had one or more strokes during the 5-year time horizon compared with 114 patients (13%) in the stenting group; a higher proportion of strokes in the stenting group were non-disabling (61% vs. 37%).

Accounting for missing data using multiple imputation, mean total costs per patient were £6762 (95% CI £6154 to £7369) in the endarterectomy group (n = 857) and £7351 (95% CI £6786 to £7915) in the stenting group (n = 853) (Table 9). In both groups, approximately two-thirds of the total costs were for the index procedure and one-third for follow-up. Values were similar for complete-case analysis (see Appendix 9).

TABLE 9

Mean utility values, QALYs and costs per patient

Cost–utility analysis

Mean NMBs for endarterectomy and stenting were £57,793 (95% CI £55,994 to £59,592) and £57,580 (95% CI £55,699 to £59,461) at a maximum willingness to pay for a QALY of £20,000, and £90,070 (95% CI £87,520 to £92,621) and £90,046 (95% CI £87,329 to £92,762) at a maximum willingness to pay for a QALY of £30,000 (Table 9). In the base-case analysis there were no significant differences in costs between the two groups (mean incremental costs for stenting versus endarterectomy £537, 95% CI –£238 to £1312) or in outcomes (mean QALYs gained –0.010, 95% CI –0.117 to 0.097; Table 10). The incremental NMB for stenting versus endarterectomy was not significantly different from zero at a maximum willingness to pay for a QALY of £20,000 (mean –£723, 95% CI –£3134 to £1670) or £30,000 (mean –£830, 95% CI –£4265 to £2605).

TABLE 10

Incremental cost-effectiveness of stenting vs. endarterectomy: base-case and subgroup analyses

We repeated the analysis several times using alternative versions of the multiple imputation process using different random number seeds to investigate if the conclusions of the analysis changed; in every case the results were qualitatively the same (i.e. there were no significant differences between the two groups in terms of costs, QALYs and NMBs).

Sensitivity and subgroup analyses

The cost-effectiveness acceptability curve shows that at a maximum willingness to pay for a QALY of £20,000 the probability that stenting is cost-effective was 0.27; at a maximum willingness to pay for a QALY of £30,000 the probability that stenting is cost-effective was 0.31 (Table 10 and Figure 10).

FIGURE 10

Cost-effectiveness acceptability curve. Cost-effectiveness acceptability curve showing the probability that stenting is cost-effective vs. endarterectomy at different values of the maximum willingness to pay for a QALY. The probability that endarterectomy (more...)

Incremental costs, QALYs gained and incremental NMBs for stenting versus endarterectomy remained not significantly different from zero when rerunning the base-case analysis without adjustment and using complete cases (see Table 10). At a maximum willingness to pay for a QALY of £20,000 the incremental NMB for stenting versus endarterectomy was most sensitive to the cost of stents, cost of operating theatre time, cost per hospital day and cost of treating stroke (Figure 11), but in every case the incremental NMB was not significantly different from zero. Similar findings were obtained using a maximum willingness to pay for a QALY of £30,000 (see Appendix 9). In women and patients aged < 70 years, the mean QALYs gained from stenting versus endarterectomy were positive, whereas in men and patients ≥ 70 years they were negative, but in all cases the differences were not significantly different from zero, neither were the incremental costs and the incremental NMBs.

FIGURE 11

Univariate sensitivity analysis. All analyses are as for the base-case analysis with univariate adjustment of the parameters listed. Results are point estimates of the incremental NMB of stenting vs. endarterectomy (circles) and 95% CIs (capped spikes). (more...)