Stroke Symptoms as a Surrogate in Stroke Primary Prevention Trials

Abstract

Background and Objectives

The use of surrogate end points can decrease sample size while maintaining statistical power. This report considers incident stroke symptoms as a surrogate end point in a post hoc analysis of asymptomatic patients from the multicenter, randomized Carotid Revascularization Endarterectomy vs Stenting Trial (CREST).

Methods

CREST assessed stroke symptoms using the Questionnaire for Verifying Stroke-free Status (QVSS) at baseline and follow-up. While the primary analysis of CREST defined “asymptomatic” as having been free of stroke/transient ischemic attack for 180 days, herein the population was further restricted by requiring no stroke symptoms at baseline. Incident adjudicated stroke was defined the same as for the primary analysis; incident stroke symptoms was defined as developing ≥1 stroke symptom in follow-up. Treatment differences between stenting (CAS) and endarterectomy (CEA) were assessed for 3 end points: adjudicated stroke, stroke symptoms, and adjudicated stroke or stroke symptoms.

Results

The cohort included 826 of the 1,181 asymptomatic patients in CREST. Adjudicated stroke events occurred in 44 patients, and incident stroke symptoms occurred in 183. Analysis of adjudicated stroke end points demonstrated a nonsignificant hazard ratio (HR) for CAS compared with CEA of 1.02 (95% CI 0.57–1.85). The corresponding HR for the incident stroke symptoms outcome was 1.54 (95% CI 1.15–2.08), and the HR for the composite outcome of adjudicated stroke or incident symptoms was 1.38 (95% CI 1.04–1.83), both significant.

Discussion

The low stroke event rates in asymptomatic patients challenges the assessment of CAS-versus-CEA treatment differences. Incorporating incident stroke symptoms as a surrogate outcome increased the number of events by over 4-fold. The analysis demonstrated a previously unreported significant difference in cerebrovascular risk with CAS compared with CEA. We propose that broadening the end points of primary stroke prevention trials to include surrogate events such as incident stroke symptoms could make trials more feasible.

Rates of stroke continue to fall in individuals with carotid stenosis coincident with advances in medical therapy. Randomized clinical trials (RCTs) using traditional clinical cerebrovascular end points have required ever-larger sample sizes to provide adequate statistical power. For trials addressing other diseases, investigators sometimes consider surrogate end points with a higher incidence to permit surmountable sample sizes.¹ For example, despite the clinical interest for stroke trials focusing on the prevention of symptomatic stroke events, some have opted to use brain MRI–detected asymptomatic cerebral infarcts as a surrogate outcome to increase the number of events and statistical power.² The use of a surrogate outcome in any trial is appropriate only if the outcome is associated with the clinically recognized outcomes of interest, and positive treatment effects for the surrogate outcome do not always imply a positive treatment effect on the more clinically important outcome.³ However, to the extent that a surrogate outcome is associated with the clinically recognized outcomes, their use can substantially reduce sample size and improve the efficiency of trials.

Another potential surrogate outcome to increase events could be stroke symptoms. The Questionnaire for Verifying Stroke-free Status (QVSS) is a standardized, validated instrument that assesses 6 key stroke symptoms.^4,5 The prevalence of stroke symptoms is high, with approximately 20% of the stroke/transient ischemic attack (TIA)–free general population older than 45 years reporting having at least one of these symptoms, as detected by the QVSS.⁶ These self-reported stroke symptoms are related to cerebrovascular health and symptomatic stroke outcomes. Compared with those not reporting symptoms, those reporting symptoms are at approximately 30% increased risk of an incident stroke,^7,8 experience reduced physical and mental quality of life,⁹ are twice as likely to develop cognitive impairment,¹⁰ and are nearly twice as likely to be hospitalized for subsequent stroke or cardiovascular disease.¹¹ That stroke symptoms as assessed by the QVSS are strongly associated with these outcomes of clinical interest supports the use of the symptoms as a surrogate outcome in trials.

The Carotid Revascularization Endarterectomy vs Stenting Trial (CREST) was a primary and secondary stroke prevention trial designed to assess the relative efficacy of carotid stenting (CAS) vs carotid endarterectomy (CEA) in the neurologically asymptomatic and symptomatic population with carotid artery stenosis.¹² For patients with asymptomatic high-grade carotid stenosis, over a 10-year follow-up, CREST reported a nonsignificantly higher stroke risk (any periprocedural stroke or postprocedural ipsilateral stroke) associated with CAS compared with CEA (hazard ratio [HR] 1.23; 95% CI 0.75–2.02; p = 0.41).¹³ In trials with 1-to-1 randomization, statistical power is a function of only the number of events and not the sample size,¹⁴ and even with 10 years of follow-up, only 64 adjudicated stroke events occurred in the 1,181 asymptomatic patients in CREST.¹³ With only 64 events of 1,181, the true treatment HR would have to be greater than 2.25 to be detected with 90% power,¹³ a treatment effect that is clinically unlikely to be achieved. Our objective was to test the hypothesis that surrogate end points that included incident stroke symptoms would increase the number of outcome events in CREST, thereby supporting the feasibility of their use in future stroke trials performed in lower-risk populations and in the face of temporal declines in incident stroke parallel with advances in medical therapy.¹⁵

Methods

Details of the CREST clinical trial design, eligibility criteria,¹² and primary results¹³ have been previously reported. Patients were enrolled from December 2000 through July 2008 and followed at 117 sites in the United States and Canada. Median duration of follow-up was 7.4 years. For the trial, asymptomatic status was defined as the absence of stroke or TIA symptoms in the past 180 days. CREST long-term follow-up assessed stroke symptoms using the QVSS at baseline and each follow-up visit (1, 3, 6, 9, 12, and every 3 months out to 120 months). The QVSS is an 8-item lifetime incidence questionnaire where the first 2 items inquire about a medical history of stroke or TIA and the subsequent 6 items (items 3–8) inquire about classical stroke symptoms (Table 1). Patients were defined as having incident stroke symptoms if they reported no stroke symptoms at baseline (i.e., negative responses to QVSS items 3–8) but reported one or more of the stroke symptoms at a follow-up visit. The QVSS does not characterize the duration of symptoms and thus does not differentiate symptoms of TIA from symptoms of stroke. Because it would likely be considered inappropriate to include patients in a future trial using stroke symptoms as the primary outcome, we excluded patients who had already experienced the trial outcome at baseline, specifically excluding patients reporting any stroke symptom on the QVSS at baseline. We further excluded patients without baseline QVSS and patients without follow-up QVSS. Patients were censored in the analysis when they died or otherwise withdrew from the follow-up. Because any patient reporting having had stroke symptoms at baseline was excluded from the analysis, this analysis represents an assessment of treatment differences in the development of incident stroke symptoms.

Table 1.

Questionnaire to Verify Stroke Symptoms

Descriptive statistics were generated by assigned treatment group: CEA or CAS. Kaplan-Meir curves were generated, and proportional hazards analysis with HRs was performed for 3 end points out to 10 years: (1) adjudicated stroke, (2) stroke symptoms, and (3) either adjudicated stroke or stroke symptoms. The time to first event was calculated using the date of stroke for adjudicated strokes and date of the follow-up visit for QVSS events.

Standard Protocol Approvals, Registrations, and Patient Consents

All participants in the CREST trial provided written informed consent. All institutions received Institutional Review Board or equivalent ethics committee approval before trial initiation. The trial is registered with ClinicalTrials.gov (NCT00004732). All participants provided written informed consent.

Data Availability

Anonymized data not published within this article will be made available by request from any qualified investigator. Data used in this report come from the completed CREST trial. The archived clinical research data set for CREST is available with a completed National Institute of Neurological Disorders and Stroke Data Request Form submitted to CRLiaison@ninds.nih.gov. More information can be found on the appropriate National Institute of Neurological Disorders and Stroke website.¹⁶

Results

Of the 2,502 patients randomized in CREST, 1,181 were asymptomatic. After excluding those without baseline QVSS, those who reported symptoms at baseline and those with no follow-up QVSS, 826 patients were included in these analyses (Figure 1). A total of 1.7% of asymptomatic patients with carotid stenosis in the trial did not have a baseline QVSS. Reasons for failure to obtain a baseline QVSS were not tracked but could include language barrier. There were 406 patients assigned to CEA and 420 assigned to CAS. In this cohort, 14,495 follow-up QVSS questionnaires were completed (mean number completed = 17.5). Table 2 shows the demographic and clinical characteristics of the 2 treatment groups. There were 44 patients with an adjudicated stroke event, while 183 developed incident stroke symptoms. Table 3 shows the prevalence for each individual stroke symptom on the QVSS, with patients assigned to CAS having a nominally higher proportion of symptoms for all 8 items.

Figure 1. Cascade Diagram Showing How the Study Population Was Derived From the CREST Trial Population.

CREST = Carotid Revascularization Endarterectomy vs Stenting Trial; QVSS = Questionnaire for Verifying Stroke-free Status

Table 2.

Baseline Demographic and Clinical Characteristics of Patients in This Subset Analysis by CEA and CAS Treatment Groups

Table 3.

Prevalence of Positive Response for Individual Items on the Questionnaire for Verifying Stroke-Free Status at First Instance of a Reported Event During Follow-up

For CAS compared with CEA, Kaplan-Meier estimates of treatment differences for each of the 3 end points are shown in Figure 2A–C. The estimated HRs are provided in Table 4. For the outcome of adjudicated stroke, the CAS-to-CEA HR was 1.02 (95% CI 0.57–1.85). For the outcome of stroke symptoms, the CAS-to-CEA HR was 1.54 (95% CI: 1.15–2.08), a significant difference. For the composite outcome of adjudicated stroke or stroke symptoms, the HR was 1.38 (95% CI 1.04–1.83), also a significant difference. The proportionality assumption was assessed using a time-dependent covariate and where there was no evidence of a nonproportionality for any of the 3 models (p > 0.05).

Figure 2. Survival Plots by Treatment Group for the End Points of Adjudicated Stroke (A), the Composite of Adjudicated Stroke or Incident Stroke Symptoms (B), and Incident Stroke Symptoms (C).

CAS = carotid artery stenting; CEA = carotid endarterectomy.

Table 4.

Estimated CAS-to-CEA Hazard Ratio for Asymptomatic Patients in the CREST Trial for Each of the End Points Considered in This Report

Discussion

In this post hoc, subset analysis of the CREST trial, the inclusion of the surrogate outcome of stroke symptoms as assessed by the QVSS indicated significant treatment differences between CAS and CEA not previously detected. The CAS vs CEA 10-year treatment differences among the full sample of 1,181 asymptomatic patients were not significant for risk of the adjudicated stroke outcome (HR 1.23; 95% CI 0.75–2.02; p = 0.41) With only 64 adjudicated strokes of 1,181 patients in the original analysis, a potential contributor to the lack of a significant difference was the relatively small number of adjudicated strokes. Even with the exclusion of patients reporting stroke symptoms at baseline that implied 30% reduction in sample size to 826, there was a greater than 4-fold increase in events from 44 to 197 events. This increase in events is sufficient to decrease the HR detectable with 90% power from 2.66 to 1.59, moving the detectable difference into the range that is clinically feasible to achieve.

Not only does the use of surrogate outcomes make the assessment of CAS-to-CEA treatment differences feasible but it could also make trials in other low-risk populations feasible, such as intervention trials in younger adults. Trials are also difficult to complete in populations where recruitment of a large cohort is a particular challenge. For example, recruitment of African Americans,^17,18 individuals with low socioeconomic status,^18,19 rural residents,^20,21 and those with low access to health care²⁰ to trials are an acknowledged challenge; use of a questionnaire-based outcome may make trials in these and other populations more feasible.

Whether stroke symptoms are a valid outcome in prevention trials merits comment. The use of a surrogate outcome is appropriate only to the extent that that the surrogate is associated with outcomes of clinical importance. Stroke symptoms have been shown to be associated with higher risk of clinically pronounced stroke,^7,8 cognitive performance,¹⁰ lower health-related quality of life,⁹ and higher subsequent risk of hospitalization for cerebrovascular/cardiovascular diseases.¹¹ Although these associations suggest stroke symptoms as an appropriate surrogate outcome, its use as a surrogate outcome still requires the presumption that reducing risk of stroke symptoms will reduce the risk of the clinical outcomes of greater interest. In addition, subclinical stroke is not necessarily asymptomatic. We hypothesize that the incidence of new stroke symptoms without a stroke diagnosis, at least in some instances, results from small infarcts. That is, some “silent” strokes may in fact be “whispering” strokes with clinically consistent symptoms that fail to result in a diagnosis of stroke or TIA.^8,22 This may be due to the symptoms being insufficiently pronounced or not persistent enough to raise concerns to the patient or family; such symptoms may not be reported to patient's physician so the diagnosis of stroke or TIA is not formalized. It is beyond argument that interventions are assessed to reduce the burden of stroke and not to prevent TIA or “subclinical” stroke symptoms. Even if prevention of TIAs or stroke symptoms is not the “goal” of the intervention, if the treatment has a similar treatment impact on these milder end points, it adds evidence to the treatment impact on adjudicated (and more severe) end points. However, this inference to the adjudicated and more severe end points rests on the assumption of a close association of the surrogate end points with these more clinically pronounced end points.

We studied the potential utility of periodically and systematically assessing for stroke symptoms in a stroke primary prevention study. There may be broader implications of our study for other areas of neurology. Other nonstroke neurotherapeutic trials rely on self-reported events. Typically, these trials seek to prevent events of apoplectic onset, often occurring outside of a hospital setting. Examples include preventing seizures after head trauma,²³ preventing falls in patients with Parkinson disease,²⁴ and preventing recurrent migraines.²⁵ Optimizing methods for ascertaining incident and recurrent symptoms in these and other neurologic conditions may improve future trial designs.

There are limitations to be acknowledged about this study. The results are based on a post hoc subset analysis of a RCT and are generally considered hypothesis-generating.²⁶ However, we tried to follow the general rules for post hoc analyses.^27,28 We would submit that guidance regarding relative risks and benefits of carotid stenting vs endarterectomy in both asymptomatic and symptomatic patients should be drawn from reports of the results of the prespecified primary aims of the main study²⁹ and the long-term follow-up¹³ of the CREST-2 trial. An acknowledged weakness of the use of self-reported stroke symptoms is an inability to adequately characterize the potential underlying lesion. Of greatest concern is a lack of ability to establish with the QVSS if the underlying lesion provoking symptoms is ipsilateral or contralateral to the “study artery.” However, even in the potential absence of a benefit to contralateral stroke risk, the inclusion of such contralateral events would only add “noise” to treatment assessment, and this noise will be more than offset by the increased statistical power from the increase in events. In addition, we used interval-censored data for the QVSS events because of these only being captured at scheduled visits, whereas the adjudicated stroke events were based on date of event, and as such, the adjudicated stroke is assessed using time as a continuous variable, while stroke symptoms are assessed only at clinic visits and as such are “interval-censored.” However, the use of “ordinary” proportional hazards analysis is likely justified given the larger number of follow-up visits.

In conclusion, we propose that broadening the outcome of stroke prevention trials to include stroke symptoms could be an approach that will allow trials to be mounted in low-risk populations Although the use of these surrogate end points does provide a substantial gain in statistical power, it does come with the price of assuming that the treatment effect of an intervention is relatively consistent across the spectrum of severity of neurologic cerebrovascular end points. There needs to be more attention and research into stroke symptoms in stroke prevention trials.

Glossary

CAS

carotid artery stenting

CEA

carotid endarterectomy

CREST

Carotid Revascularization Endarterectomy vs Stenting Trial

HR

hazard ratio

QVSS

Questionnaire for Verifying Stroke-free Status

RCT

randomized clinical trial

TIA

transient ischemic attack

Appendix. Authors

Study Funding

The study was funded by the National Institute of Neurological Disorders and Stroke U01NS080168.

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.