Transcranial Laser Therapy in Acute Stroke Treatment
Results of Neurothera Effectiveness and Safety Trial 3, a Phase III Clinical End Point Device Trial
Background and Purpose—On the basis of phase II trials, we considered that transcranial laser therapy could have neuroprotective effects in patients with acute ischemic stroke.
Methods—We studied transcranial laser therapy in a double-blind, sham-controlled randomized clinical trial intended to enroll 1000 patients with acute ischemic stroke treated ≤24 hours after stroke onset and who did not undergo thrombolytic therapy. The primary efficacy measure was the 90-day functional outcome as assessed by the modified Rankin Scale, with hierarchical Bayesian analysis incorporating relevant previous data. Interim analyses were planned after 300 and 600 patients included.
Results—The study was terminated on recommendation by the Data Monitoring Committee after a futility analysis of 566 completed patients found no difference in the primary end point (transcranial laser therapy 140/282 [49.6%] versus sham 140/284 [49.3%] for good functional outcome; modified Rankin Scale, 0–2). The results remained stable after inclusion of all 630 randomized patients (adjusted odds ratio, 1.024; 95% confidence interval, 0.705–1.488).
Conclusions—Once the results of the interim futility analysis became available, all study support was immediately withdrawn by the capital firms behind PhotoThera, and the company was dissolved. Proper termination of the trial was difficult but was finally achieved through special efforts by former employees of PhotoThera, the CRO Parexel and members of the steering and the safety committees. We conclude that transcranial laser therapy does not have a measurable neuroprotective effect in patients with acute ischemic stroke when applied within 24 hours after stroke onset.
See related article, p 3175.
To achieve a population benefit from treatment of acute ischemic stroke, approaches that could have a modest clinical benefit but can be delivered to a majority of patients would be attractive as a supplement to existing treatments, such as intravenous alteplase or thrombectomy, that even in expert centers are offered to <25% of patients, whereas population wide, lysis rates may vary between <2% to ≤12%.1–4 The holy grail of neuroprotection has been sought through a range of pharmacological approaches, none with sustained success.5,6 In a nonpharmacological approach in which near-infrared laser energy was delivered transcranially has been hypothesized to modulate biochemical changes within neural cells and to be a candidate neuroprotective approach. The putative mechanism by which transcranial laser therapy (TLT) would work was predicated on the principles of photosynthesis: near-infrared radiation is hypothesized to match the wavelength absorption spectra of copper anions in cytochrome C oxidase, which is involved in the electron transport chain of the inner mitochondrial membrane. Delivering energy to hypoxic mitochondria was postulated to have potential to rescue ischemic cells or to prevent apoptosis without necessarily altering infarct volume, even when administered many hours after onset of ischemia.7–11
This series of basic science studies preceded 2 randomized phase IIa and IIb clinical trials, the neurothera effectiveness and safety trials (NEST)-1 and NEST 2, which included 780 patients (120 in NEST 1 and 660 in NEST 2).12,13 These provided reassuring safety data in patients, but the larger trial was neutral on efficacy end points. However, the combined data were interpreted as showing positive trends that justified definitive testing.14 From the second of these trials, a post hoc analysis identified a subgroup that was hypothesized to have higher potential for efficacy and this formed the target population for the present clinical trial.13 We conducted a randomized trial to test for the benefit of TLT in improving the proportion of patients with good functional outcome (modified Rankin Scale [mRS], 0–2) at 90 days after stroke onset.
NEST 3 was a double-blind, randomized, sham-controlled, parallel group, multicenter trial intended to enroll ≤1000 patients from ≈150 investigational sites. Patients were eligible for inclusion if they were aged 40 to 80 years, had a clinical diagnosis of acute ischemic stroke, no evidence of hemorrhagic infarct exceeding petechial bleeding along the margins, had a baseline National Institutes of Health Stroke Scale (NIHSS) score of 7 to 17, premorbid functional independence, and were unsuitable for treatment with thrombolytic treatment or thrombectomy approaches. Full details of the protocol have been published.15
Randomization and Treatment
Patients were randomly assigned in a 1:1 ratio to receive either TLT or sham. After consent, an interactive voice randomization system was used, with dynamic randomization at centers to ensure balanced distribution of treatment assignments. In addition to standard of care medical management throughout the trial, patients underwent a TLT or sham procedure. First, the scalp was shaved and sites for treatment administration were identified by a cap placed on the patients head. TLT was applied by handheld probe at 20 predetermined sites for 2 minutes at each site; all sites were used irrespective of stroke location. The NeuroThera Laser System uses energy at a wavelength of 808 nm, which is near-infrared, nonionizing, and is invisible to the naked eye. The sham procedure was identical to active TLT, with the exception that no laser energy was delivered from the device. To avoid retinal damage through inadvertent application of the beam to the eye, users were trained and performed the procedure in a laser safe environment.
The study was conducted in accordance with the Food and Drug Administration/International Conference on Harmonization Good Clinical Practice guidelines and applicable local regulatory requirements. The study was designed and overseen by a Steering Committee. The protocol was approved by all relevant institutional review boards and ethics committees. An independent Data Monitoring Committee (DMC) conducted regular safety reviews based on unblinded data and prespecified interim analyses. Data management and statistical analysis during the trial were conducted by an independent contract research organization (Parexel, Waltham, MA). After premature closure of the study, all available data were provided by Parexel to the DMC and analyzed by an independent statistician at the University of Glasgow on behalf of the DMC and Steering Committee. These committees assumed final responsibility for the analysis and interpretation of the results.
Patients were assessed by raters who had been trained and certified on the NIHSS and mRS.16,17 The NIHSS is a neurological function scale that ranges from 0 (normal) to 42: scores of 7 to 17 represent moderate or moderately severe neurological impairment. The mRS is an ordinal disability index ranging from 0 (no symptoms) to 6 (death). Baseline data collected before the treatment included age, sex, demographics, time from stroke onset to treatment, prestroke mRS, vital signs, and complete medical history. Follow-up assessments were conducted on days 5, 30, 60, and 90 with the mRS on day 90 representing the primary outcome measure.
Vital signs, neurological scores, concomitant medications, adverse events, and serious adverse events were recorded from study entry until day 90. Unresolved serious adverse events were followed up for ≤30 days further. Neuroimaging assessments were completed at baseline.
The primary efficacy end point was disability at 90 days or the last rating as assessed on the mRS, dichotomized as a success (score, 0–2) or a failure (score, 3–6). The primary efficacy end point was intended to be evaluated with hierarchical Bayesian analysis incorporating relevant previous data from the NEST 2. Three baseline covariates were included in the analysis: baseline NIHSS score, time from stroke onset to time of randomization, and age. Stroke severity as measured by the NIHSS at baseline was stratified in 3 levels: 7 to 9, 10 to 13, and 14 to 17. Time to randomization was stratified in 3 levels ≤8, >8 to 16, and >16 to 24 hours. Age was stratified as <70 and ≥70 years.
Using the Bayesian approach to analysis, the NEST 3 data would be considered alongside preexisting data derived from corresponding NEST 2 subgroups. In the event that the accruing data from NEST 3 were in close agreement with results from NEST 2 in corresponding subpopulations of patients, the previous NEST 2 patients could be considered to contribute to the statistical testing and thus to enhance statistical power. However, to guard against false-positive results, the maximum contribution from NEST 2 patients would be limited through the weighting that they are awarded and this was estimated to equate to an effective sample size increase of ≤100 patients from the 353 eligible subjects in NEST 2. If NEST 3 enrolled 1000 patients, this would represent <10% of the effective statistical sample.
An interim futility analysis was planned for when 300 patients had completed follow-up. Early stopping for success was not permitted at that time point. If the predicted probability of trial success was <10%, the DMC had the option to recommend termination of the trial. A second interim analysis was planned to be undertaken for efficacy after 600 patients had completed follow-up, at which the options were to stop for futility or overwhelming efficacy. The criteria for early stopping for futility were an effect size of <3.3% and <45% conditional probability of success (ie, if the data available at this stage were to be added to a hypothetical sample of future patients taken from a population in which a treatment benefit of 7% truly existed, how likely would this combined sample be to reach the target efficacy: <45% would represent futility). A recommendation to stop for efficacy would be issued if the probability of superiority of TLT over control was ≥99.5% (to allow for the final analysis, including patients randomized but not yet followed up achieving a 95% probability of superiority). The interim analyses were based on fully adjusted data, including any contribution of data from NEST 2 and were reviewed by the DMC alongside unblinded baseline and safety outcome data.
The primary analysis of the study was to be based on the intent-to-treat population, comprising all eligible patients who gave consent to participate and who were randomized. For safety analyses the as-treated population comprised all patients in the intent-to-treat population who had received a TLT procedure, but in this case they were analyzed by actual treatment received. The per-protocol population included all patients who received a TLT procedure and had no major protocol violation (determined before unblinding by an independent adjudicator). Assuming a control rate of 44%, a sample of 1000 patients was estimated to deliver 80% power to detect a true treatment effect of 6.8%, and although this sample would achieve statistical superiority for an observed treatment effect of 4.4%, a difference of ≥5% was prespecified as necessary to be clinically meaningful.
The primary efficacy analysis was planned to be undertaken by Bayesian methods at both interim analysis and final analysis, but although the Bayesian analysis that triggered early stopping is quoted, the final results presented here instead use a frequentist approach and are restricted to the NEST 3 patients only. There were 2 reasons for this: first, limited academic resources were available to complete all analyses after the sponsoring company filed for bankruptcy; and second, the statistical team involved in NEST 3 could not take full personal responsibility for data collected in the previous trials. To control for multiplicity, the order for hypotheses to be tested was prespecified as follows:
Dichotomized outcome of the 90-day mRS score (the primary efficacy end point).
Distribution of scores on the mRS (ie, an ordinal analysis).
Binary outcome measure based on the National Institutes of Health Stroke Scale.
There were also prespecified subgroup analyses of the primary efficacy end point according to the following list:
Time to stroke onset strata (<8, >8–16, and >16–24 hours).
NIHSS baseline strata (7–9, 10–13, and 14–17).
Age <70 and ≥70 years.
Sex (men and women).
Side of stroke (left and right).
Geographical location (enrolled within United States versus outside United States).
Previous stroke and transient ischemic attack (yes, no).
Previous diabetes mellitus (yes, no).
Previous ischemic disease (yes, no).
The Sponsor of the study was PhotoThera Inc, Carlsbad, CA, a privately funded medical device company. The protocol and statistical analysis plan were drafted by the sponsor, but final approval of the protocol rested with the Steering Committee. Both documents were also submitted to and approved by the US Food and Drug Administration. Nonvoting sponsor representatives on the Steering Committee included S.P.R. and S.G.D. The sponsor and Steering Committee were completely blinded as to the treatment modality (TLT versus sham procedure) and outcome data.
Between September 2010 and October 2012, 630 patients were randomized into NEST 3, of whom 316 were allocated to TLT versus 314 allocated to sham treatment (Figure 1). Enrollment was terminated when a futility analysis using the planned Bayesian methodology found an insufficient conditional probability of 13% for success: on the basis of 566 patients with complete data at the second interim analysis, the DMC noted that 140 patients in each group (49.6% versus 49.3%) had attained a successful outcome (mRS, 0–2; Figure 2).
As a result of premature discontinuation of the trial, financial support from the sponsor and its funders was abruptly discontinued, compromising systematic collection of outstanding data on all randomized patients who were still being followed up and greatly limiting resources available for further onsite monitoring and database cleaning. However, the contract research organization received authorization to release all relevant data files to the DMC under supervision of its chairman, and analyses were conducted at the University of Glasgow according to the protocol and analysis plan, with occasional compromises as mandated by incomplete data.
Baseline characteristics of the patients were well balanced between groups with mean age 66 years, median baseline NIHSS of 10, mean delay from stroke onset to treatment 16 hours, 37% female population, 50% left hemisphere location of stroke, and with 25% reporting atrial fibrillation and 33% diabetes mellitus (Table 1).
The primary efficacy analysis based on all randomized patients at conclusion of the trial observed success in 152 of 316 patients with TLT and 148 of 314 patients with sham, giving an adjusted odds ratio for benefit with treatment of 1.024 (95% confidence interval, 0.705–1.488; Cochran–Mantel–Haenszel P=0.7674; Table 2). On the secondary end point of ordinal mRS analysis at day 90, the odds ratio for benefit was 1.008 (95% confidence interval, 0.753–1.351; Cochran–Mantel–Haenszel P=0.7683; Figure 3; Table 3). On the secondary end point of dichotomized NIHSS (attainment of 0- or 1- or 9-point improvement from baseline), the odds ratio for benefit from treatment was 0.84 (95% confidence interval, 0.599–1.180; Cochran–Mantel–Haenszel P=0.3112). Mortality by day 90 was 15 (4.8%) of 316 in the TLT group and 19 (6.0%) of 314 in the sham group, log-rank P value 0.4759. There was no prespecified subgroup in which TLT treatment showed benefit on the primary outcome measure (Table 2).
Adverse events were reported by 261 patients in the TLT group (82.6%) and 249 in the sham group (79.3%). The investigator considered 5 events to be probably or definitely related to investigational treatment in the TLT group (0.5%) and 3 events (0.25%) to be probably or definitely related to sham treatment. These events attributed to treatment consisted of pain at the application site or related to the procedure, skin laceration, or erythema. Serious adverse events were reported in 66 patients with TLT (20.9%) and 88 patients with sham (28.0%).
The NEST 3 trial showed no clinically important safety concern with the use of TLT but indicates that it has no clinical efficacy under the circumstances used. All measures of outcome that were prespecified were unequivocally neutral. When combined with data from NEST 1 and NEST 2, there was no net benefit from treatment (Figure 4).
The NEST 3 trial was designed and conducted to rigorous standards otherwise applicable for pharmacological studies and not for device studies and has been analyzed and reported according to the protocol and analysis plan with 1 exception. Instead of the Bayesian analysis used for interim review, a final frequentist analysis has been reported for the total population that includes patients still being followed up at the time of the interim futility analysis.
Although funding and organizational support were lost for the follow-up of a cohort of patients recruited after June 2011, who were still in the trial at the time futility was announced in October 2011, it is certain that this did not generate a false-negative result because the interim analysis reviewed by the DMC was based on complete data and the results for the full analysis set do not differ from the futility analysis set.
For yet another neuroprotective strategy, we must re-examine the basis for hypothesizing treatment benefit. The concept that near-infrared radiation, whether applied by laser or any other means, would offer benefits was appealing but had flaws that were not satisfactorily discounted. Penetration of the skull had been confirmed, but the energy would reach only 2 cm into the brain cortex. Because energy dissipates according to the square of the distance, the dose received by superficial tissue must be exponentially greater than that received by deeper structures. Furthermore, only a few of the 20 skull sites used for light application are likely to be adjacent to penumbral tissue and a 2 minute application of energy that would translate into improved survival or recovery of neurons requires a certain degree of credulity to accept. However, even if the mechanism of action would be valid, the delay from stroke onset to treatment initiation of 16 hours—and potentially ≤24 hours—extends well beyond the time window at which powerful treatments have been shown to confer benefit. In that context, the targeted treatment benefit of ≈7% absolute improvement in success rate must also be perceived as optimistic.
There may be some uncertainty around the biological plausibility underlying the early clinical studies. However, the combined data from NEST 1 and NEST 2 clearly justified a definitive study that would confirm or refute their findings.14 The contractual and financial arrangements under which the study was conducted, which allowed the unsurprising declaration of futility at interim analysis to precipitate a crisis and absence of organized closure procedures at all sites, raises lessons for investigators and hospital administrations. Despite these problems, with good faith on the part of the CRO, Parexel, some remaining members of the sponsor management, as well as the steering and DMC, the integrity of the reporting process has been maintained.
To summarize the clinical evidence for an effect of TLT on clinical outcome, we reanalyzed the original data from the NEST 1 and NEST 2 trials. The mRS outcomes were secondary within the NEST 1 trial, and the principal analysis had been borderline positive. Several exploratory outcome analyses were reported and it is not explicit whether full details of these analyses were defined a priori, or even which adjustment approach if any had been prespecified for the primary outcome of binary NIHSS analysis. The choice of analytic method and of covariates for adjustment can exert a strong influence on interpretation when the total sample is small, as here. We consider that the most robust approach for retrospective analysis is to apply the same method for NEST 1 as was prespecified for NEST 3. Figure 4 reflects that choice and demonstrates only limited support for a beneficial effect of TLT in acute stroke, in contrast to an earlier report.14
There were many positive features about the NEST 3 trial and its conduct. Although marketing approval arrangements for devices differ from those of pharmaceutical agents and are considered by many to be too lenient, this was a rigorously blinded, randomized trial with robust statistical approaches. The Bayesian design, which was intended to support NEST 3 using certain relevant data from NEST 2, included a hierarchical component. This meant that if results accruing in NEST 3 matched the known results from NEST 2, then the effective sample size for the final efficacy determination could be slightly expanded by borrowing previous data. However, it was not enough that the results overall should match the phase II trial: the Bayesian model also required that there should be correspondence within several strata for the outcomes between the 2 trials. This design had been subjected to extensive modeling to confirm that the type 1 error rate would be controlled at 5%. Furthermore, the inclusion of futility analyses allowed the trial to be discontinued early saving several hundred patients from enrollment and reducing the total cost by many millions of dollars.
In summary, TLT using near-infrared radiation shows no efficacy for the treatment of acute ischemic stroke.
NEST 3 Trial Group: Steering Committee—Werner Hacke (chair), Justin A. Zivin (cochair); Peter D. Schellinger, MD; Gregory W. Albers, MD; Natan M. Bornstein, MD; Bjorn L. Dahlof, MD; Rachael Fulton, PhD; Scott E. Kasner, MD; Ashfaq Shuaib, MD. Independent Data Monitoring Committee—Joseph Broderick, Anastasia Ivanova, Karen Johnston, Kennedy R. Lees (chair), Bo Norrving. Participating sites—United States: Stanford Stroke Center, Palo Alto, CA: Greg Albers (3); University of Alabama, Birmingham, AL: Andrei Alexandrov (18); Hoag Memorial Hospital Presbyterian, Newport Beach, CA: David Brown (5); Winchester Medical Center, Winchester, VA: Patrick Capone (8); The Methodist Hospital, Houston, TX: David Chiu (17); Oregon Health Sciences University, Portland, OR: Wayne Clark (14); INOVA Hospital, Falls Church, VA: Jack Cochran (3); University of Washington ST. Louis, St. Louis, MO: Colin Deredyn (3); Erlanger Health System, Chattanooga, TN: Thomas Devlin (45); Guilford Neuro/The Moses H. Cone Memorial Hospital, Greensboro, NC: William Hickling (7); Munroe Regional Medical Center, Ocala FL: George Howell (7); University of North Carolina Healthcare, Chapel Hill, NC: David Huang (6); Cleveland Clinic, Cleveland, OH: Shazam Hussain (1); Carilion Roanoke Memorial Hospital, Roanoke, VA: Sidney Mallenbaum (5); UMass Memoriol Medical Center, Worcester, MA: Majaz Moonis (14); Gewinnett Medical Center, Duluth & Lawrenceville, GA: Marshall Nash (34); Saint Lukes Hospital, Kansas City, MO: Marilyn Rymer (4); Mission Hospital/Mission Neurology Services, Asheville, NC: Reid Taylor (5); Sparks Regional Medical Center, Fort Smith, AR: Margaret Tremwel (6); Canada: Grey Nunes Community Hospital, Edmonton, Alberta: Brian Buck (6); University of Alberta Hospital W. Mackenzie Health Sc. Center, Edmonton, Alberta: Ashfaq Shuaib (28); Peru: Hospital Nacional Dos de Mayo, Lima, Peru: Julio Perez (69); Germany: Universitätsklinikum Hamburg-Eppendorf, Hamburg, Germany: Christian Gerloff (2); Neurologische Klinik Bad Neustadt, Bad Neustadt, Germany: Bernd Greiwing (22); Kreisklinikum Siegen GmbH, Siegen, Germany: Martin Grond (6); HSK Dr. Horst Schmidt Klinik GmbH, Wiesbaden, Germany: Gerhard Hamman (11); Neurologische Universitätsklinik Aachen, Aachen, Germany: Thomas Haarmeiter (1); Heinrich-Heine Universität Düsseldorf, Düsseldorf, Germany: Sebastian Jander (2); Neurologische Universitätsklinik Erlangen, Erlangen, Germany: Martin Köhrmann (52); Universitätsklinikum Münster, Münster, Germany: Martin Ritter (11); Johannes Wesling Klinikum Minden, Minden, Germany: Peter Schellinger (26); Klinik und Poliklinik für Neurologie Leipzig, Leipzig, Germany: Dietmar Schneider (33); Charite Campus Berlin Mitte, Berlin, Germany: Jan Sobesky (4); Klinikum Frankfurt-Hoechst, Frankfurt, Germany: Thorsten Steiner (1); Johann Wolfgang Goethe Universität, Frankfurt, Germany: Helmuth Steinmetz (9); Universitätsklinikum Heidelberg, Heidelberg, Germany: Roland Veltkamp (22); Universitätsklinikum Essen, Essen, Germany: Christian Weimar (3); Austria: AKH Linz, Linz, Austria: Franz Gruber (15); Sweden: Sahlgrenska University Hospital, Göteborg, Sweden: Bjorn Andersson (5); Sjukhuset Lidköping, Lidköping, Sweden: Lennart Welin (7); France: CHRU Lille-Hospital Salengro, Lille, France: Didier Leys (1); Finland: Helsinki University Central Hospital, Helsinki, Finland: Turgut Tatlisumak (22); Switzerland: University Hospital Zürich, Zürich, Switzerland: Andreas Luft (7); Universitätsspital Basel, Basel, Switzerland: Philippe Lyrer (8); Center Hospitalier Universitaire Vaudois, Lausanne, Switzerland: Patrik Michel (14); Spain: University Hospital Vall D’Hebron, Barcelona, Spain: Carlos Molina (35); Complejo Hospitalario Universitario Albacete, Albacete, Spain: Tomas Segura (21). Contract Research Organization: Parexel. Independent Statistician appointed after the trial: Rachael Fulton, Glasgow, United Kingdom.
The University of Glasgow and the virtual international stroke trials archive (Virtual International Stroke Trials Archive [ViSTA]; http://www.vista.gla.ac.uk) donated staff time (Drs Lees and Fulton) and resources for statistical analysis of the neurothera effectiveness and safety trial (NEST)-3 data. The data from NEST 3 have been lodged within VISTA for future scientific access. Parexel kindly provided the original data sets and copies of statistical programming code to facilitate independent statistical analysis. Dr Hacke designed the protocol and led the study, shared supervision of the analysis, and reviewed the article for critical revision. Dr Lees handled supervision of the analysis, wrote the first draft of the article, has access to all raw data, and takes full responsibility for the content of the article. Dr Fulton performed statistical analysis and prepared a detailed statistical report. All authors have read, had an opportunity to comment, and have approved the final article.
Sources of Funding
The study was funded by PhotoThera Inc, Carlsbad, CA.
Dr Hacke received fees and expenses from PhotoThera as chairman of the neurothera effectiveness and safety trial (NEST)-3 steering committee and as a consultant to PhotoThera. Dr Schellinger received fees and expenses from PhotoThera as member of the NEST 3 steering committee. Dr Albers received fees and expenses from PhotoThera as member of the NEST 3 steering committee. Dr Bornstein received fees and expenses from PhotoThera as member of the NEST 3 steering committee. Dr Dahlof received fees and expenses from PhotoThera as member of the NEST 3 steering committee. Dr Fulton does not declare any conflict. She is paid by the University of Glasgow and by funds from The VISTA archive. Dr Kasner received fees and expenses from PhotoThera as member of the NEST 3 steering committee. Dr Shuaib received consulting fees and expenses from PhotoThera as a member of the Nest 3 Steering Committee. Dr Richieri was an employee by PhotoThera until April 2012 and a paid consultant until termination of the study. Dr Dilly was an employee by Photthera from January 2012 until termination of the company and continued unpaid until completion of the study close down activities. Dr Zivin received fees and expenses from PhotoThera as chairman of the NEST 3 steering committee and as a consultant to PhotoThera. Dr Lees received fees and expenses from PhotoThera as chairman of the independent data monitoring committee, until the trial was closed for futility.
* A list of all NEST-3 Trial Group participants is given in the Appendix.
Guest Editor for this article was Jeffrey L. Saver, MD.
- Received April 16, 2014.
- Revision received June 23, 2014.
- Accepted July 7, 2014.
- © 2014 American Heart Association, Inc.
- 2.↵Scottish Stroke Care Audit. 2010 National Report. Stroke Services in Scottish Hospitals. Data relating to 2005–2009. http://www.strokeaudit.scot.nhs.uk/Downloads/2010%20report/SSCAReport0610.pdf. Accessed January 19, 2014.
- Schwamm LH,
- Ali SF,
- Reeves MJ,
- Smith EE,
- Saver JL,
- Messe S,
- et al
- Donnan GA
- Lapchak PA,
- Wei J,
- Zivin JA
- Oron A,
- Oron U,
- Chen J,
- et al
- Lampl Y,
- Zivin JA,
- Fisher M,
- Lew R,
- Welin L,
- Dahlof B,
- et al
- Zivin JA,
- Albers GW,
- Bornstein N,
- Chippendale T,
- Dahlof B,
- Devlin T,
- et al
- Zivin JA,
- Sehra R,
- Shoshoo A,
- Albers GW,
- Bornstein NM,
- Dahlof B,
- et al
- Lyden P,
- Raman R,
- Liu L,
- Grotta J,
- Broderick J,
- Olson S,
- et al
- Quinn TJ,
- Lees KR,
- Hardemark HG,
- Dawson J,
- Walters MR