News: Well Past Time to Stop Giving tPA for Stroke : Emergency Medicine News

Journal Logo

News

News

Well Past Time to Stop Giving tPA for Stroke

Garg, Ravi MD

doi: 10.1097/01.EEM.0000874652.75562.f0
    FU1-2
    Figure:
    EM best practice, rtPA, stroke, tPA
    FU2-2
    Figure

    A storied history of controversy surrounds the National Institute of Neurological Disorders (NINDS) rtPA stroke study, centering on whether a baseline difference in stroke severity favoring alteplase should have changed the interpretation of the results. (N Engl J Med. 1995;333[4]:1581; https://bit.ly/3IEdDjy.)

    Multiple post-hoc reanalyses were published using different statistical methods claiming the original trial results were valid with one exception written by emergency physicians Jerome R. Hoffman, MD, and David L. Schriger, MD. (Ann Emerg Med. 2009;54[3]:329; https://bit.ly/3PpJ437.)

    Otherwise, the trial has been celebrated, including its 25th anniversary at the 2020 International Stroke Conference. A principal investigator of the study also referred to it in 2021 as part of the stroke “renaissance.” (Cerebrovasc Dis. 2021;50[6]:666; https://bit.ly/3AMkPYN.)

    The impact of this trial on emergency medicine and neurology is undeniable.

    Compelling Evidence

    Randomization in a trial is intended to ensure the groups are comparable at baseline, but group differences can occur due to chance. A two-arm trial comparing 20 characteristics at baseline could be expected to yield one difference at the typical p value threshold of 0.05. The difference in baseline stroke severity in the NINDS study was assumed to be due to these chance differences after randomization. The alternative—faulty randomization—was never assessed. A detailed review of the trial suggested compelling evidence of faulty randomization and selection bias. (BMC Medical Research Methodology. 2022;22[1]:1; https://bit.ly/3IEBbEW.)

    The trial's method, which is known as stratified block randomization, failed to meet its primary objective, ensuring equal group allocations, in 11 of 16 strata. Three of eight randomization centers violated a prespecified rule regarding how many more participants could be enrolled in the 91-180-minute stratum compared with the 0-90-minute stratum.

    Typical features associated with selection bias were present in the trial in addition to the randomization errors. Envelopes used for allocation concealment in the trial are considered a weak method prone to tampering. Eight envelopes were opened without documented reasons at the end of the trial.

    Alteplase also characteristically froths when swirled for reconstitution. No reporting was done on how this frothing reaction with placebo was mimicked, leaving further room for failure of allocation concealment. The trial protocol also allowed for two opportunities to exclude otherwise eligible participants that could have facilitated biased randomization. Investigators were allowed to exclude otherwise eligible participants based on their subjective interpretation of the baseline computed tomography results. They were also allowed to exclude participants based on a prespecified restriction on enrollment in time windows.

    Compelling evidence of favoritism for the alteplase group was seen when baseline differences were evaluated on these criteria. Eight baseline differences favoring alteplase were identified, and all were prognostic factors for stroke outcome and known to investigators prior to randomization. Other facets of the trial such as a highly improbable crossover ratio also support selection bias.

    The chance of obtaining the crossover ratio in the NINDS trial is approximately the same as getting 21 heads in 22 coin flips. That nurses from the drug sponsor were present at sites collecting data (data later found to be corrupted) where participants were being randomized only further cast doubt on the integrity of the randomization. (Report of the tPA Review Committee. Aug. 25, 2004; https://bit.ly/3z5d6E6.)

    Most importantly, unlike chance imbalances, biased randomization cannot be overcome with statistical analyses limiting the conclusions made from prior reanalyses. Adjusted analysis may attempt to quantify the degree of selection bias, but a true treatment effect can only be obtained by means of an unbiased randomized trial. Remarkably, a Food and Drug Administration review found that the original authors did not perform an intention-to-treat analysis as was reported in the original publication and instead performed an as-treated analysis. (June 12, 1996; https://bit.ly/3APr1PW.) Little confidence should be placed in the original report or attempts to “correct” these methodological errors.

    ICH and Mortality

    Where does this leave thrombolytic treatment for acute stroke? The NINDS study is one of two originally positive reported randomized trials. The other, ECASS-3, has recently been shown to be negative once post-randomization residual confounding is accounted for with best analytical practices. (BMJ Evid Based Med. 2020;25[5]:168; https://bit.ly/3IDRWzK.)

    Those authors could not perform center-specific analyses assessing for selection bias as the NINDS rt-PA Stroke Study dataset did because the drug sponsor removed the center variable from the dataset (personal correspondence).

    Responding to the reanalysis, representatives from the drug sponsor have written: “As Activase in the 3-4.5-hour time frame is not FDA-approved, Genentech does not endorse the use of Activase use in this time frame.” (ACEP Now. April 23, 2021; https://bit.ly/3RyQZwG.)

    Now that evidence showed that the NINDS study suffered from selection bias, no individual randomized trials convincingly support using alteplase for stroke. Results from aggregate and individual patient data meta-analyses, which include the NINDS rtPA stroke study, should be considered hypothesis-generating similar to other meta-analyses with nonrandomized data.

    The most certain finding in the literature advocating thrombolytics for stroke is excess hazards in the form of intracerebral hemorrhage and mortality. (Acta Neurol Scand. 2022;146[1]:4; https://bit.ly/3O9HdOO.)

    Where else in medicine do we accept such biased evidence as justification for a potentially harmful treatment of patients?

    Share this article on Twitter and Facebook.

    Access the links in EMN by reading this on our website: www.EM-News.com.

    Comments? Write to us at [email protected].

    Dr. Gargis a neurologist at Loyola University Chicago Stritch School of Medicine in Maywood, IL. Follow him on Twitter@ravigarg415.

    Copyright © 2022 Wolters Kluwer Health, Inc. All rights reserved.
    • chechtmd7:21:56 PMInteresting article. This is the first time I have ever suspected that drug-company-sponsored trials might be prone to outcome tampering for the benefit of the drug company.
    • howardai1:24:59 PMGlad I am a retired EP. I was always dubious when administering tPA. I couldn't get my head to accept that a drug was recommended based on results 90 days later. In the ED, you give a drug and something happens--or not. I don't recall ever seeing immediate results after tPA! I was asked and in some institutions required to administer tPA on the vaguest of symptoms. Scary!
    • evanenglish12:12:58 PMJust like “pain is the fifth vital sign” began the opioid epidemic, Genentech with Activase has hoodwinked the public and neurology in a massive money-making scheme. How long will this travesty be allowed to go on?