Time to treatment for stroke and problems with observational data

21 Apr

Here’s a big study suggesting earlier TPA treatment for stroke results in better outcomes. This of course isn’t a new idea but this is new data here to support it:

Saver JL, Fonarow GC, Smith EE, Reeves MJ. Time to treatment with intravenous tissue plasminogen activator and outcome from acute ischemic stroke. JAMA. 2013. PMID 23780461


  • data from the get with the guidelines registry (ran by American Heart and American Stroke Associations)
  • web based data collection tool for providers to enter patients into
  • from 2003-2012 which should quite the spectrum to include pre ECASS III and post ECASS III data
  • patients got into this either prospectively as they got treated or retrospectively by someone identifying a stroke on discharge documentation and entering them into the regsitry . The retrospective data is what we should be particularly suspicious of – you could easily pick and choose the ones you want to enter into the registry. Ultimately this is a study of patients that someone decided to enter into a registry. Some patients with data relevant to this question may well not be entered and therefore we don’t know what happened to them.
  • excluded patients (for this analysis): poor documentation, sites that had few stroke pts in the registry and those who got intra arterial treatments
  • there are, as usual, a lot of conflicts of interest and indeed the registry itself is funded by a pharmaceutical company (and has been funded by several others in the past)


  • 2000 hospitals submitting 1.2 million patients. So there’s a lot of data in this registry
  • they report 6% of these getting TPA
  • they further examine this 6% but exclude all the ones with dubious times, all the ones post 4.5 hrs and all those discharged to other hospitals (for whom they say they couldn’t get good outcome data). There are various other exclusions along the way.
  • note they don’t use the standard stroke outcomes of modified Rankin scale here. They use “ambulatory at discharge” or “discharge home” as surrogates of good outcome. This is even more dubious than the usual mRS assessed by postal questionnaire.
  • ultimately after all the slicing and dicing we get 58300 pts.
  • median time at treatment was 144 mins. 10% treated before 90 mins
  • they report 8.8% mortality, 4.9% ICH, 33% walking at discharge and 38% discharged home. Remember all the exclusions that went into this before hand so these numbers may be on the slightly optimistic side.
  • by comparing all these they find +ve associations between time to treatment and all of the outcomes. Because of the massive numbers all of these are statistically significant.
  • they even calculate NNTs to compare how much better people do when they are treated quickly.


  • The basic demographic data and the number generated have important audit, governance and even actuarial importance.
  • Registry data that is only as good as what people put into it, and it’s not clear how good that data is.
  • From the colossal numbers involved any difference will be statistically significant. Numerically significant and clinically significant are hardly the same.
  • The silliness comes with the comparisons and the implication of causation.
  • The study has no way to account for the confounding factors invovled in why someone might present to the hospital earlier than someone else, or why they might get treated earlier than someone else.
  • For example, it may well be that those who presented earlier were more likely to be having a TIA rather than an established stroke and it may be natural autoregulation and fibrinolysis that resolved their stroke rather than the TIA. Indeed if there was no actual infarcted brain then it is hardly a surprise that rates of ICH were lower in the TIA…ahem… I mean earlier treated group.
  • The point is that this data cannot answer the question they have asked of it. The only data that can answer this question is the data from the RCTs – the very data that is so contentious and controversial (among emergency physicians at least).
  • The associations presented here are all very interesting but add little real science. Rather it reinforces the rhetoric that makes acute stroke the interesting, frantic and emergent condition that we see every day.

The study did, however, remind of two of my favourite bits on causation. The first from XKCD that gets rolled out every month or so on this blog:


And the most thoughtful and sensible of science writers, the late, great Stephen Jay Gould. This is from an essay critiquing the slightly naive narrative we’ve told of Galileo and his achievements in proving heliocentrism. He interestingly got a little bit carried away and claimed that Saturn didn’t have rings but had two tiny planets at either end. His supportive proof was that he had “observed” it and therefore it must be true. All a touch embarrassing really…

Utterly unbiased observation must rank as a primary myth and shibboleth of science, for we can only see what fits into our mental space, and all description includes interpretation as well as sensory reporting. Moreover, our mental spaces house a complex architecture built of social constraint, historical circumstance, and psychological hope.

P50 The lying stones of Marrakech

Stephen Jay Gould

Further Reading/Listening:

LITFL – CCC Stroke Thrombolysis

LITFL – Michelle Johnston on tPA for stroke


EMCrit Cage Match – Jagoda v Swaminathan


Busting the Clotbusters – Domhnall Brannigan

SMART EM – thrombolytics for stroke


2 Replies to “Time to treatment for stroke and problems with observational data

  1. Hi there
    I keep referring to your blog as u have a nice summary about data and research in stroke
    I’m about to conduct a study exploring local factors associated with delay in presentation in stroke patient ??nit sure if you have covered this before in your blog love to knew

  2. Pingback: The LITFL Review 135 - LITFL

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.