Monday, November 19, 2012

ECASS III Errata

This is my favorite sort of article to feature on this site – a probably-overlooked letter about a certainly-overlooked feature of a landmark trial.

This author, from Mt. Sinai, notes last year, the authors of ECASS III updated their online manuscript to change a p-value in their baseline characteristics from 0.03 to 0.003.  Since these are p-values for baseline characteristics, they're only for illustrative purposes – considering, in randomized controlled trials, all the differences do occur by "chance".  However, the conceptual interpretation of this change in ECASS III is the placebo group was inadvertently randomized to have a history of prior stroke by a 7% absolute difference – and the chance of that occurring randomly has now admitted to be 1 in 300 rather than 1 in 30.  When tremendously unlikely differences in baseline characteristics occur "by chance", it raises troubling questions regarding whether they truly occurred randomly.


Additionally, the author of this letter also makes the astute point that, because this difference in baseline characteristics did not reach statistical significance by the ECASS III authors' definition (0.004), it was not adjusted for in their data analysis.  
In his adjusted reanalysis (data not shown), the significance of outcomes favoring thrombolysis disappears (OR 1.19, CI 0.89-1.59).  Not necessarily surprising, considering the updated meta-analysis including IST-3 data published in The Lancet also makes the statistical significance of the benefit of thrombolysis disappear past 3 hours.

Thrombolysis for acute stroke remains some of the most distorted treatment data in emergency medicine, where this heterogenous patient population is being overtreated based on "eligibility", rather than "likelihood of benefit".

"Implication of ECASS III error on emergency department treatment of ischemic stroke."
http://www.ncbi.nlm.nih.gov/pubmed/23141561

No comments:

Post a Comment

Comments on posts > 10 days old will be moderated; blame spammers.