Share this article on:

Letter to the Editor: tPA Data ‘Tortured’ for Predetermined Conclusions

doi: 10.1097/01.EEM.0000446067.59393.e4
Letter to the Editor


The article “Conflicts of Interest on Guidelines Ramp Up tPA Controversy” was well-written. (EMN 2013;35[10]:1; of the controversy regarding the use of tPA is the way that the data were “tortured” to come to (predetermined) conclusions. The conclusion of ECASS III that treatment in the third 90 minutes (3 to 4.5 hours) has a “better” result than in the second 90 minutes defies biologic plausibility and occurs if you change the definition of outcomes. A lower incidence of intracerebral hemorrhage with later treatment defies biologic plausibility, and occurs only if you change the definition of significant hemorrhage. A better result than that of the initial NINDS trial, using experienced stroke neurologists, stroke neuroradiologists, and close monitoring and nursing/physician care, defies biologic plausibility.

Many of the “significant findings” are really statistical (disease-oriented outcomes) rather than clinical (patient-oriented outcomes) results. This is typically done when the results are not all that important compared with what is important, namely, an improved condition. The outcomes in ECASS III were manipulated so that the results were statistically significant; if the outcomes (the modified Rankin score) were kept the same as NINDS (the first study that was published, showing at least nominally a good outcome), then the results in ECASS III would not have been statistically significant.

There is a statistical paradox, sometimes known as the Will Rogers' effect, but more formally as Simpson's paradox, that occurs when you change definitions and the results improve in both categories. If you change the definition of TIA from a reversible event occurring within 24 hours to one occurring within one hour, the TIA group (with whatever or no treatment applied) will have a better outcome, as will the CVA group (as a higher number in this group will have had TIAs rather than completed strokes), so any intervention applied to the CVA group will appear to have better outcomes.

Publication bias has also tainted the clinician's impression of the outcomes. Even allowing for that, eight of the 10 major published studies on the use of tPA in stroke did not show clinical significance or, for that matter, even statistical significance unless you torture the data. What does that say for the value of this treatment? Not ready for prime time, in my opinion.

Michael H. LeWitt, MD, MPH

Devon, PA

© 2014 by Lippincott Williams & Wilkins