(Total Views: 405)
Posted On: 02/25/2021 2:16:45 PM
Post# of 154139
Nice. I had never seen the "fragility index" discussed systematically with a meta-analysis, especially one limited to significant trials in important journals.
The fragility index is the number of "deaths" that would have to be flipped to "alives" (or vice-versa) for the trial to go non-significant, measured on the most sensitive side of the trial.
The article says the median was 2, with an inter-quartile range of 1 to 3.5.
Obviously, our "fragile" side is the control group, as it has half the population. Hopefully, my memory is good that with 88 total deaths, you needed 37 deaths in control to reach significance. That would be a fragility index of 1.
If that's right, deaths of 38 would be a median performance on fragility, 39 or over would mean we are less fragile than most published results.
Major caveat: the median may be different if you were to look only at registrational trials.
The fragility index is the number of "deaths" that would have to be flipped to "alives" (or vice-versa) for the trial to go non-significant, measured on the most sensitive side of the trial.
The article says the median was 2, with an inter-quartile range of 1 to 3.5.
Obviously, our "fragile" side is the control group, as it has half the population. Hopefully, my memory is good that with 88 total deaths, you needed 37 deaths in control to reach significance. That would be a fragility index of 1.
If that's right, deaths of 38 would be a median performance on fragility, 39 or over would mean we are less fragile than most published results.
Major caveat: the median may be different if you were to look only at registrational trials.
(2)
(0)
