QROT-046: Dirty Stats - Removing participants' data from trials so that the substance's bad effect then becomes statistically inconclusive
Jan 14, 2022, 12:30 PM
(Another highlight from John Abramson's revelations on JRE, and analysis and commentary on it)
What if* you could just BLAB the phrase "we use 'evidence-based science,'" but then you just remove only the data about study participants who got bad side effects in your safety and efficacy study? And you pruned very specifically to make it "inconclusive" whether the drug you studied was the cause?
That is, it is not that you merely make your study smaller up front to make it less likely to distinguish effects from chance, but you instead, for instance, start with a study of 8,000 participants, but then upon identifying the worse side effect, you take away three people's data from the study. That is you directly prune from the study three people who had the worst side effect. So now, while appearing to be honest, because you do show the bad side effect, short of the last three cases you removed, you can now say the side effect "COULD be random."
But it is only potentially "random" because you showed three fewer cases then you actually had. You ended up post-designing your trial based on what math would have made your side effect statistically likely to be attributed to what you're trying to get approved.
You still get to argue "good sized trial" because you have 7.997 participants. But you've committed fraud.
And recall from previous episodes, we have set it up so that this is basically undetectable short of litigation discovery - because the "all-knowing, and all-pure" drug companies get to keep the safety data case detail to themselves.
*They can, and they do. How often, who knows, because of the last note made. But the incentive is there. Let alone how much it is there once we mandate products and pay for them via central authority.
What if* you could just BLAB the phrase "we use 'evidence-based science,'" but then you just remove only the data about study participants who got bad side effects in your safety and efficacy study? And you pruned very specifically to make it "inconclusive" whether the drug you studied was the cause?
That is, it is not that you merely make your study smaller up front to make it less likely to distinguish effects from chance, but you instead, for instance, start with a study of 8,000 participants, but then upon identifying the worse side effect, you take away three people's data from the study. That is you directly prune from the study three people who had the worst side effect. So now, while appearing to be honest, because you do show the bad side effect, short of the last three cases you removed, you can now say the side effect "COULD be random."
But it is only potentially "random" because you showed three fewer cases then you actually had. You ended up post-designing your trial based on what math would have made your side effect statistically likely to be attributed to what you're trying to get approved.
You still get to argue "good sized trial" because you have 7.997 participants. But you've committed fraud.
And recall from previous episodes, we have set it up so that this is basically undetectable short of litigation discovery - because the "all-knowing, and all-pure" drug companies get to keep the safety data case detail to themselves.
*They can, and they do. How often, who knows, because of the last note made. But the incentive is there. Let alone how much it is there once we mandate products and pay for them via central authority.