It was long ago, but here's how I remember reading my first drug safety leaflet:
"Dad, it says here that if I take this I could have a heart attack!"
"Let me see... Oh, 'in very few cases'. That probably means the chances are about one in a million. You're more likely to be run over by a car. Just go ahead and take it; your doctor knows what he's doing."
I can't find the cite, but I once read about a study which simply asked people to translate phrases like "in a few cases" and "in very rare cases" from drug safety leaflets into numerical chances. The results varied wildly, I think in some cases by a factor of 1000. Here's a simple policy proposal: You can keep your "very rare cases" all you like, but could you just add numbers? I'm aware these will be no more than estimates, and that people, including myself, are not good at reasoning about small probabilities, but it would certainly be more helpful than verbal descriptions hardly anybody understands.
Nothing as Useful as a Bad Theory
4 years ago
3 comments:
Unfortunately people are bad with numbers as well. I think they found that things like '1 in 100' (rather than 1%) were best understood. Often when they say that there are 'very rare' cases of X it means that only one or two people had the side effect and they don't really think it was due to the drug, but it was serious so they ought to cover themselves.
When there is convincing evidence for a serious side effect doctors usually tell you to look out for it.
Yup, people are bad with numbers, hence I suggest verbal descriptions plus numbers - then everybody can pick and choose.
"I think they found that things like '1 in 100' (rather than 1%) were best understood."
I don't remember that bit (and maybe they are different studies), but I'd put money on this prediction. If I remember correcly, another study I can't cite found that a substantial number of participants understood 20% to mean "one in twenty".
Gerd Gigerenzer and colleagues studied probabilities versus what he calls "natural frequencies" (e.g., "8 in 1000") in the context of Bayesian reasoning and found that natural frequencies work much better. His explanation for this is that probability theory was only invented 300 years (or so) ago, whereas we're probably adapted to thinking in natural frequencies.
"Often when they say that there are 'very rare' cases of X it means that only one or two people had the side effect and they don't really think it was due to the drug, but it was serious so they ought to cover themselves."
That's pretty much what I heard about the German situation. This has even led to a sort-of-joke in Germany: "Side effects: Spontaneously occuring death."
I've actually done a meta-analysis on a drug where we had to include a death in the drug group in our analysis of adverse events even though it was quite clearly not due to the drug (a skin cream).
Post a Comment