3 ms·
As all of these drugs are novel, the best you can do with "similar drugs" is to come up with a decent prior for a Bayesian risk model. But since each of these d
by Fomite 11y ago
As all of these drugs are novel, the best you can do with "similar drugs" is to come up with a decent prior for a Bayesian risk model. But since each of these drugs is new and untested then no, you don't have a numerical risk. You cannot say "1 in 167,234" or "1 in 1 Billion". You can say "Likely to be very, very small", but if you can conjure the risk of side effects for a drug with no data ever generated on that drug, I'd be impressed.
- Houshalter 11y agoFor the sake of argument, say there have been 100 drugs tested in the past. One of them killed someone, the rest were completely harmless. Now the next drug comes along to be tested and you know absolutely nothing about it. You need to estimate the risk, and all you know is the information above. Of course your estimate must be 1%. Unless you have more information to update that probability, 1% is the optimal baseline prior. You seem to be confused about what probability is. It's just a representation of uncertainty.
- Fomite 11y agoYour estimate is 1%. It's a good prior. It's not the actual probability of an adverse side effect.
- Houshalter 11y agoThere is no such thing as "actual probability"! An event either happens 100% probability, or doesn't happen with 100% probability. There is nothing else. If you flip a coin, it doesn't have 50% probability of landing heads. The atoms and their path through the air is predictable, and it is already determined that it will land heads by the time it's left your thumb, and perhaps long before. But you don't know that, so you can only estimate 50%. All probability is just a representation of your own uncertainty. If you bet anything other than 1% risk, over time you will lose money. After thousands of trials, 1% of them will go bad.