Agree we need to be rigorous about this. But in your example it's more sane if you flip it. Survival rate from a new medicine may be 99.99% and only allows 1.

Agree we need to be rigorous about this. But in your example it's more sane if you flip it. Survival rate from a new medicine may be 99.99% and only allows 1.0001 multiple of improvement. For severe outcomes, not allowing expansive language to describe improvements is a legitimate problem. i.e. if we insisted on this, it would be hard to motivate improvements in safety from 99% to 99.999% since they are "only <1% improvements". So in cases like this I think of it in reverse: the death rate of 0.01% can be improved by 90% (or 10x) to 0.001%. This does run the risk of inflating the importance of the fix when the base rate is extremely low. My rule is: articles which use multiples or percentages must give enough detail to exactly reconstruct the calculation method used.