Ooh, so it's super vulnerable to things like Simpson's "paradox". That is, the situation where treatment A does better with low-severity patients, and treatment A does better with high-severity patients, but when combined all together, treatment B does better overall. Not saying that this...