Reading only the abstract (I might be able to pull the full article next week), I have a couple comments.
First, this was a type 2 model (the researchers are only recording data, not performing an experimental manipulation) analysis. Most medical studies are of this type, for ethical reasons, and suffer from inability to draw inferential conclusions about strong causation. Essentially, they can never be certain that the effect observed isn't an artifact of the group, because they aren't randomizing the group, and they aren't controlling the treatment. In this case, it's possible that it's not the anticholinergic causing the dementia, but the group that is otherwise predisposed for dementia is also self selecting for anticholinergic use, or that the underlying disorder being treated with the anticholinergics is responsible for the dementia rather than the drugs themselves.
Second, using a Cox proportional hazard model where you disregard the treatment in the last segment, and continuously replace censored data points is certainly not standard, and gives me pause, but might be explained better in the full article, and I would suspect would make their estimates more conservative, so that's probably OK.
Those caveats aside, it seems reasonable to conclude there is a link between anticholenergic use and dementia/Alzheimer's disease. I don't see evidence that the effect is necessarily cumulative, but there is definitely a dose-response relationship. Interestingly, at the lowest dose level, the use of anticholenergics decreased the the prevalence of dementia vs. normal population. This could be a hormetic effect, or just an anomaly; it's hard to say.
|