A few months ago Omega-3 fatty acids were promoted as the cure for everything from reading disabilities to autism to dementia -- often based on retrospective case-control studies.
Pshaw! Give me transgenic animal model studies any day ...
Although a number of epidemiologic studies reported that higher intake of omega-3 fatty acids (largely associated with fish consumption) is protective against Alzheimer's disease (AD), other human studies reported no such effect. Because retrospective human studies are problematic and controlled longitudinal studies over decades are impractical, the present study utilized Alzheimer's transgenic mice (Tg) in a highly controlled study to determine whether a diet high in omega-3 fatty acid, equivalent to the 13% omega-3 fatty acid diet of Greenland Eskimos, can improve cognitive performance or protect against cognitive impairment. Amyloid precursor protein (APP)-sw+PS1 double transgenic mice, as well as nontransgenic (NT) normal littermates, were given a high omega-3 supplemented diet or a standard diet from 2 through 9 months of age, with a comprehensive behavioral test battery administered during the final 6 weeks. For both Tg and NT mice, long-term n-3 supplementation resulted in cognitive performance that was no better than that of mice fed a standard diet. ... While these studies involved a genetically manipulated mouse model of AD, our results suggest that diets high in omega-3 fatty acids, or use of fish oil supplements (DHA+EPA), will not protect against AD, at least in high-risk individuals...
Now that's more like it. Nice and negative.
Eons ago I used to teach evidence-based medicine. Back then we had a hierarchy of evidential goodness. The very bottom was "Dr. Schmo at Harvard loves this surgery" , the very top was randomized, double-blind case control studies. Retrospective case-control studies were lower tier but respectable given suffient statistical wizardry.
Later "meta-analysis" slipped in, perhaps higher than it deserved .
Doing it today I'd want to put retrospective case-control much closer to Dr. Schmo (low, that is), and I'd like to see transgenic animal model studies much closer to the top (wasn't even on the list in my day). Of course this is just my opinion speaking (like Dr. Schmo); I haven't seen any (retrospective) research on how well results from transgenic animal studies hold up 3-5 years later. Hope someone does that in the next year or so.
 This never goes well.
 Gets messed up by unpublished studies though now the meta-analysts struggle mightily to find the unpublished.