Causation, Correlation and Alzheimer's Research

The popular press has picked up on another questionable research result and used it to raise hopes of preventing Alzheimer's disease (AD). For example, CBS news ran a piece titled "Alzheimer's self-defense: Are two languages better than one?" (here) discussing findings from a study by Ellen Bialystok, a psychologist at York University in Toronto.

The original research (here) confuses correlation with causation, a distinction that gets lost in the popular press. Essentially, the researchers took whomever showed up at their treatment center in Toronto with AD symptoms, determined whether or not they were bilingual (this is of course interesting to the Canadians given that French and English are both the national languages) and determined how rapidly AD progressed over time for the two groups.

Epidemiological studies studies of intact populations have been a source of most ideas about AD prevention and most of the ideas have turned out to be wrong when tested by clinical trials. Consider (taken from Szekely et. al. 2007) life style factors (Mediterranean diet, physical activity, cognitive activity, etc.), neutraceuticals (Vitamins E and C, ginko biloba, Huperzine A, Folic Acid, and Curcumin) and pharmaceuticals (NSAIDs, hormone replacement therapy, anti-hypertension drugs, statins, immunotherapies, thiazolidinediones, arundic acid, etc.); none of them have proven effective in delaying the progression of AD. All were suggested by epidemiological studies.

The problem with epidemiological studies is that they do not control alternative causal explanations the way a correctly conducted randomized clinical trial would. Consider the factors implicated in AD from the causal diagram above (click to enlarge). Lifestyle factors, diet, education, genetic differences and bilingualism are all thought to affect cognitive reserve, that is, other parts of the brain's ability to deal with impairment. However, AD is thought to be a result of disease history (hypertension, diabetes, etc.), risk factors, cognitive reserve and random factors. None of these factors is controlled when we look at a non-random sample of patients who enter a Canadian clinic for AD treatment. What's especially important here is that Canada is supposedly a bilingual country but many citizens only speak English. That choice alone my define different populations with different risk factors.

Now, Ellen Bialystok is a respected researcher who specializes in the psychological study of bilingualism. Given her specialty, she probably feels compelled to report any conferred advantages of bilingualism. Unfortunately, just like all the many other dead-ends generated by epidemiological studies, the result is unlikely to replicate. That's how science works: results that don't replicate are forgotten.

What is unfortunate is that the distinction between correlation and causation is lost on the popular press and is probably lost on most of the readers. As statisticians, although we might have succeeded in teaching students how to calculate a correlation coefficient, we have evidently failed to successfully communicate (even to the editors of medical journals) the difference between correlation and causation.