Epidemiology: friend or foe?

[Danny]

A good little read.......

BMJ 2004;329:467 (21 August), doi:10.1136/bmj.329.7463.467

reviews
PERSONAL VIEW

Epidemiology: friend or foe?

Geriatric research is rapidly growing. Twenty years ago it was nearly impossible to find a trial that included old people. The major subject heading (MeSH) term "aged, 80 and over" wasn't introduced until 1986. But in the past 10 years more than 10 000 randomised controlled trials were published that included this age group.

Alzheimer's disease is a good example of a geriatric disease that has long been ignored. It took us a while to realise that this devastating and incurable disease is about to become a pandemic. In the past decade we started a frantic search for clues that might lead to its cure. Since no one really knew where to start, we asked our friend epidemiology for help. Large cohorts of elderly people were studied. Epidemiologists looked for correlations, hoping to identify factors associated with a reduced risk of dementia. And with success. They gave us, among others, hormone replacement therapy in women and two groups of drugs, statins and non-steroidal anti-inflammatory drugs (NSAIDs). People using either of these types of drug were less likely to have Alzheimer's disease, they said. These hopeful findings were welcomed like a drop of water in the desert, and they were published in high ranking journals.

Friends aren't always right, but they do make life more interesting


All this inspired more research. The gold standard of evidence based medicine, the randomised controlled trial, was applied to confirm the data in a prospective setting. And in laboratories worldwide sophisticated in vitro and in vivo studies were designed to discover the pathophysiological mechanisms by which oestrogen, statins, and NSAIDs protect against Alzheimer's disease.

To our surprise, the randomised controlled trials gave us bad news. The drugs had no effect at all on Alzheimer's disease. How can we explain this? It seems that the original observational studies suffered from the epidemiologist's biggest enemy: bias. Even worse, with hindsight this bias could and should have been easily identified. Doctors do not readily prescribe cholesterol lowering agents to patients with dementia. Likewise, patients with cognitive impairment are not likely to see their doctor to have their cholesterol checked. This explains why patients without dementia are more likely to use a statin than patients with dementia. The same holds for the use of hormone replacement therapy in women. In the case of the NSAIDs it was chance that played a trick on us. Chance findings are a recognised risk of doing multiple regression analysis on large volumes of variables in population studies: you can end up finding that patients born on a Monday are more likely to develop dementia than patients born on other days of the week.

When we think about these events it seems that epidemiology?or rather its careless use?provided us with false clues. This led to a waste of valuable time and money. When scientists are so eager to find results, epidemiology can present as a friend but behave as our foe.

What came next is puzzling. The laboratories started to publish their data. They had discovered that inflammatory processes are involved in the pathogenesis of Alzheimer's and that NSAIDs can modify these. They also identified several mechanisms of action for statins. For example, statins reduce the formation of amyloid, one of the hallmarks of Alzheimer's. These data were enough to indicate that both drugs do have the potential to reduce the risk of the disease in certain patients.

So, where does this leave us? Were the epidemiological findings true associations after all? This would mean that the randomised controlled trials have led us astray. This is troubling news for a gold standard. It is, however, not inconceivable that the design of these trials was inadequate. The process resulting in Alzheimer's disease probably starts many years before clinical symptoms present. If the therapeutic effect of statins and NSAIDs lies at a point 10 to 20 years before the onset of symptoms, any trial that starts at a much later stage will never find an effect.

The alternative explanation is that the problem lies in the interpretation of the data. Maybe the inflammatory response seen in Alzheimer's is a non-specific reactive process and modifying it with NSAIDs has no true benefit. And maybe the amyloid that is reduced by statins is merely an innocent bystander and not a factor contributing to the disease. I think we're simply not clever enough at this point to have the broad view needed to solve the puzzle of Alzheimer's. It is like trying to describe a painting when your view is limited to a corner of it.

We started out looking for clues to a cure for Alzheimer's disease. With the help of epidemiology we found several. Unfortunately, not one of them has turned out to be useful, despite lots of research effort. What we are left with is a number of new hypotheses. So how should we judge epidemiology in this case: friend or foe? I still vote for friend. Friends aren't always right, but they do make life much more interesting.

J A H R Claassen, consultant and researcher in geriatric medicine

[injuryupdate]

Very good article, but the title is misleading. Epidemiology is overwhelmingly a fantastic tool that contributes massive amounts to human well being (e.g. proof that smoking is bad for you, led to cures for many infectious diseases, prevention of cervical cancer etc. etc.). Just because studies can have flaws or limitations or can be interpreted incorrectly does not mean that we should ever stop doing surveys or research. A bit like just because some medications have side effects, are they good or bad for you - obviously in the vast majority of cases prescription medicines are more helpful than harmful, just like epidemiology.