Tuesday, July 19, 2005

data collection without theory is not science

Noted epidemiolgist , Dr. Diana B. Petitti, wrote in 1988 (from Kenneth Rothman's book, Causal Inference, p 149-152) that " the literature of epidemiology increasingly is becoming an archive of information derived from mechanical applications of multivariate analysis. ..tabulating every exposure against every disease...investigators are more interested in the mechanics of data analysis than in the substance of the issue being addressed."
It seem even more the case now that in 1988. A non-systematic, non random plunge into a recent ( May 1, 2005) issue of Internal Medicine News found two articles which may illustrate this point. One analyzed incidence cases of type 2 diabetes. Moderate drinkers of alcohol fared best in terms of not becoming diabetic, heavy drinkers and light drinkers had about the same risk. No comment was made re mechanisms. The second article is announced with a headline saying " Bone density screen tied to fewer hip fractures". This is presented as evidence that screening decreases fractures, not that preventive treatment perhaps arising out of the screening prevents fractures.It seem to prevent falls as well ( 16% versus 20%). Participants were screened for 31 variables.
Contrast these with an article from a recent Archive of Internal Medicine from Holland. The researchers postulated that based on the physiology of bladder contraction and the pharmacology of NSAIDs that there might be a relationship between acute urinary retention and NSAID use which is what they found in their data analysis. They studied the issue, formulated an hypothesis and tested it and found supporting data. The study did not falsify their thesis. A Popperian approach was followed in the NSAID article. Petitti says the opposing approach is the "inductivist" approach wherein " knowledge resides in cumulating data, and more data means more knowledge"
She continues " The crucial distinction is the difference between scientific knowledge and factual knowledge. Science is better described as a system of abstract theories than as an agglomeration of factual observations."
John L Casti in his book " Reality Rules" says "...in many ways there is nothing more useful and practical than a good abstraction. This calls to mind Hilbert's dictum that "there is nothing more practical than a good theory' ". I remember cramming for my internal medicine written exam ( back when there was a written and a oral for IM certification.) One of the other pulmonary fellows looked up from a pile of yellow, legal pads on which he had hundreds of notes and announced a epiphany. "Every thing is associated with everything else and whatever finding is described in whatever illness happen 20- 80% of the time".

No comments: