Wednesday, July 06, 2005

Statins may be put in the water supply yet

If we can believe the epidemic of case-control studies, we may yet have to put statins in the drinking water because they seem to do everything. NEJM recently published a case control study that indicated protection from colon cancer with the use of statins. More recently, Dr. Vikas Khurana of the VA Medical Center in Shreveport, La. reported an analysis of a large database that claimed a reduction in the risk of 7 types of cancer by 50% including colon cancer. The records of 1,400,000 veterans from 10 VAs were analyzed.
These were, of course, retrospective case control studies and the author admitted that the duration and dose and type of statin was not factored in the analysis.
I believe the comments I made regarding the NEJM article apply here as well.
It has now been 20 years since NEJM published back to back articles dealing with the incidence of cardiovascular disease in women taking hormone replacement therapy. One paper was from the Framingham group, the other from Harvard reporting from the Nurse's Health Study. One reported a two fold increase in risk, the other a 50% decrease. Had either paper been published alone in such a high profile, widely read and quoted journal the results would have likely been widely acccepted. The editors, however, did the medical community a great service by their article juxaposition choice. Dr. John C. Bailar iii has a excellent chapter in his book " Medical Uses of Statistics" ( NEJM Books, Boston, 1992) in which he discusses Uncertainty about Confounders. Bailar asks the question why did these articles disagree so sharply.His answer was he did not know, an admission he also made in the editorial that followed these two articles. He could not give a specific reason for the difference, but in his book he says that the general reason involves unrecognized and perhaps unrecognizable differences in the study groups. You cannot control for unrecognized confounders in a case control or observational study.The scientific process involves drawing inference from information that is inevitably subject to error. The usual statistical measures of uncertainty ( p values and confidence limits) capture only that part of the uncertainty attributed to random variability in the context of the particular statistical model used.
Every study - even the cream of the crop, the RCT -is subject to those limitations of the statistical method. Of course, the RCTs- unlike the case control studies- control for even unrecognized confounders by the randomization process. So, case control studies and observational studies are much more vulnerable to limitations and the results of those studies need to be viewed with appropriate skepticism and restraint. Analyzing large data bases with outcomes against a battery of potential risk factors or risk modifiers always yeilds a set of statistically significant correlations. The trick is to know what to do with them and what leads to follow. So it may still be premature to add simvastatin to the water supply.

No comments: