As background, here's how the study was performed (from MedCo's press release (pdf)):
The Clopidogrel Outcomes Study investigated medical and pharmacy claims data of 16,690 patients whoNo prospective data, mind you, but the results were "highly statistically significant."
were taking clopidogrel following a stent procedure and tracked the study subjects (editors note: using pharmacy and medical claims data) for a 12-month period from 2005 to 2006. The study compared a group of 6,828 patients who were concurrently taking a PPI and clopidogrel to a group of 9,862 patients who were only taking clopidogrel. When PPIs were examined individually, all of the associations were highly statistically significant.
Now, is this evidenced-based medicine or correlationally-based medicine? Not that these data might not be important. But when we use medical claims data to guide medical treatment recommendations, as Medco seems to be suggesting, we tread on a potentially very slippery slope. For instance, the presence of "myocardial infarction" on a medical insurance claim may be there to assure payment for services rendered, rather than to identify a new heart attack. As such, claims data are notoriously poor at conclusively identifying clinical endpoints. Should physicians change patient's medication regimens based solely on datamining studies like this? Or would prospective randomized trials be more appropriate to make treatment recommendations? Can the claim that this interaction "increased hosptial costs by 40%" (as their press release claims) be substantiated? Or is this a means of marketing Medco's services to their clients, the largest of which might become the federal government?
It is interesting to note the rush to datamining as the United States implements rationing policies in the name of "personalized medicine." But we only have to harken back to the days of PVC suppression in my field where we thought that reducing PVC's must inherently be helpful at reducing sudden death risk. These assumptions started on the basis of correlational analysis. Only though a carefully-designed prospective, randomized trial (the CAST Trial) did we find the opposite was true: that antiarrhythmic medications designed to suppress PVC's actually increased mortality.
While general correlations are helpful to identify areas for future study, we have to be cautious not to leap to full blown treatment recommendations based on retrospective correlational studies derived from database mining alone.