As I previously noted here, a study by Graham et al, examining the cardiovascular risks of Vioxx, was being considered for publication in the Lancet when it was prematurely published on the FDA’s website, much to the Lancet’s dismay. Today, that same study, with some modifications, has been published online at the Lancet. It is being widely quoted in the press, particularly the assertion that Vioxx is shown to be responsible for 88,000 to 140,000 excess cases of serious coronary heart disease in the United States between 1999 and 2004. Let us take a closer look at this assertion.
The study itself is from Kaiser Permanente. It is a nested case-control study of 1,394,764 persons who were exposed to various NSAIDs between 1999 and 2001. The subset of patients who had myocardial infarction or sudden death from coronary disease were identified and constituted the cases. They were each matched to 4 controls from the overall group (matched for age, sex and health-plan region). In the Lancet study published today, the various types of NSAIDS were then evaluated for the odds-ratio of coronary events compared with remote use of NSAIDS (basically similar to non-users) and compared with celecoxib. This study found that Vioxx users had an increased risk of coronary events of 1.34 compared to remote users, and 1.5 compared to current users of celecoxib. This risk increase was particularly significant for high dose Vioxx (3.00 compared to remote users and 3.58 compared with celecoxib users).
Apart from the the usual caveats that apply to non-randomized cohort studies, I don’t have much of a problem with this part of the study. It becomes more interesting when the excess morbidity from Vioxx is considered.
In the original report published on November 2 at the FDA’s website, the authors state in their discussion:
High-dose rofecoxib conferred a 3.7-fold increase in risk and standard-dose a 1.5-fold increase compared with celecoxib, the most frequently prescribed COX-2 selective agent… From 1999 to 2003, there were an estimated 92,791,000 prescriptions for rofecoxib, of which 17.6% were high-dose. Combining this with data on mean prescription length, we estimate that the increased rofecoxib risk observed in this study would yield an excess of 27,785 cases of AMI and SCD in the US over the years 1999-2003, with 53.4% due to standard-dose use.
Now compare this statement, which estimated an excess of 27,785 cases of AMI and SCD, with the following statement in the conclusion of today’s Lancet paper:
From 1999 to September, 2004, an estimated 106.7 million rofecoxib prescriptions were dispensed in the USA, of which 17.6% were high-dose. In two Merck-sponsored randomised clinical trials relative risks for acute myocardial infarction of 5 for high-dose rofecoxib and 2 for the standard dose were recorded. The background rate for acute myocardial infarction among control groups from studies of cardiovascular risk in NSAID users varied from 7.9 per 1000 person-years in CLASS to 12.4 per 1000 person-years in TennCare. Using the relative risks from the abovementioned randomised clinical trials and the background rates seen in NSAID risk studies, an estimated 88,000–140,000 excess cases of serious coronary heart disease probably occurred in the USA over the market-life of rofecoxib.
Note that here, the authors use relative risks for MI of 5 and 2 for their calculations, which they did not derive from their own data but rather from two trials that have nothing to do with their study. They then use these numbers to come up with estimates of excess cases of serious coronary disease between 88,000 and 140,000, rather than the 27,785 from their original analysis (the slight increase in prescriptions used in the calculations only accounts for a small fraction of the difference).
It is very possible that the latter numbers more accurately reflect the truth than the former ones. And I am quite convinced of the excessive cardiovascular risk of Vioxx. Nevertheless, it seems disingenuous, at best, to publish a large trial whose primary purpose is to estimate the cardiovascular risk from Vioxx, and then, in the conclusion, calculate the number of patients harmed using risk estimates from totally different studies, without even mentioning that this is what is being done. And, of course, the number 140,000 is what gets picked up.