While other people contemplate the politics of this, I’ll contemplate the statistics

No doubt that the “ongoing tiff”:http://pipeline.corante.com/archives/2006/02/22/nejm_vs_its_contributors_round_two.php#comments between NEJM(New England Journal of Medicine) and the co-authors of the VIGOR study has analysts pontificating and pharma scientists scratching their heads. I’ll leave the political issues to people who are closer to them than I.

For me, the issue might involve a central practice in my field. There were three cardiovascular deaths that occurred after a prespecified cutoff date in the Vioxx™ arm, and one stroke in the naproxen arm. These four deaths were not reported in the publication of the VIGOR study.

We have a reason for this practice in statistics. We do it to avoid any “data-driven hypotheses”There actually is a statistical methodology called “data-driven hypothesis-testing.” However, to do this correctly you have to lay out _exactly_ what you are doing before the data comes in. We detail our analysis, including the main analysis, what cutoff dates are, how to handle missing data, and so forth _before_ we know who belongs to which treatment group, and then we follow that plan. This is to avoid saying “Gee, this other analysis will make our drug look better” after seeing the data.

So my question is this: will this practice become unethical? Are we going to be ethically bound to change some of the details of our analysis based on new data?

*Update 2/26*: So, I’ve been thinking about this some more. I realize that someone could say, “I believe we should amend ethical practice to report all deaths, even if they happen 20 min before paper submission deadline.” Why stop there, why not all serious adverse events (usually hospitalizations or medical intervention needed to avoid death, life-long sequelae, or birth defects)? How about anything adverse, especially if it makes the drug look bad? Isn’t that the more conservative approach?

I don’t think so. Data such as the three deaths get reported to the FDA as part of a submission, and gets updated periodically through safety reporting mechanisms already in place. For a paper, you’re looking for information dissemination, not marketing approval. I think in this case, as long as the cutoff dates were established before they knew who received what, they should stick with it.

PS Professor Stemwedel “states the point”:http://scienceblogs.com/ethicsandscience/2006/02/does_extra_data_always_get_you.php more clearly than I can.

%d bloggers like this: