Saturday, September 5, 2009

COI vs. Investigator Bias: Clarifying the Issues

A recent paper (subscription required to access) coming out of the departments of epidemiology and statistics at UCLA offers a new take on conflicts of interest in pharmaceutical research, plus some valuable conceptual clarification.

To lay a bit of groundwork, most approaches to conflicts of interest in medical research, and whether they should be either disclosed or avoided, starts from one of two positions. The position I favor begins with the ethical values of professionalism and scientific integrity and looks askance at competing interests that might distract physicians or investigators from their primary duties to serve the patient, the public health, and scientific truth. In reply, defenders of close industry ties have appealed to the values to collaboration in the service of scientific innovation. The latter group has championed the notion of "intellectual conflict of interest" as a way of arguing that many things besides taking money from commercial outfits can bias the scientific investigator, and since in fact COI is all over the place, and essentialy unavoidable, stop obsessing about it and let the cash flow.

To this mishmash Sander Greenland brings a refreshingly empirical perspective which is value-free to the extent that anything in medical science can be value-free (another discussion I won't attempt here). Basically, Greenland asks about the broad category of investigator bias, which he defines as "the biasing of study results towards results expected or desired by the investigator." He approaches the problem as if he were the evaluator of a research study--such as would be the case if he were a physician reading a medical journal and wondering whether or not to apply the results of that study to his own practice. He describes his role as placing bets on the probability that the results of any study would approximate the scientific "truth" (again, whatever that is). He then asks the core question: If I am that evaluator placing bets on the truth of study results, would I be assisted by the disclosure of one particular sort of investigator bias, that which results from financial ties to the sponsor?

Greenland notes that both Type 1 and Type 2 errors are possible in placing bets. If he pays too little attention to commercial bias, he may accept study results that are untrue and misleading for care of patients. But if he goes overboard in attaching too much weight to COI, then he might reject true findings, which would have a similarly bad impact on patient care.

Greenland then proceeds to jump through a bunch of mathematical and statistical hoops where the likes of me cannot follow. Where he ends up after those gymnastics seems to be this. First, he agrees with folks like me that commercial COI-related bias is not something we just dreamed up because we hate the drug industry; it is real and substantial. If he didn't take it into account he couldn't place his "bets" rationally. Second, Greenland agrees with those urging disclosure of COI as a minimum requirement. As a rational bettor, he would be significantly aided in making his bets by having that information available.

Put slightly differently, Greenland disagrees with some earlier commentators who have criticized mandatory disclosure of financial COI, because it unfairly taints the way we read a new research study, where we should focus exclusively on the actual study methods to decide whether or not the results are trustworthy. And he does so without making any value judgments whatever about the personal intentions of any actors, or the goodness or badness of taking any form of money from industry. So far as he is concerned, everyone engaged in these practices might have the noblest intentions and the purest motives, and it would make no whit of difference for his calculations and results.

So to sum up, one thing the Greenland paper seems to offer us is an empirical approach to defending the propriety and indeed the necessity of full disclosure of financial COI in medical research publication. The second, I suggest, is a valuable conceptual clarification. Greenland seems to be suggesting that the proper umbrella concept is this thing called "investigator bias." Under that large category, bias resulting from financial COI forms a subcategory. As I have said in previous posts, buit without the fine-tuning of this language, those who try to introduce the concept of "intellectual conflict of interest" err in equating the specific problems associated with financial COI with the much more general problem of research bias overall. If we see that the general category is investigator bias, and the subcategory is financial COI, we have no need to introduce a new (and I would argue, spurious) concept such as "intellectual COI."

Thanks to Roy Poses at Health Care Renewal blog for calling this valuable paper to my attention.

Greenland S. Accounting for uncertainty about investigator bias: disclosure is informative. J Epidemiol Community Health 63:593-98, 2009.

No comments: