I’ve worried that disclosing conflicts of interest may be counterproductive ever since we did an experiment that showed that readers of articles with declared conflicts discounted not only the believability of the results, but every aspect of the paper, including its importance and originality.  Now my worries are increased by a paper in JAMA and an editorial in PloS Medicine providing evidence that disclosure of conflicts of interest may increase bias, the opposite of what is intended.

George Lowenstein and two colleagues in JAMA describe the experiments that led them to this uncomfortable conclusion. They tried to mimic the doctor patient relationship by making people advisers (doctors) or estimators (patients). The estimators had to estimate the value of coins in a jar or house prices but with inadequate information. Advisers were given more but still incomplete information and provided estimators with suggested estimates. Estimators were paid based on the accuracy of their estimate, and advisers were either paid based on the estimators accuracy (so not conflicted) or paid based on the overestimate of the estimators (so conflicted). Sometimes the conflict was disclosed to estimators.

As expected, conflicted advisers gave higher (more biased) estimates, but the key finding was that the estimates were still higher (more biased) when the conflicts were disclosed. With disclosure advisers made more money and estimators less—which translates to patients suffering poorer outcomes while doctors made more money. Think of those KOLs (key opinion leaders) feasting in the South of France while their patients back home receive worse treatment.

Why might disclosure produce these disturbing effects? One explanation, write Lowenstein and colleagues, is strategic exaggeration—because conflicted doctors know their advice will be discounted they bias it even more. (This all happens unconsciously, which is how conflict of interest always works.) Another explanation is moral licensing—feeling that because the conflict is disclosed it’s acceptable to give even more biased advice. (Moral licensing, the authors point out, may work with the whole profession—feeling that because we are now more serious about disclosing of conflicts we can give more biased advice.)

So if all this is true where does it leave us? It can’t mean that we return to not declaring conflicts of interest. It surely means, however, that in certain circumstances—perhaps the drawing up of guidelines—we should work hard to have more advisers without conflicts. (This is tricky, however, because I don’t believe that it is only financial conflicts of interest that matter, and even financial conflicts are not all about  drug companies).

Lowenstein and colleagues do have two practical pieces of advice from research. Firstly, second opinions can help mitigate conflicts of interest, and in their American context they suggest that insurers might cover the cost of a second, unconflicted opinion. (It crossed my mind that the insurers might pay more for an initial unconflicted opinion—but that would in itself provide a conflict.)

A second response is based on disclosure working better when it comes from third parties. The authors advocate “a unified web-based universal online disclosure form” (which the International Committee of Medical Journals has done), “storing and making available detailed explanations about payments received” (which the Sunshine Act in the US is mandating), and “bringing in intermediaries such as consumer watchdog groups to help patients make sense of the information.” A national database open to everybody—such as MPs have—might be a good idea for Britain.

My main conclusion comes from poetry—like an increasing number of my conclusions: “World is crazier and more of it than we think, Incorrigibly plural.”

Richard Smith was the editor of the BMJ until 2004 and is director of the United Health Group’s chronic disease initiative.