Physician Industry Consultants Will Put Their Mouths Wherever Their Money Is
This theory is well-exemplified by a widely cited study by Stelfox et al. (1998). The authors used a survey to correlate the presence or absence of cardiologists’ connections with makers of cardiovascular drugs and their published opinions on the dangers of calcium-channel blockers (CCBs) whose safety was controversial at the time. The dangers associated with shorter-acting, immediate-release CCBs was well appreciated. What was controversial: dosing and whether the risks of short-acting CCB should be extrapolated to longer-acting CCBs?
Stelfox and colleagues found that cardiologists who consulted for industry were significantly more likely to discount the dangers allegedly associated with CCBs than those without connections to industry. The authors concluded that “the medical profession needs to develop a more effective policy on conflict of interest.” They proposed that journal editors erect a stronger bias filter for manuscript submissions by COI-burdened industry consultants. This suggestion would bear fruit. In 2005, JAMA instituted a discriminatory policy burdening industry-sponsored submissions with the cost of securing a sign-off from academic statisticians. Evidence for doubting the validity of this imposition will be discussed below.
Remarkably, the Stelfox study found that the cardiovascular drugs made by the companies for which some of the cardiologists consulted, bore no consistent relationship to the consultants’ opinions about the risks of CCBs. Consultants to companies manufacturing cardiovascular medications unrelated to calcium-channel blockers were as likely to discount the risks of CCBs as those consulting for companies that produced these drugs.
How does the theory of risk-emphasizing relationship bias (COI) adopted by Stelfox et al. explain this? Is it plausible that cardiologists who consult with makers of any cardiovascular drugs publish favorable comments about all cardiovascular drugs, including CCBs made by companies who do not pay them? Or, might it merit investigation to determine whether the risk-discounting, industry-consulting cardiologists were overall more knowledgeable than the non-consulting cardiologists and had pharmacological justification for claiming that CCBs (properly dosed, longer-acting agents) are not inherently more dangerous than other drugs used to treat hypertension?
Lumping all CCBs together and associating favorable commentary regarding them as a therapeutic class with industry consulting misses important pharmacological distinctions. Industry consultants may have had access to pertinent information from industry scientists that was not yet generally available. They may have already appreciated the relevant distinctions. That confounder was not controlled.
Since “CCB” is ambiguous, industry-consulting experts might have assumed that others would appreciate the ever-present puzzle about dosing and ambiguity between short-acting and longer-acting agents. Knowledgeable people make the mistake regularly. On the other hand, naive COI-suspicious auditors may blow off any medical subtleties and assume the consultants’ risk-discounting was driven by financial COI.
Stelfox and colleagues prejudiced their inquiry by biased risk framing (COI). A corruption narrative is powerful. It substitutes an easier inquiry (into conflicted relationships) for a harder one that would have plumbed pharmacological nuances. The substitution of an easier inquiry for a harder one colored the value of industryconsulting cardiologists’ consulting opinions. It may also have delayed use of safe and effective drugs with some, albeit unknown, patient detriment (e.g., forgone benefit).
Subsequent studies have corroborated the safety and efficacy of properly dosed, longer-acting CCBs (Opie and Schall 2002; Epstein et al. 2007). Extrapolating risks from shorter-acting, immediate-release CCBs to longer-acting compounds was inappropriate. Sometimes a sense of smell detects a skunk in the woodpile, but sometimes it detects a fox, which smells similarly. Overreliance on “the smell test” shows willingness to take to the bank the risks of “thinking fast.”
Mukherjee (2010: 276) succinctly describes the central weakness in using statistical methods to identify previously unappreciated risk factors. By their very nature, these methods, though powerful, are descriptive and associational, not mechanistic and causal. They rely on a degree of foreknowledge. To run a classic case-control trial to identify an unknown risk factor (e.g., whether one is more likely to get biased advice from paid industry consultants), paradoxically, the investigator must already know the question to ask. Biased framing (COI) does that job. Confirmation bias steps in to do the rest.