By Christopher Robertson
Scholars and policymakers have long been concerned that the biomedical science literature — and thus the practice of medicine — is biased by the companies who fund research on their own products. Prior research has shown that industry-funded studies tend to produce results favorable to their company sponsors. One solution is disclosure of industry funding, so that physicians and other consumers of the biomedical literature can weigh scientific findings accordingly.
My prior work with Aaron Kesselheim, Susannah Rose, and others has found that adding such disclosures to biomedical abstracts could make a big difference — physicians understand them and will rely upon them. Nonetheless, most journals bury the disclosures at the end of articles, which are often hidden behind paywalls and not nearly as salient as the methods and findings displayed in the abstract. For the Institutional Corruption Lab of the Edmond J. Safra Center, I worked with a team of hackers to create a browser extension that proves the feasibility of adding those disclosures into PubMed, a Federal government database of the scientific literature.
Thankfully, that browser extension is becoming obsolete, as the National Library of Medicine (part of the NIH) has begun implementing such disclosures themselves, right in PubMed. A search reveals that nearly 80,000 abstracts now have such tags. While a lot in absolute terms, it is a small minority of the 17 million abstracts covered by PubMed. Commentators have suggested that as much as 70% of the funding for clinical trials comes from industry, so we should expect millions of abstracts to have such disclosures.
Thus we are still a long way from comprehensive and effective disclosure. There are two problems.
One limitation is that PubMed relies on journal publishers to provide the Conflict of Interest (COI) disclosures tagged as such. In contrast, our browser extension (appropriately called, “Unearth”) actually mined the full text of articles to grab whatever funding or financial relationships might plausibly be relevant. (Alas due to the complexity, the browser extension is not currently functional.) Going forward, we need policies (perhaps from ICMJ or Congress) that require journals to collect and provide such disclosure information, and a system for retroactively compiling that information from full text articles already published.
A second problem is the design decision to put the disclosure outside the abstract itself, below the keywords, PMID identifiers, and even social media links, along with some white space, and then put it behind a [+] toggle that defaults to hiding the information. Thus, a PubMed user has to scroll down, find the toggle, and click to reveal whether there is any COI information hidden therein. These design choices are not consistent with our knowledge about the biasing effects of the financial conflicts and the proven importance of disclosures to readers. At the very least, the disclosure should be moved up above the PMID identifier and the default should be to display the information.
With PubMed’s embrace of disclosures, we have made great progress. Yet, until we fix these two problems — coverage and salience — the biomedical literature and PubMed may continue to mislead physicians and scientists.