Increasingly, research in academic and industry settings overlaps as for-profit companies fund institutional studies. Trust in scientific results depends on knowing the biases that go into it — including the motivations of a company that stands to benefit from some outcomes more than others — so researchers are usually required to disclose those conflicts of interest (COI) when they publish their papers in a journal.
Usually, COIs appear in a clearly-identified section toward the end of the paper, which is typically labeled something like "conflict of interest statement” or “competing interests.” But reporting standards are inconsistent across journals, and those disclosures don't offer a complete historical picture of conflicts. Researchers often stop reporting conflicts after a certain time period, don't always report conflicts that aren't directly relevant to a study (i.e. they could be funded by a company that competes with a company being studied), and don't always have to report funding given to their institution rather than them as individuals.
Science as a whole is good about transparently sharing conflicts of interest on a case-by-case basis, but it's impossible to use them to inform trust in a body of work when individual reports with overlapping relevance are locked up in millions of pdfs. This is a problem that's solvable, and that STAT wants to tackle in a subset of scientific literature focused on health and life sciences. We imagine a program that can identify conflict of interest sections from papers, extract the named entities and relationships between them, and ultimately deliver them into a database that links author names with their conflicts, including where and when they were disclosed.
As a whole, this information could be used to vet sources of scientific and journalistic expertise for potential bias that would otherwise go unnoticed. It would also be possible to see which institutions and groups of researchers are most commonly tied to specific companies and funders, pointing to broader trends relevant to the study of research funding. A simple searchable database would be one great output of this project, but a tool that allowed the easy viewing of such network effects could be a stretch goal.
