Quantifying structural vulnerabilities and resilience to research integrity risks in biomedical research networks

Retractions of scientific articles reflect a rising research integrity risk in biomedical research, and continued citation of retracted articles poses a downstream risk to research that built upon unreliable studies. Few studies have examined the structural risk factors embedded in how the research enterprise is organized. Here, we apply an analytical risk assessment framework to structural risk factors associated with retracted articles and the propagation of their citations to ten broad fields of biomedical research, collectively accounting for 14% of biomedicine. We pronounced field-specific differences in retraction rates and citations to retracted articles, and show that multiple retractions from single authors occur far more often than chance would predict. By subdividing fields into finer-grained topics with machine learning network clustering, we find that individual authors can reach large proportions of the literature within their topics through citations, and that retraction risk is positively associated with author productivity. Across these fields, 22% of the literature cites authors with at least one retracted article, and 6% cite work from authors with multiple prior retractions. Together, these factors lead to a concentration of risk within topics, which is only partially explained by the uneven distribution of authors with multiple retractions. Despite these quantifiable vulnerabilities, the risk has not yet been fully realized: most topics cite retracted work no more than baseline. Our findings thus reveal both a latent network vulnerability to the rapid dissemination of questionable results and a measurable resilience that has so far kept this possibility in check. The scientific community would particularly benefit from targeted efforts to test and strengthen reproducibility in high-risk scientific topics.