The scientific journal The Lancet has announced plans to change its article-review process in order to prevent the publication of papers based on interpretations of large data sets that cannot be audited. The new rules are designed to correct flaws highlighted by the publication in May of an important paper that concluded that hydroxychloroquine was ineffective against the novel coronavirus and could cause heart problems. In June, the article had to be retracted after it emerged that the origin of its primary data was so dubious that nobody could be sure if they really even existed.
The study was based on medical records attributed to 96,000 patients who tested positive for the SARS-CoV-2 virus at 671 hospitals in several countries between December 2019 and April 2020. The data was collected and analyzed by Surgisphere Corp, a small healthcare analytics company based in Chicago. But inconsistencies were quickly detected, such as a high number of patients from regions not greatly affected by COVID-19, as well as doubts about the company’s ability to obtain information from hospitals all over the world. When asked to provide access to the data, the company refused, claiming it was confidential. It was therefore impossible to confirm the results, leading to a retraction of the article (see Pesquisa FAPESP issue no. 293).
The Lancet will now require that more than one author of a scientific article has accessed and verified the data on which the study is based, and the names of these authors must be provided to the editors. In the Surgisphere scandal, only the company’s owner, Sapan Desai, who was one of the article’s authors, had access to the primary data. The other coauthors, including cardiac surgeon Mandeep Mehra, a researcher at Harvard University, declared that they never saw the raw data and simply trusted Surgisphere’s analysis.
From now on, the authors of all articles published in the journal must state which data will be shared and how to access them. This was previously required for the results of clinical trials, but not for studies that analyze multiple data sources, as was the case in the retracted articles. Finally, extra attention will be given to this type of paper. Articles based on large patient data sets will only be accepted for publication after at least one of the reviewers has evaluated and commented on the strengths and vulnerabilities of the data. Experts in statistics and data science will also be asked to give their opinion, especially in relation to data integrity and ethics.
Republish