Privacy concerns in preventing fraudulent publication

I understand the need for scientific journals to take every reasonable step to prevent the publication of flawed or fraudulent research,¹ but, as Ontario's privacy commissioner, I have serious concerns about the proposed solution of publishing all of the data on which research findings are based.

The CMAJ editorialists state that the ethical and legal obstacles in doing so, such as “anonymizing” the data, can be overcome. However, rendering a data set anonymous is not a trivial matter. My experience in dealing with privacy issues in the context of research indicates little consensus within the research community as to what variables need to be stripped from a data set to render it truly anonymous. If anonymization is not done in an extremely conservative manner, the data set could be used alone or linked with other data to re-identify individuals. There is also the risk of re-identification of individuals through the publication of small sets of data.

It is my understanding that most researchers treat all data sets — anonymized or not — as confidential personal information, a highly desirable practice. Further, to the extent that anonymous data sets may be used to re-identify individuals, their publication may be a violation of Ontario's Personal Health Information Protection Act (PHIPA) (and possibly other health privacy statutes in other provinces) and/or the requirements imposed by research ethics boards.

Even if data sets could be truly anonymized, it is not clear that their publication would be useful for validating research, as suggested in the editorial.¹ In many cases, there would be insufficient information in a stripped-down data set to replicate findings or conduct further analyses.

The publication of data sets would pose a serious threat to the privacy of individual research subjects. In the absence of clear evidence that the publication of anonymized data sets would deter the publication of flawed or fraudulent research, and in the absence of a clear standard for anonymizing data sets to ensure that individuals cannot be re-identified, alternative means of validating research findings should be considered.