The BMJ is now requiring authors to share the computer code behind their published medical research, but almost nobody is doing it. This matters because medical research relies on computer code to clean data, run statistical tests, and produce results. Without that code, other scientists cannot check whether the findings are correct, spot errors, or build on the work. Current estimates put code sharing in biomedical science as nearly non-existent. Even when code is shared, it often cannot be run independently. The UK House of Commons Select Committee on Science, Innovations and Technology has recognised this as a problem. The researchers at Oxford, working with the University of Rennes and *The BMJ*’s editorial staff, are developing a checklist called DECODER that specifies exactly what authors should share: data extraction code, codelists, operational code, documentation, and the computing environment. They will pilot it at *The BMJ* and interview authors and editors about what works and what does not. If successful, the DECODER checklist could be adopted by other journals and institutions. That would make published medical research verifiable and reusable, strengthening the evidence base that underpins clinical guidelines and public health decisions.
View original technical description
The computer code used to collect, prepare, and analyse data is fundamental to modern clinical and epidemiological research. These scripts and programs contain key information on data management, study methods, analytic choices, and allow independent parties to reproduce, validate, error-check, and build on existing findings. Even when data cannot easily be shared, due to regulatory, privacy, or confidentiality concerns, as is common in health research, code and related documentation can almost always be shared. However, this rarely occurs with most current estimates putting the prevalence of code sharing in the biomedical sciences as nearly non-existant. Even when code is shared, computational reproducibility may be lacking if shared code cannot be independently run or consistently reproduce original results. It is of paramount importance to understand how code sharing can be improved and provide value across the biomedical sciences. However, stakeholders throughout science, including the UK House of Commons Select Committee on Science, Innovations and Technology, are increasingly recognising the importance of code sharing. Many journals, including The BMJ, are either encouraging or requiring the code underlying published analyses to be shared alongside original research articles. As we begin to move to a more open and transparent scientific endeavour, it is crucial that we understand how and what is needed to facilitate efficient and effective code sharing. The recent launch of this comprehensive code sharing policy at The BMJ provides an opportunity to explore code sharing behaviour in the biomedical sciences. We propose the development and pilot of the DECODER (Data, Extraction Code, Codelists, Operational Code, Documentation, Environment, Repository) checklist to support and audit code sharing, a descriptive and comparative analysis of code sharing practices among authors at The BMJ, and a formal evaluation of The BMJ’s code sharing policy. We propose four work packages as part of our project: A structured development process involving a scoping exercise followed by expert consensus meetings to refine and build consensus around the DECODER elements prior to its pilot at The BMJ. A formal pilot of DECODER at The BMJ followed by an analysis of code sharing behaviour and an assessment of the impact of the DECODER checklist. A qualitative assessment of both DECODER and The BMJ code sharing policy among authors and editorial staff at The BMJ. Dissemination of the project outputs and support for broader adoption of the DECODER checklist. This project will be led by researchers at The Bennett Institute for Applied Data Science at the University of Oxford in collaboration with our international partners at the University of Rennes I and the professional editorial staff at The BMJ. The outputs of this research will be used to promote the adoption of code sharing requirements, and the DECODER checklist, by other publishers and interested institutions.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know