JMIR Publications

ABSTRACT

Developing or independently evaluating algorithms in biomedical research is difficult due to restrictions on access to clinical data. Access is restricted because of the cost of data curation and distribution, privacy concerns, the proprietary treatment of data by institutions, concerns over misuse, and the complexities of applicable regulatory frameworks. The use of cloud technology and services can address many of these barriers to data sharing. For example, researchers can access data in high-performance, secure and auditable cloud computing environments without the need for copying or download. An alternative path to accessing datasets requiring additional protections is the “Model to Data” (M2D) approach. In M2D researchers submit algorithms to run on secure datasets that remain hidden. M2D is designed to enhance security and local control while enabling communities of researchers to generate new knowledge from sequestered data. This approach has been successfully used when technical or legal constraints have precluded other methods of sharing. Successful implementation of M2D, however, requires significant resources and commitment from data stewards, and places non-trivial limitations on scientific freedom, reproducibility, and scalability. M2D should therefore be adopted with care and should supplement rather than replace existing data sharing approaches. While M2D does reduce concerns over data privacy and loss of local control when sharing clinical data, it is not an ethical panacea. The values and processes that guide open science, including reproducibility, respect for data subjects, systems of ethics oversight, and security, all need to be reevaluated in light of M2D’s promise and limitations.