Bringing Code to Data: Do Not Forget Governance

摘要: Developing or independently evaluating algorithms in biomedical research is difficult because of restrictions on access to clinical data. Access restricted privacy concerns, the proprietary treatment data by institutions (fueled part cost hosting, curation, and distribution), concerns over misuse, complexities applicable regulatory frameworks. The use cloud technology services can address many barriers sharing. For example, researchers high performance, secure, auditable computing environments without need for copying downloading. An alternative path accessing sets requiring additional protection model-to-data approach. In model-to-data, submit run secure that remain hidden. Model-to-data designed enhance security local control while enabling communities generate new knowledge from sequestered has not yet been widely implemented, but pilots have demonstrated its utility when technical legal constraints preclude other methods We argue make a valuable addition our sharing arsenal, with 2 caveats. First, should only be adopted where necessary supplement rather than replace existing data-sharing approaches given it requires significant resource commitments stewards limits scientific freedom, reproducibility, scalability. Second, although reduces loss data, an ethical panacea. Data will hesitant adopt guidance how do so responsibly. To this gap, we explored open science, security, respect subjects, ethics oversight must re-evaluated context.