Bringing Code to Data: Do Not Forget Governance.

Area of Science:

• Biomedical Informatics
• Data Science
• Computational Biology

Background:

• Accessing clinical data for algorithm development is challenging due to privacy, institutional, and regulatory concerns.
• Existing data sharing methods face limitations in protecting sensitive biomedical information.
• Cloud computing offers solutions for secure data access without direct data transfer.

Purpose of the Study:

• To introduce and evaluate the model-to-data approach as a secure method for algorithm development on sensitive clinical data.
• To address barriers in biomedical research data sharing.
• To explore the ethical and practical considerations of implementing model-to-data.

Main Methods:

• The study discusses the model-to-data paradigm where algorithms are submitted to run on sequestered datasets.
• Cloud-based secure computing environments are highlighted as a complementary approach.
• Pilot studies demonstrating the utility of model-to-data in constrained environments are referenced.

Main Results:

• Model-to-data enhances data security and local control, enabling knowledge generation from hidden datasets.
• Pilots show model-to-data is effective when other sharing methods are infeasible.
• The approach necessitates significant resource commitment and may impact scientific freedom and reproducibility.

Conclusions:

• Model-to-data is a valuable addition to data sharing, best used supplementally where necessary.
• Responsible adoption requires guidance on open science, reproducibility, security, and ethics.
• It reduces privacy risks but is not a complete ethical solution; careful re-evaluation of research ethics is needed.