A web-based federal registry of information about clinical trials first became publicly available in 2000. That registry, ClinicalTrials.gov, first provided descriptive information about interventional trials performed under US regulatory oversight and then expanded to include information about study findings. In 2024, such registration is one step needed to assure transparent access to information about clinical trials, and several tools now facilitate analysis of registry data.
Currently, standards, regulations, and oversight for many such technologies lag the rapid advance of these technologies. Therefore, this creates a 2-fold danger: without meaningful oversight and evaluation, patients may be exposed to ineffective, biassed, or even harmful technologies.
Although some reporting guidelines for health AI-such as the recent SPIRIT-AI and CONSORT-AI guideline extensions-outline elements and provide checklists for reporting on protocols and results from interventional trials that incorporate AI,1 this by itself will not ensure that every adopter of the said technologies has all the information needed for the responsible management of these solutions in clinical care.
The Case for Local Registries of Health AI Technologies:
Transparency in health AI technologies is a fundamental social imperative and forms the thrust of recent policy and regulatory initiatives arising from government. A necessary precondition for such transparency has been a registry of the various health AI technologies that are in development, have been deployed, or have been retired from use within an institution or health system at any time.
While the need for trial registration improved transparency in the conduct of clinical research, AI registries would serve to do the same: clinicians, administrators, technology developers, patients, and the public all have reasons for wanting to know where and how these technologies are being used.
The technology developers could follow the diffusion of their work and appreciate how they were being deployed and adapted. Thirdly, an outward-facing part of the registry will help patients understand the use of health AI technologies in their care-this is especially important for compliance with federal transparency requirements.
Finally, increasing use of health AI applications will most probably affect medical liability. The national federated registration system would be helpful in highlighting safety and quality of the product.
Categories for entry to the registry would be through nationally agreed standards through partnerships between public-private, for example, Coalition for Health AI5 and local registry information would source from several sources, most importantly, key characteristics of a given solution, results of independent testing.
In order to get the most and best out of this developer-provided information, algorithm transparency requirements for vendors of electronic health records should extend to all developers of health AI technologies.
Reference:
Pencina MJ, McCall J, Economou-Zavlanos NJ. A Federated Registration System for Artificial Intelligence in Health. JAMA
