KOR gov’t has started to establish AI infrastructure…case in Denmark referenced

HIRA shared cases of Danish Medicines Agency (DKMA) at the international symposium

Director ​Claus Møldrup said, "AI can be adopted in related fields, but humans must make the final decision"

MOHW "Will adopt AI in basic healthcare…will provide GPU support to the public healthcare sector"

HIRA "Plans to use AI in transforming appropriate care beyond regulatory approval automation"

While South Korea takes initial steps to build artificial intelligence (AI) in public healthcare and debates data standardization and infrastructure development, Denmark already uses AI across diverse regulatory functions, from drug review and authorization to post-marketing pharmacovigilance.

Aligning with this global trend, the Ministry of Health and Welfare (MOHW) and the Health Insurance Review and Assessment Service (HIRA) have announced plans to drive a full-scale AI transformation across the public healthcare sector based on the government's recently announced "Basic Strategy for Medical AI."

​Claus Møldrup, Director of Data and Analytics at the Danish Medicines Agency (DKMA).

​Claus Møldrup, Director of Data and Analytics at the Danish Medicines Agency (DKMA), shared how the European Union is utilizing AI in regulatory affairs during an international symposium hosted by HIRA on the 28th, under the title "The Great AI Transformation and Healthcare Innovation".

​The Danish Medicines Agency established a principle against utilizing general commercial AI platforms, such as ChatGPT or Claude, for official regulatory tasks, including marketing authorization reviews.

​This is because such tools risk leaking data to external servers and operate as "black box" systems with untraceable training datasets.

​The agency reasoned that commercial models of ambiguous origin are unsuitable because regulatory bodies must clearly substantiate the rationale behind their decisions to pharmaceutical companies.

​Møldrup said, "Generative AI models such as ChatGPT and Claude function like black boxes where the underlying data and training sets remain unknown," adding, "If regulators submit draft reviews generated by such AI to pharmaceutical companies, the companies could legitimately question the evidentiary basis of those determinations."

Consequently, the Danish Medicines Agency developed and currently operates an AI model deployed within its private on-premises network using open-source large language models, engineered to generate responses strictly bounded within specific regulatory guidelines and standard operating procedures (SOPs) designated by review officers.

​In addition, the agency instituted a "human-in-the-loop (HITL)" control principle, ensuring that specialized reviewers in the respective domain directly review and finalize determinations.

Møldrup emphasized that staff are directed to use AI only within their respective areas of expertise, explaining that personnel cannot effectively control the system in unfamiliar domains and that only subject-matter experts can screen out AI-generated hallucinations.

However, AI's operational scope remains broad. Beyond marketing authorization reviews, it is also being utilized in ▲post-marketing pharmacovigilance ▲prescription optimization support.

​Møldrup explained, "While a patient taking a new epilepsy medication might report hormonal fluctuations to a physician who might not officially log it as an adverse event," adding, "The agency's system, capable of accessing comprehensive national healthcare data in Denmark, can rapidly uncover such unreported adverse reaction cases."

​The agency is also pursuing pilot demonstration projects that analyze real-world clinical data to guide patient-tailored prescribing.

​Møldrup added, "Approximately 10% of patients prescribed metformin discontinue their regimen within 90 days due to factors such as taking five to six concurrent medications," and further stated, "If AI can predict treatment failure patterns and identify high-risk populations in advance, clinicians could be guided from the outset to prescribe alternatives, such as once-weekly formulations, for elderly polypharmacy patients projected to discontinue therapy."

​MOHW "Outcomes-based reimbursement"...HIRA "Transforming appropriate care beyond review automation"

​Jung-Hwan Park, Director General of the Division of Medical AI and Data Policy at the MOHW.

​In line with the AI transformation trends across global regulatory authorities such as Denmark, the MOHW and HIRA are substantiating strategies to integrate AI across public healthcare and broader claims review and evaluation systems.

​Following the government's recent announcement of the "Basic Strategy for Medical AI," both entities maintain that they will fully fulfill their roles as the competent ministry and overseeing institution.

​However, practical hurdles such as data standardization, substantial infrastructure expenditure, and the establishment of reimbursement frameworks must also be resolved.

​Jung-Hwan Park, Director General of the Division of Medical AI and Data Policy at the MOHW, stated, "The government should act as a key player in the development of medical AI," adding, "Active investments will be made spanning from the initial development of AI models to their integration into the public healthcare system.

Park emphasized that the government will provide state-owned GPU resources to ensure that a lack of computing capacity does not impede AI utilization in public healthcare.

However, the ministry maintains that reimbursement frameworks for healthcare AI utilization in clinical settings must be established after rigorous post-implementation evaluations.

​Moo-Sung Kim, Director General of AX Innovation Department.

Park said, "Simply providing financial incentives proportional to utilization frequency could undermine the financial sustainability of the National Health Insurance (NHI) fund," pointing out that "High usage volume does not automatically translate into patient benefit."

Park added, "The MOHW is carefully deliberating how to structure appropriate compensation based on rigorous post-hoc assessments of whether meaningful clinical efficacy was achieved."

​HIRA announced its commitment to proactively adopting AI to advance public health and support the fiscal sustainability of National Health Insurance finances.

​Despite constraints related to private data security, data standardization, and substantial infrastructure costs, the agency plans to pursue AI integration that establishes a feedback loop to promote appropriate clinical care.

​Moo-Sung Kim, Director General of AX Innovation Department, said, "Ultimately, the answer to why adopting AI is necessary lies in enhancing public health and optimizing NHI expenditure," and Kim emphasized that "The need for a feedback system that enables healthcare institutions to deliver appropriate care in real time rather than merely deploying AI for routine claims review tasks."