HONG KONG – South Korea is actively working to expand its artificial intelligence (AI) capabilities in health care, but outdated regulations and concerns over privacy and profit sharing are proving to be significant stumbling blocks.
In 2016, Incheon-based Gachon University Gil Hospital started using Watson for Oncology developed by IBM Corp. Since then, various AI-based medical devices have been developed and launched in Korea.
Also, to date, South Korea's Ministry of Food and Drug Safety (MFDS) has approved several AI-based medical products by domestic developers. One is an AI-based cerebral infarction magnetic resonance imaging diagnosis system by JLK Inspection Inc., headquartered in Cheongju. Another is a bone age measurement system for children based on X-rays that was developed by Vuno Inc. based in Seoul. Third is a deep learning-based chest X-ray interpretation and lung disease diagnosis system that was developed by Lunit Inc. based in Seoul. All three have earned MFDS approval over the past two years. Lunit's breast cancer diagnosis software and JLK Inspection's brain aging measurement solution received approval this year. The products are already being used in radiology departments of hospitals in Korea.
While South Korea is keen to more widely tapping into the power of AI, current regulations may be slowing down the introduction of AI-based devices.
A key concern is linked to the national health insurance system. Medical institutes get paid for diagnosis and treatment from patients and the government. The Health Insurance Review and Assessment Service (HIRA), a state-run organization, determines whether a medical procedure is necessary and deserves to be covered by the national health insurance. If insurance covers the procedure, the government pays part of the medical fees. It normally takes as many as 500 days for an application for marketing approval to determining insurance coverage.
For the time being, a procedure using an AI-based device may not be covered by the government. None of the three products approved by MFDS have been included in applications for public health insurance coverage because the devices "support" existing medical procedures.
"It is not simple to evaluate the value of AI-based machines as they assist doctors' medical practices. This is common worldwide," Joon Beom Seo, chairman of the Korea Society of Artificial Intelligence in Medicine and professor in the Thoracic Radiology Department of Asan Medical Center of Korea, told BioWorld MedTech. "The unique challenge in Korea is that the public health insurance has a limited budget to pay for certain medical practices. If a practice is newly applied for the insurance, some other practices get supported less than before. This may cause interest conflicts, thus it is not easy for innovative medical technologies to enter the market."
"Medical institutes use our AI-based devices in Korea, however, AI-based solutions would spread much faster if covered by the national insurance. If the national insurance covers medical practices using AI-based technology, it will spur the technology's quality enhancement," a spokeswoman at Vuno told BioWorld MedTech.
Draft guidelines for coverage emerge
To facilitate the use of AI-based medical solutions, HIRA has announced draft guidelines to provide insurance coverage for AI technologies. The guidelines, focused on AI-based radiology, classify the contribution of the technology to the diagnosis and treatment in four levels: improve efficiency of medical consultation, improve diagnosis that is clinically meaningful, improve treatment and verify cost-effectiveness.
According to HIRA, level three and four (improve treatment and verify cost-effectiveness) have a high possibility of public health insurance coverage. The government has not decided when it will start carrying out the guidelines yet.
Another block to the faster development of AI in health care is personal information regulations. Korea's Personal Information Protection Act Article 19 prohibits companies from sharing personal information or using it for any other purpose than the one for which the information was originally collected.
The Act makes it hard to build and use big medical datasets. Even though the hospitals have medical information from numerous patients, they cannot transfer the data without patient's agreements to AI companies that need big data. For example, to collect 100,000 patient records to use in the development of an AI, a company would need approval from each patient.
Another challenge is how to share profits from the use of the device that relies on personal information, Seo said.
"It would not be possible to share the profits with all the personal information providers. If a company commercializes an AI-based device using 1 million patients' X-rays, it will be almost impossible to distribute the revenue from the device with all the patients. The system for this issue has not been set up yet," Seo noted. "Korea needs a system that supports patients to actively agree with providing their information for commercial purpose and companies to use the information without much regulations. Once it is set up, sharing profits from the AI-based devices should be properly carried out."