South Korea Releases Guidance for Software using Big Data, AI, and Machine Learning
Published September 18, 2020
Providing Transparency to the Process
The South Korean regulator, the Ministry of Food and Drug Safety (MFDS), recently released multiple guidance documents related to software as medical devices (SaMD) using AI, Big Data and Machine Learning. As one of the newest and fastest growing sectors of the healthcare industry, there are still many questions as to how machine learning, big data, and AI supported products will be regulated. As these technologies become more common across a variety of industries, governments have yet to produce a cohesive strategy. As the US FDA works to produce official guidance, countries like Singapore and Korea are trying to lead by already releasing guidance to help provide clarity and transparency to the regulatory requirements.
South Korea’s MFDS has identified AI, big data, and machine learning software products as vital to the future of their healthcare industry. These technologies not only offer benefits to patients therapeutically and practitioners professionally but can also save a lot of money from misdiagnoses and time saved by clinicians. In the last year, the MFDS has issued multiple Guidance documents to help provide clarity into the registration process of Software as Medical Devices (SaMD) and Software using AI and Big Data, including Guidance for review/approval of a medical device employing Big data and AI (artificial intelligence) technology and Guidance for evaluation of the clinical efficacy of a medical device employing AI. (English translation available upon request.)
Types of Software Output
The growing trend for classifying and evaluating Software that uses AI and/or Big Data is determined by the type of decision-making power the product has. If a product automatically diagnoses, predicts, monitors, or treats something, it is considered a “Sole Determinant” and requires more clinical evidence than “Informing” or “Driving” products. Informing and Driving products are only designed to assist healthcare practitioners by analyzing clinical data and providing insight and recommendations, which are then confirmed through additional tests and/or professional evaluation
|Analyze vital signs measured and integrated in the emergency room to predict emergency situations such as breathing difficulties that gives warnings such as alarms.||Software that calculates the probability of a specific cancer based on medical information such as a biopsy and electronic medical record (EMR).
Screening software that detects and displays abnormal areas through upper CT image analysis.
Analyzes medical images to determine quantitative values for specific areas of blood vessels such as blood flow rate and vessel diameter.
|Software that predicts hypoglycemia in advance by analyzing information such as blood sugar data, food intake, and insulin injection.
Software that diagnoses or predicts arrhythmia using ECG measurement results.
Come grow with us in the US in Thailand in China in Korea in the Philippines in Taiwan in Hong Kong
US: +1 512 898-9222
SG: +65 3138-4148
Substantially Equivalence and Clinical Evidence Requirements
As part of the application process, manufacturers will need to either a) show that their product’s “indications for use” or “mode of action” or both are substantially equivalent to a predicate device registered with the MFDS, or b) provide a sufficient clinical trial report. Products with a predicate can register through the Substantial Equivalence (SE) route, saving a significant amount of time and money. A comparison of the “mode of actions” will need to include a comparison of the type of machine learning and of the data characteristics. Components of a clinical trial report include prospective and retrospective studies, reference standard information, parallel design(s), etc.
Related guidance documents:
– Guidance for review/approval of a medical device employing Big data and AI (artificial intelligence) technology (Link)
– Guidance for evaluation of the clinical efficacy of a medical device employing AI (Link)
Image Analysis Software
A crucial emerging category, image analysis software, is revolutionizing how clinicians and health practitioners read and diagnose a patient’s data. Not only do these products reduce the time spent assessing data, they also allow non-experts better understand the data without needing to rely on specialists, like radiologists.
The MFDS has initially identified the following 7 categories of software used for Image analysis that uses Big Data, AI and/or Machine Learning. On August 28th, the MFDS issued a draft amendment that intends on creating several new subcategories for software based on functions. Until final guidance is released, an MFDS consultation is recommended to ensure the product is correctly classified.
|Medical Image Analysis Software||Software that can be used for simulation treatment, simulation procedure, and diagnosis by acquiring and analyzing medical images.|
|Radiation Treatment Software||Used to determine radiation simulation treatment and simulation using acquired medical images.|
|Medical Image Detection Aid Software||Assist medical personnel in making better diagnoses by marking them with color or leader lines|
|Medical Image Analysis Device||Device with software that acquires and analyzes medical images and can be used for simulation treatment, simulation procedures, and diagnosis.|
|Medical Image Detection Aid Device||After detecting abnormal areas in a medical image. The device assists medical personnel in making diagnoses by marking them with color or leader lines|
|Medical Imaging Assistance Software||Software used to aid in diagnostic decisions. Uses medical images to determine the presence or absence of a
disease, the severity of the disease, or the degree of likelihood of the condition, etc.
|Medical Imaging Assistance Device||Medical imaging is used to determine the presence or absence of a disease, the severity of the disease, or the degree of likelihood of the condition, etc. is automatically displayed to a device used to aid in making diagnoses.|
Non-Image Analysis AI Software including Wearables
Arguably the fastest growing AI-using medical device segment is wearable devices. A wearable device, like a smart watch, will need to apply under multiple parameters depending on the input method, analysis method, and alert system. For example, a watch that can utilize big data and artificial intelligence to monitor a person’s blood pressure and provide crucial alerts would apply as a patient monitoring device (A26090.01) classified as an item within the examination device (A26000). However, this practice will be updated in the near future as MFDS is planning to release a list of software product types. The public feedback period was closed on September 16, 2020 and the draft list can be found here.
In this example, a wearable blood pressure monitor would be a Class 2 device, requiring 25 working days for a 3rd party to review the technical documentation and 5 working days for the National Institute of Medical Device Safety Information (NIDS) to grant a final approval if it has a predicate. If it doesn’t have a predicate, the application is reviewed by the MFDS for clinical evidence with an official timeline of 80 working days. Manufacturers will also need to submit a valid KGMP certificate as part of either application process.
Importing and Invoicing
Medical device logistics and customs clearance can be a complicated process in Asia. It is important to work with a trusted partner to not only ensure timely delivery of imports but to also maintain compliance and market access. SaMDs add an additional level of complexity as these products are typically sold online and don’t require formal customs clearance. Instead, the Korea Medical Devices Industry Association (KMDIA) requires all manufacturers who sell software downloaded from the cloud or an Appstore, submit reports of the number of downloads or activations. Furthermore, individuals can purchase and download a non-physical SaMD from the Appstore or a website without paying importation taxes.
Examples of AI Products NOT Considered Medical Devices
Not all products using AI sold to healthcare facilities fall under the definition of a Medical Device in Korea.
The following are examples of products not considered a medical device.
- Software that supports administrative affairs of medical institutions (hospital and inventory management, electronic procedures, insurance claim processing, etc.)
- Software for exercise, leisure and daily health management purposes (promote healthy lifestyle, nutritional intake monitoring, etc.)
- Software for educational and research purposes only
- Software for the purpose of managing medical records irrelevant to disease treatment and diagnosis (Electronic Medical Records, Prescription Delivery Systems, etc.)
- Provide medical personnel with tools to organize and track patient health information or treatment information. Software that provides or helps you easily access medical information.
- Software that makes it easy to access medical information related to the patient’s health status or treatment
- Software that searches for standard treatments and clinical literature and summarizes the contents
- Software that searches drug interactions and allergic reactions to prevent drug side effects
Asia Actual is experienced in registering innovative products, including those utilizing AI software, in South Korea and throughout the region.
Please contact Asia Actual if your product uses AI and would like to learn more about the regulatory requirements and market potential, in South Korea and throughout the region.
Even though a product is not classified as a medical device, the product may still be required to adhere to other guidance unrelated to medical devices.