Necessity, accessibility and acquisition cost of unstructured medical data for South Korean medical device companies

Characteristics of South Korean medical device companies

Regarding the type of development, 45% of companies developed only medical software, 22% of companies developed only medical hardware, 26% of companies developed both medical software and hardware and 7% of companies developed other products (figure 1a). Of the managers, 29% belonged to a research and development team, 46% belonged to a licensing team, 13% belonged to a quality control team, 5% belonged to a business planning team and 7% belonged to another team. Of the managers, 37% were staff, assistant managers or deputy section chiefs, 55% were section chiefs, deputy department managers or department managers, and 8% were directors or more senior. Of the managers, 64% had less than 5 years of service, 18% had 5–10 years of service and 18% had more than 10 years of service (figure 1).

Necessity of structured medical data

Among structured medical data, managers of medical device companies felt that history of past diseases was most necessary, followed by information on hospital use and then demographic information (figure 2). These three types of structured data are closely related to each other. It is easier to develop personalised precision medical services for a patient if their past disease information, hospital usage history and sociodemographic characteristics are obtained than if information about their blood tests, urine tests, health behaviour and eating habits is obtained. South Korea’s Personal Information Protection Act is conservative and has many restrictions on the use of personal information; therefore, it is difficult to secure these three types of data. This is because legal problems such as the dualisation of online and offline laws, sanctions and punishment-oriented laws, and imbalanced regulations leaning towards online private operators have not been resolved.6

Accessibility of structured medical data

Managers of medical device companies felt that urine test data were most difficult to obtain, followed by blood test data and then medication data. One reason why these three types of structured medical data are difficult to obtain is that medical device companies have no way to obtain them unless they enter a consortium for research and development projects with hospitals. This is because these data are generated only by hospitals (figure 2).

Acquisition cost of structured medical data

Managers of medical device companies felt that the acquisition cost of blood test data was highest, followed by medication data and then urine test data. These types of data, which cannot be obtained unless a consortium for research and development projects is established with hospitals, require a large amount of money and qualification according to the Bioethics Act7 and can be used to solve problems in very complex processes and directly recruit subjects. The development of new medical devices is hampered by problems associated with both time and cost due to very inefficient processes (figure 2).

Necessity of unstructured medical data

Among unstructured medical data, managers of medical device companies felt that CT and MRI were most necessary, followed by X-rays and then patient progress records (figure 2d). CT/MRI and X-ray data are widely used in the digital healthcare market. Advances in the use of AI computing techniques have enabled computers to learn from these data, replacing direct determination of diseases by doctors.8 The utilisation of these types of data is only at the beginning stage and these data have great potential because they can be used to develop more diverse and precise diagnosis and prediction services. Patient progress records are useful for diagnosis and predicting the future via continuous analysis of text written about the disease course of each patient (figure 2).

Accessibility of unstructured medical data

Among unstructured medical data, managers of medical device companies felt that microbiome data were most difficult to obtain, followed by cancer cell images and then human genetic data. These three types of data can only be generated from human samples and therefore can only be obtained after recruiting donors to provide samples and evaluating these samples via biological processes.9 Although the cost of sequencing has decreased,10 many medical device companies still find the process and cost burdensome. Therefore, few medical device companies in Korea develop digital healthcare products using genetic data. In addition, even if data are secured, it is difficult to acquire high-quality human resources to analyse them and create value (figure 2).

Acquisition cost of unstructured medical data

Among unstructured medical data, managers of medical device companies felt that the acquisition cost of cancer cell images was highest, followed by CT and MRI and then microbiome data (figure 2f). Cancer cell images can only be obtained from hospitals where cancer surgery is performed; therefore, it is difficult to obtain these images and particularly to obtain a large number of them. Therefore, the acquisition cost may increase. CT and MRI are slightly easier to acquire, but high-resolution image data processing and advanced storage methods to prevent digital decay are required.11 12 In the case of microbiome data, biological processes and sequencing of human samples explain the high costs.

Differences in need, accessibility and acquisition cost between structured and unstructured medical data

Managers of medical device companies felt that the need for unstructured medical data (3.78 points) was higher than that for structured medical data (3.27 points). There was also a statistically significant difference (p<0.001). The accessibility of structured (3.71 points) and unstructured (3.69 points) medical data did not significantly differ (p=0.747). Managers of medical device companies felt that the acquisition cost of unstructured medical data (3.75 points) was higher than that of structured medical data (3.62 points). There was also a statistically significant difference (p=0.024) (figure 3).

Mutual necessity, accessibility and acquisition cost of structured and unstructured medical data

The mutual necessity for structured and unstructured data according to managers of medical device companies was assessed and data types with high mutual necessity (r=0.7 or higher) were identified. Video and sound data had high mutual necessity for image data, and sound data had high mutual necessity for video and ECG data.

The difficulties associated with mutual acquisition of structured and unstructured data by medical device companies were investigated. This revealed that it was difficult to mutually acquire unstructured data overall (r=0.7 or higher). Image, video, natural language and sound data were most difficult to mutually acquire (r=0.9 or higher).

The mutual acquisition costs of structured and unstructured data for medical device companies were assessed, and the data types with high mutual acquisition costs (r=0.7 or higher) were identified. The mutual acquisition costs of structured data and natural language data as well as image data and video and genetic data were high. Natural language recording, sound and ECG data had high mutual acquisition costs. Video data had high mutual acquisition costs with genetic, natural language recording, sound and ECG data. Genetic data had high mutual acquisition costs with image, video and natural language transcription data. Sound data had high mutual acquisition costs with image, video and ECG data. ECG data had high mutual acquisition costs with image, video, natural language recording and sound data. The data types with the highest mutual acquisition costs (r=0.9 or higher) were image and video data as well as natural language recording, sound and ECG data (figure 4).