Data Provision from Medical Institutions for SaMD/AI Development

Abstract

Various kinds of medical information are required in large quantities for AI development. In this manuscript, we presented data provision cases of Kagawa University Hospital (KUH) which is a typical regional national university hospital in Japan, and addressed the issues we recognize in the current situation. Most important issue is the conflict with personal information protection. Even if anonymization is applied, there are still constraints. We have introduced the KUH’s workflow for extracting medical information from electronic medical records (EMR). The purposes for extracting data from EMR are for clinical practice, research, and education. Extraction for research purposes requires approval from the ethical committee. When obtaining ethical committee approval, preparation such as informed consent forms related to opt-in/opt-out options is necessary, and specifics such as data storage and recipients need to be identified. The research plan specifies the extraction criteria based on the planned study. The medical information department follows the extraction criteria for extraction requests, extracts the data, and preserves the records of the extracted data. We have several cases of data provision for Software as a Medical Device (SaMD) and/or AI development. Data can be provided either from the hospital’s core EMR system or directly from department systems. In the latter case, there is a possibility that there won’t be a record of the data provision in the hospital’s management department. Data collection for registry studies is also useful for SaMD and AI development. We efficiently conduct registry studies using a regional medical network called Kagawa Medical Information eXchange R (K-MIX R). Such methods need to be considered for studies that require a large amount of data.