Usefulness and Challenges in Developing Disease Extraction Algorithms in Japanese Large-Scale Data-Driven Databases.

In the electronic medical record (EHR), few data items accurately represent the disease pathology itself. Therefore, we need algorithms to extract various diseases with the highest possible accuracy from the EHR. Therefore, we created and validated outcome definitions to identify six diseases from MID-NET, a Japanese large-scale infrastructure database of Data-driven medical studies established for drug safety by PMDA (Pharmaceuticals and Medical Devices Agency). Each definition was practical enough to use for a large-scale clinical database. In addition, PPVs and sensitivities differed between medical facilities, allowing us to identify issues in creating and operating definitions. Our research is expected to contribute to developing transnational outcome definitions.