High-quality data is essential when medical information from multiple institutions is combined for analysis. In Japan, the Medical Information Database Network, known as MID-NET, gathers large-scale medical information including electronic medical records and administrative claims to support advanced safety measures for pharmaceuticals and other products. The network uses a common data model and standardised codes so data from different institutions can be handled consistently. Daily hospital operations, however, rely on local codes, and those codes change as new drugs, laboratory reagents, laboratory methods and system updates are introduced. Earlier work within MID-NET identified several sources of data quality problems, including data transfer errors, mapping errors, local electronic medical record rules, system customisation, non-numerical laboratory entries and date interpretation errors. A governance framework was introduced to strengthen code standardisation across cooperating institutions and support more reliable use of real-world data.
A Central Framework for Code Governance
The governance framework centred on a governance centre established at Kyushu University Hospital in March 2017. Its operational structure included laboratory technicians, pharmacists, health information managers and system engineers. The centre specified mapping tables, assigned standardised codes and developed a code difference output tool to detect daily changes in code registration at each medical institution. Full-scale governance for drug, laboratory test and disease codes ran from July 2020 to December 2021, with feedback to institutions beginning in August 2020.
The code difference output tool compared local codes and local names with the corresponding standardised codes and standardised names in mapping tables. It extracted daily change logs whenever codes or names were newly registered or modified. The governance centre defined unified specifications for mapping tables and output formats, and electronic medical record vendors implemented the required functionality in MID-NET institutions. Because the implementations followed shared specifications, institutions could generate consistent outputs regardless of vendor. The tool was introduced across 18 MID-NET cooperating medical institutions.
Each extracted difference moved through the secure MID-NET network to the governance centre. The centre then reviewed the outputs and manually assigned the most appropriate standardised codes using predefined mapping rules, master data and domain expertise. No AI-based or automated algorithms were used. Governance focused on HOT for drugs, JLAC-10 for laboratory tests and ICD-10 for diseases. By targeting differences rather than entire code sets, the process supported continuous validation with a lower operational burden.
Must Read: Platform Advances AI-Driven ECG Analysis
How Review and Classification Were Managed
The governance process followed a defined workflow: extracting chaAgency andollecting and transferring them to the governance centre, accumulating the logs, assigning standard codes, sending mapping tables to the Pharmaceuticals and Medical Devices Agency and returning mapping tables to each medical site. A management procedure manual supported this work, and governance activity was recorded through uniform procedures.
The framework used four mapping statuses. A code could be classified as correct initially when the local code already matched the appropriate standardised code. It could be classified as correct after revision when the governance centre proposed a revised code and the institution later implemented it. It could remain a proposed standard code when a mismatch was identified but the institution had not yet adopted the suggested code. It could also be classified as out of scope when the item did not fall within the MID-NET governance framework.
This structure enabled systematic review of change logs. When a standardised code was already present, the process checked whether that code existed in the relevant standard master and whether the local name matched the standardised name. When no standardised code was present, the item was assessed to determine whether it fell within scope. Domain experts then reviewed in-scope items and proposed the appropriate code. In laboratory testing, this required attention to specimen type, measurement method and result identification, because similar local test names could correspond to different standardised codes.
Results Show Progress and Persistent Barriers
Between July 2020 and December 2021, the governance centre collected change logs weekly and tracked results over time. For drugs, 43,387 change log codes were collected, the largest number among the three categories. HOT codes were ultimately assigned to 15,463 cases. Of these, 14,699 had been assigned initially by institutions and 764 were registered after governance centre proposals. Another 25,727 required standardisation and 2,197 were out of scope. Out-of-scope items included in-hospital preparations, Kampo medicines and medical devices.
For laboratory tests, 4,090 codes were collected, but most were outside the MID-NET scope. The focus narrowed to 1,091 codes within the clinical laboratory test scope. JLAC-10 codes were ultimately assigned to 311 cases, including 218 initial assignments and 93 added after governance centre proposals. Another 687 required standardisation and 93 were out of scope. For diseases, 16,694 codes were collected and ICD-10 codes were ultimately assigned to 11,252 cases. Disease codes were more often mapped correctly from the outset than drug and laboratory test codes. Another 833 required standardisation and 1,608 were out of scope.
Registration rates during the study period remained 36% for drugs, 29% for laboratory tests and 67% for diseases. Monthly proposals from the governance centre did not lead to large increases in registration. Contributing factors included limited specialised personnel, electronic medical record specifications, update cycles, local operational policies, competing hospital priorities and low immediate operational benefit. Laboratory code registration was particularly difficult because JLAC-10 uses a 17-digit structure built from five elements.
The governance framework verified whether local codes were converted into standard codes at each clinical site and enabled continuous monitoring through accumulated differential data. It created a practical structure for reviewing mapping tables, proposing revisions and maintaining oversight across multiple institutions. Standardised code use could be tracked in detail, while persistent operational and structural barriers remained, particularly for laboratory tests. Continuous governance of mapping tables, supported by expert review and sustained operation, offers a workable model for centralised data repositories linked to local electronic medical records.
Source: BMC Medical Informatics and Decision Making
Image Credit: iStock
