Developing test data for quality assurance in anonymized emergency health data registry in germany

Abstract Issue Fragmented data systems across the emergency chain in Germany hinder comprehensive analysis of quality, resource allocation, and patient outcomes. To address this issue the Bavarian Emergency Health Data Registry was introduced by the Bavarian Ministry of the Interior, to enable the digital recording and consolidation of patient data and structural data across the entire rescue chain. Multi-layered plausibility checks were implemented to ensure input data quality. Description of the problem To validate the performance of the necessary plausibility checks of input data and testing the data stream, testing data was required for each of the data sources emergency dispatch centers, ambulance services and hospitals. Due to data protection regulations synthetic test data had to be generated, ideally containing realistic, naturally occurring errors. Development of the registry was planned from early 2022 to the end of 2024. Results For preclinical data, anonymized real-world data was permuted to ensure conclusions could not be drawn about any individuals while maintaining testing requirements and authenticity of the data. For the hospital input, synthetic data was generated based on statistical distributions and dependencies of attributes. Data production began in 06/2023 and was continuously refined through 04/2025 to meet evolving test scenarios. The data were generated based on specified XSD schemas for each data source. To this end, 126 Datasets with over 3 million individual emergency cases were created. Lessons The development of test data was based on an iterative process between further adaptation of the data and improvement of the plausibility checks. This approach makes it possible to develop plausibility checks without the need to access individual records of real input data. This is essential to ensure that erroneous or implausible data submissions are either rejected or appropriately flagged without compromising data privacy. Key messages • For the initial development and evaluation of plausibility checks in an anonymized registry, hybrid test data consisting of permuted real data and synthetic data were well suited. • Due to the lack of precedent projects, expertise in emergency health data quality and plausibility is essential for generating meaningful and reliable requirement-driven test data.