Notre approche de la cartographie des systèmes pilotée par l'IA pour les données HL7 vers FHIR

In the medical field, diverse data formats often present a significant challenge for healthcare organisations. Data from one device may not correspond seamlessly with target instances, leading to a need for data mapping. Usually, this is a manual, time-consuming process. We decided to take on an initiative that automates this job, focusing on mapping data between the HL7 standard and FHIR. This article explores the hypothesis that artificial intelligence (AI) can effectively support the process.

À propos des normes de données HL7 et FHIR

HL7, ou Health Level Seven, is a set of international standards for exchanging, integrating, sharing, and retrieving electronic health data. It primarily operates at the seventh level of the OSI (Open Systems Interconnection) model, focusing on application-level communication in healthcare systems. HL7 is widely used for clinical and administrative data exchange in information systems such as Electronic Health Records (EHR), hospital management systems, and other tools used by doctors, nurses, and hospital administration to exchange patient information efficiently and securely, ensuring proper patient care.

FHIR (Fast Healthcare Interoperability Resources) is a standard developed by HL7 International. It leverages web technologies and data formats like JSON and XML to facilitate efficient data interchange in the healthcare industry. FHIR is part of the HL7 standards family but differs significantly from its predecessors, such as HL7 versions 2 and 3, in its approach and ease of implementation. FHIR facilitates the integration of different healthcare systems, offering a more accessible and flexible way to exchange patient information between healthcare providers.

The key differences between HL7 and FHIR lie in their data formats, structure, and interoperability challenges due to these differences. Thus, to use data that is in different standards in one system, it is necessary to unify it, and this process is called data mapping.

Qu'est-ce que la cartographie des données dans le secteur de la santé ?

Supposons que l'on dispositif médical collects and stores a patient’s data in a database. The patient goes to a hospital, and the staff needs their previous health results. A doctor or a nurse sends a query via an application, but the hospital system operates in the FHIR standard and the medical device from before stores data in the HL7 format. To make use of the results of the examination taken earlier, the data has to be “translated” from one format to another so the hospital system can understand it.

La cartographie des données conduit àinteropérabilité des données médicales; it’s a process of matching fields from one database or dataset to another, creating a link between the elements. It’s a crucial step in data integration and migration projects, ensuring that data from one format is accurately translated and transferred to the target system.

Usually, data mapping in the healthcare sector is conducted manually, which limits real-time data access possibilities. This method involves domain experts who analyse and align data elements from one standard to the other. This approach, while thorough, significantly slows down the data exchange process and burdens healthcare IT professionals.

Une étude de cas d'un système de cartographie des données piloté par l'IA

The healthcare data mapping system we developed responds to the need for automation in the patient data exchange process. By automating data mapping from one system to the other, as an example HL7 to FHIR, the system aims to shorten the time needed to transform the data, thus relieving people involved in the process.

Le projet a été entrepris dans le cadre du programme de R&D de Spyrosoft, Innovation Lab. Répondant au besoin du marché, nous avons décidé de mettre en œuvre intelligence artificielle en tant que composant essentiel d'un système de cartographie des données de santé.

Spécialistes impliqués dans l'initiative

Ce projet a réuni une équipe d'experts issus de trois domaines :

• Expert métier – une personne possédant une connaissance approfondie des normes médicales. Il a été impliqué pour fournir des conseils aux développeurs et pour effectuer des contrôles qualité sur les résultats ;
• Développeurs logiciels qui ont construit l'application et assuré son efficacité. Ils étaient également responsables des tests et de l'amélioration de la solution pour répondre aux exigences du projet ;
• Des spécialistes de l'IA qui ont accompagné les développeurs dans des domaines critiques tels que l'apprentissage et le fine-tuning des modèles. Ils ont veillé à ce que les développeurs disposent de l'assistance nécessaire pour obtenir le meilleur résultat possible.

Technologie utilisée pour l'échange de données de santé

The system’s backend is developed using C# in the .NET environment and features an integrated ChatGPT model. We chose to use a commercial AI model rather than build one from scratch because it had already implemented both HL7 and FHIR standards. Using a commercial model complies with medical regulations since our system only processes data structures and schemes, not patient data. The ChatGPT feature is hosted in the cloud, and training was conducted there, too.

Entraînement de modèles d'intelligence artificielle

We chose to work on observational data such as blood test results, blood pressure, or weight. There is also coded information on the type of examination, when the test was taken, and if the data is still valid. Many of those labels differ between the HL7 and FHIR standards.

We generated data that complies with both formats to train the model. While training, we observed that when asked the same question many times, the model generated different answers. At first, programmers were feeding the model all the information in one query, and that was the moment AI specialists came with help, introducing the development team to fine-tuning. Fine-tuning is about building the context and working on the results for longer, pointing out good answers and wrong answers and providing examples of correct responses rather than sending single, complex queries.

Validation des données

For now, the system generates several data mapping proposals, and the user chooses the one that best meets their expectations. The user checks if the structures seem valid and manually completes the mapping in areas that were too complex for the model to handle.

Plans de développement futurs

The next idea is to use AI to validate data that the previous model had come up with. Now, the model gives three or four different results, and a person has to choose one. This model would pass all mapping proposals through a validator, refer them to standards, find errors, correct them, and pass again.

Résultats de mise en œuvre à jour

Currently, at about 70% operational level, the system has demonstrated that AI can significantly speed up the medical data mapping process. So far, the model has only been trained on artificial data; feeding it the target information (e.g., client’s medical data) will enable it to learn even better and increase its efficiency level.

Pour nous, le résultat le plus significatif de cette initiative est la connaissance et l'expérience que nous avons acquises. Cela ne se limite pas à la conversion des données de HL7 vers FHIR, car il est possible de construire un système propulsé par l'IA qui fonctionne avec d'autres données d'entrée et de sortie.

Exploitez la puissance de l'IA dans le secteur de la santé

Nos experts vous conseilleront sur la manière d'appliquer l'IA à votre cas d'usage spécifique. En savoir plus sur notre services dans le domaine de la santé et des sciences de la vie ou contactez-nous via le formulaire ci-dessous.