The Smart Problem List: AI Agent enriches EHR Problem Lists with Clinical Precision

A complete and up-to-date problem list is essential for quality care. It serves as a central source for clinical decision-making, medication monitoring, and collaboration between healthcare providers. For nurses, a comprehensive problem list provides direct insight into the patient’s status and history, increasing the safety and efficiency of care.

Our latest solution, the Smart Problem List Agent, is an AI solution that automatically supplements problem lists in the electronic health record (EHR) based on medical notes and letters. The agent recognizes diagnoses, procedures, and clinical findings, and translates these into standardized SNOMED CT codes, supporting the reference sets of both the Netherlands and Belgium.

Today, we are proud to announce the results of the extensive clinical validation. This validation was carried out in collaboration with ZAS (Ziekenhuis aan de Stroom in Antwerp). ZAS is a large general hospital in Antwerp with 14 campuses and over 1,000 medical specialists, and it has a strong focus on patient safety, data quality, and digital innovation. The results and the agent were presented at the SNOMED CT Expo (23–25 October). The solution will soon be available to all hospitals in the Netherlands and Belgium.

Project phases

The development and validation program ran from January to August 2025 and followed a phased approach:

• Phase 1 – Preparation: defining scope, legal frameworks, and technical setup.
• Phase 2 – Training: collecting and annotating data (over 10,000 SNOMED CT concepts) and training the model.
• Phase 3 – Validation: assessment by physicians for seven specialties: nephrology, hematology, cardiology, pulmonology, intensive care, rheumatology, and geriatrics.
• Phase 4 – Generalizability: testing on data from other hospitals and expanding to new specialties.

Training & results

The project started with an analysis of clinical workflows and the desired clinical output. Together with ZAS, principles for relevance, correctness, and usability of suggestions were established. A dataset of approximately 1,000 medical documents from seven different specialties was compiled.

The AI agent was trained to recognize relevant information not yet on the problem list. During the proof-of-concept phase, the model was evaluated and achieved the following positive results:

• Completeness (recall): 90%
• Accuracy (precision): 93%
• Correct SNOMED CT coding: 95%

The Smart Problem List Agent analyzes notes and letters in real time and generates suggestions to add to the problem list within seconds. The responsible physician only needs to review the results before importing them into the EHR with a single click. This makes the patient record complete and more accurate, saves time for healthcare providers, and prepares hospitals for the transition to SNOMED CT and EHDS.

Validation and generalizability

The AI agent has now been validated for seven specialties at ZAS. The solution is currently being validated in other hospitals in the Netherlands and Belgium and for additional specialties. The first signals are very positive. Final results and a scientific publication will follow later.

Relevance for EHDS

The Smart Problem List Agent actively contributes to the European Health Data Space (EHDS) by enriching problem lists with standardized and structured data. This facilitates safe and interoperable data exchange between healthcare institutions and supports secondary use of data for research, policymaking, and population health.

About ZAS

ZAS is the largest healthcare organisation in Belgium. It has 10,000 employees, including 1,000 doctors, and provides 1,400,000 consultations, 140,000 day admissions, and 8,500 births annually, across 14 campuses spread throughout Antwerp.

About Coforix

At Coforix, we believe every patient deserves care based on the most up-to-date insights. Our goal is a connected healthcare system where best practices and the latest AI models are continuously shared. We leverage AI to make healthcare more efficient, innovative, and patient-centric.

With our Open Agentic AI platform, we provide a single vendor-neutral solution that brings together AI solutions from multiple (commercial) providers, integrates these seamlessly into existing workflows, and meets the strictest EU privacy and security standards.

Responsibility is at the heart of everything we do. We develop our solutions in close collaboration with CMIOs and clinicians and validate them extensively in clinical practice before making them commercially available. Transparency, reliability, and usability are guiding principles in our approach.