For shared decision making about personalized care, it is important to consider not only purely clinical outcomes, such as expected survival rate, but also quality-of-life (QoL) aspects, such as experienced pain, mobility, or nausea. But high-quality QoL data is not easy to obtain and often requires patients to fill in questionnaires. However, there is an abundance of information in free text in Electronic Health Records, for example, clinical documentation that – combined with information about medication use – can suggest a relationship between medication changes and side effects: “Yesterday the subcutaneous morfine administration was increased from 3dd 5mg to 3dd 10mg and patient feels worse today than yesterday”. This information is in natural language, uses synonyms, is often incomplete, and as such not easy to use; this is where large language models (LLMs) may be useful.

As health care is a high-risk AI application area, quality control over the procedure of constructing decision support tools and the underlying models, but also absolute guarantees about privacy and transparency about the source and development of the LLMs is crucial. In collaboration with IKNL and Lareb, two AI students will study both the technical aspects (at IKNL) and the organisational, medical-ethical, and legal aspects (at Lareb) of using the new Dutch LLM, GPT-NL, for side-effect data collection. IKNL or the Netherlands Comprehensive Cancer Organisation is the national organisation that provides open access to high-quality cancer data. Lareb or the Netherlands Pharmacovigilance Centre is the national organisation that collects data about side effects of medication, vaccination, and other health products and has an important public role in increasing knowledge and understanding about side-effects.

Jointly, both projects contribute to the GENAISIS-NL case study in the ELSA lab for Decision Support. In this case study, we explore how the new (and soon to be published) national guidelines for the use of generative AI in healthcare apply in a data acquisition context. The student projects will help provide information and experience that may further sharpen these guidelines in case they provide insufficient guide in this particular setting.

link to GENAISIS-NL: https://www.guideline-ai-healthcare.com/