News Jobs Contact
Login
Deutsch

AI search strategies for paediatric cancer research

24.09.2026 – reference case University Children's Hospital Zurich

insightsPageview({ aktuelles_topic: 'AI search strategies for paediatric cancer research', aktuelles_category: '', aktuelles_date: '24.09.2026' }) { "@context": "https://schema.org", "@type": "NewsArticle", "headline": "AI search strategies for paediatric cancer research", "image": "https://www.ergon.ch/dam/jcr:3f86076d-0038-4df9-8a1e-8bad59a6d315/Referenz_Kinderspital_Zuerich_1880x1300.899bb7c67751185f2f7c5fdb9abbf9d2.jpg", "datePublished": "2026-09-24", "dateModified": "2026-10-08", "author": { "@type": "Person", "name": "reference case University Children's Hospital Zurich" } }
How can artificial intelligence help find relevant literature on potentially effective drug combinations for children and young people with cancer more efficiently? Today, doctors and researchers search the scientific literature by hand for findings on drug combinations that might be relevant to a specific patient case. The University Children's Hospital Zurich (Kispi) set out to explore how this kind of literature search could be made more efficient and effective in future. Together with Ergon, and with support from Google Cloud, Kispi systematically investigated the potential of AI-based research.

The University Children's Hospital Zurich is one of Europe's leading centres for paediatric medicine. In paediatric oncology, a specialist team of biomedical scientists, bioinformaticians and physicians develops methods for functionally testing cancer cells against different drugs – an approach known as "drug response profiling". The aim is to develop and select individually tailored therapies for children and young people with cancer.

Manual research as a bottleneck

Until now, doctors and researchers have had to search manually for the drug combinations that are effective for a specific patient case. Given the sheer volume of medical publications, this approach is laborious and time-consuming.

Comparing search strategies

Together with Ergon, the Kispi team developed a research prototype: a cloud-based web application that automatically applies and compares different AI-based search strategies. The technical foundation is built on Google Cloud Run, Vertex AI (Gemini) and Cloud SQL. An agentic AI refines search queries iteratively, extracting relevant results from PubMed, a search database for medical and life-science literature. The researchers optimised and tested this approach on synthetic datasets drawn from published case studies. To evaluate the results, the application presents the identified articles to researchers in pairs for assessment. An Elo scoring system then shows which search strategies are the most effective.

"Working with Ergon allows us, for the first time, to reliably evaluate which AI methods genuinely support our research and literature searches."

Fabio Steffen Head of Functional Precision Oncology, University Children's Hospital Zurich

Sources at a glance

The infrastructure and software were built using modern software development methods. Ergon supported the Kispi team with workshops and detailed documentation on topics including infrastructure as code, identity federation, secrets management, container-based applications, integration testing and deployment pipelines. The application presents its sources transparently, automatically highlighting the relevant passages of text. This allows doctors and researchers to quickly see which passages an answer is based on.

A prototype that makes AI's potential measurable

The project combines AI technologies, objective evaluation and professional software development. This gave rise to an initial, practice-ready prototype that helps researchers find relevant literature. The project demonstrates the potential of AI-based approaches for biomedical research, while also providing valuable insights for the further exploration of AI-supported approaches in paediatric oncology.