It identifies clinical trials suitable for the patient based on the medical report or a short questionnaire, and prepares a recommendation for the coordinator with a single click.
The problem. A doctor wishing to offer a patient a place on a clinical trial currently has to manually go through lists of ongoing trials. For each one, they then compare the patient’s diagnosis, disease stage and biomarkers with the inclusion and exclusion criteria. This takes a great deal of time, which is why a suitable trial is often not suggested to the patient at all.
Solution. TriaLink is a web-based application that identifies all open trials for which a patient might be eligible.
Study database. The application retrieves an export of MOÚ studies from the SONIS system. Furthermore, it synchronises weekly with the public registries ClinicalTrials.gov and EU CTIS, thus covering ongoing oncology studies throughout the Czech Republic. Input from the doctor. The doctor enters an anonymised medical report. A local language model (Qwen) extracts key parameters from it: diagnosis, stage, histology, biomarkers and previous treatment. The second option is to complete a short decision-making questionnaire. This allows the doctor to access the results within 30 seconds. Evaluation. The app compares the patient’s parameters with the inclusion and exclusion criteria for each trial. The result is a list of trials showing which criteria the patient meets and which still need to be verified. Recommendation. With a single click, the app generates a recommendation email for the principal investigator and the trial coordinator. Patient mode. Patients can view active trials and complete a simplified questionnaire. The application filters these for them in the same way as it does for doctors. This means they can find out for themselves whether they are eligible to take part in a trial.
Data protection. Patient data is processed solely by a model running on our own server, and only in memory. It is not stored anywhere. We use an external model solely for formatting public texts from registers.
Technologies. React and Tailwind on the front end, FastAPI and SQLite on the back end, and vLLM with the Qwen model on a GPU server.
- Had before
- nic
- Use of AI
- Claude/Codex/Copilot/Kimi/GLM, čo prišlo pod ruku. Správnosť kontrolujeme sledovaním agentových tool calls a manuálnou kontrolou kódu. Už v tom máme prax.
- Impact
- Lekári. Lekár dnes ručne prechádza zoznamy klinických štúdií a porovnáva diagnózu pacienta s podmienkami každej z nich. S našou aplikáciou stačí vložiť lekársku správu alebo vyklikať krátky dotazník. Aplikácia za pár sekúnd ukáže všetky štúdie, do ktorých by pacient mohol byť zaradený. Odporúčací e-mail koordinátorovi štúdie lekár odošle jedným klikom. Z hodiny práce sú 2 minúty.
Pacienti. Pacient si sám prezrie aktívne štúdie a vyplní rovnaký dotazník, aký používa lekár. O možnosti zapojiť sa do štúdie sa tak dozvie, aj keď ju jeho lekár sám nenavrhne. Prístup k inovatívnej liečbe potom menej závisí od toho, v ktorej nemocnici sa pacient lieči.
Klinický výskum. Ľahšie vyhľadanie štúdie znamená viac odporúčaných pacientov a rýchlejší nábor. Štúdie sa tak skôr naplnia a nová liečba sa dostane k pacientom rýchlejšie. - Who benefits
- Patients and their families, Doctors and nurses, Hospitals and care providers
- InterSystems technologies
- No
Machine translation from Czech
- Team
- Podbranč — David Bujzaš, Jakub Komanický, Jaroslav Podmajerský, Marek Hric
- Challenge
- TriaLink · Brno