Revolutionizing Myeloma Care: AI Matches Patients to Trials | Jorge Hurtado Martinez, MD | EHA 2025

Hi everyone, I'm Jorge Arturo Tadal-Martinez and I'm the Senior Manager of Clinical Research at the Health Foundation. This time I'm going to present the poster we're presenting at IHA at Milan. This poster is related to clinical trials. So as we know right now clinical trials have a lot of barriers and it's almost a labyrinth to navigate which becomes ever more important as advances occur and as patients start to have less options as their journey with maloma progresses and with other diseases. So we at Health Tree try to find new ways to approach these problems. We use a data-driven approach where we use electronic health records to try to match patients to these clinical trials. So the first challenge that you find is that the clinical trials at clinicaltrials.gov, although they made a huge change in the platform, it's even more user-friendly than before. In terms of the data structure, there is not a clear standard. So you see one criteria shown in multiple ways and sometimes this can be difficult for physicians and patients to navigate. How to match this criteria that there are sometimes 70 different criteria just to get into clinical trials. What we did is that our development team led by Juan Capdevilla and Aurelio looked at a bunch of clinical trials from clinicaltrials.gov and found out that 125 clinical trials produced several thousands of clinical trials with a little bit of criteria. But from these criteria they can only generate 100 rules in terms of how patients can be matched with these a little bit of criteria. To navigate these different criteria easily, we built an AI engineering project where we use Chagypti to structure these eligibility criteria into clear rules that we can match to electronic health records. We focused on age-related eligibility criteria, ECOG, and other criteria that we could structure easily and that we could build benchmarks from these clinical trials. And we found out that with these models we had about a 95% accuracy in terms of the generation of these criteria to match to electronic health records. When you have thousands of these criteria and you need to navigate patients, I think these projects help us get patients much closer to getting into a clinical trial and increasing not only the quality of life of the patient by getting them into a new treatment much faster but also accelerated research for all clinical trials.