CancerTrialsBR: An open-access, collaborative platform to improve visibility and access to cancer clinical trials in Brazil.

Abstract

1546 Background: Lack of timely and understandable information is a major barrier for cancer trials recruitment, particularly outside high-income countries. ClinicalTrials.gov (CTGov) has been widely used to address this need, but language barriers and inconsistent identification of research facilities limit its use. Large language models (LLMs) now offer opportunities to address these limitations. We describe CancerTrialsBR, a pilot platform that integrates CTGov, Google Maps (GMaps), and LLMs to provide Portuguese-language geographically contextualized cancer trial information in Brazil. Methods: Clinical trials data and listed research sites are retrieved from CTGov on a daily basis with condition as “cancer”, location as “Brazil”, and status as “recruiting”. Using prompt engineering techniques (CoT and few-shot), trial information is processed by LLM to generate a Portuguese brief summary, ideal patient profile, and trial descriptors according to primary site (N=39), treatment settings (N=8, i.e. first-line), and biomarkers (N=18, i.e. PDL1). For each site, its address components (i.e. name, city) are queried through the GMaps API for standardized facility identification. A LLM performs facility-matching validation, and invalid cases are manually reviewed. This enriched database is made available as an open-access clinical research platform of the Brazilian Society of Clinical Oncology since November 4th, 2025 (https://pesquisaclinica.sboc.org.br), where users can suggest real-time status updates. We report facility identification metrics, database statistics, and site interactions since launch. Results: Of 2,873 unique CTGov addresses, 654 (22.76%) retrieved zero results from GMaps, of which 612 (93.57%) were subsequently identified by zip code. For the 2,219 GMaps retrieved locations, LLM validation demonstrated an accuracy of 85.78%, with positive and negative predictive values of 96.9% and 59.0%. Overall, only 93 (3.2%) addresses remained unidentifiable after human review. As of January 19th, 2026, the database comprised 272 trials conducted across 191 research facilities, encompassing 22 of 27 Brazilian states and 76 cities. The median (IQR) of clinical trials per facility was 2 (1-9) with 8 facilities hosting 40 trials or more. São Paulo (260/272, 95.6%), Rio Grande do Sul (195/272, 71.6%), and Rio de Janeiro (108/272, 39.7%) had the largest availability of trials. The platform had 2,320 visits from 220 cities within 76 days, and the leading searched cancer was breast. Users feedback resulted in 34 of 57 (59.6%) suggested status updates incorporated in the database. Conclusions: CancerTrialsBR demonstrates the feasibility of integrating LLMs and GMaps to deliver accessible, real-time cancer trial information in Brazil, with early adoption highlighting unmet need and offering insights into the country's cancer research infrastructure.