Yandex's neural networks help clinicians start clinical trials dozens of times faster
N.N. Petrov National Medical Research Center of Oncology (NMRC) implemented a solution powered by a YandexGPT model to process clinical trial documents. These include protocols, instructions, scientific evidence, and other documents. The initial data verification now takes minutes instead of weeks. The solution automatically analyzes hundreds of pages, checks compliance with requirements, identifies errors, and suggests how to fix inaccuracies. This helps medical professionals and pharmaceutical companies prepare accurate documentation more quickly and obtain approval for their research. As a result, patients gain earlier access to new drugs, medical devices, and treatment methods. The solution is planned for use in oncology centers in BRICS countries and institutions conducting clinical trials.
The solution, developed by the Yandex Cloud Technology Center for Society (part of Yandex B2B Tech), Raft, and the NMRC, is already being used by the local ethics committee (LEC) at the NMRC. The LEC is responsible for reviewing the ethics of each clinical trial by assessing the risk-benefit ratio and ensuring that the rights of research participants are upheld. The LEC handles about 1000 applications per year.
Artyom Poltoratsky,
Head of Preclinical and Clinical Research at the NMRC
A single clinical trial application includes hundreds of pages: graphs, tables, descriptions of procedures, scientific evidence, and results of previous studies. All this must strictly comply with both Russian and international requirements on safety, ethics, and research design. Manually analyzing such a volume of data is a difficult task: an expert spends dozens of hours, and the review can take a week or even two. The first tests showed that with the YandexGPT neural network, it now takes only minutes. And the complete approval cycle can be reduced from several months to five to ten business days.
The solution helps speed up initial document evaluation and error correction — over 35% of applications are rejected due to required revisions. Applicants then have to wait a month for the next LEC meeting to have their application reviewed again. These delays mean patients wait longer to begin treatment, and pharmaceutical companies take longer to bring drugs and medical devices to market. With YandexGPT, applicants receive feedback much faster and can fix issues in their documentation before the next LEC meeting.
The solution is a web service running on the Yandex Cloud platform where document scans are uploaded. YandexGPT 5 Pro analyzes documents, divides them into 10 types, extracts information, and automatically checks them against a set of criteria developed by experts at the NMRC. For example, it evaluates research design and structure, participant recruitment methods, and the manner in which participants are informed. LEC experts track the status of applications in Yandex Tracker. In 95% of cases, the initial document review is performed without human intervention. A specialist only needs to check the data accuracy and decide whether to approve or reject the submission.
The solution developed by Yandex and Raft can also be utilized by banks, insurance companies, and government agencies to process a wide range of multi-page documents.