Beijing Yifan Fengshun Pharmaceutical Technology Co., Ltd., a subsidiary of Waterdrop Inc., has recently been granted a national invention patent for its independently developed "Intelligent Patient Matching Technology for Clinical Drug Trials." This patent covers methods, devices, and computer equipment for matching patients to clinical trials. Waterdrop had previously developed its own large AI model, Waterdrop Shou, to enhance operational efficiency across its business. The new patent technology integrates deep neural networks with natural language processing to address key industry challenges in traditional patient recruitment for clinical trials, such as high workload, low efficiency, and error-proneness, thereby accelerating pharmaceutical innovation.
Waterdrop is one of the earliest companies in the industry to apply large AI models, investing nearly 300 million yuan annually in research and development. The company has now accumulated over 100 patents related to artificial intelligence technologies. Its self-developed Waterdrop Shou model serves as a central nervous system driving business efficiency and is widely used in various scenarios, including product innovation and user services, improving the overall process efficiency and quality of insurance and health services.
Clinical drug trials are a critical step in bringing new drugs to market, and accurately recruiting eligible patients is essential for trial progress. In traditional recruitment methods, trial eligibility criteria are often described in unstructured natural language, while patient information is scattered across various medical records. Staff must manually compare and analyze this data, which is not only time-consuming and labor-intensive but also prone to inaccuracies due to information discrepancies and human error, significantly delaying trial timelines. Industry statistics indicate that over 80% of clinical trials face delays due to recruitment challenges.
The newly introduced intelligent matching technology employs an innovative dual-track approach combining "numerical matching" and "semantic embedding matching" to achieve end-to-end precision in pairing patients with trial protocols. The system first breaks down patient information into numerical data (such as age and lab results) and clinical visit details (such as medical history and treatment records), while similarly deconstructing trial eligibility criteria. A tri-element extraction algorithm converts numerical data into value-unit-meaning triplets for precise comparison and rapid screening of core matches. For unstructured clinical and descriptive information, techniques like one-hot encoding and BERT language models are used for feature extraction and normalization, enabling deep semantic matching within a unified feature space. The final matching results are derived from combining both tracks.
Compared to traditional manual methods, this technology offers three key advantages: First, a dramatic increase in efficiency, reducing matching tasks from several days to minutes and boosting overall recruitment efficiency by over 300%. Second, higher accuracy, with a matching precision rate exceeding 92%, achieved through globally optimized end-to-end training that avoids error accumulation from traditional steps like word segmentation and relation extraction. Third, strong adaptability, allowing compatibility with diverse medical record formats and varied trial eligibility descriptions, meeting clinical trial needs across multiple therapeutic areas including solid tumors, hematologic diseases, cardiovascular conditions, and autoimmune disorders.
A research and development lead at Beijing Yifan Fengshun Pharmaceutical Technology Co., Ltd. stated that the application of this technology will not only save pharmaceutical companies significant time and labor costs in recruitment but also enable eligible patients to quickly access high-quality clinical trials, offering new treatment hope for those with complex conditions. Moving forward, the Yifan Pharma R&D team will continue to optimize the technology model, expanding capabilities such as multilingual adaptation and specialized matching for rare diseases, to provide more efficient technical support for global pharmaceutical innovation.
The technology is currently being piloted in clinical trial projects at several top-tier hospitals and pharmaceutical companies, having already completed nearly 1,000 patient matches and helping innovative drug trials achieve recruitment targets 3–6 months ahead of schedule. Yifan Pharma, Waterdrop's digital clinical patient recruitment platform, leverages the company's extensive patient network and digital AI capabilities to address industry pain points like slow recruitment and difficult matching, assisting both domestic and international pharmaceutical companies in enhancing their R&D efficiency.