WeDoctor AI Research Institute: 'Knowledge-Embodied Intelligence' Emerges in AI-Powered Health Management

Xu Hongxia, Chief Scientist of the WeDoctor AI Research Institute, on April 27Invited to Attend VCBeat“AI Large Models Empowering Health Management” Online Roundtable Discussion: In-Depth Exploration with Industry Experts on the Hot Topic of “Multi-Angle, Multi-Stage Empowerment of Health Management by AI Large Models”. Addressing the industry's widespread concerns about mitigating hallucinations in general-purpose large language models, wearable device applications,Topics such as the commercialization pathway of AI in healthcare were the focus of the event, highlighted by Micro Medical Group’s pioneering explorations and Dr. Xu Hongxia’s in-depth insights.

In the fields of AI-driven healthcare and health management, WeDoctor Holdings has established an early presence and achieved remarkable results, consistently remaining at the forefront of the industry. Xu Hongxia introduced that, leveraging industry-leading AI infrastructure—including the WeDoctor Medical Large Language Model and five major intelligent agents: AI Physician, AI Pharmacist, AI Diagnostic Testing, AI Health Management, and AI Intelligent Control—the Tianjin AI Healthcare Community has significantly improved patient outcomes. From January 2023 to June 2024, the rate of glycated hemoglobin (HbA1c) compliance among diabetic patients under its management rose from 17.8% to 44.2%, blood pressure compliance increased from 19.5% to 61.5%, and blood lipid compliance grew from 24.8% to 27.9%. To date, the number of patients managed by the Tianjin AI Healthcare Community has exceeded one million.

During the roundtable discussion, Xu Hongxia addressed the widely debated issue of hallucinations in large AI models. She pointed out that hallucinations in general-purpose large models are currently unavoidable; however, in healthcare applications, such errors are unacceptable, necessitating a two-pronged approach to mitigate them. Citing the pioneering efforts of WeDoctor Holdings as an example, she provided an in-depth analysis from the perspectives of technology and collaboration, as well as product and operations.

“On the technical front, it is necessary to optimize the capabilities of large language models (LLMs) by adjusting model parameters and employing techniques such as fine-tuning with specialized medical knowledge and Retrieval-Augmented Generation (RAG). Meanwhile, a professional audit model is indispensable, which tests practitioners’ industry accumulation,” revealed Xu Hongxia. She disclosed that the Weiyi Medical LLM adopts a multi-base architecture, integrating external general-purpose LLMs to form an industry-specific reasoning engine base. Simultaneously, through proprietary technologies, it focuses on the deep integration of real-world clinical diagnosis and treatment data with clinical decision pathways. For instance, in collaboration with national-level expert teams, it has developed unique medical rules, including 13,000 specialized rules for specific diseases and chronic conditions across 308 categories, 470,000 rational drug use rules, and 2.22 million medical insurance audit rules across 377 categories. This embeds medical rigor into the entire AI decision-making process. The model demonstrates outstanding performance across various evaluation dimensions, including medical knowledge Q&A, medical language generation, complex medical reasoning, medical language understanding, and healthcare safety and ethics. It establishes an intelligent decision-making chain from “open exploration” to “precise targeting,” ensuring evidence-based compliance and thereby achieving breakthroughs in the clinical application of medical large language models.

On the product operations front, Micro Medical Group (Zhejiang) Co., Ltd. has implemented a dual-safeguard strategy through dedicated professional staffing. Within the Tianjin AI Health Community, the company has established the critical role of Health Manager to effectively reach and manage elderly patients. With AI assistance, Health Managers not only efficiently deliver health prescriptions and provide exercise and dietary reminders but also prompt users to schedule regular follow-up visits. By delivering on-the-ground services at primary healthcare institutions, they successfully bridge the gap between AI applications and patients. This approach is a key factor behind Micro Medical Group’s significant breakthroughs in enhancing quality and efficiency in patient health management within the AI Health Community.

In user health management, the integrated application of large language models with hardware terminals such as smart wearables is garnering increasing attention. Xu Hongxia specifically outlined four core scenarios.

First is out-of-hospital health monitoring. By using smart wearable devices to collect data such as blood pressure and heart rate in real-time, and synchronizing it with the health management system, abnormal warnings can be achieved.

Next is daily exercise management. Large language models generate personalized exercise plans, while smart wearable devices act as monitors to dynamically adjust the plans based on user adherence and data outcomes.

Prescription medication reminders are essential. Smart wearable devices can serve as medication supervisors; by integrating the physician’s medication regimen into the device, they can provide timed alerts to patients, thereby improving medication adherence.

Adherence Incentives Are Also a Major Scenario in Health Management. In this scenario, mechanisms such as point-based rewards and health rankings can motivate patients to adhere to health plans over the long term, thereby improving health management outcomes.

Xu Hongxia pointed out that the four major scenarios span the entire process of chronic disease management, helping patients shift from “passive treatment” to “proactive health,” thereby improving the efficiency of chronic disease management. As whole-process health management becomes more robust and comprehensive, its target population can expand from individuals with chronic diseases to the general healthy population, ultimately achieving family doctor-contracted health management for the entire population.

These four core scenarios represent the key directions that WeDoctor Holdings is currently exploring in the field of health management. Dr. Xu Hongxia revealed that in recent years, WeDoctor Holdings has been collaborating with several well-known domestic smart wearable hardware manufacturers to develop customized solutions, achieving integration between “in-hospital” and “out-of-hospital” care and bridging the “last mile” across the full lifecycle of “prevention, screening, diagnosis, treatment, management, and rehabilitation.”

In Xu Hongxia’s view, large language models have a predictable development trajectory. On one hand, they will continue to evolve toward multimodality; on the other, they will integrate with various tools and give rise to “new species.”

“Currently, the concept of embodied intelligence is gaining significant traction. In the realm of large language models and health management, ‘cognitive-embodied intelligence’ is actually taking shape.” Xu Hongxia believes that while embodied intelligence represents the intersection of artificial intelligence and robotics, “cognitive-embodied intelligence” is the product of integrating artificial intelligence with healthcare. The term “cognitive” comprises two components: on one hand, specialized medical knowledge, and on the other, the large model’s perceptual processing of patient data.

She explained that large language models specialized in the medical field possess extensive medical knowledge, while smart wearable devices can capture patient data to feed back into these models. The combination of the two is poised to bring about a transformation in health management that may be no less significant than the impact of artificial intelligence on robotics. In this domain, WeDoctor Holdings Limited is well-positioned to continue leading industry exploration.

In recent years, Micro Medical Group’s outstanding achievements in the field of AI healthcare have become a highlight of the industry. In 2015, Micro Medical Group pioneered China’s first internet hospital, the Wuzhen Internet Hospital, and successfully facilitated the establishment of regulatory standards for the internet healthcare industry. In 2017, Micro Medical Group partnered with Zhejiang University to co-establish the Ruiyi Artificial Intelligence Research Center, aiming to promote the research, development, and application of artificial intelligence technologies in the healthcare sector. In 2020, leveraging its accumulated AI capabilities, Micro Medical Group began constructing an AI-powered hospital based on the Tianjin Internet Hospital. Focusing on medical service capabilities that cover the entire process from pre-diagnosis to post-diagnosis—including examinations, diagnosis, medication, health management, and cost control—Micro Medical Group’s AI Hospital has developed five intelligent agents: AI Doctor, AI Pharmacist, AI Examination, AI Health Management, and AI Intelligent Control.

In 2022, in accordance with the deployment of the Tianjin Municipal People’s Government, WeDoctor AI Hospital took the lead in joining forces with 266 primary healthcare institutions across Tianjin to establish the AI Health Community, jointly building a health accountability system based on disease-specific and capitated global budget payment models. Among these efforts, WeDoctor AI Hospital assumed three core functions: “overall technological empowerment, overall health accountability, and overall performance coordination,” serving as the engine of the AI Health Community—the General AI Hospital. Empowered by the General AI Hospital, the Tianjin AI Health Community has achieved a “four-party win-win” scenario for primary healthcare institutions, physicians, patients, and health authorities along with public medical insurance funds, offering a “Chinese solution” to addressing this global challenge.

Dr. Xu Hongxia is primarily dedicated to the interdisciplinary integration and industrial translation of medicine and artificial intelligence. In recent years, she has published more than 50 SCI-indexed research papers and conference papers in international journals or at top-tier conferences, filed over 10 invention patent applications, presided over one project funded by the National Natural Science Foundation of China, served as the principal investigator for one topic under the National Key R&D Program, and participated in seven projects, including General Program projects of the National Natural Science Foundation of China and projects funded by the Zhejiang Provincial Natural Science Foundation. She was awarded the Second Prize of the China Institute of Electronics Science and Technology Award in 2022 and the Second Prize for Innovation Achievements in Industry-University-Research Cooperation, among other honors.

The “Medical Large Language Models” Special Series of Live Streams, initiated by VCBeat, has garnered widespread attention from both within and outside the industry. This series invites leading enterprises in China’s medical large language model sector to engage in in-depth discussions focused on specific application scenarios, aiming to address implementation challenges, connect ecosystem resources, and facilitate greater ecosystem collaboration and supply-demand matchmaking.