Shanghai healthcare data moves from aggregation to operations, with three types of scenarios—real-world research, health insurance, and clinical agents—reaching scale

In the past two decades, Shanghai's core action on medical data has been "aggregation."

Data shows that, under the guidance of the Shanghai Municipal Data Bureau, the Shanghai Municipal Health Commission, together with the Shanghai Municipal Medical Insurance Bureau, the Shanghai Municipal Medical Products Administration, and the Shanghai Shenkang Hospital Development Center and other relevant departments, has promoted the construction of the Shanghai Municipal Health Data Platform. It has collected and integrated more than 140 billion records from over 400 major public medical and health institutions across the city, achieving "unified standards, unified collection, one source per data item, and shared use" for the city's health data. At the same time, it has built the Municipal Health Data Laboratory, which has connected with 41 development entities and cumulatively engaged with 69 application scenarios. Relying on this platform, out-of-domain delivery has been completed for a cumulative total of 18 data products in the commercial insurance field.

At the sub-forum on "Releasing the Value of Medical Data Elements and Co-building the Industrial Ecosystem" at the 2026 Global Data Merchants Conference, how data becomes products and how cash flow is generated has become a new focus.

But the particularity of medical data lies in the fact that the higher the value, the stronger the sensitivity. If medical data is to be priced, traded, and reused, the prerequisite must be trusted circulation. As the city's "one network" of medical and health information becomes denser and denser, and the institutional framework for authorized operation of public data has basically been established, the real question has become: after data can be supplied, how can it flow, be used well, and be managed effectively?

From "openable" to "operable," security governance comes first

Rather than a technical issue, this is an industrial organization issue.

The use of medical data in Shanghai strictly implements a three-tier gatekeeping process of "trusted data space + ethical review + review of results leaving the domain." Operating institutions provide full-process accompanying services such as scenario plan guidance, data exploration, tool training, and security compliance training; scenario plans must be reviewed by third-party ethics experts to assess scientific validity, security compliance, and reasonableness of data use; after data product development is completed, before out-of-domain delivery, a review of results leaving the domain must also be conducted to ensure that the out-of-domain content is consistent with the declared content.

This mechanism transforms "security" from a mere constraint into verifiable and auditable credit. When every call, authorization, and out-of-domain result is traceable, hospitals dare to continue supplying data, enterprises dare to invest in development, and regulators dare to approve at scale.

Liu Xiao, head of data operations at Shanghai Data Resource Operation Co., Ltd., emphasized in his sharing that the key to truly releasing value lies in transforming data resources into "high-quality data supply and practical applications." This means that raw data cannot directly enter the market; it must undergo anonymization, structuring, labeling, and quality inspection before it can become data products that can be called upon by pharmaceutical companies, insurers, and research institutions.

The development paradigm given by Shanghai is: enterprises bring algorithms on site and complete exploration, training, and validation within the domain; what can ultimately leave the domain can only be model scores, statistical reports, disease-specific labels, and interface services, not detailed medical records. This model has already been proven at the Zhangjiang Science Big Data Innovation Laboratory, incubating scenarios such as real-world research, insurance pricing for patients with pre-existing conditions, and intelligent underwriting.

In other words, the path for the development and utilization of health data has been clarified: data stays put, computing power moves; raw data does not flow out, but value flows out. This is both a requirement of the security bottom line and a prerequisite for the commercial logic to hold.

The development and utilization of three major scenarios is scaling up

The returns from the development and utilization of medical data are quantifiable. Liu Xiao believes that three types of scenarios already have the conditions for large-scale development and utilization.

The first is real-world research. This is the direction with the greatest demand and the broadest scope of application. Taking cardiomyopathy research as an example, the platform can call upon information such as diagnosis and treatment, examinations, costs, and medical insurance enrollment over seven consecutive years to retrospectively evaluate the economic burden of the disease and provide real-world evidence for innovative drug access and pre- and post-market decisions. For pharmaceutical companies, this can replace part of the cost of traditional research and shorten the R&D decision-making cycle.

The second is the restructuring of health insurance. Traditional insurance provides insufficient coverage for people with pre-existing conditions, while insurance products for patients with pre-existing conditions require refined assessment of the disease burden and costs of groups with prior illnesses. Shanghai has already proven scenarios for insurance for patients with pre-existing conditions such as high-incidence cancers in women and lung cancer, with outputs in the form of data reports and statistical analysis rather than raw data. Another type is the application of "two underwriting checks" in health insurance. In July this year, through the laboratory incubation model, intelligent underwriting was proven, and under the premise of personal informed consent and a complete authorization chain, personal diagnosis and treatment records are called upon for analysis, outputting health risk scores to assist insurance companies in underwriting decisions. This model can reduce the pressure of manual review while improving the efficiency of converting high-quality customers.

The third is clinical agents. This is the implementation direction with the greatest imagination. Baidu Health demonstrated the application of tools such as AI pre-consultation, intelligent appointment addition, and intelligent follow-up in hospitals; AI triage and guidance at Wuhan Union Hospital has cumulatively served more than 260,000 patients, with more than 120,000 AI appointment addition services, and the adoption rate by doctors of content generated by AI pre-consultation reached 93.4%. Zhongshan Hospital, meanwhile, has proposed the goal of an "AI-native hospital," turning traditional business systems into MCP-based ones to form more than 400 data and service tools, allowing agents to directly assist doctors in operating systems, writing medical records, and consulting data.

The common feature of these three types of scenarios is that data-use results are measurable, auditable, and closed-loop, making them suitable for forming subscription-based or usage-based billing under the authorized operation framework. They have become representative scenarios for medical data moving from resources to assets.

Blockages shift to "who supplies data, who shares the money, and how to sustain it"

After the scenarios were proven, the problems were not completely solved. The biggest current blockage has shifted from "whether it can be done" to "who supplies data, who shares the money, and how to sustain it." In interviews at the exhibition hall, many exhibitors and industry insiders also supplemented the views of the guests at the conference.

The core contradiction on the hospital side is the lack of motivation to supply data. Data comes from hospitals' daily diagnosis and treatment systems, and supplying data means additional governance costs and compliance pressure, while the benefit-sharing mechanism has not yet been clarified. Xue Wei, deputy director of the Urban Digital Transformation Division of the Shanghai Municipal Data Bureau, also proposed in his speech that it is necessary to "string together like beads into a chain" hospitals' data resources, enterprises' development capabilities, and clinical research needs, turning "case-by-case negotiation" into "systemic breakthroughs."

On the enterprise side, they face the problem of being unable to obtain raw data and can only process it within the domain, and there is still uncertainty about the pricing, reuse, and secondary development of data products. "We have to undertake data desensitization and product development work, and also invest a large amount of computing power. These upfront costs are very high," a professional said.

On the personal data side, there are dual constraints from the authorization chain and compliance boundaries. Liu Xiao specifically mentioned that in personal authorization scenarios, authorization does not mean that diagnosis and treatment details can be delivered; current exploration still mainly relies on "deploying models on site," outputting scores, verification results, and labels, and delivering them in the form of API calls.

In response, Shanghai is promoting the evolution from system building to large-scale operation and value conversion. The next key areas of focus include: clarifying the revenue-sharing mechanism so that every link—hospital data supply, platform operation, data vendor processing, and enterprise reuse—has a predictable share of revenue, solving the problem of "daring not to supply"; establishing a due diligence exemption framework to provide institutional protection for hospitals' data supply behavior under the premise of compliance; and promoting scenario standardization, turning already proven scenarios such as real-world research, health insurance, and clinical agents into replicable application templates and compliance processes to reduce repeated review costs.

Being able to supply is the starting point, being able to flow is the process, being well used is the result, and being well managed is the prerequisite. When hospitals dare to supply, enterprises dare to buy, models dare to be trained, and regulators dare to approve, medical data can truly complete its transformation into a production factor.