Why preventive care needs
a verifiable evidence chain

A model may identify people at high risk. That does not show that an intervention reached them, let alone that their health outcomes improved. Hospitals, insurers and public-health programmes need a connected record of who was served, what was delivered and how outcomes were assessed.

Policy supports continuity, not automatic procurement

China's 15th Five-Year National Health Plan calls for closer integration of medical care and prevention, digital health management, and stronger links between commercial health insurance and health management. A separate National Health Commission policy on AI in healthcare addresses chronic-disease screening, risk stratification, personalised interventions and follow-up. Neither document gives any particular product automatic procurement or reimbursement status. Sources 1–2

For product design, the question is therefore not only how many people a model flags. Institutions must also be able to see who entered a programme, which clinician-reviewed service was offered, when follow-up occurred and what outcomes can be observed reliably.

External evaluations connect service delivery with results

The US Centers for Medicare & Medicaid Services' Accountable Health Communities (AHC) model screened more than 1.1 million people for social needs and connected eligible participants with community resources. Its final evaluation estimated approximately $220 million in net savings after programme funding costs. This is an estimate for a specific US model, not an average return on preventive care and not a projection for ReHealth AI. Source 3

CMS's AHEAD model also links preventive care and chronic-disease management to population-health accountability and payment reform. Both examples point to the importance of defining the population, service delivery, quality measures, observation period and cost methodology before discussing financial results. Source 4

Four layers separate model performance from programme impact

A 2026 systematic review of AI in cardiovascular care reported improvements in some engagement or health-promotion measures, while noting short follow-up, limited hard endpoints and gaps in external validation across trials. Engagement, risk-factor change and fewer cardiovascular events should not be presented as interchangeable outcomes. Source 5

Useful data must also be lawful, reliable and traceable

US FDA guidance on real-world evidence for medical devices focuses on whether data are relevant and reliable enough for a regulatory question. It is not a Chinese insurance payment rule. The EU's European Health Data Space likewise addresses data reuse alongside individual control, permitted purposes and secure processing, with major requirements phased in over time. Sources 6–7

For an institution-distributed preventive service, an invitation code or QR code is an activation channel, not blanket permission to access all health information. Authorisation should correspond to a defined purpose, appropriate access and necessary processing.

Five questions to settle before a programme begins

Who is eligible? Who reviews and delivers the intervention? What data are used under which authorisation? What are the observation period and primary outcomes? Who reviews the results, and under what agreement are costs discussed? Answering these questions early reduces reliance on retrospective explanations.

For methods, see propensity score matching in preventive care, real-world evidence and why correlation is insufficient for settlement.