Medical AI is entering an era of 'reward based on improved health,' as the new US public health insurance system encourages the use of AI.
One of the biggest obstacles hindering the widespread adoption of medical AI has been the question of 'who will pay and how if AI improves a patient's condition?' ACCESS, a program launched by the Centers for Medicare & Medicaid Services (CMS) in the United States on July 5, 2026, will pay based on actual improvements in health, such as lower blood pressure, reduced pain, and improved depression and anxiety, rather than the number of consultations or phone calls. TechCrunch has reported that ACCESS 'is laying the groundwork for payments toward the full-scale implementation of medical AI.'
ACCESS (Advancing Chronic Care with Effective, Scalable Solutions) Model | CMS
Medicare's new payment model is built for AI, and most of the tech world has no idea | TechCrunch
https://techcrunch.com/2026/05/12/medicares-new-payment-model-is-built-for-ai-and-most-of-the-tech-world-has-no-idea/
Traditional Medicare primarily relied on a fee-for-service model, where payments were based on the time, consultations, tests, and treatments provided by doctors and healthcare professionals. This fee-for-service model lacked a system to reward activities such as AI agents calling patients between appointments, confirming medication pickups, and connecting them with housing and nutritional support.
The 'performance-based payment' system used in ACCESS is not simply based on the fact that a service was provided, but rather on how much the patient's health indicators improved. ACCESS makes ongoing payments to participating organizations that manage the patient's condition, and the full amount is paid when the health condition improves beyond a certain level. CMS gives the example of lowering the blood pressure of hypertensive patients by 10 mmHg. Targeted diseases and symptoms include hypertension, diabetes, chronic kidney disease, obesity, chronic musculoskeletal pain, depression, and anxiety.
However, ACCESS is not a panacea for medical AI. Participating organizations handle highly sensitive patient information such as housing, illness, and mental health, leaving challenges such as how to manage the information entered into the AI, obtaining patient consent, accountability in the event of a data breach, and how to prevent misjudgments or biases by the AI. In particular, housing insecurity and mental health status are not only medical information but also information related to life difficulties and social vulnerabilities. Therefore, even if AI can streamline support, it cannot be widely implemented without protecting patient privacy and implementing human verification systems.
Furthermore, there are financial concerns regarding the CMS innovation program. In 2023, the Congressional Budget Office estimated that the CMS Innovation Center's activities during its first 10 years failed to meet initially expected spending cuts, and instead increased federal spending by $5.4 billion (approximately 852.6 billion yen).
Another challenge is the low payment amount. According to TechCrunch, the monthly per-patient payment that CMS pays is below the expectations of many participating organizations, making it difficult to break even with a human-centered operation. On the other hand, Neil Batribara of 'Pair Team,' an organization selected to participate in ACCESS, says that the payment amount must be low in order to truly encourage the use of AI. The lower the payment amount, the more difficult it becomes to continue patient support with human staff alone, and organizations that can use AI to streamline patient communication and coordination of support will have an advantage.
TechCrunch reports that the essence of ACCESS is not 'medical AI itself,' but 'a system in which payments are made when medical AI improves a patient's health.' Batribara states that a transformation of the payment model is occurring that was impossible with the conventional system. The success or failure of medical AI depends not only on whether a highly accurate model can be created, but also on whether the improved health can be measured and linked to compensation.
Related Posts:
in AI, Posted by log1d_ts