Some artificial intelligence breakthroughs happen in computer science labs or tense televised board games between a person and a machine. The latest advance in medical AI has less glamorous origins: the depths of US government bureaucracy.
The US Centers for Medicare & Medicaid Services (CMS) recently said it would pay for use of two AI systems: one that can diagnose a complication of diabetes that causes blindness, and another that alerts a specialist when a brain scan suggests a patient has suffered a stroke. The decisions are notable for more than just Medicare and Medicaid patients—they could help drive much wider use of AI in health care.
Both products are already cleared by the Food and Drug Administration and are in use by some providers. But new devices and treatments generally aren’t widely used until the US government authorizes payments for Medicare and Medicaid patients. Private insurers often take their cues on whether to cover a new invention from CMS, although they usually pay higher rates.
Last month, CMS began paying for the use of AI software called ContaCT, from San Francisco startup Viz.ai, under a program that encourages adoption of new technologies. ContaCT is installed in a hospital emergency department to alert a neurosurgeon when algorithms see evidence on a CT scan that a patient has a blood clot in their brain. Speeding up stroke diagnosis and treatment, even by minutes, can dramatically reduce a person’s disabilities and recovery time.
The agency has also said it will soon pay for software called IDx-DR, which analyzes photos of a person’s retinas to diagnose diabetic retinopathy, a complication of diabetes that can cause blindness. CMS proposed in August to pay for the software, which was created by Digital Diagnostics of Oakdale, Iowa.
Viz and Digital Diagnostics both received FDA approval in 2018, making them pioneers in convincing regulators that AI can improve health outcomes. IDx was the first AI product approved to diagnose disease, a clinical call previously made only by human physicians. Convincing CMS that taxpayer dollars should be spent on medical AI could be considered a similar milestone.
“This is very important for everyone in AI,” says ophthalmologist Michael Abramoff, the CEO of Digital Diagnostics. The proposal to pay for IDx would also cover other AI tools that diagnose diabetic retinopathy.
Government willingness to pay for use of AI tools could be good news for other companies working on medical AI products. Analysts CB Insights reports that $4 billion was invested into AI health care startups in 2019, up from nearly $2.7 billion in the previous year.
CMS valued the brain and eye scans very differently—highlighting the complexity of US health care and a challenge for new AI technology. In each case the agency faced a question with philosophical dimensions: How do you value the work of a piece of software that by design performs a task typically done by a highly skilled human? It came up with two very different answers.
In the case of Viz’s ContaCT, CMS ruled that hospitals should be eligible for up to $1,040 for using the software for certain patients, citing evidence that it substantially improves stroke treatment. The agency pondered whether AI software that merely speeds work typically performed by humans was novel enough for a program reserved for new technologies, noting that “human intelligence and human processes are not FDA approved or cleared technologies.”
Chris Mansi, CEO and founder of Viz, says CMS’ approval has already encouraged more hospitals to sign up for ContaCT. It was previously used in about 500 major hospitals, who went ahead without the promise of government reimbursement because identifying stroke patients faster can increase the number of lucrative, time-sensitive surgeries they perform. “The people benefiting were major hubs,” Mansi says. “More hospitals can now use the software and get paid for it.”
CMS has proposed that hospitals get paid much less for retina-checking software like IDx. Abramoff says it would be generally less than $20, though the amount varies depending on geography and other factors.