This Serial Entrepreneur Wants The FDA To Approve His AI Doctor |
Martin Varsavsky has trouble keeping track of all the ventures he’s started. There are more than a dozen of them, including a handful that became worth more than $1 billion. But Certuma, which launched quietly this winter, may be his biggest idea yet: He plans to build the first FDA-approved AI doctor.
“What’s happening now is everyone you know, and probably you yourself, are checking your medical problems with AI. But then what happens when you want action? The AI, after giving you a wonderful, accurate diagnosis of what’s wrong with you, says, ‘I am not a doctor,’” Varsavsky tells Forbes. He ticks off all the questions it might answer this way, from getting a prescription to scheduling imaging. “I want to fix the ‘I am not a doctor’ problem by building AI that is recognized by the FDA and recognized by the states.”
AI doctors could help solve an important problem, much like telemedicine did during the Covid-19 pandemic. There simply aren’t enough physicians to serve all the people who need them, especially in rural areas. The shortage is only getting worse. More than 100 million people in the United States face barriers to accessing primary care. Meanwhile, some 46% of counties don’t have a cardiologist; in rural counties, that number rises to 86%.
Certuma has an early target list of 25 medical conditions, including urinary tract infections and sore throats. All are common health problems that take up a lot of primary care doctors’ time, but are treated with a standard protocol. “We look for conditions that are low medical risk, and have very well defined management decision trees,” says Dr. Armando Cuesta, the startup’s chief medical officer. The goal would be for Certuma’s AI doctor to both diagnose these conditions and provide basic treatment for them.
The biggest danger of an AI doctor is that it will provide inaccurate advice or incorrect treatment that harms – or, worse, kills – a patient, and that any safeguards will fail to catch the errors. The concept also raises questions about what being a doctor actually means: while it may be possible to automate parts of a doctor’s job, it’s an uncomfortable idea to replace a human as the decision-maker.
Varsavsky is clear-eyed that developing safeguards on the technology is only half the battle. The other will be gaining regulatory approval. To launch Austin, Texas-based Certuma, he raised $10 million in seed funding led by Joe Lonsdale’s 8VC at a valuation of $60 million. The company is working simultaneously in the United States and in Varsavsky’s native Argentina, where any clinical trials would cost a lot less and where he has connections with its libertarian President Javier Milei. It’s a long way to go from there to an AI doctor approved in the U.S., but Varsavsky has a decent shot at it. Previously, he built the largest chain of fertility clinics in North America, and has substantial experience starting businesses at the intersection of medicine and technology that require FDA oversight. Forbes estimates that he is worth at least $700 million.
“I want to fix the ‘I am not a doctor’ problem by building AI that is recognized by the FDA and recognized by the states.” Martin Varsavsky
“I want to fix the ‘I am not a doctor’ problem by building AI that is recognized by the FDA and recognized by the states.”
The path to FDA approval for an AI doctor isn’t entirely clear yet. The agency has approved algorithms for specific clinical uses, such as analyzing radiology reports and flagging strokes quickly. But approving AI as a “doctor” that could order tests, write prescriptions or diagnose multiple conditions would likely require building out each use case, one by one. That might mean getting the AI approved for urinary tract infections first, say, and then moving on to another condition.
It could get even more complicated. In the U.S., doctors are licensed by the states, not by the federal government, so oversight of certain bits of doctoring, like prescribing, may reside with local authorities. That could set up a showdown between the states and the federal government. Finally, as with any new technology in medicine, there’s the question of figuring out how to get insurance to cover it.
“It’s going to be a painful, long, difficult process,” Varsavsky says. “But many things I do are painful, difficult and long, and then they pay off.”
Before turning his attention to AI doctors, Varsavsky, 65, built his fortune in telecom, biotech and fertility. Forbes detailed his early days with Prelude Fertility – his seventh company, while he was having his seventh child – in a 2016 magazine cover story (“End of the Biological Clock”).
He was born in Buenos Aires, where his father was an astrophysicist and academic. In his liberal Jewish family, Varsavsky says he was “kind of the black sheep” for his interest in business. After his cousin was murdered by the Argentine military in the 1970s, a teenage Varsavsky immigrated to the U.S. with his family. He launched his first business, converting industrial buildings in New York’s SoHo neighborhood into residential lofts, while still in graduate school at Columbia University. In 1991, he started his first telecom, Viatel, to provide low-cost, long-distance calling at a time when overseas calls were extremely expensive. “I wanted to be an entrepreneur, a capitalist, a builder,” he says.
After moving to Madrid in 1995, he launched a string of telecoms, including Spanish broadband operator Jazztel, which was ultimately acquired by Orange for $4.4 billion in 2014. Not everything worked: Varsavsky points to European cloud computing company Einsteinet that “failed miserably” and cost him some $50 million.
“You’ll see hundreds of people with headaches, and lurking among them is someone with meningitis.” Dr. Robert Wachter, chair of the department of medicine at UCSF
“You’ll see hundreds of people with headaches, and lurking among them is someone with meningitis.”
At the time he started Prelude, in 2015, he and his wife Nina were struggling to conceive. He soon saw an opportunity to shake up the large and fragmented infertility industry. By 2016, Prelude, which was backed by Lee Equity Partners, was profitable on an estimated $35 million in revenue. In 2019, it merged with Inception Fertility to become the largest network of fertility clinics in North America. Today, it has 2,300 employees in 83 cities. His early idea of marketing fertility services to younger women who want to freeze their eggs to gain control of their lives is now mainstream. And he has launched two other fertility firms: Overture Life, which is working to automate embryology labs, and artificial womb startup Gameto.
Certuma is the only one of Varsavsky’s companies where he is currently CEO. “Every three or four years, when I have a compelling idea for a business, I just can’t stop thinking about it,” he says. “I was put into this world to do this job. It’s what I like to do—to turn ideas into companies. It’s the mission of my life.”
Varsavsky says he first came up with the idea of Certuma when he had a dermatological problem while at his second home in Miami. “I knew what I needed, but I couldn’t get an appointment,” he says.
He soon became obsessed with how he might use AI to build a new type of doctor who would always be available. Late last year, he took his nascent idea to 8VC’s Lonsdale, and wound up speaking with Sebastian Caliri, head of the firm’s healthcare team, for two hours. “When I met Martin I immediately recognized a fellow traveler who was aligned on the technology, but was also practiced on the regulatory conversations that need to happen to introduce this technology to the country,” says Caliri, who worked on the health team at Palantir before joining 8VC. “We are going to need to pioneer a path to FDA approval.”
Dr. Robert Wachter, chair of the department of medicine at UCSF and author of a new book on how AI is transforming healthcare, argues that while AI doctors are needed to solve the primary care crisis, the bar needs to be high for what they’re allowed to do. “It is important to figure out the boundaries of this. Proving the AI can refill your Lipitor is a far cry from the AI being a doctor,” he says. Even for seemingly simple cases like a patient who has a headache or a sore throat, there will be outliers that the AI needs to be trained to recognize and triage appropriately. “Lurking in people with sore throats is someone with a diagnosis of throat cancer,” he says. “You’ll see hundreds of people with headaches, and lurking among them is someone with meningitis.”
“It’s going to be a painful, long, difficult process. But many things I do are painful, difficult and long, and then they pay off.” Martin Varsavsky
“It’s going to be a painful, long, difficult process. But many things I do are painful, difficult and long, and then they pay off.”
Varsavsky believes that AI will one day give patients better outcomes than human doctors because of its ability to synthesize vast quantities of data. But for now AI still hallucinates in ways that can be dangerous to patients. “The AI is good now, but it is not good enough to get approved, and we need to get it approved,” Varsavsky says. “The Hippocratic oath is ‘do no harm.’ I want an AI that complies with the Hippocratic oath.”
To put that directive into practice, Certuma is building a second layer of safety into its AI, which incorporates multiple LLMs. Rather than generative AI, this safeguard is deterministic, based on clinical data that applies explicit rules and works to detect red flags – and to avoid improvising if something is unclear. In the beginning, Certuma will have a licensed physician review each AI-generated decision before it goes to a patient. Over time, its goal is for the AI to work increasingly autonomously with physicians focused only on the complex or ambiguous cases.
Especially for those who lack access to primary care, treating simple health problems like sinusitis and pink eye with AI should, in theory, keep people healthier, while reducing stress on the medical system. Varsavsky also sees it helping patients who have been overlooked by the medical establishment, like Hispanics who face language barriers in receiving care.
Varsavsky believes that AI doctors are coming sooner than you think—perhaps as soon as six months for certain primary care tasks. In Utah, startup Doctronic is already piloting AI to renew a limited set of prescriptions, the first real case of an AI acting as a doctor. Other states like Texas and Arizona have also set up programs to test out AI experiments (though none have yet announced AI doctor initiatives). Meanwhile, the federal government through its healthcare-focused moonshot funding program Arpa-H has invited teams to develop AI for cardiovascular care. Certuma is both applying to Arpa-H, and talking with Utah about “a way to put this into practice with those conditions that are low-hanging fruit,” Cuesta says.
Beyond an AI doctor that could diagnose or prescribe, Varsavsky has an even grander vision that would ultimately combine Certuma’s AI software with autonomous robotic tools. He imagines a booth where a patient could simply step in to receive full imaging or get their blood drawn autonomously, or perhaps a humanoid robot performing such medical care at a pharmacy. That, of course, is a lot further off.
In the more immediate term, Certuma is building out its AI and figuring out its strategy for clinical trials. “The FDA is still pondering how to do this,” says Jenni Main, Certuma’s CFO who previously was chief operating officer at the Centers for Medicare and Medicaid Services. She thinks a likely way for Certuma to approach regulatory approval would be to chunk off different pieces for submission and then amass as much clinical data as possible to be clear about the risks.
It helps that the Trump administration may be particularly receptive to AI doctors, Varsavsky says. He figures that while gaining regulatory approval could cost hundreds of millions of dollars, which the startup would need to raise from investors, it’s the best way to gain an edge.
“This is going to be a highly competitive field. We want to build the best AI and the first one approved,” he says. “The approvals are the moat.”