A stab in time saves nine, they say — and a blood thinner in time saves a trip to the emergency room for a heart attack, like Diagnostic robotics hopes to show. The company’s machine learning-driven preventative care aims to predict and prevent dangerous (and expensive) medical crises, saving everyone money and hopefully staying healthier overall — and it’s raised $45 million to scale.
It’s important to explain at the outset that this particular combination of AI, insurance, hospital bills, and “predictive medicine” isn’t some sort of technotopic nightmare. The whole business is based on the fact that it’s both better for you and cheaper if, for example, you improve your heart health instead of having a heart attack.
That’s why your doctors tell you to cut down on red meat and maybe even take cholesterol-lowering medication instead of saying “well, if you’re having a heart attack, just go to the emergency room.” It’s just common sense and it also saves patients, hospitals and insurance companies money. And don’t worry, these kinds of predictions can’t be used to increase your premium or decline care. They want you to make monthly payments – they just don’t want to pay for a $25,000 surgery if they can help it.
The question is, what about less obvious conditions, or conditions for which patients have not undergone specific testing? This is where machine learning models come in; they are very good at teasing a signal out of a large amount of noise. And in this case, the AI was trained on 65 million anonymized medical records.
“We see what people look like before the problems – all we do is preventative care,” said Kira Radinsky, CEO and co-founder of Diagnostic Robotics. “It’s all about providing the right intervention, at the right time, to the right patient.”
She noted that health care providers often target the most expensive patients to reduce costs, for example someone with advanced heart disease. But while acute and maintenance care remains important to them, that money is already gone. On the other hand, diagnosing someone with early signs of congestive heart failure can prevent it from progressing further and save money and possibly even a life. And the technique applies beyond things that can be detected in labs.
“Suppose the challenge is to find patients who suffer from depression or anxiety but are not on drugs,” Radinsky suggested. “How do you recognize someone with depression or anxiety based on medical records? We identify the entropy of their visits – many providers, many complaints – that is a strong signal. Then you do specific questions, a medical triage, and you link them to a psychologist or psychiatrist, and they don’t deteriorate anymore.”
The company claims it can reduce ER visits by three-quarters, which outweighs the direct benefits to a person and their provider; First aid and emergency care are overwhelmed in the US, paradoxically because of the pervasive fear of huge medical costs.
In many cases, she said, medical providers or insurers will offer medications or treatments for free or at a nominal cost because they know they will save themselves a higher bill later on. Sure, it’s all self-serving, but that means you can trust them.
Tel Aviv-based Diagnostic Robotics just raised a $45 million B-round led by StageOne investors, with participation from Mayo Clinic, Technion (Israel Institute of Technology) and Bradley Bloom. Radinsky said this will help the company work more directly with healthcare providers, taking on more holistic health goals in addition to specific high-risk conditions. (The company currently tracks about 20.)
A pilot test of this broader approach was recently validated in a study of several hundred patients, in which the AI-created health plan was tested. statistically indistinguishable from that of a doctor. The company already serves millions of patients in some capacity, in Israel, South Africa and the US, with Blue Cross Rhode Island.
If they expand to your carrier, don’t expect robot research, although the name clearly suggests it.
“You get calls from health care managers who offer complementary treatments, free or almost free,” Radinsky said. The AI has already done its job, and maybe your test results and location suggest you’re at risk – and you’d do well to take these recommendations seriously. AI may still have a lot of room to grow, but it’s good at detecting statistical correlations.
She cautiously added that they are also actively working to find, define and reduce biases in the algorithms, whether they are the result of data bias or human error elsewhere down the line. “What the algorithm is trying to do is see who benefits the most,” Radinsky explained, but as with other forms of AI and machine learning, only careful monitoring will reveal whether the idea of who benefits, corresponds to the real world.