The Center for Data Innovation spoke to Daniel Nathrath, co-founder and chief executive officer of Ada, a London-based health startup that has developed an artificial intelligence (AI) tool to help people interpret their symptoms. Nathrath talked about how AI can help doctors to make diagnoses more transparent to patients, and how the technology could improve global access to healthcare in the future.
Nick Wallace: Ada is an AI-driven chatbot that helps people understand their symptoms. How does it differ from the familiar scenario of typing your cold symptoms into a search engine and being told you are probably dying?
Daniel Nathrath: That’s a good description of what goes on. We know that nine out of ten people Google themselves before they go to the doctor’s, so for most people—including some doctors themselves—that’s really the start of the patient journey nowadays. And as you point out, it’s quite inaccurate. Search engines are great tools to look for information if you already have a general idea about what you’re looking for, but when it comes to assessing your health, they’re not ideal because you need more pieces of information than one or two search terms if you want an individualized assessment of your health. A pregnant woman with diabetes, an 80 year-old man, and a two-year-old child might all share the same symptoms, but could have completely different problems. Because every human is different and every constellation of symptoms can mean something different, Google is not the right tool to get definitive answers about what you have.
So what Ada does, in contrast, is work on a conversational basis to collect information from the user, very much like a doctor would. The first thing that happens when you go to the doctor is they ask you a series questions, which is called taking the patient history. A lot of diagnoses made by doctors are on patient history alone—this is estimated to be over 80 percent of diagnoses. What we’ve done over the last six-and-a-half years is develop an extensive medical knowledge base and a sophisticated reasoning engine—the combination of which you could call AI, if you want to—and we’ve been testing and refining them.
We started with specialist doctors, and then broadened the medical knowledge base and hired a lot more doctors. We trained them to work on the software, basically as medical programmers. Then finally, at the end of last year, we started to make available the same technology in a very easy-to-use interface, which is the chatbot for patients, and then launched that on iOS and Android. Some of the most common comments we get in reviews from users are thinks like, “this is so much better than Googling my symptoms,” and “I’m no longer worrying that I’m going to die.”
Wallace: You also produce a version for health services to use: what is the value to them of deploying an AI health assistant?
Nathrath: The value is really for health systems and health professionals. Doctors actually tell us they’re surprised when they get a patient who hasn’t Googled their symptoms, and they have to spend a lot of time explaining to patients that the information they find on Google doesn’t necessarily apply to them. There’s a shortage of about 7 million doctors worldwide. Typically, doctors only have about eight minutes with each patient. With Ada, doctors get a briefing about the patient that saves them having to ask some of these questions. So there’s an efficiency gain for each doctor.
But it’s also more efficient from a health system perspective. In the United Kingdom, you have the National Health Service’s 111 helpline for non-emergency calls, but that doesn’t triage people very well, so a lot of people who don’t need to still end up in the emergency room. That is a very bad allocation of resources, because the people who really need emergency care get prevented from accessing it as quickly as they should because people who shouldn’t be there are using it. So if you have something like Ada that can triage patients properly and send them to the more appropriate next step in their care journey, then this is more desirable from a health systems perspective.
The other aspect is that if you have a more accurate diagnosis faster, that’s desirable not only for the patient but also for the health service, because a misdiagnosed patient leaves a very significant cost trail in the system: they visit more doctors, have more tests, take more medication, and see their health deteriorate further, leading to additional healthcare costs. That’s why we provide the app to doctors as well as patients, because we believe we’re the only such service that has both in mind.
Wallace: Does Ada ever spot things that a human might not, and how can doctors learn from that and make use of it?
Nathrath: Doctors tell us that the mere process of using Ada to go through patients’ symptoms with them helps them to become better doctors, and some doctors also use it to explain things to the patient. That’s why we make sure Ada isn’t a “black box.” We make the reasoning behind diagnostic suggestions transparent, so a doctor can see how the combination of symptoms points towards a certain medical condition being the cause of those symptoms. We put an emphasis on making that reasoning transparent, which is what doctors really like, because if the app does spot something, the doctor can understand and trust the finding and it can help train them to spot it themselves next time.
Wallace: What have you learned about how people interact with Ada generally? Are we all hypochondriacs deep down?
Nathrath: We have over two million cases in the system now, so obviously we have a lot of data and a lot of feedback. When we were starting out with Ada, some of my self-described hypochondriac friends said things like, “don’t give it to me, I’ll never do anything else, I’ll be constantly checking to see whether I’m about to die!” They were joking, but that was a serious question we had to ask ourselves: would this lead people who tend to be worriers to be even more worried? But we get feedback from people who say they feel less anxious now because they can better identify where their issues are coming from and prepare themselves better to decide whether they should see a doctor, and what kind of doctor. They say using Ada calms them down.
Another funny thing is that although we try to avoid personifying Ada too much with things like avatars, it obviously still has a woman’s name. We do get people saying things like, “when can we meet, Ada?” and “I love you Ada,” and “will you marry me, Ada?” but I don’t think these people are entirely serious.
Wallace: What impact do you think AI is going to have on healthcare in the long term?
Nathrath: We are strongly convinced that Ada, and technologies like Ada, will help improve access to healthcare, and lead to more evenly distributed access to high quality information and care.
I mentioned earlier that there’s a shortage of about 7 million doctors in the world. Right now, there’s about a billion people in the world who will probably never see a doctor in their life, because either there is no doctor in their area or if there is, they can’t afford to see them. But if you can give these people a technology that can diagnose them as well as a doctor through a cheap android smartphone—which someone somewhere around them will have—then that really is a democratization of healthcare, because suddenly the villager can get the same diagnostic quality as a rich banker in New York or a venture capitalist in Silicon Valley.
Ada, by all our benchmarking tests, is already more accurate than any individual general practitioner we tested working from the same information. So we are already starting to realize that vision. There are other issues to solve—like how do you get treatment to those people once they’re diagnosed. But we’re looking forward working with NGOs, for instance, on problems like that. So I really hope it will help people in the developing world.
But I hope it will help people in the developed world too: even in the United States, people often have to wait weeks to see a doctor. You don’t have to see Ada, so as a starting point, we hope this will help people to better manage their health from home.