It’s a too familiar scene — 2 a.m. and the computer screen is open to a Google search leading the user down every possible scenario and health scare. The patient then has to decide based on whatever search has come up whether or not a trip the emergency room is in order.
That’s where AI chatbots have come in to play. More and more digital health companies are focusing on providing a place for patients to triage symptoms and then find help.
“It is really difficult for people to understand whether to go to the ER, the urgent care center, the retail clinic, telemedicine, the nurse call line — any of those,” Dr. Andrew Le, CEO and cofounder of Buoy Health, told MobiHealthNews. “It’s hard for people to know which of those options are appropriate because at the end of the day people didn’t get trained medically to triage their cough.”
Triage chatbots are presented as an alternative to googling the internet for symptoms. Typically in a patient facing chatbot the user can put in information and will then be asked a number of questions based on their presenting symptoms. At the end of a session the bot can give the user a little bit more information about their condition and suggestions about next steps. Some chatbots lead to video chats or patient booking systems, while others just give suggestions about where to turn for help.
The main idea is to help folks find resources. Le was inspired to start his company when his father became sick during Le’s last year of medical school.
“My dad had a mini-stroke and he didn’t go to the doctor and I asked him why didn’t you ask me. And he said well you were working,” Le said. “So I said ‘well why didn’t you google it?’ and he said ‘well I don’t trust what google is going to tell me.’ That was the emotional tipping point telling me to leave [school.] I started Buoy with three other people. And we became obsessed with this concept of fixing the first step of healthcare.”
For the most part AI triage chatbots aren’t looking to take doctor’s jobs anytime soon. In fact, many aren’t even diagnosing.
“We are not diagnosing people,” Le said. “We are just trying to help them understand most likely what is going on so that they get to the right type of care at the right time.”
Many bots are looking to support doctor’s work rather than replace it. The idea is to relieve the burden on doctors and medical teams.
GYANT CEO Pascal Zuta said the company’s triage chatbot handles some of the tasks that lead to provider burnout, such as charting and documentation, replying to patient emails and post-visit follow-ups.
“We want to use AI to improve a provider’s ability to diagnose consistently and accurately, see more patients and, most importantly, help their patients get the care they need,” Zuta said.
UK-based Babylon lets users answers a number of questions via the chatbot and if need be hands the case off to a clinician via video call. The idea is to take some of the pressure off of the system.
“In our healthcare system we already divvy up healthcare concerns between people with different skills. Not all patients will receive their care or advice by a doctor, many will be dealt with by excellent nurses or even by administrative staff that may hand out a leaflet,” Nils Hammerla PhD, Machine Leaning Specialist, Babylon, wrote to MobiHealthNews in an email. “We see the AI helping with some of these roles, where it is appropriate. Nobody expects AI-based services to substitute for a 'real' doctor, but there are many other tasks that they could very well help with. They could take your history before you actually see the GP, keep track of how you are doing to spot if things go wrong that you haven’t even noticed, give you medical advice, or even lifestyle advice in order to improve your long term health. We see AI as a chance to improve the efficiency of the healthcare system, to support our doctors and other medical staff, and to improve accessibility all around the world to healthcare services.”
But most systems still strive for accuracy. GYANT was originally trained on three million EHR physician diagnoses. That was then combined with the patient exchange interactions on the system since 2016.
Meanwhile, Buoy used 22,000 pieces of clinical literature to help the system understand the relationship between risk factors, diagnosis and symptoms. The system was also trained to factor in certain risks such as smoking. It also learns off of patient interactions.
In June Babylon made news when it announced that its technology outperformed new doctors in a simulation version of the UK’s Member of the Royal College of General Practitioners (MRCGP) exam. The company said that its technology recorded an 81 percent score on the recreated test, which was 9 percent higher than the average passing score for UK medical students.
“I would want to mention how much emphasis we at Babylon put on the medical validation of all the automated tech that we deploy in our app or in other products and services,” Hammerla wrote. “We have a lot of doctors on staff that continuously look at potential risks and hazards of our technology, and who help us design suitable procedures to mitigate any risk that they discover.”
AI and chatbot partnerships
Leaders in the triage chatbot space have started teaming up with major providers and health systems. Many of these deals feed the bot’s data and assist the health system in some way.
In August Buoy announced a new partnership with Boston Children’s Hospital. As part of the deal Buoy will be creating a version of its product for the hospital’s website and work with doctor’s from the hospital to get help evaluating its pediatric algorithm.
Meanwhile, in the UK Babylon has made strides with the NHS. In January of 2017 Babylon first teamed up with the NHS in a six month trial in north-central London, which covers the span of 1.2 million citizens. The company’s direct-to-consumer offering starts with an AI-powered chatbot. Triage via Babylon requires about 12 text messages and takes about a minute and a half. While users of the service can typically funnels patients into a video visit if necessary, in this pilot the chatbot advised users if they need to get in touch with their GP.
During the pilot the chatbot replaced NHS 111, a non-emergency phone hotlines staffed by call center workers who are in general not medical professionals.
Eventually the pair rolled out the NHS GP at Hand service, which lets patients see a practitioner through the app once they have switched from their current GP. According to the website, the service is available for people living or working within 40 minutes of one of their clinics. If needed a patient can see a physician in person. Currently there are five clinics all within Greater London.
Chatbot technology is not without controversy. Babylon in particular has had its share of criticism in the last year. After announcing that its triage chatbot outperformed new doctors in a simulation version of the MRCGP exam, the company faced major backlash.
"The potential of technology to support doctors to deliver the best possible patient care is fantastic, but at the end of the day, computers are computers, and GPs are highly-trained medical professionals: the two can't be compared and the former may support but will never replace the latter,” Martin Marshall, vice chair of the Royal College of General Practitioners, said in a statement. “No app or algorithm will be able to do what a GP does. Every day we deliver care to more than a million people across the UK, taking into account the physical, psychological and social factors that may be impacting on a patient's health; we consider the different health conditions a patient is living with, and medications they might be taking, when formulating a treatment plan. Much of what GPs do is based on a trusting relationship between a patient and a doctor, and research has shown GPs have a 'gut feeling' when they just know something is wrong with a patient.”
That ‘gut feeling’ that Marshall is referring to is a key argument point in healthcare and technology and has been studied alongside artificial intelligence.
In July, a study from MIT computer scientists suggested that artificial intelligence can’t provide the same quality of care as doctors. Researchers analyzed doctors' written notes on intensive care unit patients, the researchers found that the doctors' "gut feelings" about a particular patient's condition played a significant role in determining how many tests they ordered for the patient.
However, for the most part triage chatbots are directing people to doctors care rather than totally replacing that care. But some say that even giving advice can be problematic.
“It’s a balancing act, on the one hand, you don’t want to stifle innovation. On the other hand, these are so-called low-risk medical devices,” Michael Millenson, president of Health Quality Advisors LLC and adjunct associate professor at Northwestern University’s Feinberg School of Medicine said, while speaking specifically on dermatology diagnostics. “I think it’s a worrisome situation where you tell someone ‘this is only advice’ and implicitly there’s a little wink there. If people should see a doctor, you should be putting out a directory of dermatologists, instead of an app that lets me photograph my lesion.”
What’s next for chatbots?
On the developer side, Zuta over at GYNT sees the uses for triage chatbots expanding and sees the technology working its way into the standard of care.
“AI-based triage bots will be the new standard for accessing care in the near future. It will be the first step into the care journey for most people. Whether it's helping out with a medication refill or helpful homecare tips, bots will be able to efficiently triage and process a large number of every-day requests currently handled by over-worked physicians, nurses and front-office staff. This will allow providers to spend more quality time with their patients needing attention and care,” Zuta wrote. “Over time, they can personalize their recommendations for individual patients taking into account the patient's medical history. “
In terms of the business model, Le said he sees there being opportunity for future collaboration with the payer and employer. While the free model of Buoy aims to help folks figure out what is wrong with them and where to go — the company is now teaming up with employers so that patients can see what facilities are in their network and where to go.
“AI chatbots have been getting a lot of press recently and its interesting because we all get lumped in together,” Le said. “But if you start to dig one layer deeper the goal of each chatbot is really slightly different. Some have telemedicine tied into the end because they are a telemedicine appointment and the goal is to get someone to talk to a doctor and other programs are to help book appointments and the end provider is to get a service booked.
Others are to help navigate benefits and answering questions about what is a deductible and copay… Buoy is about how patients navigate the healthcare system.”
While the focus of these bots has been on countries with some of the most cutting edge medical care in the world, some developers are pointing out the possibilities for these systems in places that are difficult to reach.
“People in industrialised countries don’t usually appreciate that, according to the WHO, the majority of people in the world have practically no access to adequate healthcare at all. It is our mission to address this problem — how can we make healthcare accessible and affordable to people that normally would have to travel for a whole day to speak with someone about their health? Why not provide triage advice and other medical information to people via technology they have available?” Hammerla wrote. “This doesn’t have to be a smartphone. We are exploring other ways of accessing these services, for example via SMS messaging, or by simply calling the AI via the phone.”
But whether you are for or against the emerging technology, many in the field are saying the technology is inevitable therefore it’s best to work with the tools than against them.
“NHS is trying to get a good balance. As you know the Babylon app has taken a lot of criticism as a direct to consumer app,” Millenson said. “On the other hand as being used as a way to contact your doctor it’s perfectly appropriate. What the NHS has done I support is they’ve said, let’s curate these, knowing that they’re out there. I think that is a really good approach.”
Correction: Nils Hammerla's quotes were updated to accurate reflect he was the speaker.
Focus on Artificial Intelligence
In November, we take a deep dive into AI and machine learning.