Artificial intelligence can power healthcare and empower patients
Founder and CEO of Ada Health, Daniel Nathrath, discusses AI in healthcare
Artificial intelligence (AI) in healthcare offers exciting opportunities for powering future healthcare, while empowering patients and clinicians. Daniel Nathrath, Ada Health CEO and co-founder, spoke with Dr. Ed Fitzgerald, KPMG's Global Healthcare Executive, about Ada Health and their technology at KPMG's recent Global Healthcare conference in Berlin.
Ada is an AI-powered personal health guide that helps people understand and manage their health. The original idea for our AI product was quite different: it was purely to create a tool for supporting doctors. But changing healthcare often comes from changing consumer demand, as well as innovative healthcare professionals. So we decided a couple of years ago to make our AI technology available to patients directly. At that time, the concept was a bit novel and a harder sell; now, it's easy. On average, 1 patient assessment is completed by Ada every 3 seconds!
A diagnostic assessment is a complex process of assimilating a constellation of symptoms with a knowledge of risk factors and prevalence. My co-founder Dr Martin Hirsch (the grandson of Nobel Laureate Werner Heisenberg - creator of the uncertainty principle) had been researching how human thinking worked for decades. Martin studied theoretical medicine and wanted to apply his findings, and so began to explore diagnosis and the use of technology to support human thinking and decision making. Martin's initial goal was to build a machine to mimic the thinking of `the perfect doctor', which could be used to support and assist them.
Doctors look at a presentation in terms of probabilities. Take appendicitis: there is no blood test, but there are a number of common symptoms, and it's clinically a bit of a `yes/no/grey area' situation. Doctors form their diagnosis using an imperceptible process in their head, which aggregates their experience of when those `typically appendicitis' symptoms and signs are more likely to cumulatively say yes or no.
Ada replicates this decision-making process. The core behind our product is a combination of constellatory thinking and pattern recognition. If an experienced doctor sees symptom A and B and C, they know that symptom C may mean something different in the presence of A and B than symptom C means on its own.
We now have what we think is the only fully developed clinical reasoning engine, based on decades of research into human thinking. We've developed a large medical knowledge base, covering over 1,500 conditions and thousands of symptoms and real-world cases, and have been refining our algorithms over the last seven years with more than five million completed assessments. Ada's depth of medical knowledge and accuracy are now unparalleled.
An experienced doctor sees certain things when they are assessing someone. So an important aspect for us was to have explainable AI. We didn't want it to be a black box. To gain doctors' trust, we have to show how we arrived at the conclusions derived from our software.
So we've tried to visualize (albeit to a less detailed extent in the patient-facing version) how strongly the presence of a symptom contributes to a machine recommendation that this may be this or that. We have experimented with different visualizations over time and there are different explanations provided for patient users and doctor users.
I was concerned that when patients started taking the Ada app to their doctor and showing them its suggestions, doctors could be irritated, but that hasn't yet been the reaction. Once they see what our technology can do and how well it works, most see the potential benefits. There is also an option for a patient to send their doctor a PDF of the Ada assessment. If a GP is using our system, the patient's app consultation data is already fully integrated, so their systems would have the full patient history.
So one way Ada can be very useful is in terms of saving time. Of the average 8 minutes for a GP appointment, they spend 2-4 minutes getting the patient's history. The potential efficiency gain of having undertaken this in advance of the consultation, with the relevant questions answered is huge.
After years of barely anyone (including our relatives!) understanding the depth of what we were doing, most of these days I get many requests from governments, health systems, payers, and chief digital officers who want to use our technology. Ada now looks like it's at an inflection point; it was the fastest growing medical app in Europe last year, and it's attracting huge and growing interest and awareness. This is obviously exciting for us, and we're thinking about how we take it to scale. Part of our challenge now is to prioritize.
From a market perspective, Ada is currently available in German, English, Portuguese and Spanish; Hindi and French are the next languages on the list. China is also showing a lot of interest, and is obviously a fascinating market: a single 1.4-billion-person market for data with lot of activity in AI. China doesn't have a significant primary care provision in their healthcare system, so there's potential for this approach as a solution to that. In a developed market like the US or UK, Ada can help patients and GPs have more efficient consultations. In China, there's a potential for Ada to digitally support that GP role as gatekeeper and guide.
There are many ways to turn something into revenue that some people argue is like having a doctor in your pocket 24/7. We're confident we can commercialize this at a large scale, and that our value will deliver shared benefits and savings to all stakeholders within the broader healthcare ecosystem. We anticipate some challenges around whether Ada creates extra demand by meeting a previously unmet need. But we are very confident that in practice, Ada reduces costs significantly while offering better outcomes for patients. We are starting to work with different stakeholders to make Ada more accessible and offer the highest range and quality of care.
In the UK, we tested putting Ada on a tablet computer into patients' hands while they were waiting to see a doctor at a large NHS clinic. Patients entered the reason for their visit, and we asked them after their consultation whether they would use and recommend Ada, and we got a Net Promoter Score of approval in the 90% range. We then asked them `if you'd accessed this at home, would you have come to see your GP?' and 15% of the sample replied that using Ada beforehand would have given them sufficient peace of mind to avoid the GP visit, due to the assessments' outcome including information for self-help and guidance towards next steps. Every doctor is under a lot of pressure, and by providing more detailed health assessments and integrating with medical health records, Ada can offer earlier relief to patients while optimizing clinicians' time.
Another large benefit Ada could bring to systems and payers would be to reduce instances where people go to the wrong clinician or end up with an inappropriately high level of care (like urgent care/emergency room). Ada can help triage patients to the right next step in the care journey: right care, right place, right time. Introducing Ada as the initial consultation option can save health systems a lot of money - and patients a lot of wasted time.
Furthermore, with the help of Ada, people can get to the root of their condition sooner. In particular, rare diseases, for which the average time to reach a diagnosis could be 5-7 years. It means so much to hear someone say, `it took 20 doctors and 10 years until I saw the right specialist last year. I wanted to test your system, and it took 5 minutes to arrive at my condition, not 20 doctors and 10 years'. We have nearly 100,000 ratings and we receive hundreds of these types of reviews and emails every day, and I take the time to read them. From a payers' perspective, those patients leave a huge cost trail, and from those individuals' viewpoints, think about the unnecessary worry, time and suffering.
Integrating Ada with health systems usually involves a phased approach when we agree to partner with a system or payer. Customizing is key, in order to direct patients to the most appropriate next step in their care journey. There is often legacy software including electronic health records (EHRs) we need to work with. Systems like those in the US are very interested because we can do this process relatively quickly.
In the next phase, they usually want Ada functionality embedded within their own app, to give patients a seamless experience. So we co-brand and embed our functionality, to give users a single sign-in experience.
The third phase of implementation is about a bi-directional exchange of data if patients wish to do that. At this stage, we can combine the legacy clinical EHR with Ada data, and at that point, patients can do home self-assessments and can share it with their doctors' system.
The fourth phase of implementation takes us to a prevention model, combining those data sources with census data. At that stage, you could potentially bring in a small home-based blood test monitor for long-term conditions (LTCs), to prevent worsening of conditions and perhaps spot others before they develop. Combining all these data sources and Ada make sense for the benefit of patients and health systems as it helps to saves cost.
Ensuring data security is a very important topic for a company like ours. We work with an industry standard renowned cloud provider, and we've just passed ISO certification, which shows that we fulfil industry standards for data security. In healthcare, data security is even more important than in banking.