Three Steps To Get Closer To AI-Driven Healthcare

Featured in the World Economic Forum Agenda

By: Tom Robinson and Aditya Lingampally
This article first appeared in the World Economic Forum Agenda on December 2, 2020.

The coronavirus pandemic is accelerating the adoption of AI-driven primary care chatbots. In the United States and Europe, healthcare networks, hospitals, insurers, and startups have rushed out these “doc-bots” to help millions of people diagnose on-demand for COVID-19 and receive information on next steps.

Since May, for example, a doc-bot offered by Sutter Health, one of the United States’ largest health networks, has identified whether more than 30,000 people have COVID-19, referring 11% of them to a video consultation with a physician.

The Centers for Medicare and Medicaid Services (CMS) has “Clara,” a bot that assesses whether people should self-isolate at home or go see a doctor. Across the Atlantic, 75,000 London residents have selected an AI-powered chatbot offered by the National Health Service (NHS) as their “primary care” provider.

The coronavirus-related surge in popularity of these doc-bots signals that a future of affordable and convenient AI-driven primary care may be closer than previously predicted.

Doc-bots have the technical ability to assess patients 24-7 and recommend whether they should see a doctor on video or in person. Bots can even perform differential diagnoses and provide patients with different possibilities along with the probability of each disease.

Soon, AI-driven doc-bots will likely be able to determine if patients need lab tests, x-rays, follow-up examinations, and medications. One AI system recently developed by researchers at Google and the University of California anticipated a physician’s prescriptions with 75% accuracy.

There are still some challenges, however, and for AI-driven doc-bots to succeed in bringing significant wins for primary healthcare, these three steps need to happen:

1. First, there is a need for more concrete evidence that doc-bots consistently give high quality diagnoses to patients across a broad range of medical conditions so that consumers and other industry stakeholders feel as comfortable with them as with physicians whose white coats signal trust.

2. Second, they should be integrated into the rest of the healthcare system to ensure that care is delivered from beginning to end seamlessly.

3. Finally, the healthcare industry needs to reimburse doc-bots for their services so they will be attractive to health systems.

Trust me, I’m a doc-bot

To win over both health systems and consumers, doc-bots need the equivalent of the certifications that physicians procure over years of schooling to demonstrate their expertise.

In our advisory work, healthcare providers still doubt the accuracy of doc-bots’ algorithms, questioning if they take into account all of the variables that a human physician considers. There are also concerns that AI-driven apps may be inaccurate across ethnicities given the training data set is not representative of the diversity of the population and it reflects racial biases exhibited by human doctors.

Doc-bot creators have started to have their algorithms peer-reviewed. Babylon, for example, has published more than 20 peer-reviewed papers that demonstrate its bot can reliably interpret symptoms from what patients tell them. Competitor Ada has participated in dozens of clinical studies in which patients have rated their AI-driven chatbot highly.

Doc-bot startups are beginning to share their degree of accuracy for different types of diagnoses in their apps and are involved in initiatives to establish industry guidelines and reporting standards. Ada is working with the World Health Organization, the International Telecommunication Unit, the Food and Drug Administration, and The Lancet medical journal to establish benchmarks for AI-driven medical symptom checkers as part of an initiative called Artificial Intelligence for Health.

Finally, like AI-driven smart speakers, consumers will likely trust doc-bots more as they become better at displaying empathy. One bot developed by San Francisco-based startup GYANT found patients were more engaged because their bot could display “algorithmic empathy” across a broad range of medical conditions. If a patient mentions they have a serious disease like cancer, the bot knows to answer differently than with a conversational “that’s too bad.”

Integrated part of the healthcare journey

To become ubiquitous, doc-bots will need to become true practitioners of primary care, by taking on more of human doctors’ duties. Physicians monitor patients’ health histories and risk factors and address health concerns. They direct care to specialists via referrals, order and perform routine exams, interpret test results, write prescriptions, and generally quarterback care across all life stages. Doc-bots are not there yet.

Given their limitations, doc-bot treatment recommendations need to be forwarded to human physicians for review and implementation. For example, after answering a few short questions, 98point6’s doc-bot connects patients with a live physician who confirms the bot’s diagnosis and treatment, and orders the required prescriptions and labs. Babylon has its bot hand off treatment to human doctors via in-app video visits.

Health systems we work with find doc-bots help to ensure patients receive physical treatment as efficiently as possible. For example, Ada’s bot not only evaluates patients’ conditions for Sutter Health, but also sends them to the most appropriate venue for treatment.

Using geolocation services, the doc-bot shows patients the closest facilities and helps them book appointments with human doctors. Behind the scenes, the bot shares important data with Sutter physicians such that the entire encounter is seamless for the patient from start to finish.

A Sutter patient can ask for advice from a chatbot on the Sutter web site, come into a walk-in clinic without an appointment, have a telehealth visit, or be seen in a traditional doctor’s office – and, while the experiences may feel different, they’ll all be powered by the same underlying health record.

Doc-bot fee for service

A big reason more health systems do not offer primary care chatbots is that insurers don’t reimburse providers for doc-bot services. Even if doc-bots provide affordable, convenient, and effective advice, primary care providers prefer an in-person model to hold up their revenues.

Telehealth utilization soared after CMS expanded telehealth reimbursement during the pandemic. Doc-bots might see a similar jump once health systems are financially incentivized to offer software-based diagnoses.

“Value-based” systems, like Kaiser Permanente, should be pre-disposed to primary care chatbot adoption because if someone is treated more efficiently, the organizations profit.

But recent estimates suggest only 34% of US healthcare spend is tied to delivering value. For the rest of the fee-for-service system, health networks and doc-bot creators need to agree on a rate and allow the doc-bots to bill for the value they provide, whether for an appropriate referral or telling a patient to take Advil for a headache.

GPS for healthcare

Once these challenges are overcome, doc-bots will become even more common both as tools that support physicians and potentially primary care providers on their own, especially in places where there’s a shortage of medical services.

Doc-bots have some unassailable advantages over their human brethren: they never tire, they continuously learn, improve, adapt, and they are economically compelling, with every additional unit of care they deliver being close to free.

Once they become more trusted, integrated, and valued, doc-bots could become as ever-present as the Global Positioning Systems (GPS) in our cars and as integrated as the personal assistants in our pockets.