Artificial Intelligence Healthcare

Artificial Intelligence is on Track to Improve Medicine

Image via Unsplash

Artificial intelligence is set to transform healthcare. AI companies are receiving large amounts of venture capital funding for tools that help manage a range of healthcare challenges, including directing patients to proper treatments, helping doctors diagnose and cure various diseases, and enabling a better understanding of Covid-19.

“Healthcare organizations have plugged AI and other tech tools into existing workflows, focusing on automation and execution,” according to Accenture’s Digital Health Technology Vision 2020 report. AI systems “are powering chatbots that help healthcare providers screen and triage patients,” the report says. It also reports that AI is “enabling the rapid reconfiguration of supply chains impacted by Covid-19.”

These sorts of developments are emerging not just in the US, but also around the world in low- and middle-income countries. In such places, says the New England Journal of Medicine, AI “has the potential revolutionize health and healthcare by addressing the large knowledge and judgment gaps that make care delivery poor.”

AI has multiple uses in medicine, says Dr. Kaveh Safavi, senior managing director of global health consulting at Accenture Health. One is the capability to do core analytics that help researchers study a wide range of diseases, with Covid-19 being a very recent success. “Very advanced forms of analytics led researchers through the sequencing of the virus’ gene, and the creation and formulation of vaccines,” Safavi says.

AI is also widely used to derive meaning from unstructured documents. “Something like 80 percent of the information stored as clinical data is unstructured–meaning handwritten, PDF, etc.,” says Safavi. “Healthcare systems spend an enormous amount of time reading these documents to get structured concepts upon which they can make a decision about something. This includes research, where AI plays a significant role.”

One of the biggest criticisms of electronic medical records is that the time a doctor spends typing while listening to a patient reduces productivity and creates a distraction from the dialogue, says Safavi. In a medical office, AI-based ambient listening—similar to what consumer devices such as Amazon’s Alexa do—can now capture information a doctor would otherwise type into an electronic health record. It can offer a diagnosis or propose tests, streamlining a process that usually starts on a piece of paper and that historically went through several steps before being entered into a computer.

Humana deploys IBM’s Watson AI system to help its interactive voice response system address account queries from administrative staff at healthcare providers about plan benefits and eligibility. Previously IBM transferred them to human agents, which costs far more than handling them electronically. Most of the calls are related to routine, specific issues that have well-defined answers, says a case study on IBM’s website. The solution “relies on AI to understand the intent of a provider’s call, verify they are permitted to access the system and member information, and determine how best to provide the information requested.”

Privately held Buoy Health offers a similar service, targeted directly at patients but without the voice element. Buoy offers what it calls triage—the first step in helping patients find medical care. The company recently announced a $37.5 million Series C funding round led by Cigna Ventures and Humana.

“There is a lot of work to make telemedicine the new front door to healthcare,” says Buoy CEO Andrew Le. “But what that ignores is that the sidewalk to the door is Google. 72 percent of Americans start their search for medical care there. During this self-triage period, they have to navigate the system like a health insurance expert and they end up making haphazard decisions. Telemedicine is great to scale a doctor’s reach, but having them use it for triage takes them away from more important work. Our system chats with you like a clinician.” Buoy hires doctors to write educational content that augments SEO efforts to draw people to its website and away from potentially less-reliable sources of information.

Buoy’s system goes beyond traditional decision trees, which continually bifurcate. For example, “If A, go here, if B, go there. Then answer the next question.” Such systems can’t possibly account for all medical situations, says Le. With the Buoy Assistant, a user begins by entering a symptom. After a back-and-forth of several minutes, the assistant has narrowed down a range of possibilities and can help the patient take the next step towards care. According to Buoy’s website, “It’s a more dynamic map, switching from one piece of information to another.”

Because electronic health records can’t provide the kind of in-depth data that such a system would need to succeed, Buoy turned to primary literature and medical textbooks, and built a statistical map for the AI based on the thousands of questions a clinician might ask. The system is updated in real time, and symptoms are re-ranked accordingly. Every transaction makes the system better. “Being able to reason dynamically, based on real evidence-based medicine is an immediate improvement over the status quo,” says a video on the company website.

The Allen Institute for Artificial Intelligence (AI2), launched by the late Paul Allen, co-founder of Microsoft, works closely with physicians and medical experts to drive its research work. Among the organization’s current projects is Core-19, a “freely available resource of scientific articles about Covid-19 to accelerate new research insights.” Core-19 seeks to make scientific papers on Covid available to doctors and computing experts so that they can build tools to help clinicians and experts discover information.

“The number of publications related to Covid is enormous. There are several hundred published each day, as well as clinical trials and other tools and mechanisms put in place to understand the pandemic,” says Lucy Lu Wang, a researcher at AI2. “It’s overwhelming.”

Despite the fact that scientific papers are more widely accessible on the internet than ever before, they are not necessarily more discoverable, says Wang. That’s because the content, the substance of each paper, remains locked up—there is no easy way to retrieve it short of someone manually reading them. AI helps extract what is truly relevant for a medical situation.

Another area in which AI shows great potential, says Wang, is the ability to assist in evidence-based medicine, or to diagnose based on the most recent and accurate information available. “AI can integrate information from a clinic about an active patient and link it to a massive repository of information, hospitals, and local guidelines in order to arrive at the best treatment,” she explains. The final treatment is almost always interpreted and ordered by a doctor, whose decision-making is speeded by the assistance from AI. But there remains much work to be done, Wang says: “Getting this right is an enormous challenge—all of these things are dependent on the availability of data and whether it’s trustworthy.”

Accenture’s Safavi agrees about the challenges. “The potential for clinical use of AI is strong, but near-term gains are likely to be non-clinical. We often think of AI as substituting for humans. But right now, it’s better as a way to augment what humans are already doing.”

Leave a Reply

Your email address will not be published. Required fields are marked *