zapp2photo - stock.adobe.com
Seeing through the hype: Where AI in healthcare shines
While AI in healthcare shines in clinical areas like radiology, it is currently more widely adopted in nonclinical areas, such as medication development by pharmaceutical companies.
Artificial intelligence has been touted for years as the future of providing extensive data analysis and useful insights to healthcare providers.
The technology has proved most valuable in radiology, with vast amounts of digitized, standardized data to work through, but AI, a broad term for technology that aims to mimic human intelligence, is emerging in other areas such as predicting patient deterioration.
"AI is really being tested everywhere," said Jeff Becker, senior analyst at Forrester. "Right now, there aren't many areas of the healthcare ecosystem -- payer, provider, pharma -- that aren't deploying AI somewhere to try and either improve outcomes, cut costs, or improve population health. All of those are seeing investments in AI."
Indeed, the U.S. Food and Drug Administration is looking to create a new regulatory framework for AI in medical devices, and the Centers for Medicare and Medicaid Services recently launched a competition encouraging innovators to demonstrate the predictive capabilities of AI tools.
But according to healthcare experts, the current adoption of AI in the clinical setting, such as in assisting providers during direct patient care and patient data analysis, is far slower than in nonclinical settings like medication development by pharmaceutical companies. Still, the technology is poised to make its mark.
AI in healthcare: Where it's most successful
AI is making its mark in healthcare the same way it has in consumer products. AI algorithms can be trained to perform sophisticated functions such as image analysis. Unlike identifying and learning from shopping patterns on the internet, the algorithms are used to detect anomalies in medical images.
In this use case, the AI screens hundreds of images a radiologist might look at per day and highlights areas a radiologist should take a closer look at.
"There are a number of players building medical image analysis algorithms," Becker said. "Essentially what they'll do is screen an image, looking for signs of diabetic retinopathy or any number of things. The AI is capable of highlighting something like a lung tumor and zeroing the radiologist in on that so they can process images more quickly and accurately."
Jeff Becker Senior analyst, Forrester
Robert Challen, M.D., researcher at the University of Exeter in the U.K., said AI is becoming more broadly adopted in radiology because the data sources used to train an algorithm are standardized. That means the data is in a common format across the healthcare industry.
Chilmark senior research analyst Alexander Lennox-Miller echoed Becker and Challen's statements about AI being deployed effectively in imaging evaluation. But he also pointed to areas like pathology for cancer detection as another successful use case.
For Becker, that's about the current extent of AI use in the clinical setting. Its most successful deployments are happening in multiple areas outside of direct clinical care.
Lennox-Miller pointed to pharmaceutical companies using AI for medication development, as well as healthcare organizations applying AI to the back office or to help with patient flow. He described use cases like these as "areas that are maybe not healthcare delivery but enable healthcare delivery."
Becker pointed to EHR-embedded virtual assistants as another growing area for AI in healthcare. The technology uses a component of AI to search EHR data, find relevant information and present it to the provider.
Another area that's showing signs of success are chatbots, which use natural language processing and generation. Becker said chatbot-infused chronic disease management platforms using AI algorithms are starting to generate value and pointed to call center augmentation in the form of a chatbot for insurance providers.
AI adoption poised to grow in clinical care
A nascent but emerging set of AI capabilities centers around predicting and forecasting worsening symptoms such as migraines, epilepsy, depression and suicide before they happen, Becker said.
"Kaiser Permanente and the VA are both using forecasting models to pick up on patients at risk of suicide," Becker said. "And then the patients that really are the most likely to act on those compulsions -- identifying them and changing their course of treatment based on that forecast model."
The most common use case of predicting worsening symptoms is sepsis, Becker said. Hospitals use AI tools to monitor lab values coming into the EHRs for signs of early onset sepsis. That application is gaining ground.
"You're seeing a lot of efforts to get in front of deterioration," Becker said.
Overall, Becker said analyzing medical images and forecasting worsening symptoms are two talked about AI applications in healthcare, but the reality is, most healthcare organizations are using AI, whether they realize it or not.
"If they're using M*Modal or Nuance for the doctors to do their voice transcriptions, they're using AI," Becker said.
For healthcare CIOs interested in adopting AI tools, Becker advised they focus on how AI algorithms improved outcomes in other organizations, as well as the problem they're trying to solve rather than the technology.
But AI isn't infiltrating all facets of healthcare just yet. In part two of this series, Becker and Lennox-Miller focus on what's hindering AI adoption in direct clinical care.