The promise of Artificial Intelligence in health IT
Artificial Intelligence (AI) has been around a long time, but it is a newer concept within healthcare. AI holds a lot of promise, particularly in the areas of population health management, healthcare access and quality. At Becker’s Hospital Review Health IT + Clinical Leadership 2018 event, I served on a panel that talked about the promise and possibilities of AI.
What is AI?
AI and machine learning are hot topics in healthcare that are often used interchangeably, but they mean different things. AI is about making our technology “smarter,” so that it uses curated knowledge to automate and improve function. Machine learning is more focused on coding technology to process information the way humans do, taking in a vast amount of data and “learning” from it to make predictions.
When AI is working correctly, its algorithms are working for you in the background. For example, Facebook gives us “People You May Know,” and Amazon presents products “You Might Also Like” (and we often do). Many clinicians would love to have that personalized experience when they open their electronic health record (EHR), tailored to meet their specialty-specific workflows and documentation styles.
7 tips for success with AI in health IT
Several keen insights came from my fellow panelists, Dr. Lily Henson, Chief Medical Officer at Piedmont Henry Hospital (Stockbridge, Georgia) and Dr. Jeff Hoffman, Chief Medical Information officer at Nationwide Children’s Hospital (Columbus, Ohio). Here are some of the highlights from our panel discussion:
1) Make insights relevant to the caregiver
Cool algorithms are great, but they must be something the end user wants and can use. If the insights from AI are not embedded in the technology they’re already using, it’s not going to be successful.
2) Be prepared to act on findings
Dr. Hoffman shared an example of an algorithm that can help predict which asthma patients are at the greatest risk for an emergency room visit within the next six months. Unless there is an intervention program in place to help address this prediction, the information is not impactful.
3) Look for proactive applications in population health management
There is a vast number of opportunities to use AI to improve population health. Dr. Henson shared examples about how the next level of intelligence has helped her organization be more proactive in more quickly identifying sepsis and protecting against surgical infections.
4) Build trust with clinicians
Adoption and acceptance of these technologies is still an issue. Dr. Hoffman referred to AI as “a third type of evidence” that are not as familiar to clinicians as clinical trials, medical journals or other traditional knowledge bases. Exposing the “how” and “why” for AI algorithms to the clinicians who will use them will help build trust.
5) Understand causal relationships vs. associations
For AI to be a trusted source of information, data scientists must draw correct inferences from the data. Dr. Hoffman shared an example of a predictive model to determine which children are at greatest risk for cavities before age 2. The highest predictor surprised clinicians: if the child had been referred for a blood test for lead, but never received the test, these children were at greatest risk for cavities. Obviously, not getting a blood test for lead does not cause cavities, but it is a proxy to identify children living in high-risk communities with a pattern of poor compliance. The data can reveal unexpected insights, but we must be careful when we apply them.
6) Improve access to care for people who need it most
Equity and access to care are fundamentally important, especially in today’s resource-constrained healthcare environment. Physicians are overworked, and AI can ease the burden by helping stratify patient populations for more targeted, appropriate levels of care. Dr. Henson noted that her daughter, who is in her 20s, needs different resources and care than her mother, who is in her 90s. Technology is a great way to “divvy up” the resources appropriate for each and every patient.
7) Try it
Dr. Hoffman concluded with this simple piece of advice: Try it. Many organizations don’t fully understand AI and may not have the resources for it. But in five to ten years, AI will be the norm. Those who haven’t started incorporating it will be behind the curve.
Ultimately, insight from all of this data is only worthwhile if we deliver it – in a consumable way – to the point of care. Physicians are taking on increasing amounts of risk, and AI can deliver the patient-level predictions they will need to personalize care.