AI Adoption in Healthcare: ‘You Need to Trust But Verify’ Says CIO

By Christopher Cheney for HealthLeadersMedia

“If you do not have the right clinical champion, I don’t care if it is the best AI in the world, it will be dead on arrival.”

KEY TAKEAWAYS

• Healthcare leaders such as CMOs should thoroughly understand what an AI tool is doing and make sure that an AI tool does what its designers claim it can do, a health system CIO says.
• When adopting new AI technology, health systems and hospitals should have a robust process to review adoption, including engineers and clinical champions.
• One of the emerging areas for AI in healthcare are tools that are patient-facing.

When it comes to adoption of AI tools, health systems and hospitals need to be cautiously optimistic, says Doug King, MBA, senior vice president and chief information officer at Northwestern Medicine.

There is an enormous amount of hype surrounding AI in healthcare. Health systems and hospitals are adopting AI tools from partners or developing their own AI tools internally.

“You should be optimistic but realistic,” King says of AI adoption. “There is a lot of hype around AI, so you need to understand what an AI tool is actually doing. You need to validate that an AI tool does what it claims to be doing. You need to trust but verify.”

Northwestern Medicine has adopted several AI tools, and the health system is open to using any AI technology that is clinically relevant and that adds value to patients or clinicians, King explained.

“We are primarily focused on two key areas when we are evaluating AI,” King says. “The first area is around early disease state detection, and the second area is around clinical efficiency.”

AI tools that can achieve early disease state detection provide value for patients, according to King.

“If we can identify diseases through algorithms, such as looking at images or looking at radiology notes,” King says, “it is better for the patient because we can identify diseases earlier and intervene with care to keep them healthier.”

AI tools that achieve clinical efficiency provide value for clinicians, according to King.

“We want to find AI that boosts provider efficiency because there are not enough doctors and not enough nurses,” King says, “so anything that we can do to make them more efficient so they can see more patients through AI is helpful.”

Northwestern Medicine has a robust process to evaluate AI tools that the health system might adopt, according to King.

“We have a team of engineers that understands AI,” King says. “They validate the technology. They understand how the algorithm works. If we are working with an outside partner, we have technical calls with the partner’s AI team to make sure that the algorithm works.”

Clinicians also play a crucial role in the adoption of AI tools at the health system, according to King.

When adopting AI tools, Northwestern Medicine looks for clinical champions who understand the technology, who want to own the initiative, and who can drive the technology with other clinicians throughout the entire health system, according to King.

“If you do not have the right clinical champion, I don’t care if it is the best AI in the world, it will be dead on arrival,” King says.

When a health system or hospital is working with an outside partner, safety needs to be a primary concern in the adoption of AI tools, says David Atashroo, MD, CMO of Qventus.

“You have got to make sure that you are working with a partner that has a track record of building AI models and doing it in a manner that is safe and secure,” Atashroo says. “That does not come overnight. It takes a lot of experience to do that well.”

Health systems and hospitals should also be able to determine a return on investment from AI tools. According to Atashroo, an AI tool should be designed to address a specific problem.

“At the end of every problem, there is a potential outcome that can generate a return on investment,” Atashroo says. “The only way you can validate and substantiate the benefit of an AI tool is insofar as it is deployed to solve an acute problem, where you can measure the impact on the back end and determine the financial return.”

Patient-facing AI tools

One of the emerging areas for AI in healthcare are tools that are patient-facing, according to King.

“Patient-facing AI is just starting to come out,” King says. “We will see it become more and more robust over the next few years. No. 1, AI will allow the patient to engage in the healthcare system the way they want to engage it.”

Patient-facing AI will allow health systems and hospitals to personalize the patient experience, according to King. For example, if a health system knows that a patient has a preference to go to a particular location, a patient-facing AI tool will be able to automatically serve up times for appointments at that location.

“If the patient wants to have a dermatology check, we will be able to offer particular locations and times for that appointment using a patient-facing AI tool,” King says. “It will allow us to personalize the patient experience as much as possible.”

Northwestern Medicine is piloting a patient-facing AI tool that helps clinicians respond to messages from patients, according to King.

“If you send your physician a note, whether you have a question about something in your chart or a question about your prescriptions, AI reads that and generates an automatic draft response for the physician to use,” King says. “That allows the physician on average to be about 30% more efficient as far as sending those notes out.”

The health system is planning to roll out another patient-facing AI tool by the end of the year, according to King.

A patient will be able to start a conversation with a chatbot symptom checker. The symptom checker will ask the patient a series of questions, then the responses will lead to an understanding of the patient’s clinical need, and the AI will point the patient to a physician who can address the clinical need. The AI will get the patient to the right place and look for appointments, so the patient can make an appointment without having to call an office.

“That is an entire interaction between the patient and the health system without any human interaction,” King says. “It goes from what the patient is experiencing, what the patient is feeling, then making an appointment that works for the patient. It is going to be sophisticated.”

AI tools are becoming more patient-facing because they are on-demand as soon as a patient needs access to them, according to Atashroo.

“As a patient, it can be frustrating when they need an answer to a question, but the clinician is not available or on call, so they have to wait,” Atashroo says. “AI tools can provide immediate access when a patient needs it versus when a clinician is available.”