Northwestern Medicine implements AI imaging workflow to improve patient care
Introducing Artificial Intelligence into an existing healthcare platform requires a considerable team effort, system-specific data, and patience to streamline the workflow and improve patient outcomes.
Northwestern Medicine spent two years implementing a system-wide AI tool built in-house, specifically for Northwestern’s 11 clinical centers. The tool is built into the radiology imaging process and can trigger an alert signaling the need for follow-up care. Specifically, the program flags findings that are sometimes noted by radiologists but go unnoticed or are not of use to the provider who originally ordered the imaging. Even within the EHR, physicians overlook these details, and it can lead to delayed follow-ups by patients, with potentially serious consequences down the line.
“This problem is a case study of preventable harm. There is a documented, but unrecognized finding that could have led to a meaningful intervention, but instead the patient’s disease develops unchecked,” said Dr. Mozziyar Etemadi, medical director of advanced technologies at Northwestern Medicine.
Etemadi notes that most physicians do not look at actual radiology images for results but the reports that the radiologist generates. And up to 5% of the time, that is an oversight. Combined with missed appointments, etc., it leads to a lack of follow-up by the patient in the recommended amount of time.
The AI tool used by Northwestern Medicine was built in-house using the system’s data because “our own providers and nurses know our data better than anyone else,” Etemadi said. He notes that the hardest part of the process was finding the right spot to integrate the alert into the workflow. And, of course, integrating the new tool into an old system, which first required some upgrades to Epic and cooperation with Google to harness the correct data.
Etemadi says that systems like the one created at Northwestern Medicine can ease the burdens of physicians. As an anesthesiologist, he understands that knowing a computer program can back up over-worked physicians and ease his worries that something serious could slip through the cracks.
So what about sharing the AI tool with other healthcare systems? While Etemadi hopes that other systems will learn from Northwestern Medicine, he’s not sure that this exact AI tool can be transferred as is. He notes that healthcare tech needs to ask where things can be more generalized for broader use and where centralizing a tool can introduce biases into the healthcare system. As he points out, no two patient populations will be identical, meaning what works for one may not work for another.
Etemadi notes that once an AI system is in place, it needs frequent monitoring to check that the tool is making a difference for providers and patients while not introducing new biases.
While Etemadi sees great potential in the future of AI in healthcare, he warns that the biggest challenge is “over hyping it too soon.” He fears that healthcare companies are rapidly introducing new technology and tools into the workflow without stopping to collect information on the outcomes. As an example, he also talks about the use of telemedicine, which rapidly grew during the pandemic. While there are great benefits, it’s worth noting that the digital platform has limitations as to when it can deliver the best results for patients in the long run.
Jacqueline Renfrow is a journalist with more than 20 years of experience reporting on and writing about the intersection of healthcare, education, and retail with technology. Living just outside of Washington, DC, she enjoys exploring all that the nation’s capital has to offer with her husband and three children in tow.