Real-Time Data Delivery in Remote Patient Monitoring
David Ebert, chief AI and data science officer at the University of Arizona, also uses the term “big-picture view” to describe what AI and wearables can do for RPM. The true power, he says, comes from the processing capabilities embedded in today’s wearables and implantable devices.
Several years ago, a patient with a pacemaker needed a purpose-built home monitor. Now, pacemakers have Bluetooth sensors that connect to smartphones, aggregate data and send notifications to a patient’s care team.
“We’re taking advantage of the capabilities that people are carrying around on a chip,” Ebert says. “We can do machine learning and predictive analytics on the device.”
There are two keys to making this work. One is continued focus on the efficiency of AI models. Data compression will save bandwidth, and the ability to “pull out the signals” will make a device’s output more valuable to clinicians who don’t have time to look at raw data.
“We don’t want AI models to drain the battery or take up a lot of processing time,” Ebert says. “We don’t want to have bandwidth challenges that exacerbate the digital divide.”
The other important step is integrating streams of data and insights from devices into electronic health record and clinical alerting systems. Otherwise, he notes, clinics will need additional equipment and the resources necessary to set it up.
DISCOVER: Here are the wearable technology trends shaping healthcare.
How to Implement and Scale RPM With AI and Wearables
Mahajan says that ease of integration is important. “The solutions that tend to be effective and adopted as seamlessly as possible don’t create unnecessary work for clinicians.”
Getting this right may require upgraded data ingestion pipelines that can accommodate high-frequency data streams, Mahajan notes, along with tools that normalize data as it’s aggregated. “Organizations have to shift from systems built for episodes to systems built for continuous data,” he says.
Ebert says another consideration is using devices that have evolved from application programming interfaces to agentic AI interfaces. That way, devices can be deployed, monitored and updated using software instead of specialized hardware, which comes with an upfront cost and need for specialized skills that pose a barrier to adoption. “That’s a game changer for rural hospitals,” he says.
Another common obstacle, says Mahajan, is the single-use predictive model or clinical decision support tool: “Health systems aren’t willing to take on 100 different tools. They’re looking for platforms or systems.”
Of course, there’s also the concern that AI models will replace clinicians. That’s not an issue for Dr. Sairam Parthasarathy, director of the Center for Sleep and Circadian Sciences at the University of Arizona.
Licensed providers are few and far between, he says, and “there are so many people who need our help. People shouldn’t have to get sick before we give them health advice,” and data from wearables and insights from AI models can ensure that won’t happen.
