AI tool offers cure for scattered medical data

A patient in the ER, ICU, and other care environments is often connected to monitoring equipment such as cardiac monitors or ventilators, which capture a range of medical data points: heart rate, respiratory rate, oxygen saturation levels, body temperature, and more. Studying these numbers over time can yield vital information about the body’s physiological patterns indicating imminent deterioration such as cardiac arrests, respiratory depression, and stroke.

Unfortunately, in most cases, medical professionals are not able to leverage such data because most information from medical devices is transient. Very little of the bedside device data makes its way to the EHR, and the rest is deleted once a patient is taken off of the monitor. When a patient is transferred to a different unit, there is no easy way for members of the care team to relay historical data to the new care team. While nurses or physicians might record notes of events, detailed physiological information is lacking. In effect, medical professionals can only tell that an event happened; they can’t unravel the why’s or how’s.

Integrating data across the patient journey

One of the primary goals of Medical Informatics Corp (MIC) is to aggregate this patient data to both enable remote access and give early warnings to clinicians about impending events. This ambitious goal ran into early issues when MIC’s founder, Emma Fauss, PhD, discovered that existing data collection systems each use proprietary formats that make them unwieldy to integrate. Data gathering is not done en masse. “There was also no way to take an algorithm and operationalize it into an existing hospital infrastructure, so you can actually deploy it at scale across thousands of beds with standardized workflows,” Fauss said.

MIC discovered they had to solve an end-to-end problem — from (clean) data acquisition to software-based remote monitoring to the development of scalable AI that can then be deployed back into a standardized workflow. This is what the Sickbay clinical platform, MIC’s scalable, FDA-cleared Real Time Clinical Surveillance-as-a-Service (RTCS) Solution does: it integrates medical device data, unlocks it for flexible virtual care across all service lines, and uses AI (both supervised and unsupervised) to deliver insights to clinicians.

“We specifically focus on time series device data, including the waveforms and physiological patterns and data coming from the patient. It’s a space that’s entirely untapped,” Fauss said, “It’s sort of like a blue ocean area for exploration and development.”

AI models and monitoring use cases

MIC’s AI algorithms train on patient data, while abiding by the protocols set by the Health Insurance Portability and Accountability Act (HIPAA). While MIC can work with both supervised and unsupervised models, algorithms predicting health events such as cardiac events rely on supervised models. Working at the Texas Children’s Hospital with physicians from Baylor College of Medicine, MIC helped develop an analytic that can predict cardiac or respiratory arrest in children with single-ventricle hearts one to two hours in advance.

MIC is committed to developing patient-specific AI and providing tool sets for hospitals and enabling health care systems to create their own — understanding physiological patterns associated with specific conditions and then developing algorithms that can be used to monitor patients with qualifying risk factors.

The AI is good at detecting rapid deterioration of key physiological indicators and also filling in the gaps when patients are being monitored for days or weeks. “In units, if there’s a shift change every 12 hours, if the deterioration is slow enough over many days, you may not see that it is happening because your time watching the patient is too short. That’s just the nature of shift work,” Fauss said. AI helps fill in those gaps and delivers risk score calculators for patients based on integrated device data.

The Sickbay clinical platform enables not just single patient monitoring, but also remote monitoring for multiple patients at scale. “It could mean that you have a virtual command center with virtual monitors that are monitoring analytics being run on patients. You don’t have to be bedside for routine monitoring,” Fauss pointed out. It’s an additional protection layer that clinicians value. Houston Methodist, for example, launched a virtual intensive care unit in 2020 that allows monitoring of all ICU patients remotely. MIC’s algorithms — the hospital runs close to 20 different ones — and data visualization enable the medical facility to track vitals carefully and be notified of problematic events well before they happen. The AI augments decision-making and helps care teams intervene faster when needed.

MIC’s Sickbay integrates with many medical devices and provides a consistent set of data for analysis. It’s a complex problem given the variety of medical devices, tech stacks, and methods for saving electronic medical records. Fauss hopes that the Sickbay orchestration layer makes access to AI applications that much easier and faster. “If we can break down the barriers to adoption of this data-driven, patient-centered technology, we can move the standard of care forward,” Fauss said.