The "Learning" System: AI & Virtual Care

Machine learning is critical to effective and scalable virtual care; allowing clinicians to simultaneously improve outcomes and reduce the cost of care.

With the proliferation of sensors and wearables in the home setting, a new host of data is now available for clinicians. And yet, no human can feasibly and economically make sense of this “deluge” of data. Enter machine learning, which according to Nature, is already showing significant promise augmenting clinicians’ ability to treat Type II Diabetes, analyze skin lesions, and electrocardiograms. According to Accenture, machine learning will save $150 Billion a year in healthcare costs by 2026. 

And yet, today’s care model, as is perpetuated by telehealth providers, struggles to adapt and learn from each patient interaction as any learned knowledge that can benefit a population, is effectively lost when clinicians press “end call” after each session.

Machine learning will unlock clinicians' ability to deliver personalized care at population scale. 

Effectively “bridging” a capacity gap, machine learning is critical to understanding the “deluge” of data coming from multivariate sensors in the patient’s home.

Enabling clinicians to scale personalized care to thousands of patients, machine learning will not only allow clinicians to practice “top of license” it will also foster a “learning system” that gets smarter with each patient interaction. 

Underpinning this opportunity must be an emphasis on transparency and accountability into the drivers of recommendations coming from any machine learning system.

On any given day, a human with a network of wearable sensors in the home may provide over 100+ features on a patient with a chronic condition like heart failure; less than 10 of which a clinician can reasonably be expected to act on. 

Enter machine learning. 

Machine learning, when fully integrated within a virtual care system, can parse and organize complex multivariate data requisite to understanding the day-to-day health status of patients with complex chronic conditions. 

Augmenting the current standard of care, machine learning surfaces patients with acute risk of hospitalization that would otherwise be missed. For instance, machine learning can understand the risk of decline in patients with chronic conditions like heart failure and COPD. Myia’s own modeling continually surfaces patients at risk of hospitalization 10-12 days in advance. And this is just the beginning. 

Under existing care models, it is increasingly challenging for clinicians to offer personalized care plans to their roster of 10-20 patients a day. And given the ambition of many health systems to scale virtual care to potentially monitor thousands of patients, it will be near impossible for clinicians to approach anything remotely resembling “personalized.” 

Machine learning is the crucial tool in the arsenal to solve this challenge. Integrating machine learning, clinicians will be able to focus less on administrative busy work and more on personalized patient care; thus enabling them to practice “top of license.” Machine learning’s ability to filter and parse data will ensure that countless hours are not lost by poring over unnecessary data. 

In a world beset by clinician shortages, where simply adding more data and more personnel is not an option, machine learning can provide significant benefits to augment the skills of those on the ground. Namely: 

• Personalized Care: Timely and personalized follow-up between clinician and patient; optimized according to the patient’s condition;
• Personalized Therapy: Guiding clinicians to intelligently optimize Guideline Directed Medical Therapy (GDMT) to a patient’s conditions and goals over time; and 
• Personalized Attention: Knowing just the right time to bring a human clinician or even alternative resources like lifestyle coaching into the loop. 

And yet, machine learning has the potential to go even further by allowing providers to effectively “learn” from each interaction. Recording each patient’s baseline and continually augmenting its database with each successive interaction, the system becomes more skilled to understand longitudinal trends, gaps in care, and response to therapy.. This has a compounding “network effect”; the more population data is exposed, the better and more nuanced insights the system will be able to deliver.

Clinical apprehension towards “black box” machine learning applications are well placed. Clinicians fear that without some form of understanding of the reasoning behind the machine learning recommendations, they may be abandoning their professional and moral responsibilities. 

To counteract this, explainability and transparency must be paramount when integrating machine learning to any virtual care system. A virtual care system must easily explain the underlying inputs and drivers of risk assessments and recommendations for every patient interaction. This must be easily demonstrated in a clinician workspace that is understandable and gives everyone confidence in any system derived recommendation.

To see how Myia Health is building the operating system for data driven virtual care, please click here. 

This article is part of a content series from Myia Health. Click here to read more.
 

Myia Health is the operating system for virtual care. Myia’s platform ingests a wide range of real-world data from curated sensors and sources, transforming it through applied machine intelligence into actionable and objective clinical insights. Myia equips clinicians with the precise information they need to both optimize care and prevent unplanned medical events. For more information, please visit: Myia Health.