In this special guest feature, Jack Kreindler of Sentrian discusses how remote patient intelligence (RPI) uses sophisticated machine intelligence to emulate the same decision process and pattern recognition in a doctor’s head – only faster, and on a larger scale. Jack Kreindler, MD is Chairman, Chief Medical Officer and Co-Founder of Sentrian. Also the founder of London’s Center for Health and Human Performance, Dr. Kreindler lectures internationally on the future of medicine and is a guest expert presenter for CNBC, Sky Sport, BT Sport and the BBC.
I’m a ski-mountaineer and a physician specializing in high altitude medicine. For years, I’ve used medical-grade biosensor data to help improve my own performance and that of other extreme sports athletes at my institute, CHHP, the Center for Health and Human Performance, in London. A few years ago, just before the ‘wearables’ buzzword started buzzing, I saw that advances in the consumer and medical grade biosensors would lead a revolution in the remote monitoring and care of people who’ve never had the chance to visit a mountaintop – the millions worldwide who suffer from chronic illness.
Of course, we have been hearing about the promise of remote patient monitoring for nearly a decade. Yet the technology has so far failed to be widely adopted. That may be about to change. Ever cheaper and better biosensors, ubiquitous smartphones and connected devices, and breakthroughs in machine learning and big-data analytics now hold the key to effective, economic management of patients at home.
My hope is that we can use these technologies to eliminate all avoidable hospitalization and in the process solve the trillion-dollar problem of chronic disease, which is crippling our economies.
A grand challenge, indeed. But one that’s grounded in the reality of some pretty astonishing advances.
Let’s recap why it has taken us so long to get to this point.
Although biosensor devices have been around for decades, they were simply too costly for widespread adoption. With the popularity of Fitbit and other wearable fitness monitors, we’ve seen orders-of-magnitude improvements in the affordability and precision of such devices. Tiny stick-on cardiac monitors that once cost tens of thousands of dollars, now sell for tens of dollars and deliver far broader types of clinical grade data, wirelessly.
Another challenge that has slowed the adoption of remote monitoring: Many medical professionals have been slow to embrace device-generated data, even from FDA-approved devices prescribed by a physician. That’s because the vast streams of data these devices generate can quickly overwhelm clinicians untrained in making sense of personalized physiological data streams. They have no way to siphon out clear signals that a patient is headed for trouble and needs help now to avoid an ER visit. Even if they could use biosensors to predict patient deterioration, clinicians would still be unable to affordably scale their use across entire patient panels, let alone whole populations of the chronically ill.
So the vital questions are, how do we make sense of these new, more efficient and vastly cheaper sensor data? How do we transform enormous streams of data from individual patients and also whole populations? How do we, in other words, make it easy for clinicians to affordably monitor thousands of patients at a time, be alerted to deteriorating health, and intervene early to prevent avoidable hospitalization?
The answer may lie in a new approach to analytics that enables clinicians to tame the storm of biosensor information. This approach “tunes” the data to alert clinicians only to the most critical patient information, thereby circumventing the showstoppers of false alarms, alarm fatigue and missed issues. It also has the potential to become truly sensitive to the needs of its users, patients, professionals, loved ones and caregivers.
Here’s how it works. Using a hybrid approach that combines natural language authoring with artificial intelligence – an approach we’ve coined “Human Augmented Machine Learning” – clinicians can build the clinical rules or models that trigger alerts, without needing IT professionals to write computer code. These models analyze any number of data in real time with actions triggered when an anomaly occurs. Unlike the simplistic rules of current monitoring technologies – for instance, ‘if blood pressure rises above 200, call an ambulance’ – these rules are personalized to every patient from the start.
For example, a clinician can tell the system to “alert a care manager when their patient’s blood pressure rises more than 5 percent, over 3 consecutive reliable recordings, above each individual’s baseline levels, from a moving average of the last 5 days.” Or perhaps we would want the system to send an SMS (first), email (second) and voice message (third) to the nearest, most responsive loved one, asking them to check up on how their relative is doing when their normal cycle of physical activity starts to fall significantly. Only the patients who are most critical or deteriorating most rapidly are triaged as a priority. Patient’s location and proximity to primary care is known to optimize home visit logistics.
Adding to the system’s agility, it continually learns about which rules and interventions worked best for which patient. This is accomplished using another novel concept called “Adaptive Physiological Modeling” that lets the system learn from patient data and case manager feedback, thus improving its accuracy and reducing false alerts. For instance, if a case manager reports that an alert was a false positive, the system uses machine learning techniques to immediately recommend adaptations to the physiological model to correct the problem, again in natural language. Clinicians can use the system to test the new rules on a randomized set of candidate patients with controls, and once the evidence solidifies, apply the new Adaptive Physiological Models to the respective patients.
This approach marks a significant paradigm shift in healthcare, potentially enabling the limited number of disease management professionals across the country to easily monitor, and manage, millions of patients at a time.
Together, human augmented machine learning and adaptive physiological modeling promise to overcome many of the key barriers to wide adoption of remote monitoring in healthcare, turning monitoring into effective management. The melding of pioneering analytics and exponentially improving remote biosensor technology may be our best chance for avoiding tens of millions of needless hospital bed-days every year, saving hundreds of billions of dollars, and keeping the sickest people in our society healthier and happier in their own homes.
The technology is being tested now in controlled studies with thousands of COPD, CHF, cancer and frail patients. We have a long climb before we summit this grand challenge, but thankfully we’re getting closer every day.
Sign up for the free insideAI News newsletter.