Changing the paradigm: Moving from an episodic to a digital and continuous health care system

As a society, we have rapidly transitioned into a world of digital know-how. Every day digital technologies have the opportunity to revolutionize our lives. Today, nearly every smartphone or smartwatch has the capacity to produce 1 million measurements per day for every person on earth, but most of this data is not used to its full capacity. Harnessing this technology in health care has a high potential for impact in the early detection of age-related disease. Igor Dzhekiev and his co-founders at Insubiq Inc believe that digital data streams from sensors using smart devices can be a new source of information that can turn our gadgets into invisible health sentinels.

Insubiq Inc has domain expertise in digital biomarkers, which are consumer-generated physiological and behavioral data points collected through digital devices, and AI-powered algorithms. They apply these technologies to various therapeutic areas, and recently won a Healthy Longevity Catalyst Award from the National Academy of Medicine (NAM) for their Digital Biomarkers Platform and smartwatch app, which detects early signs of Parkinson’s disease.

The app is able to find the unseen abnormalities in our daily and long-term health through continuous biomarker tracking. It works in the background and monitors vital signs, analyses data to alert users without distracting people from their routine lifestyles. This is especially critical for the senior audience, for whom using and interacting with new gadgets or apps can be challenging.

“One symptom of Parkinson’s disease is uncontrollable tremors, most common is the hand tremor. Usually, it appears in the late stages of the disease when treatment is not very efficient,” said Dzhekiev.

Detecting the biochemical changes that trigger Parkinson’s using a consumer device is not easy at the onset of this disease. To work with available technology, the team hypothesized that they could identify the initially invisible micropatterns of abnormalities using high-frequency sensors inside consumers’ smartwatches and via the AI-powered app.

“The tricky part of this disease is that the symptoms are sporadic, and it is unknown when they might occur. It complicates laboratory studies and limits the detection of new findings. To work around the uncertainty, we hypothesized that training AI algorithms over long-term periods, routinely would solve this obstacle,” Dzhekiev continued.

The project team gained support from a well-known neurological clinic and had patients opt-in to share digital data to train their algorithm. Patients wore the smartwatch for several days under usual lifestyle conditions. After assembling about 400,000,000 digital segments, the Insubig team trained their neural network to classify Parkinson’s disease signatures.

By performing a basic test validation the team was able determine how well the algorithm could recognize signs of Parkinson’s disease in patients with distinct hand tremors. As tremors are a pronounced parameter and well recognized by the sensors, the task was relatively uncomplicated and had high recognition accuracy.

In subsequent trials, the team complicated the task for the algorithm by excluding the data on tremors from the analysis and the algorithm was still able to recognize the signs of the disease with a high accuracy. Typically, similar experiments with several patients led to similar outcomes, which led the team to believe that the algorithm can detect some other patterns not related to tremors. 

They tested their theory on another distinctive, yet more challenging sign of Parkinson’s disease, a change in the patient’s gait, which becomes stiffer due to changes in body balance. Though this stiffening is not difficult to determine visually, it was not obvious what conclusion the algorithm would reach, as a change in motion is a complex change in body motility without an obvious parameter to analyze, such as a tremor with a known frequency.

To determine the ability that the algorithm could be successful, the team challenged the algorithm to learn how to classify the gait of a healthy patient and a patient diagnosed with Parkinson’s disease. This time, the algorithm had to interpret complex pictures of movements through a smartwatch. Once again the team was able to achieve a high quality of recognition. The team’s next step is to determine how the algorithm classifies the parameters to successfully identify changes in gait.

“We hope that AI algorithms will someday be able to recognize disease symptoms at a level comparable to human health care professionals. For example, some neurologists can guess the signs of Parkinson’s disease from a gait. Some pulmonologists can suggest the type of respiratory disease from the sound of a cough,” said Dzhekiev when asked about the goals of the project.

Digital Biomarkers and AI open up new frontiers for all of us

This opportunity provided a unique confirmation of the potential of digital biomarker technology. Though the team had to suspend active patient trials due to lockdown, they were able to attempt further application of their algorithms for respiratory diseases and created the Cough Tracker App.

The Cough Tracker App, which is the first in its class, automatically detects cough and classifies its parameters (dry/wet, frequency, severity, etc.) via the user’s smartphone. The algorithm works in the background without interrupting the person from his/her activities. Classifications of cough are very well-known symptoms for many illnesses, but it was previously impossible to monitor it 24/7. Most commonly, respiratory professionals ask the patient about the cough during the appointment, although often this data is subjective and cannot be relied upon.  It is especially crucial to investigate nocturnal cough, which is difficult to control while patients are asleep. To monitor cough at night, the team developed a specific version of the Cough Tracker App for asthma with an advanced algorithm that can predict asthmatic crises with an accuracy of 94%. Hundreds of people from all over the world have already installed the Cough Tracker App, which is available on the Apple Store.

The pandemic and popularity of remote telehealth indicates that the time for transformation to a technology-embedded, data-first approach to health care has come, as we are all equipped with gadgets that are full of sensors. Some experts believe that the Remote Patient Monitoring market is five years behind Telehealth and will soon repeat the success that Telehealth apps have seen.

When asked about the future integration of this technology into the health care system, Dzhekiev shared more about the trends that shaping the future of care. First is the explosive popularity of telehealth around the world, which increased by nearly 50 percent in the U.S., as well as the widespread use of AI algorithms in health care settings. Also in the U.S., there have been new reimbursement codes added that can offer more remote patient monitoring technologies as part of care. The Food and Drug Administration has even made efforts to regulate these digital technologies for more advanced patient care.

Seeing the trends and need for advancement in this field, Dzhekiev said, “We hope that this pandemic will give a significant boost to remote medicine and that digital biomarker technology can be a significant part of this progress and Healthcare transformation especially for the elderly.”

This article is part of a series of profiles on the 2020 winners of the National Academy of Medicine’s Healthy Longevity Catalyst Awards — a component of the Healthy Longevity Global Competition, a multiyear, multimillion-dollar international competition seeking breakthrough innovations to improve physical, mental, and social well-being for people as they age 

Sign up for updates