Wearable tech fuels AI research
At the end of February physIQ and the US Department of Veteran’s Affairs published data from a clinical trial demonstrating the use of physIQ’s machine-learning algorithm to predict when an individual patient with previously diagnosed heart failure will likely require a readmission to hospital.
In the study, heart failure patients were given disposable, wearable sensors, which collected physiological data that was streamed to physIQ’s pinpointIQ continuous data collection platform. Results from the trial, which was reported in Circulation – Heart Failure, showed that the AI-based algorithm could predict from the sensor data whether a patient was likely to be re-hospitalised up to 10 days before they were either readmitted, or had to go to the emergency room.
‘These results exceeded our expectations and demonstrate the power of our FDA cleared AI to give clinicians a fighting chance in keeping late-stage chronically ill patients out of the hospital and at home. Most clinicians will tell you that, with several days of lead time to intervene, their chances of changing the trajectory of any disease exacerbation are very high,’ reported Gary Conkright, physIQ founder, chairman and CEO.
Publication of the paper comes just weeks after physIQ was granted two new US patents for patient monitoring AI-based data analytics. One of the new patents covers use of deep learning to make estimates of cardiopulmonary functional capacity using data from sensors on wearable devices. The second new patent is part of an expanding suite of patents relating to physIQ’s personalised physiology modelling technology.
PhysIQ is exploiting its scalable cloud-based platform and proprietary AI technology to enable personalised physiology analytics, for what it terms proactive care delivery models. The FDA 510(k)-cleared data analytics technology is designed to process multiple vital signs from wearable sensors, and create a personalised, dynamic baseline for each individual.
Subtle deviations from this baseline identified by the algorithm can indicate worsening disease, or a change in health, and alert the physician. PhysIQ is leveraging the platform for applications in both healthcare and to support clinical trials. ‘The technology generates real-time continuous data that can provide far greater insight into health and disease than, say, an ECG, blood test, or blood pressure measurement taken at periodic intervals in the physician’s office,’ commented Conkright. ‘This 24/7 monitoring is critical, given that many physiological parameters have a daily rhythm, and so deviations from the patient’s baseline may not be evident if only measured at certain times of day.’
The algorithms underpinning the physIQ technology were originally developed at the US Department of Energy-funded Argonne National Laboratory, which is run by the University of Chicago, explained Conkright. ‘The AI technology was initially developed to detect anomalous behaviour in nuclear power plants. I was brought in by the University of Chicago to see how we could commercialise the technology more widely, and a company, SmartSignal, was set up in 1999 to develop an AI platform for use in a wider range of industries, from aviation – our first client was Delta Airlines, which used our product to assess the health of 2,600 jet engines – to power generation, petrochemicals, windmills, etc.
‘SmartSignal was subsequently acquired by GE, in 2011 but physIQ retains an exclusive licence to apply the IP to signals from biological systems. To date, we have filed and received another 16 global patents, have another eight pending and about another dozen in process, taking this technology to a much higher level of sophistication and performance,’ added Conkright.
To think about how AI can be applied in the physiology field, it’s useful to take a step back, Conkright noted. ‘One of the problems with deep learning, for example, is that it takes massive amounts of data to train the algorithm, and that is just not practical with the type of data that is available from today’s wearables. However, we have developed a novel way of leveraging our existing 2+ million hours of high-fidelity wearable data to significantly reduce the inputs needed for a deep learning generated algorithm. Some of this is described by our new patent; some we keep as trade secrets.’
The potential for using this level of continuous physiological data to change the way health status is measured are considerable, he commented. Consider how we may measure the efficiency of a cardiovascular drug, for example. ‘This is traditionally done in two ways. Either using the “gold standard” six-minute walk test, which generates a very ‘noisy’ signal, or by measuring VO2max, which is a time-intensive laboratory test that is uncomfortable for the patient, costly, and not really feasible to carry out every month,’ said Conkright.
Both tests aim to measure quantitatively the level of cardiac function, but neither is ideal. ‘Imagine if you could obtain the same insight using data from a sensor that is worn either as a watch, or as a patch on the chest, and which continuously collects data, from which a machine learning algorithm can generate a measure that is highly correlated to the VO2max test... that is what is covered by one of the latest two patents, and for which we will seek FDA clearance to use in clinical settings.’
As physIQ continues to develop the platform, additional AI can be added, to give the analyses an even greater depth of understanding and insight into health status. ‘The last FDA-cleared AI that we added was an atrial fibrillation algorithm that is agnostic to the wearable device that produces the data. For example, Apple’s A-fib is only cleared for data from its smart watch.
Ultimately, the aim of AI in the healthcare field is to enable exactly that, ‘health care,’ he suggested. ‘A lot of what doctors do today is treat sick people, either in or out of hospital, so in essence they are practicing “sick care”. AI will enable the three ‘Ps’; personalised, precision and proactive healthcare, which we believe will help to keep people out of hospital, and in their homes, improve quality of life, and also potentially reduce medical costs. This is the best chance we have in bending the cost curve in healthcare, empowering proactive patient care by treating people individually, not as a population, through personalised medicine,’ stated Conkright.
‘leverage AI’ in battle on Covid-19
Commenting on the coronavirus situation, Conkright said, “I believe the current COVID-19 pandemic will soon illustrate the need to leverage AI and remote sensing in delivering quality care. In most other industries, technology has freed humans from the drudgery of our day-to-day work, improving output and efficiency. We do not have enough hospital beds in the US to care for everyone that will get sick with COVID-19, and we do not have enough nurses to care for the sick population, especially as they get exposed to the virus and have to self-quarantine, a very real probability. Monitoring high-risk COVID-19 patients remotely, and allowing AI to identify which patients require our resource-limited healthcare staff so that they can focus their time and critical skills on the highest impact area, is not only going to be required for managing COVID-19 – but is a prescription for healthcare in the future.
The same is true for clinical trials. Why require a clinical trial subject to come into the clinic every six weeks to get a snapshot of his/her condition at that point in time? Avoid those trips and exponentially increase the amount of data you are able to collect, for the purpose of building a better understanding of the new therapy’s impact in a much shorter time frame. Everyone gains when a new lifesaving drug gets to market sooner. Once that drug is on the market, continuous monitoring with AI-enabled surveillance will be able to provide early warning of an adverse side effect, enabling the physician to either switch therapies or change the dosing. In the case of chemotherapy, where every day matters, avoiding a side effect that causes hospitalisation not only saves an enormous amount of money, but it could also have a lifesaving impact.
‘AI and continuous monitoring’s time has come,’ concluded Conkright.