Disclaimer: This blog is independently written and published by me. The opinions expressed herein are my own personal opinions and do not represent my employer's view in any way.
For many Americans, wearables like Fitbit, Garmin, Apple watch, and other fitness trackers are already capturing useful information about our health and behavior. But this wealth of information is rarely realized to generate insights that inform healthcare decisions. Information that could assist a physician with diagnosis and treatment is lost in data silos and a sea of data nonconformity. For example, a Fitbit device has its own proprietary format and can’t send or exchange data with a patient’s healthcare record. Instead, physicians largely rely on infrequent ‘snap shots’ of vitals signs as health indicator when a patient comes into the doctor’s office.
In 2018, Apple brought broad public awareness to the power of wearable data and Artificial Intelligence (AI) used for medical purposes when it received FDA clearance for its EKG wearable with arrhythmia detection. This release hinted at a future where patients and physicians have greater insight into often invisible aspects of one’s health utilizing internet connected devices, apps, and artificial intelligence.
Advances in high quality, yet inexpensive biometric sensors, have made smart devices smaller and more portable and into many households. There are more consumer-grade and medical-grade devices available to the patient and healthcare industry than ever before. Data from devices that previously required a visit to a physician‘s office to physically hand over data is beginning to reach physicians over the internet in near real-time. As a collection, these devices are often referred to as the “Internet of Medical Things” or “IoMT” for short. The most important distinction of an IoMT, as opposed to an IoT, is that the device transmits protected health information (or PHI), adhering to very stringent government requirements.
Internet of Medical Things data captured by apps and wearables indeed holds tremendous promise to generating clinically significant insights. The promise of using these devices in combination with AI is to empower clinicians with timely insights so that patient outreach and interventions can occur. The aim is that AI applied to this data could assist physicians in real-time, giving them answers to the following types of questions:
Has the patient’s disease severity changed? (e.g. cancer progression)
Does the patient have any new health problems? (e.g. detect early onset of Parkinson’s disease)
Is the patient at risk / susceptible to additional health problems? (e.g. combine weight, activity and demographics to estimate risk of heart disease)
Is the patient responding as expected to a medication, treatment, or an environmental agent? (e.g. epilepsy patient’s response to anti-seizure medication)
Is the patient heading to a bad state or highly likely to have an event? (e.g. a patient with asthma uses a smart peak flow device and it detects an upcoming episode of worsening of asthma before the patient can feel it)
Is the patient sufficiently adhering to the physician’s orders relating to their activities and responsibilities? (e.g. a CHF patients adherence to weighing himself daily using a connected scale)
The insights generated from IoMT data provide clinicians with specific, accurate, and timely information about a patient’s condition and behaviors allowing for highly targeted and timely treatment modifications and interventions and bolster the patient/physician relationship. IoMT not only extends the clinicians visibility into a patient's health beyond the walls of the clinic or hospital, it also provides data at orders of magnitudes richer than the snapshots we have today. Instead of comparing against a population normal, you can compare against a patient’s own normal, and build upon computer generated insights from data from hundreds of thousands of patients who have been in similar contexts or who have similar genetic profiles. Thus, beyond the benefit to the individual patient, there is an amplification that can be realized by combining this massive corpus of information. When brought together, IoMT data will have the ability to power the next generation of medical AI.
Given the tremendous potential of this technology, I expected to see ongoing research studies proving out the concept, similar to drug trials where interventions are studied to assess their outcomes against a control. But when I searched ClinicalTrials.org looking for studies over the last 5 years using Internet of Medical Things devices there were so few that I was able to count them on one hand. Given the potential to save human lives or thwart adverse health events like strokes or heart attacks, I became passionate about understanding the barriers to utilizing this potentially life saving technology.
I spoke to healthcare providers, researchers, and hospital systems about their desire to try out connected devices with their patients. Though they are often excited, they don’t know the path to get there. Most providers and healthcare systems are simply not equipped with the foundational technology, know-how, or resources for managing and making IoMT data useful. Getting from hypothesis to better outcomes is a long and costly journey filled with obstacles. To go off on their own and build a custom solution for connected devices and remote patient monitoring could take years, along with significant capital for IT resourcing and consultants. And it still might very well be a huge failure technically, reputationally, or financially.
THE Technical challenges
Designing and building the technical infrastructure needed in order to run a clinical study utilizing data from a wearable paired with an app and then transport that data to the cloud cannot be understated. Protected health information (PHI) from these devices in the cloud must be exchanged securely in a HIPAA compliant cloud. Secondly, when dealing with streaming data with sporadic connectivity, due to being out of range of a hub or Bluetooth disconnecting, data from IoMT can arrive late and not necessarily in sequential order. A third major problem is that of interoperability of IoMT device data with the rest of the healthcare data, such as clinical and pharmaceutical records. To be utilized, IoMT data must be speaking the same language as other healthcare data. So for example, if a chemotherapy patient has an infusion, the data collected from a wearable will need that context, and to do that it needs to be able to communicate with the patient’s medical record.
Hospitals and many practices need to be able to use multiple kinds of devices, such as wearables, pacemakers, glucose monitors, sleep apnea machines, asthma inhalers, smartphone apps. Developers must handle with significant challenges dealing with formats from a range of devices and schemas, from sensors worn on the body, ambient data capture devices, applications that document patient reported outcomes, and even devices that only require the patient to be within a few meters of a sensor. Devices may speak different languages. For example, one might call heart rate “HR”, another “pulse rate”. The messages might be formatted and grouped differently as some devices send a packet with one biometric (say, heart rate) in sequential order in time while others will send a packet with heart rate, respiratory rate, and temperature all for the same point in time. Whether you’re a researcher looking to do a study with a single IoMT device or a hospital setting up infrastructure to support an ecosystem of devices, building out the technical infrastructure to utilize IoMT in a secure, compliant, and usable way with patients takes significant expertise and resources.
THE CLINICAL CHALLENGES
There are many third party solution providers who have tried to fill the gap in connected patient monitoring and offer end to end solutions for both remote patient monitoring and in-hospital monitoring. Many solutions provide pre-packaged kits typically including a single IoMT device and a bedside monitor or smart-device that acts as a gateway to the internet. Many offer a very rudimentary scoring system called Early Warning Scores (EWS) in which alerts are triggering based on simple thresholds. And while it can yield benefits, clinicians are easily frustrated by their lack of precision. These simple heuristics offered by solution providers often suffer from over triggering and induce alert fatigue in clinicians. The thresholds are based on what’s normal for a population rather than what’s each person’s unique normal and what’s normal for their circumstance.
Another critical issue with existing solutions is that the data lands in a data silo controlled by the solution provider. Information is not exchanged with the patient’s medical record, and remains with the third party, not with the health system where it can use it for improving machine learning models. While the solution providers benefit from these silos, patients and health systems do not.
Furthermore, the workflow starts to break down when more than one solution is applied. Providers will use all sorts of connected medical-grade devices, such as wearables, glucose monitors, sleep apnea machines, asthma inhalers, smartphone apps, etc. New devices will be coming in to the market regularly and clinicians will want to know if these new devices yield better results. Healthcare systems quickly realize that it is not feasible for them to use 10 apps and 10 clinician dashboards, one per device, to monitor their patients - that is in no way scalable, efficient, or cost effective. Clinical attention is the most valuable resource that a hospital has, and these one-off solutions promotes the same problems that they try to address, adding to an already inefficient and exhausting clinical workflow.
The successful adoption of IoMT is enormously complex and requires a large investment on the part of the hospital including logistics, engineering, IRB studies, machine learning, training care teams, and funding. Not every hospital will be up for this risky proposition. There are also ethical and liability concerns to worry about – what if the data is not acted upon? What if there is an outage? What if the device detects a patient is about to have a seizure and he or she is driving on the highway? Do you alert the patient? What if the alert causes an accident?
Several studies to date show a positive return on investment in terms of ROI and outcomes with the EarlySense device in-hospital. But the feedback from the clinical staff is not always positive for these devices. There are a lot of false positives and noise and the information is only available at the bedside monitor or at the nurse’s station – places that are not helpful for the nurse.
To get hospitals interested typically the IoMT vendor must demonstrate more than just reliability of the device:
Value: What is the actual clinical value of the data being provided from a care perspective? Is the supporting data reliable?
ROI: What is the return on investment? What is the compensation, remuneration associated with it? How long does it take to reach break-even? Is there a return on time/efficiency? Are there secondary returns?
Time: What is the time it takes for the initial set up and the time it takes each day for staff to produce the proposed value?
Feasibility: What is the cost and time it takes to set up? What skills are the skills? Is it possible from an engineering and regulatory standpoint? Training/workflow/adoption standpoint? How easy is it to measure outcomes?
Usability: For both patients and care team: is the device usable? What training is required? Will the experience produce clinician burnout?
Roadmap: Does the solution align with the long-term objectives of the hospital? How do you set up an architecture that supports long term needs such that work is not reproduced?
A SOLUTION
Given these enormous challenges, my team at Microsoft set out to make the technical hurdles easier for the IoMT engineer. We released the IoMT FHIR® Connector for Azure, a new open source tool that securely ingests, normalizes, and persist protected health information (PHI) from IoMT devices in the cloud. It handles high frequency, high volume, and sometimes requires sub-second measurements. The data is converted to the HL7 FHIR Standard (Fast Healthcare Interoperability Resources) which has emerged over the last few years as a solution for interoperability for healthcare data. FHIR is rapidly becoming the preferred industry-wide standard for exchanging and managing healthcare information in electronic format and has been most successful in exchange of clinical health records.
The are several benefits to developers. Firstly, there are enormous time savings as the infrastructure is built for them. Secondly, they don’t need to know FHIR®. All they have to do is to send the data objects to the IoMT Connector to FHIR® for Azure and it will transform it. Third, they don’t have to spin up their own cloud or maintain it, simply leverage FHIR® server for Azure and maintain all of the data in the hospital’s own Azure subscription. The data sits in a world-class Microsoft Azure cloud that handles all security and compliance, including HIPAA, HITRUST, and GDPR.
Another key benefit of the IoMT data in the cloud is the ability to easily do machine learning. For IoMT data to be useful, AI has to be trained. That means, a repository of historical data needs to exist for the models to detect meaningful patterns. Only then can it be applied to detect these patterns and tested to see whether the models result in better patient outcomes. Algorithms can be applied to the data to do simple trends such as noticing that your blood pressure has been steadily increasing over the last year, to more advanced AI recognition of strokes or seizures. It is critical for the models to be reliable and demonstrate efficacy in clinical studies.
For clinicians to be able to utilize IoMT data is a huge step forward toward proactive care. Given the IoMT FHIR® Connector for Azure, Healthcare systems don’t need to commit to a multi-million-dollar deal to enable IoMT scenarios. Small-scale experiments can be done quickly. Developers can get a device up and running in a day. A few devices are tested out and just like a lean startup. Perhaps one device works better than another or one device is too uncomfortable for patients to use. For the ones that work well and look promising, you move forward. As your needs grow, the service scales up as needed to support a study or wide-scale rollout to patients.
THE FUTURE OF HEALTHCARE
In the not too distant future, it will be standard practice for providers to remain “connected” to their patients. Healthcare will increasingly become proactive rather than a reactive practice and it will be agnostic of patient and provider location. With your permission, your provider will have greater visibility into changes in your health. You won’t have to wonder if something is wrong with your health, your doctor will call you and say, “Jean, it looks like your asthma is worsening, please use your inhaler” or “Alexandra, we’ve noticed you haven’t used your kitchen in the last 24 hours, are you okay”?. In the future, Predictive AI will be key and based on data that proves increased visibility of the patient’s real-time status and detect meaningful changes that lead to timely interventions and better outcomes.
The relationship between patient and physician serves as the backbone of healthcare delivery. Health outcomes rely greatly on the ability of providers to accurately understand their patients' conditions and to know when meaningful changes occur in real-time and over time. The goal is not to replace physicians with AI, but rather to bring to light changes in a patient’s status that a physician may not have had the ability to spot without the pattern detection of machine learning and transform healthcare from reactive care to proactive care. Opportunities for increased visibility into the patient’s status are being bolstered by the rise of Internet of Medical Things (IoMT) devices and medical applications that transmit patient health information (PHI) and the use of AI.
My personal belief is that IoMT represents one of the largest technological revolutions changing the we deliver healthcare. I see a future where devices monitoring patients in their daily lives will be used as a standard approach to deliver cost savings, improve patient visibility outside of the physician’s office, and to create new insights for patient care. My personal mission is to uncover and remove barriers in the industry so that this life-saving technology can be adopted and I am privileged to be in a position to do so through my work at Microsoft.