Role of AI and Machine Learning in Chronic Disease Management During COVID-19
The management of chronic conditions, such as diabetes, asthma and congestive heart failure, have long required judicious oversight and management from a savvy healthcare provider, along with - in many cases - daily actions on the part of the patient, to execute the evidence and guideline-driven instructions of their healthcare providers to remain in the “safe zone.” Provided the trust behind this arrangement is maintained, there’s the usual cadence of physician office visits - say 3-4 times per year - when the physician-patient team can check in on progress and make any required course corrections. And in more dire situations, or in ones where new variables may be introduced (e.g., the initiation of insulin for a patient who is naive to the drug and to injections as a whole), that frequency of interaction between the patient and provider may increase in order to maintain a safe health trajectory.
COVID-19 has turned this paradigm on its head. For months now, patients cannot see their healthcare providers in the usual manner, as offices may have been either closed or imposed restrictions for emergency visits only. In some counties in the US and in some countries globally, this persists today. And that’s not all. COVID-19, apart from its immediate impact on the continuity of care in chronic condition management, has further uncovered underlying and systemic problems in healthcare. As mentioned, the patient-provider cadence of meetings has been interrupted. While telemedicine services can help to address this issue, COVID-19 has further uncovered societal disparities in the basics required for telehealth to even work. Access to reliable high-speed internet, smartphone use and provider practices equipped with telehealth technology are barriers for telehealth in underserved communities. Security and privacy are risks with the rapid introduction to telehealth. When the first wave of tele-consults due to COVID-19 constraints manifested, many shifted to and adopted tools. Adding to the complexity, , some of these tools faced significant challenges as it related to ensuring privacy and security. So while the healthcare community was scrambling to deal with the new impositions of COVID-19, many “holes in the bucket” were discovered.
The truth is no one has ever experienced anything like this, and certainly not since the pandemic in the early 20th century. It could take well over a year to develop a vaccine, and even though Dr. Anthony Fauci - the respected and renowned Director of the NIH National Institute of Allergy and Infectious Diseases and arguably, one of the world's top 5 virus pandemic experts - recently expressed cautious optimism about a vaccine by the end of the year, is that the only solution? Do we sit still in the interim?
Enter stage left - Innovation. There is an incredible opportunity for innovation to unlock alternate strategies to help cope with these impositions of COVID-19 on chronic conditions and that too, well beyond our pandemic has subsided. We are beginning to witness this through telehealth and the increase in virtual care. On March 27, 2020, the Coronavirus Aid, Relief and Economic Security (CARES) Act was enacted into law. Amongst other things, the bill provided a further expansion of telehealth services. McKinsey recently surveyed consumers and unearthed that 11% of US consumers used telehealth in 2019 versus a whopping 46% who are using telehealth today due to COVID-19 altering healthcare visits and breaking that continuity of care. This can only help so much until it faces the foundational constraints we’ve already mentioned. How do we take this a step further? What else can be done with digital technology and more importantly, the data it generates therein.
First, some somber statistics about how dire the risk is for those who suffer from chronic conditions in the midst of COVID-19. According to CDC case surveillance in the US, people with chronic conditions are 6 times more likely to be hospitalized and 12 times more likely to die from COVID-19. Diabetes is one of the most common underlying health conditions accounting for 30% of cases. That’s nearly three times the normal incidence of Type 2 diabetes in the US! What experts now believe is chronic inflammation, immune response impairment, and potential direct pancreatic damage is the association between COVID-19 and diabetes. The most recent study published examines specific characteristics before and at the time of hospital admission, which predict worse outcomes among people with diabetes. One of the key conclusions: BMI is a major risk factor. This research determined one in ten people with diabetes hospitalized for COVID-19 die within a week. Thus, chronic conditions must be a part of this conversation and we must move beyond just telemedicine in order to address the problem.
The advent of digital health and digital therapeutics - those digital health solutions that generally meet a higher level of rigor associated with evidence, claims, randomized control studies, FDA clearance, cybersecurity certification and clinical workflow integration - has been accelerated by evidence-driven health apps, wearable technology, and advanced sensors. These digital health solutions are already impacting chronic disease management in ways that both traditional care and telemedicine cannot solve alone. Add to that advanced, artificial intelligence and machine learning techniques that can be used to turn patient-generated health data from those who suffer from chronic conditions into behavior patterns which inturn, can aid in the response to their condition amidst this unprecedented pandemic.
Digital health solutions can analyze patterns in heart rate, blood pressure, blood glucose, physical activity, weight, sleep, medication adherence, diet, age, gender, you name it. This data can be both self-entered, and also brought-in via a series of technology-enabled (e.g., BLE, NFC, etc.) protocols, to take out the “adoption and engagement friction” associated with manual data entry. In one quick move, digital health improves what we refer to as the four “V’s” of data: Variety, Velocity, Volume, and Veracity. The variety is the data patterns mentioned above. The velocity is near-real time, and the volume therefore is high. With trusted sensor integration, or even integrations with labs and EMRs, the veracity of the data is increased. All of these contribute to perhaps the most important “V” - the fifth one: Value.
So what can we do with this data? We can run primary analysis, to address basics like “what’s the pattern for a male vs. a female, younger vs. older, tech-savvy vs. non-tech savvy person, etc. That primary analysis is facilitated by tools such as Tableau, Microsoft Power BI, R and others, where data vectors can be input, cleansed (according to data governance best-practices) and then plotted to inform on “patterns of interest.” These patterns can be a good event (such as a patient remembering to take their medication) or a bad event (such as a low blood sugar value for a diabetic or rapid weight gain for a patient with the onset of CHF). Then, add to this the more advanced statistical pattern recognition techniques of clustering (as an example, K-nearest neighbor (K-NN)), or discriminant functions (as an example, linear discriminant functions) whereby the data isn’t just answering “what” event happened, but more importantly, “why” it happened. Once a pattern is discovered, probabilistic and predictive models can then be used to understand “when” such an event may occur again. There are predictive models - the likes of Gaussian, Bayesian, Markov and others - that can be used to predict the trajectory of a patient. And finally, these data then fuel adaptive models, which apply machine learning algorithms to “fine-tune” and adapt the model to track, trace and predict a patient’s pathway based on the data they are entering. The output of such analysis can inform providers to then apply their clinical training to guide the patient to the best therapy pathway. Without the data and advanced algorithms, the provider will have little to go on, but their ability to optimize the treatment pathway for the patient is illuminated by data and analytics.
Welldoc has used such techniques to predict the onset of hypoglycemia, develop models that isolate feature usage within the digital health solution that are predictive of a successful pathway, and continued persistence of engagement to ensure the patient remains in the “safe zone.” All of this while not being able to physically or virtually “see” their healthcare provider.
What we learned through our research is early identification of individuals at risk of either succeeding or failing with a specific therapeutic pathway. Leveraging data and advanced analytics associated with digital health can therefore increase the efficacy of telehealth services to be more patient and condition-specific.
We can apply these methods to COVID-19 too and learn clues from patients who are suffering to better inform our response. What are good and bad symptoms that impact the onset or recovery? It’s important to note that this can be very individualistic. Hypothetically, we may discover that an unrelated medication mitigates progression for many patients, but the taking of certain actions can have a completely different effect for two different individuals.
The bottom line is digital health solutions offer hope and can help us address this conundrum when a doctor’s office is empty and patients can’t leave their homes. Remote capture of real-time data coupled with predictive analytics are key during this pandemic and beyond.
The capability exists to remotely monitor critical health conditions, predict future outcomes and utilize data analytics to help drive the course of patient actions to a more healthy future.