How to Truly Solve the Patient Engagement Problem with AI

By Waqaas Al-Siddiq, CEO and Founder of Biotricity Inc.
Twitter: @biotricity_inc

The most urgent problem facing the U.S. healthcare system today is cost. Traditional factions of political party affiliation, socio-economic status, and patient-payer-provider categorization all agree that the astronomical costs of the current system are not sustainable—and only increasing.

Paradoxically, a staggering amount of these expenses are either preventable or reducible. Most are simply due to an inefficient allocation of resources, both financial and medical. This fact is typified by the patient engagement problem, which is estimated to consume half a trillion dollars including $300 billion for medication adherence alone. Other notable examples include non-urgent Emergency Room admission and chronic care treatment, with the latter accounting for over 80 percent of the system’s costs.

These statistics and others explain the renewed interest in healthcare solutions involving Artificial Intelligence (AI), which is touted as a viable means of simultaneously controlling costs and achieving the elusive goal of patient adherence. A number of contemporary solutions in this space are using rudimentary AI techniques based on static datasets, simple decision-tree rules engines, and wearable devices to achieve this objective.

There is a profound distinction between such inert systems and those which truly embody AI’s most renowned quality: the ability to dynamically learn and improve over time. When this manifestation of AI’s full potential is paired with medical grade wearable devices, and when medical applications of this technology mature in the healthcare industry—patients will suddenly find a host of new, verifiable data to increase their healthcare engagement.

The underlying neural networks, deep learning, and traditional machine learning algorithms facilitating AI require abundant quantities of data for optimization. Medical grade wearable devices that continuously stream patient data are ideal for this purpose, as they provide the data necessary to create algorithms for personalized medical treatment. In the wake of such individualized experiences, compliance is sure to rise, resulting in a commensurate decline in the overall costs to the healthcare system.

The Extent of the Problem
The notion of patient engagement is so prominent within the healthcare industry today because of its reverberating effects within the system itself. The billions of dollars attributed to patient non-adherence are predicated on the fact that millions of patients do not adhere to their practitioners’ instructions between visits. Whether willful or not, non-adherence commonly includes ignoring or forgetting directives for ingesting medication, exercising, or eating properly. Adherence issues also involve not staying within predefined parameters for temporal, dietary, or quantity concerns. The preventable nature of such expensive neglect is apparent; if patients would simply follow defined regimens outside of care centers, they would be more likely to achieve healthy goals and reduce systemic costs. Engagement, therefore, can be as basic as timely reminders for patients to take responsibility for their own care management, which is a pivotal starting point for holistic improvement of the entire healthcare system.

The correlation between patient engagement and chronic care management is both symbiotic and useful for illustrating the full extent of the patient adherence problem. On the one hand, regimens sustained over lengthy time periods (such as taking medication) are frequently associated with low adherence. Anticoagulants to reduce the likelihood of strokes are examples of chronic condition medications with insufficient levels of patient adherence. Since chronic care costs account for over three-fourths of the money spent in the current healthcare system, the amount of funds dedicated to them could be considerably reduced by boosting adherence. Reciprocally, patients are more likely to incur chronic conditions by not adhering to practitioner instructions which, in turn, creates more chronic care costs. Improving patient engagement could simultaneously reduce costs associated with both chronic conditions and patient compliance, which overlap and exacerbate one another.

The final factor associated with healthcare costs is less directly related to patient engagement, yet just as significant. High expenses for ER treatment are justified by the time-sensitive nature of the services provided; but high systemic costs are worsened by the numerous instances in which ER patients could have opted for less timely and costly treatment options. A research study indicates that approximately 15 to 30 percent of ER visits could have been performed at alternative care centers and clinics, saving over $4.4 billion. A study of the 10 most common ER outpatient conditions indicated that injuries for strains and sprains topped the list—with some treatments priced at over $24,000.

The soaring costs of an over-reliance on ER services are three-fold. Many patients check into ERs for conditions that don’t necessarily require immediate attention, including sprained ankles or headache symptoms. According to the University of California at San Francisco, there are also “the ranks of uninsured patients, who disproportionately rely on the emergency room for non-emergency care.” In these instances, patients may not be knowledgeable enough to discern when an urgent care out-patient clinic would suffice over a visit to the ER. Finally, there are anxious caretakers of children or the elderly who overreact to minor symptoms, opting for the nearest ER instead of less expensive, more conventional treatment options. Most of these three ER patient types could benefit from less immediate treatment in urgent care centers or clinics to decrease total ER costs.

Today’s Solutions
Today, fledgling AI solutions are deployed to manage all of these variations of the patient engagement problem. Some have demonstrated empirical evidence for increasing patient adherence. A solution targeted at anticoagulation therapy proved 50 percent better at achieving patient adherence than the control in the study did. Still, most of these options deploy infantile AI techniques much less sophisticated than the true incarnation of these technologies. They offer a promising glimpse of the full scope of AI’s potential to solve the patient engagement issue.

The primary limitation of these solutions is they are too general. Most involve historic, static datasets, which are licensed from research groups such as Mayo Clinic or the American Heart Association. These data sets form the basis for either basic algorithms or rule-oriented decision trees, which are useful for categorizing patients, symptoms, and various conditions. Although the data for these rudimentary AI techniques are often times current, they are nonetheless immutable and devoid of the vibrancy characteristic of mature AI solutions such as IBM Watson. Despite their lack of intuition, these techniques are adept at categorizing patient conditions based on responses to preconfigured questions, enabling them to triage patient needs with varying degrees of effectiveness. This functionality links non-critical ER visits to the patient engagement problem; many patients are incorrectly triaged into ERs by hospital systems designed to “fast track” them via remote assistance. In Nevada, tens of thousands of dollars have been spent by patients incorrectly triaged (and subsequently helicoptered) into ER for non-urgent procedures. Similar simple AI methods are employed to use historic data as protocol for follow-ups. The data provides the basis for the number and nature of reminders to patients, as determined by previous effectiveness in getting patients to suitably follow-up practitioner visits.

In almost all these early healthcare-oriented AI efforts, the patient interaction involves either a voice-based solution, an avatar (digital assistants with visual components), or mobile apps. Some companies specialize in voice-based reminders that create a set of rules around care facility policy for patient follow-ups or adherence. By using third-party voice platforms such as Amazon Echo (a smart speaker system), they provide the software for facilities to create their own set of spoken reminders. Customers merely drag and drop their rules for the software to work; patients interact with the solution via an application on their mobile devices. Other companies offer similar solutions with and without avatars that go through decision tree oriented patient engagement questions based on the underlying licensed data. Again, most are implemented through mobile devices.

There are several positive aspects of the evolution of infrastructure and architecture required to deploy these solutions. They unambiguously indicate the relatively recent ability to traverse massive amounts of data and transmit them through the cloud with continuous connectivity for unlimited access. They also illustrate the cloud’s storage capacity, underscore the power of mobile technologies, and exemplify the enhanced processing speeds of contemporary computing devices. Essentially, they’re paving the way for tomorrow’s AI healthcare tools.

Tomorrow’s Solutions
What these solutions are lacking is AI’s chief value proposition: the propensity to learn based on constant data streams to improve efficiency at completing machine-generated tasks. The aforementioned “intelligent” solutions aren’t capable of learning because they’re based on historic data. True learning systems evolve to heighten effectiveness based on experience, but this requires newly generated datasets to constantly collect and compile information to provide the basis for true machine intelligence. Augmenting historical data with real-time data transmissions will enable tomorrow’s AI options to transition from general to specific feedback for patients, which could prove the difference in eradicating the patient engagement issue.

That difference is based on the type of feedback AI requires. Machine learning and deep learning algorithms can offer demonstrable, measurable feedback which is far more convincing than simple reminders. Proven results are a cogent means of motivating people attempting to lose weight. After seeing an initial decrease in pounds, participants are further motivated to continue weight loss programs to see more results. Comparably, the movement for organic food and healthy eating championed by retailers such as Whole Foods has gained momentum because these dietary products help people eat better and, in turn, feel better. Those feelings translate into renewed emphasis for achieving healthy outcomes, and can produce the same effect for patient adherence when begat from mature AI healthcare solutions.

What’s truly required to resolve the patient engagement issue is for AI to tailor its learning potential to the healthcare industry. Initially, neural networks and traditional machine learning algorithms must acquire domain knowledge by parsing through historic datasets for specific healthcare conditions. Data from the American Heart Association, for example, can provide fodder for a host of cardiac applications. IBM Watson has specialized in several areas of terminal illnesses, and AI tools for mobile devices can do the same. Next, those algorithms must engage in an analogous “residency program” in which they perform analytics on real-time patient data overseen by a trained specialist. The dual experience will simultaneously enable the algorithms to learn from the data and from the specialist’s oversight. This approach was utilized by Uber with drivers accompanying autonomous vehicles during its pilot program. Clinically, the machine intelligence will either make accurate predictions from the data or learn from erroneous ones.

The long-term trajectory for medical grade AI solutions will culminate in an enhanced capability for personalized patient engagement experiences. The realization of this objective will signal a complete transition from generalized interactions to truly personalized ones, according to the maturation of AI technologies within the mobile healthcare device space. Numerous factors must coalesce to realize this goal. Mobile devices transmitting patient metrics must deliver clinically accurate, medically verifiable data upon which to base AI’s algorithms. Otherwise, the ensuing analytics results are worthless.

Architecturally, these devices must be able to connect this remote data to algorithms in a centralized location for aggregation of patient biometrics with other relevant datasets, such as those for a specific cardiac condition. Doing so involves a hybrid of both distributed and centralized architectural paradigms, which is characteristic of numerous Internet of Things applications. Finally, it’s necessary to issue real-time feedback to patients for action that lets them more effectively manage their care. Such feedback could include details about greater or lesser effectiveness of exercise or medication for coronary heart disease versus the previous week’s metrics, which could spur patients to adhere or to consult with practitioners between appointments to improve regimens and future results.

AI’s potential for personalization represents its greatest utility for solving the patient engagement problem because of its capacity for learning individual human characteristics. These technologies can understand what makes each person unique based on their individual biometric data. When that personalized understanding is paired with engagement systems offering demonstrable evidence of how one’s biometrics are affected by patient adherence, it can resolve one of the costliest healthcare problems today.