IS IT POSSIBLE TO FORMULATE THE DRIVERS THAT MOTIVATE PATIENTS TOWARDS ADHERENCE?
Updated: Nov 26, 2020
When psychology models and mathematics meet
Dr. Jackie Assa - Research and Data Science manager, Well-Beat Technical Advisory Board Member.
Is it possible to formulate patient adherence, specifically through a methodological model that will enable a machine to understand and recommend activities that increase adherence?
How close is it to Netflix’s recommendation engine?
What are the challenges, dangers, and solutions?
In today’s day and age, it isn’t enough to tell patients – ‘just do it’. We need to dig deeper into their behavior patterns, barriers, and drivers. We need to understand why patients don’t show up to each appointment or don’t take their prescribed medication on time. We need to create a remote personal engagement, especially during the COVID-19 pandemic. We need to support the overworked caregivers while helping patients.
Through a recent interview with Dr. Jackie Assa, Research & Data Science Manager, we learned about combining a unique machine learning technology with psychology that increases adherence to treatment. The exact type of advancement in healthcare to the world expects.
We recently interviewed Dr. Jacki Assa about Machine Learning (ML), Artificial Intelligence (AI), and his work with Well-Beat to drive patient engagement and adherence to treatment.
Dr. Jackie Assa is highly familiar in the field of ML, having extensive experience at various enterprises and high-tech companies including big label organizations such as eBay, Intel, CitiBank, and others.
“I was approached by Well-Beat’s founders - Ravit Ram Bar-Dea, Keren Aharon, and David Voschina - several years ago, to assist. The idea was to formulate patient adherence through a methodological model that will enable a machine to understand and recommend activities to increase the adherence,” he says. “We were facing different challenges – such as how to understand patients, how to build a more accurate patient profile, and how to represent the profile to a machine so that it can learn to increase the adherence of the patient to treatment, and convince them to stick with it. People have different motivations, triggers, and reactions, and don’t always have consistent and rational behavior.”
There is a methodological path!
“Physical activity is vital to the success of your health plan. Take a walk today.” This is one example of a motivational message used for a structure type of patient. However, not every patient will respond positively to it. On the contrary, a significant number of patients will feel antagonized. The challenge is to understand which message (and content) is right for which patient, at what time, and in what tone.
Successful behavioral patterns have already been implemented by machines, but in the healthcare environment, it becomes highly complex, given the constant changes in variables as well as barriers that weren’t there initially.
The idea to utilize previous users’ results, or patients in our case, is already known, especially in the ‘recommendation’ arena. It is based on ‘Success’ and ‘Failure’, but is currently being used in a completely different domain – sales. An example of one popular usage of a recommendation engine is Netflix. The system will recommend a movie based on other customers’ history or previous views.
Why not use the same system? And how can we use it for adherence and engagement in the healthcare world?
The challenge: starting from scratch
We encountered a challenge that is specific to adherence but not found in the sales domain.
Starting from scratch – no previous data. This means there is no ‘Success’ or ‘Failure’ information to base our initial suggestions on until a large number of statistics kick in.
This problem is known in the traditional literature as “cold start” or bootstrapping a recommendation engine. Once it gets going, the engine begins to work and suggests more relevant options, using similar profiles.
The Well-Beat team needed to produce data and accurate recommendations, without data. In the marketing and sales world companies conduct A/B testing. But we can’t afford to do this for healthcare. There is a big difference.
Going back to the example of Netflix - when we just open a new account and look for recommended videos without Netflix knowing anything about our preferences or profiles, getting a less accurate or more generic recommendation isn’t damaging. A slight disappointment from a movie selection will probably not prevent us from being a Netflix customer. However, this data can be used to gather data which will kick-start the recommendation engine by determining the details on our viewing profile.
For healthcare adherence, a cold start is a big issue. If we don’t get it right, the effect can be damaging. In fact, it might do the opposite of what we want and get patients to stop treatment. If you motivate to stop treatment, especially during the initial stages, it becomes dangerous. Therefore, we must be a lot more careful during cold starts for healthcare in comparison with ‘cold starts’ in sales or marketing.
A novel idea direction
How can the challenge of cold start in a recommendation engine for treatment adherence be solved safely? Assuming we have an initial patient profile, how do we know which message works, the tone, the frequency, and on which user (patient)?
The concept for our solution was to generalize. We can look at a cluster of patients and determine some common factors. The initial high-level characterization doesn’t have to be exact, but we can use it to ask behavior experts, like psychologists, to explain the person’s characteristics.
We wanted to harness the experts’ wisdom and find out how to build a motivational profile for patients. We also examined all the relevant papers and methods in order to find the most important personal traits.
Using the experts gave a kick start for the recommendation engine. Based on the previous phase, we were able to build the recommendation engine and create dynamic profiles.
How does it work
Motivating each patient personally and avoiding a ‘one size fits all’ approach, requires a deep understanding of the individual and their persona. To achieve this, we needed to recognize the sensitivities and needs (including the unconscious needs) of the patient, sort out the barriers indirectly, and learn which interventions are effective in different circumstances. We could then use the ML mechanisms, including monitoring consistent ‘Success’ factors. The result – a creation of human and digital interventions, used in different personalized ways for each patient, motiving them to action, consistently and over time, in ways that enable the healthcare organization great flexibility in real-time.
Human behavior isn’t steady, given the highs and lows people experience throughout the treatment process. Therefore, the recommendation engine must also consider and estimate the patient’s stability and measure their abandonment risks. Imagine a world where the caregivers (or management) know in advance who is at high risk for abandonment and react in a proactive manner. When we know in advance who is at high-risk, we can provide a different treatment, using the available resources effectively.
Well-Beat offers another unique capability – monitoring patients dynamically and adjusting the messages accordingly. The platform provides frequent data updates about each patient. The changes reflect the processes, advancement, emotional journey, environmental factors, evolving history, and personal style. The engine then updates the profile daily, depending on the patient’s activity, and adjusts the guidelines accordingly.
To date, the company has conducted several pilots at leading healthcare organizations across the world, including the Cardiac Prevention & Rehabilitation Institute at the Sheba Medical Center. The results were remarkable and showed a tripled treatment adherence among patients, as well as 85% persistence prediction in the long run. These are breakthrough numbers given the current worldwide data pointing at high abandonment rates for rehabilitative patients, and their need for more serious procedures in the future, as a result.
What does the future entail?
Motivating patients towards adherence is only a small part of a much bigger picture. Chronic patients and the elders are, and will continue to be, the focus of the healthcare industry. These are large populations that demand services throughout the remainder of their lives. In order to provide such a long-term service, and deal with denial and despair, we must be able to understand each individual personally and offer the appropriate intervention.
In fact, this challenge is already here, given the exhausted resources available. We must equip healthcare teams with tools that assist them to see patients quickly, and create instant trust, even without the need for ongoing personal contact. This can be done through digital motivational messages that are generated automatically using a recommendation engine, while using the existing resources effectively.
It is an enormous challenge, but the journey must begin with small steps.
Well-Beat is a participant in the IP² LaunchPad of the Taiwanese Innovation to Industry i2i organization. The program aims to assist Israeli companies in penetrating Asian markets through pilot partnerships, and strategic business connections.
Originally posted on NewTech Magazine 2020