Checkmate: Using artificial intelligence for diabetes health coaching

The recipe for a healthy lifestyle is to eat well, stay active, reduce stress and take the prescribed medications. For someone with type 2 diabetes, small changes in any of these categories can have a big impact — positive or negative.

Living with diabetes requires ongoing access to diabetes care to manage the condition, so patients meet with their healthcare team about every three months. But what if they struggle between appointments?

Hamilton Health Sciences (HHS) researcher Diana Sherifali wanted to see if it was possible to use artificial intelligence to prevent a small problem from becoming a bigger one. She takes the first steps to develop a health coaching algorithm to help people with diabetes. This algorithm could be added to existing fitness or wellness apps that already track diet and exercise.

Diabetes Health Coaching

As a clinical nurse specialist at HHS, associate professor at McMaster University’s School of Nursing, and associate scientist at the Population Health Research Institute (PHRI), Sherifali has been investigating whether a type of artificial intelligence called machine learning could be the solution. PHRI is a joint institute of HHS and McMaster University.

After all, a computer can learn winning strategies for almost any situation.

“People living with diabetes deal with the disease every day. That means at least 95% of diabetes management happens outside of the healthcare system,” says Sherifali, who is also a certified diabetes educator and understands the benefits of health coaching. “It is unrealistic that their care teams help them on a daily basis. But if a machine learning algorithm is developed and applied to existing technology that these people are already using, it could provide some of the support needed between appointments.”

Type 2 diabetes occurs when your pancreas doesn’t produce enough insulin, which regulates sugar in the body, or your body doesn’t respond well to insulin. There is no cure, so blood sugar levels must be controlled. This is done through diet changes, exercise, maintaining a healthy weight, monitoring blood sugar levels, and possible medication or insulin injections.

Digital Healthcare Experts

When the health coaching algorithm is applied to a wellness tracker, it can use existing nutrition and exercise data. Once people add their weight, blood sugar levels, and medications, the algorithm can determine if changes need to be made and make recommendations on what to do.

“With my basic knowledge of artificial intelligence, I knew I had to work with CREATE.”

To further explore and develop this idea, Sherifali has partnered with digital health and data science experts at the HHS Center for Data Science and Digital Health (CREATE).

“With my background in artificial intelligence, I knew I had to work with CREATE, so I approached them with the idea from the start,” says Sherifali. “I was excited when told it was worth exploring.”

Teach the computer

When you play chess on your phone, sometimes you wonder how the computer knows how to play? By automatically playing thousands of games and being rewarded for wins and penalized for losses, a computer can eventually learn winning strategies for almost any situation. This approach is called machine learning reinforcement learning and is leading to breakthrough advances in many artificial intelligence applications, including self-driving cars.

The same approach was used by CREATE for computer-based diabetes health coaching, says Jeremy Petch, founding director of CREATE.

“We feed the algorithm health data, a doctor’s recommendations, and the results—good and bad,” says Petch. “That way, it learns the best strategy under all circumstances, just like the computer opponents you play against on your phone.”

In this case, the data includes blood glucose levels, medications, diet, physical activity, weight, and stress.

After developing and testing the algorithm, the team found that it provided accurate initial recommendations.

More data makes a stronger algorithm

“Now that we’ve determined that the algorithm can learn the appropriate recommendations for common problems faced by people with type 2 diabetes, we need more detailed data to refine it further,” says Petch.

In the next phase of the study, the algorithm will learn how to make accurate recommendations with more complex data. Then it is finally tested with clinicians and finally with patients.

“The challenge is that there is more and more data,” says Sherifali. “The algorithm is not intended to replace in-person appointments, so we need to determine at what stage there is enough data for the algorithm to effectively coach individuals through many different issues, while leaving complex scenarios to a medical team to address.”

Since fitness and wellness apps are already well established for storing and tracking the kind of data that people with type 2 diabetes are already monitoring, implementing the algorithm into those apps won’t be the most difficult part of the project. The first steps – determining if the algorithm works are actually the most difficult steps.

Sherifali says she’s thrilled that this first phase is already showing results. This means health coaching for people with type 2 diabetes could be available in the near future.

related posts