Diabetes is a pervasive and expensive public health threat. Close to one in ten Americans has diabetes, it is the 7th leading cause of death in the United States, and its total costs amount to $245 billion each year. Diabetes typically refers to two different diseases: type 1 diabetes, which is an autoimmune disease, and type 2 diabetes, which is a metabolic disease. Ninety-five percent of diabetics have type 2 diabetes. Despite a decade of efforts to bolster diabetes awareness and prevention, the prevalence of type 2 diabetes continues to grow, and is increasingly common among young adults and teenagers.
To address this growing threat, researchers at Columbia University’s Department of Biomedical Informatics (DBMI) have developed a comprehensive program for investigating ways to ease the burden of managing the disease for the people who are affected by it. This includes developing new informatics solutions for optimizing diabetes treatment and novel tools for diabetes self-management.
Unlike some diseases, diabetes requires individuals to actively engage in self-management. Often, this means making lifestyle changes such as diet modification and introducing regular exercise. However, the varied nature of type 2 diabetes requires that individuals find targeted ways of managing the disease that not only keep their blood glucose levels under control, but also help them maintain quality of life. The American Diabetes Association recognizes this personalized approach to self-management with its recommendations for nutritional management in diabetes. “In a way, each individual with diabetes needs to become a detective looking for clues and solving puzzles”, says Lena Mamykina, an Assistant Professor at DBMI. This process can take years of trial and error, and often leads to frustration and burnout.
Mamykina and her colleagues at DBMI are taking advantage of new self-monitoring technologies that help individuals with diabetes keep track of their meals, physical activity, and blood glucose levels. “Our goal is to find new ways to use these data to discover personal triggers for each individual with diabetes, help them experiment with changes to their diet and different types of exercise, and find their own way of managing their disease.”
Turning patients into detectives
Before individuals with diabetes can take advantage of the data collected with self-monitoring, they need to adopt an investigative mindset and develop problem-solving skills. Working with her co-investigators and Certified Diabetes Educators Arlene Smalldone and Patricia Davidson, Mamykina interviewed dozens of individuals with diabetes, trying to understand how they engage with their disease and develop self-management strategies. As a result of this work, the researchers developed Diabetes Detective, a web-based app that helps individuals with diabetes adopt “detective thinking”.
Diabetes Detective grew out of a collaboration between DBMI, Georgia Institute of Technology, and Clinical Directors Network, with funding by the National Institute of Diabetes and Digestive and Kidney Disease (NIDDK). Using the app, individuals with diabetes log their blood glucose readings at different times of day, particularly before and after meals. The app visualizes daily blood glucose patterns and highlights readings that are consistently higher or lower than ranges considered safe. Then, the app helps users reflect on their behaviors and examine possible reasons for spikes or drops in their blood glucose.
“For some people, the problems come because they include too much pasta or other carbohydrates in their meals, or perhaps not enough protein, or they may skip meals altogether. Or perhaps it has nothing to do with food, but is a result of stress or illness. All these factors can affect blood glucose,” says Dr. Mamykina. The app recommends small and specific dietary and behavioral modifications, helps its users monitor their future blood glucose, and recognize positive response to changes in behaviors.
To develop the app, the team worked together with clinicians and patients of several Community Health Centers in New York City that provide care to economically disadvantaged communities and to ethnic and linguistic minorities with high prevalence of diabetes. Researchers relied on participatory design methods to make sure the app reflects the needs and preferences of its users. Now the team is carrying out a randomized control trial to measure the effectiveness of Diabetes Detective for 240 participants. The study is still ongoing, but the initial results are encouraging. “Some of our participants see dramatic improvement in their daily blood glucose readings. We hope this will translate into lasting benefits for their health,” says Dr. Mamykina.
A personalized glucose forecast
Dave Albers, another DBMI faculty member, is working to incorporate more advanced data science methods into personalized diabetes feedback. “Nutrition and diet are an important part of diabetes self-management,” he says. However, because the same foods can have wildly different impacts on different people or even for the same person in different circumstances, developing nutritional guidelines that work for everybody is not possible. “Among people with type 2 diabetes, there is great variation in the effects of food choices across people and over time. No two individuals are the same,” says Albers. “Each person may have different challenges related to their current metabolic state. For example, insulin production and resistance varies over time even for a single person.” As a result, predicting how an individual’s blood glucose will change in response to a specific meal is a challenging task, even for those who have had diabetes for years. “For many people with diabetes, every meal comes with questions what will happen after I eat it, and what will it do to my blood glucose levels,” says Mamykina. This problem motivated Dr. Albers to partner with Dr. Mamykina to develop the machinery to provide a personalized, carbohydrate-based glucose forecast.
Albers’s training in dynamical systems and mathematical physics allows him to manage this complexity. He and his colleague Matt Levine, a research associate at DBMI with a background in biophysics, are using advanced data science methods and data assimilation to forecast individuals’ reactions to nutrition based on their past glucose measurements and carbohydrate consumption. “Our approach is similar to weather forecasting we integrate mechanistic modeling with observations to form a glucose forecast, using methods similar to those meteorologists use to predict weather,” says Albers. “But here, we can not only predict the future glucose level based on carbohydrate consumption, but more importantly give people tools to change their glucose levels by making different nutrition choices based on the glucose forecasts.” The early results of the models are impressive. In a small pilot study conducted at DBMI, the models matched or outperformed trained diabetes educators in their ability to predict how particular meals would affect blood glucose levels of individuals with type 2 diabetes.
The next step in this work is to use predictions generated by the models in personal decision support tools. “Imagine a tool that allows you to snap a picture of a meal and then not only tells you what your blood glucose is likely to be after you eat this meal, but also helps you to tweak the meal and see the change in the projected post-meal blood glucose,” says Mamykina. “Such a tool could be a real life-changer for many people with diabetes.”
Social tools for managing nutrition
Learning from one’s own history is a powerful way to improve self-management. Learning from others is another. “People often learn vicariously, by observing others, their actions, and outcomes of these actions,” says Mamykina. In their latest work, the researchers are exploring new frontiers in nutrition management that harvests the power of observational learning. “In our recent study, we asked participants what apps they use for nutrition management,” says Mamykina. “The two apps that came on top of the list were Instagram and Pinterest, neither of which were designed for health or nutrition.” Using these social computing apps, people upload pictures of their meals and look at pictures uploaded by others to get new insights and ideas. Many subscribe to nutritional photo blogs and create personal libraries of meals to try in the future.
Now, Mamykina and her colleagues are developing a new breed of social computing platforms for social management of nutrition. “We want to help people learn from each other, and learn about nutrition in a group,” she says. The project is funded through the EAGER grant from the National Science Foundation, a special funding mechanism for risky research with high potential impact. “You know you are out there when even NSF thinks your research is risky,” jokes Mamykina. The project is just ramping up, so the researchers are looking for new minds to join their groundbreaking diabetes research effort.