How AI Can Help Patients with Diabetes Lose Weight

Matt Hoffman
NOVEMBER 08, 2017
lark,artificial intelligence,michigan state,healthcare analytics newsArtificial intelligence (AI) is showing promise in yet another corner of healthcare. This time, researchers said, a conversational tool appears adept at helping patients with type 2 diabetes (T2D) shed excess pounds.

Natalie Stein, MPH, MSc, a professor in Michigan State University’s college of human medicine, led a study that examined how an automated text-based coaching service affected 239 patients with T2D who were overweight or obese (body mass index [BMI] ≥25 kg/m2). On average, patients lost 2.38% (standard error, 0.69%) of their baseline weight, according to the findings.

Nearly 70% of American adults have prediabetes or are overweight and at risk for T2D. The country, meanwhile, is up against a lack of healthcare providers. That means face-to-face behavioral counseling isn’t on the table for everyone.

The new technology, called the Lark Weight Loss Health Coach AI, emulates health counseling typically performed by professionals. The app includes features like a dashboard with the ability to monitor activity, sleep, food, weight, and notes.

In sample conversations, the bot asked questions and replied to statements.

In 1 case, it asked a user, “What did you have for breakfast?” Depending on the answer, the health coach’s response offered either positive reinforcement or constructive criticism and tips on how to eat better. For example, when a user claimed to have eaten eggs, bacon, toast, and coffee, the AI listed how many times that individual ate bacon over the past week and recommended a leaner source of protein. But when the user replied, “Chicken, veggies, and rice,” the bot praised the healthy choice.

Although the study started with 239 people, 159 installed the app and entered baseline height and weight data. Then, 83 recorded a second weight, and 75 did not. Researchers targeted 70 participants who appeared “active” enough to analyze, deeming another 13 “inactive,” according to the study.

Throughout the study period, users averaged 103 sessions, and 75.7% of them (n = 53) experienced weight loss. The mean decrease was 2.4 kg, while the number of healthy meals eaten increased by 31% at 21 weeks, according to the report.

An in-app trust survey returned a 100% response rate, achieving 7.9 out of 10 in average user satisfaction with the program, a disappointment level of 8.3 if not offered the program, and 6.73 in their satisfaction with the outcome.

Together, the study results suggested that health coaching AI is “available for real-life applications,” Stein and co-author Kevin Brooks, MS, PhD, noted.

From here, researchers plan to study how the AI coach can boost diet and other components of chronic disease management, including hypertension and prediabetes.

This, of course, is but AI’s latest entry into healthcare. Along with patient-facing roles, it is also enhancing clinical trials, academic research, and even physicians. AI’s potential to become a predictive powerhouse has also sparked much conversation. But, as experts have said, it might take some time before Amazon’s Alexa begins doling out medication information.

Images have been resized. Courtesy: Stein N, Brooks K
A Fully Automated Conversational Artificial Intelligence for Weight Loss: Longitudinal Observational Study Among Overweight and Obese Adults
JMIR Diabetes 2017;2(2):e28
DOI: 10.2196/diabetes.8590


Become a contributor