Developing an automated virtual health coach

An interdisciplinary group of researchers at the University of Illinois at Chicago has received a grant from the National Science Foundation to create an SMS-based virtual health coach that would help overweight people set and achieve healthy eating and exercise goals. The $1.2 million, four-year grant will enable the development of an automated virtual health coaching system whose benefits would be accessible to a much larger population than those currently using human health coaches — whose services can be expensive — to help them achieve their exercise and weight-loss goals.

The virtual health coach would interact with clients via text messaging.

A health coach helps clients identify their health goals and develop strategies to achieve those goals. The health coach also often plays a supportive role, helping clients stay on track and troubleshooting problems if they arise. While most health coaching is done in person or over the phone, the personalized attention can be very expensive. Currently available automated SMS- or text-based health coaching is mostly one-sided. For example, a personal health tracker may simply send random texts to take more steps, or congratulate the user on reaching a daily step goal.

“A true automated health coach that can appropriately set goals and help clients reach those goals while providing encouragement would be a first and would help the service reach more people,” said Brian Ziebart, associate professor of computer science in the UIC College of Engineering and principal investigator on the grant.

Ziebart, together with his co-investigators from the UIC College of Medicine and College of Pharmacy, will base their automated virtual health coach on data gleaned from texts sent between patients and their human health coaches.

“We will use natural language interpretation and sentiment analysis to develop an autonomous health coaching system that learns to automate an increasing amount of coach-patient interactions for setting goals,” Ziebart said. “The system will learn directly from the human health coach’s interaction with participants, but when there is uncertainty on how to respond, it will notify the human health coach, who can intervene and communicate appropriately with the patient. The system would then refine its internal models for dialogue and coaching strategies so that it better imitates the human health coach.”

To generate the initial texts for analysis, the researchers will recruit 30 overweight to obese individuals who receive care from the University of Illinois Hospital & Health Sciences System (UI Health). Over an eight-week period, participants will receive Fitbit tracking devices to self-monitor their physical activity through steps taken. In addition, they will text with an assigned health coach about setting goals that are specific, measurable, attainable, relevant and timely, or SMART.

The researchers will use a unique texting platform developed for health behavior interventions by Dr. Ben Gerber, professor of medicine in the UIC College of Medicine, and Lisa Sharp, professor of pharmacy systems, outcomes and policy in the UIC College of Pharmacy — both co-investigators on the grant. The app, called mytapp, has been used in more than 14 research studies. Mytapp will allow the researchers to analyze the text messages for specific characteristics that will help them develop their automated virtual health coach, including mood and sentiment, how goals and timing of goals are discussed, and use of informal language.

The researchers will then recruit an additional 120 participants to help them further hone and develop the automated virtual health coach. To participate, the people must be determined to be overweight or obese UI Health patients and will use the new virtual health coach for two to three months.

“Eventually, the virtual health coach will decide when to send messages to participants — either in response to messages received or by taking the initiative and sending messages first. It will also determine when to solicit the input of a human health coach,” Gerber said. However, before the virtual coach can have this level of autonomy, health coaches will screen outgoing messages before they are sent to ensure that no incorrect health advice is provided. Aside from this restriction, the virtual coach will guide communication.

The researchers will evaluate the efficacy of the virtual health coach based on participants’ perceived success in changing health behaviors, the number of steps taken during the study, and participant interviews after the study.

“If successful, the automated virtual health coach can be further adapted to improve other health behaviors among populations where access to health coaches is limited,” Ziebart said. “It might be used to help people quit smoking, follow a specific diet, or adhere to medication regimens.”

Computer science professors Barbara Di Eugenio and Bing Liu in the UIC College of Engineering are co-investigators on the grant.