to patients. Still, most health professionals have extremely limited time to spend with patients.

By emulating human relationship-building behavior to create and maintain a trusting therapeutic alliance, virtual characters can maintain engagement for longitudinal inventions. Bickmore and his team simulated the patient–provider interactions they observed with as much fidelity as possible in automated health care providers. Patients need to like and trust a virtual character so they will keep talking to it over time and follow its recommendations. Thus, the characters rely heavily on nonverbal cues such as facial displays of empathy, simulating closer proximity, orientation toward the patient, more facial animation, more direct gaze or smiling, and other cues that would be difficult or impossible to create in a text-based system. Bickmore and his colleagues have even conducted studies where the character has a human backstory, and patients react positively and log in more frequently to home-based interventions when the character has this human dimension (Bickmore et al., 2010).

Virtual Discharge Nurse

Bickmore demonstrated such interactions with an automated virtual discharge nurse developed with clinicians at Boston Medical Center (Bickmore et al., 2009). A computer that contains information about a patient is wheeled next to the patient before discharge. The patient then has a half-hour conversation with the virtual nurse about self-care procedures at home. The nurse talks using synthetic speech, and the conversation is dynamically assembled from the patient’s medical record. The simulated nurse has hand gestures, gaze cues, and body posture shifts that are synchronized with speech. The patient interacts with the nurse by choosing options on a touchscreen. Patients can be trained to use the system in seconds, after which they can use it on their own.

The system has been used with about 200 patients in a safety-net hospital where patients typically have low levels of computer and health literacy. When asked whether they would rather have their discharge instructions from the automated nurse or a real nurse, 70 percent choose the automated character, saying that it provides them with a more relaxed environment to get the information they need and have questions answered. Indeed, the highest levels of acceptance are in patients who have low computer health literacy.

Health Behavior Change Interventions

With a variety of collaborators, Bickmore’s team has used the system for various health behavior change interventions, from medication adher-



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement