Training virtual patients to interact with learners

FEB 13, 2020 

Chatting with robots has become pretty common place whether it be over the phone when you are checking a credit card balance or typing a question to a chatbot on a company’s website. With advances in artificial intelligence (AI), you may not even realize you are talking to a robot in some cases. This type of technology is starting to be used in health care education by way of virtual patient programs.

Once it’s perfected, it could be a game changer for the way we train our medical students. As of now, we mostly use lecture followed by simulations with standardized participant actors to prepare learners for certain scenarios, such as delivering bad news. And we use manikin-based simulations for resuscitation training. These are still the gold standard of experiences students can receive, but a lack of resources makes it difficult to give everyone that opportunity.

We are testing the use of AI-enabled virtual standardized patients in simulation to discover whether we can actually teach these avatars to have meaningful conversations through Natural Language Processing, and if it’s an effective way for students to learn. The idea is not to replace our actors, but to supplement that experience using virtual patients for pre-training purposes. We believe this could improve the time spent working with real people and faculty.

Advance Care Planning

The first type of case we are testing this technology on is Advance Care Planning (ACP), a model that nurses and social workers use to help people discuss future health care decisions with their families and care team. Jump Simulation assists in training facilitators how to compassionately discuss care options available to patients using standardized participants, people who have been highly trained to portray real patients, their families, or members of the care team to provide realistic clinical encounters.

This program has been successful in not only positively shaping ACP facilitators, but in improving the documentation of end-of-life wishes. However, we believe use of virtual standardized patients could help learners better understand the basics of the training model before having direct conversations with our real actors.

So, instead of working on fundamentals with the actors, such as “What form is used for that?” “What topic needs to be covered?” or “What order should I ask these questions in?”--they can delve into the subtleties of their discussions, such as being able to figure out whether a patient’s selected power of attorney can serve the role effectively, or whether their goals of care mesh with a palliative approach.

Over the last year, our education and research teams have worked on building a basic ACP simulation case into a virtual standardized patient program. Then we’ve had faculty and small groups of learners interact with the virtual patient over standard flat screens and head-mounted displays. They had a chance to ask questions in a variety of ways in hopes of helping us find issues, and to build the language skills of the software.

What we’ve learned so far is there is still a lot of work to do. The virtual patients can only answer what’s programmed in the software. And learners can’t ask the computer complex questions. We also want to overcome some technical barriers of determining which headset works best, how big the computer screen needs to be, how much orientation learners need and how to deliver post event surveys.

Once we have finalized those details and the training of our virtual standardized patients, we will be ready to test this capability with our learners and ask whether it improves performance.

Potential benefits

Because so much of communication relies on subtleties of body language and tone of voice, virtual patients are unlikely to ever replace humans as patient or clinical actors, but they can walk novice learners through the steps of a conversation.

If we can train virtual standardized patients to have useful conversations with our learners, we believe we can free resources to more easily spread some of the in-person simulations we do. That includes the ability to practice consenting patients for procedures, basic customer service, giving discharge instructions or delivering bad news.

We have thousands of Mission Partners who could take advantage of improving their conversation skills with patients by simply accessing this program over their computers from anywhere within the Ministry. They could repeat cases as many times as they need, review how they performed, look at their transcripts and get automated feedback when they submit their results.

We see this as an opportunity to make better use of our faculty and standardized participants and to reach more learners throughout the organization. This means a better one-on-one experience for the patients we serve no matter where they are located.

William Bond, MD is the Director of Simulation Research at Jump and is instrumental in shaping the research agenda. Dr. Bond is committed to many areas of patient safety, improving health care for professionals and supporting the structure of the work surrounding simulation at Jump.