Virtual Standardized Patients

Virtual Standardized Patient (VSP) simulations allow for a unique approach to assessing essential skills such as history taking and clinical decision making. Our Virtual Patients are avatars representing human standardized patients that students can interview using natural language. Students take a medical history and develop a differential diagnosis of the VSP, much as they would a standardized or actual patient. The distinguishing feature of our VSP system is the artificial intelligence (AI) that controls the conversation between the student doctor and the virtual patient avatar. The VSP understands typed or spoken questions and responds with complaint-specific answers. The system provides immediate feedback on student performance allowing students to rehearse professional behaviors and interviewing skills prior to working with real patients.

Here are some of our ongoing projects using our Virtual Standardized Patient technology.

Virtual Patients to Assess Information Gathering and Clinical Reasoning in Medical Education
Douglas Danforth Ph.D., Kellen Maicher M.F.A., Alan Price M.F.A., Michael White Ph.D., Eric Fosler-Lussier Ph.D., William Schuler Ph.D.

This project was initially funded by the Edward J. Stemmler Foundation. The project was designed to create a Virtual Standardized Patient (VSP) system that would enable students to practice their history taking skills and receive immediate feedback. VSPs are created in Unity and the dialogue is controlled by ChatScript, a pattern matching (rule based) natural language program. Students take a patient history, including history of present illness, past medical history, family history, and social history. At the conclusion of the encounter, the system provides students with feedback on their questions along with an expert’s approach to the case.

Towards a Conversational Assistant for Patient Prep
Michael White, Ph.D., Douglas Danforth Ph.D., William Schuler, Ph.D., Subhankar Chakraborty, M.D., Ph.D., Eric Fosler-Lussier, Ph.D.

This new project is designed to create an automated conversational assistant that can help patients properly prepare for procedures such as colonoscopies. We will use our existing expertise in Virtual Patient natural language dialogue systems and take advantage of recent advances in developing task-oriented dialogue (TOD) systems that make use of large-scale, pre-trained neural language models (NLMs) to bootstrap a baseline dialogue system for conducting patient prep conversations.

Virtual Reality: A Unique Means to Teach the Reality of the Patient-Centered Medical Home.
Douglas Danforth Ph.D., Kellen Maicher M.F.A., Milisa Rizer M.D., Laurie Belknap D.O., Allison Macerollo M.D., Holly Cronau M.D., Douglas Post Ph.D.

This project was sponsored by the Department of Health and Human Services Health Resources and Services Administration (HRSA). The two primary goals of the project were to create a student version of our Electronic Health Record, and to create a cohort of Virtual Standardized Patients presenting with a variety of chief complaints. E-Learning modules were created describing management and treatment of common complaints including how they would be addressed in the PCMH setting. At the conclusion of each module, students interviewed a VSP and then documented the encounter in the EHR. Approximately 10 different VSPs were created, presenting with common complaints such as back pain, abdominal pain, chest pain, etc.

Using Automatically Generated Paraphrases and Discriminative ASR Training to Author Robust Question-Answering Dialogue Systems
Michael White Ph.D., Eric Fosler-Lussier Ph.D., William Schuler Ph.D., Douglas Danforth Ph.D.

This National Science Funded project focuses on the natural language understanding component of our Virtual Standardized Patient system. Our VSPs use a combination of pattern matching rule based software (ChatScript) and natural language understanding (NLU) systems trained using machine learning (ML). A common challenge using ML is that adequate training requires a very large number of sample inputs. Typical VSP dialogue encounters generate a relatively small number of dialogues on which to train. This project is designed to explore novel approaches for generating paraphrases for training our VSP NLU systems.

Creating the Complete Virtual Standardized Patient: Integrating Natural Language Ability into Clinical Reasoning Cases to Assess Information Gathering and Clinical Reasoning
Douglas Danforth Ph.D., Kellen Maicher M.F.A

This project funded by Aquifer INC (formerly Med-U), is designed to combine our dialogue driven VSP system with the clinical reasoning cases provided by Aquifer. The primary goal is to asses whether obtaining the virtual patient’s medical history using active learning (asking questions) as opposed to being given the history (passive learning impacts the learner’s clinical reasoning and performance on the case. A “Complete Virtual Standardized Patient” system may allow students to practice their history taking and clinical reasoning skills in a single encounter.

Artificial Intelligence based Virtual Reality (VR) Simulation of Provider-patient interaction to enhance cultural competency using a simulated patient with limited English proficiency.
Douglas Danforth Ph.D., Kellen Maicher M.F.A., Sheryl Pfeil, M.D., Camilla Curren M.D., Michael White Ph.D., Eric Fosler-Lussier Ph.D., William Schuler Ph.D.

This project funded by the Department of Medicaid is designed to create a patient with limited English proficiency. Gathering information from an immigrant or refugee who has difficulty understanding questions is significantly more challenging than simply taking a medical history from a native speaker. Enabling Medicaid providers to practice their medical interviewing and history taking with a patient for whom English is not their first language, and to receive immediate feedback on their encounter, is a novel approach towards achieving cultural competency.

To maximize the immersiveness of the interaction, and provide a more realistic and authentic environment, the encounter will be simulated using Virtual Reality so the trainee is virtually in the room with the patient. Using VR we can readily simulate a doctor’s office, hospital setting, or patient home depending on the type of encounter desired.

Virtual Patients for Medication Assisted Treatment of Opioid Use Disorder.
Douglas Danforth Ph.D., Kellen Maicher M.F.A, Kristen Rundell, M.D.

This project, funded by the Department of Health and Human Services Health Resources and Services Administration (HRSA), is  focused on the specifics of Medication Assisted Treatment MAT of Opioid Use Disorder, including information and resources for obtaining a data waiver for each of the medical disciplines. As part of this project, we are creating a Virtual Reality (VR) Simulation of Provider-patient interaction to simulate the initiation of MAT therapy in a multidisciplinary team.

Virtual Standardized Patients for End of Life Care
Kellen Maicher M.F.A., Douglas Danforth Ph.D.

This project is in the early design stages and will involve the creation of Virtual Standardized Patients to allow health care providers to practice difficult conversations with a patient at the end of life.