Virtual reality simulations reveal potential biases in patient care

Sonja Soo: Can you tell me more about the background of the study?

Susan Persky: Researchers are increasingly using genomic data in the clinic for diagnosing rare health conditions. But for common health conditions, information such as polygenic risk scores are not currently as accurate at predicting risk. Various direct-to-consumer companies are making these risk scores available to anyone. But researchers know very little about how these new polygenic risk scores will be used or received in medical clinics.

For this study, we wanted to predict how these polygenic risk reports will be applied in the healthcare system and understand whether introducing this information will influence how a doctor might view a patient and make medical recommendations.

We know from previous research that primary care providers generally don’t have a lot of experience with genetic and genomic information. When there is uncertainty in a situation, people typically lean back on their mental shortcuts, such as biases, heuristics and stereotypes. One area where that is applied often in medicine is in racial categorization of patients. So, with this study, we wanted to know what happens when polygenic risk information is brought to a medical clinic and how that is affected by the patient’s apparent race.

Soo: What are some of the key findings from this study?

Persky: Medical students who received these polygenic risk reports made different recommendations to the virtual patient based on the additional genomic information. For example, medical students are more likely to recommend observing kidney function and screening for diabetes in response to the patient’s symptoms when they had the genomic risk information.

These medical students also viewed the patients with polygenic risk information as less healthy and in need of more care. We’ve seen that in the past, in which genomic risk information is thought of almost as a pre-diagnosis.

Lastly, we saw differences in clinical recommendations based on patient race. For example, we found that participants were more likely to recommend assessing diabetes when the patient was Black, which is in line with racial disparities in diabetes rates in the United States.

In a couple cases, the polygenic risk scores seemed to be used differently depending on the patient’s race. For example, medical students were more likely to recommend monitoring kidney function when taking over-the-counter medication if the patient was white and had the genomic risk information available.

Soo: What does it mean to have these unintentional biases with genomic data in healthcare?

Persky: We all need to be aware of our biases in our day-to-day life. For the medical student participants, their daily work involves learning how to make medical decisions. It’s important to consider any assumptions that we make about another person and how that can influence our thoughts and decisions.

But, like most things in healthcare, the responsibility is not solely towards the individual. Educating healthcare providers is part of it, but we are going to need systemic approaches. For example, we can consider how we are communicating this genomic risk information to healthcare providers and how to help them use it equitably. Our knowledge of how polygenic risk information is used in the clinic is not at the stage where we can make specific recommendations, but our study has found potential pitfalls and places where information isn't — but should be — used equitably.