Using virtual reality to investigate bias and discrimination in clinical decision making: A proof of concept study
1Department of Psychological Medicine, Institute of Psychiatry, Psychology & Neuroscience, King’s College London, United Kingdom
2Centre for Longitudinal Studies, University College London, United Kingdom
3Department of Psychology, Institute of Psychiatry, Psychology & Neuroscience, King’s College London, United Kingdom
4NIHR Maudsley Biomedical Research Centre, South London and Maudsley NHS Foundation Trust, London, United Kingdom
5ESRC Centre for Society and Mental Health, King’s College London, United Kingdom
†Corresponding author: Helen Welsh, Department of Psychological Medicine, King’s College London, 16 De Crespigny Park, London, SE5 8AF, United Kingdom, helen.e.welsh@kcl.ac.ukAbstract
Background
Biases held by healthcare practitioners can shape clinical interactions, leading to discrimination and poor patient outcomes. Traditional methods, such as vignettes or self-report measures, often lack ecological validity.
Objective
We evaluated a novel VR simulation designed to explore bias and discrimination in clinical decision-making.
Methods
Thirty-five healthcare practitioners across 14 NHS trusts completed an in-person VR simulation and provided feedback on realism, immersion, and usability via surveys and open-ended responses.Patients in the simulation had intersecting sociodemographic characteristics (race, gender, migration status).
Results
Most participants rated the VR environment as at least moderately realistic (88.6%), with 42.9% rating it very or extremely realistic. Nearly half (45.7%) reported consistency with real-world experience. Open-text feedback highlighted strengths in realism, immersion, and educational potential, with suggestions to improve dialogue flexibility.
Conclusions
VR is a feasible, immersive, and scalable tool for investigating bias in clinical decision-making.
Graphical Highlights
- Virtual reality (VR) is feasible for studying bias in clinical decision-making.
- VR simulations capture intersecting patient characteristics, including race, gender, and migration status.
- Participants rated the VR environment as realistic, immersive, and usable.
- VR provides a scalable, controlled platform for research and professional training on bias and discrimination.
Article notes
Competing Interest Statement
The authors have declared no competing interest.
Funding Statement
This paper represents independent research funded by the Wellcome Trust [203380/Z/16/Z] and the Economic and Social Research Council [ES/V009931/1]. J.O. is part-funded by the NIHR Biomedical Research Centre [BRC121520018] at South London and Maudsley NHS Foundation Trust. S.L.H. and R.R. are supported by the Economic and Social Research Council Centre for Society and Mental Health at King's College London [ES/S012567/1] and by UK Research and Innovation [MR/Y030788/1] as part of Population Health Improvement UK (PHI-UK), a national research network that seeks to transform health and reduce inequalities through change at the population level. S.L.H, R.R, J.O and H.W are supported by the Wellcome Trust [308556/Z/23/Z], S.L.H. currently receives funding from Impact on Urban Health part of Guy's & St Thomas' Foundation [EIC210605 and EIC221208], the Swedish Research Council [2023-05959] and the Wellcome Trust [28117/Z/23/Z & 223486/Z/21/Z]. The funders had no involvement in study design, data collection, analysis, interpretation or the decision to submit for publication. The views expressed are those of the author(s) and not necessarily those of the funders.
Introduction
Background
In healthcare settings, biased attitudes and beliefs held by practitioners towards different groups of patients can shape clinical interactions, outcomes, and levels of service satisfaction. This can also lead to direct and indirect acts of discrimination. In England and Wales, discrimination is defined as treating a person unfairly because of characteristics protected by the Equality Act 2010, which, in a healthcare context, can result in poor care outcomes for patients [1–3].
Research on identifying and reducing clinician bias and discrimination in healthcare has largely relied on self-report questionnaires and clinical vignettes [3, 4]. These methods lack ecological validity (i.e., the degree to which a study’s methods reflect the complexity of real world situations) and are unable to capture the full range of clinician–patient interactions that occur in practice. While some studies have attempted to address these limitations using video-based cases and role-play scenarios, these approaches remain limited in its ability to standardise patient behaviour across all aspects of encounters [5, 6].
Traditional approaches to clinical education and bias training through simulated patients, role-play, and 2D video case studies also present limitations. Live simulation provides high fidelity but is resource-intensive and costly to scale [7,8], whereas 2D video lacks immersion and interactivity. Virtual reality (VR) offers a middle ground: studies suggest it can achieve comparable learning outcomes to high-fidelity simulation but at lower cost, while enhancing engagement and immersion compared with video-based learning [9 -12].
Bias and discrimination have typically been investigated using written vignettes, standardised patients, and role-play exercises [13,14]. While valuable, these methods often lack scalability and ecological validity. Emerging applications of VR in health professions education have demonstrated promise for enhancing realism and directly influencing bias related attitudes and behaviours. For example, embodiment in avatars of different racial backgrounds has been shown to reduce implicit racial bias [15, 16, 17]. This underscores the potential of VR as a novel platform for both investigating and addressing bias in clinical decision-making.
VR may be particularly advantageous for studying socially undesirable or harmful behaviours, such as discrimination, since well-designed simulations can evoke responses that closely mirror real world behaviour [18]. VR can also be used to examine how intersecting sociodemographic characteristics (e.g., gender, race, ethnicity, and class) shape clinical interactions. These intersecting characteristics influence not only how clinicians perceive and interact with patients, but also the treatment recommendations they make [19, 20].
Despite these benefits, VR remains underutilised in non-experimental social science studies of bias and discrimination. Given the potential of VR to generate ecologically valid evidence on this pressing source of health inequality, it is important to establish the feasibility of this method for researching bias and discrimination in clinical decision-making.
Aims
The Tackling Inequalities and Discrimination Experiences in Health Services (TIDES) study developed a VR simulation to address limitations of existing methods for studying bias and discrimination. The overarching aim of TIDES is to understand how witnessed and experienced discrimination contributes to health inequalities. This study specifically aimed to evaluate the feasibility of VR in simulating clinical scenarios and gathering participant feedback on realism, usability, and its potential for investigating bias.
Methods
Study Setting
The study was conducted at King’s College London by the TIDES study team (Department of Psychological Medicine) and the VR Research Lab (Department of Psychology). A trained researcher facilitated VR sessions, supported by a VR lab developer who managed software updates and exported headset data.
Participants
The VR study built on the 2019 TIDES survey of harassment and discrimination among staff in London NHS trusts [21]. The survey included 931 healthcare practitioners. Purposive sampling ensured diversity across occupation, gender, ethnicity, and migration status. Participants consenting to be recontacted were invited to join the VR study. Exclusion criteria included seizure or vestibular disorders or uncorrected hearing or vision issues [22, 23].
VR Equipment
The simulation, built by Dan Archer of Empathetic Media using Unity3D, ran on Oculus Quest headsets. Participants interacted using a hand-held controller.
VR Simulation Development
The VR simulation was developed iteratively with the TIDES team, clinicians, and VR specialists, following guidelines for immersive VR in cognitive neuroscience [24]. Realism was a central goal: the simulation recreated a primary care consultation room from a first-person perspective, with interactive elements such as virtual hands and a keyboard (Figure 1). Dialogue trees co-developed with clinicians comprised 32 questions across eight categories.
Patient avatars reflected intersecting sociodemographic characteristics, including gender, race, and migration status (Figure 2). Intersectionality was considered so participant decisions could be observed relative to overlapping social identities. The design balanced technical fidelity (graphics, responsiveness) with social and cognitive realism (interaction with avatars, clinical decision-making) [25, 26]. Participant feedback guided refinements in conversational flow, empathy, and usability.
Procedure
Participants were first provided with a digital information sheet and could ask questions via email before participation. Those interested attended a King’s College London campus of their choice, where they received a paper copy of the information sheet and consent form. The researcher reviewed these documents with each participant, answered questions, and emphasised that data would be stored securely at King’s College London and accessed only by the TIDES team. Participants who agreed signed a digital version of the consent form.
They then completed a short sociodemographic survey on a tablet, covering age, education, household composition, job title, NHS Trust, occupational group, pay band, income, years in healthcare, patient contact, and NHS Equality and Diversity training (Supplementary Table 1). Participants received a brief description of each virtual patient (Figure 2) and estimated expected consultation length. Meanwhile, the researcher sanitised and prepared the Oculus Quest headset and controller. Because the headset obscured physical surroundings, care was taken to ensure a safe environment free of obstacles. All participants remained seated throughout the simulation.
After adjusting the headset for comfort, participants completed a short tutorial to learn the controller and navigation. The main simulation lasted 30–45 minutes. Each virtual consultation was capped at ten minutes, displayed on a virtual wall clock (Figure 1). Participants assessed each patient and recorded treatment recommendations using a virtual dictaphone.
Following each consultation, participants completed a short in-VR survey rating the patient’s personality, probable behaviour, social roles, and overall health. After the simulation, the researcher assisted participants in removing the headset, which was sanitised again.
Participants remained seated briefly to readjust and were offered refreshments to mitigate potential motion sickness [28]. Finally, they completed a feedback survey on a tablet, taking approximately five minutes.
Measures
The feedback survey consisted of the following three questions which participants responded to using a five-point Likert scale:
- “How real did the virtual world seem to you?” (0 = Not real at all to; 4 = Extremely real)
- “How much did your experience in the virtual environment seem consistent with your real-world experience?” (0 = Not consistent; 4 = Extremely consistent)
- “How aware were you of the real-world surroundings while navigating the virtual world? (i.e. sound, room temperature, other people, etc.)?” (0 = Not aware at all; 4 = Extremely aware)
Analysis
Descriptive statistics were calculated to describe participants’ responses to the three quantitative items in the feedback survey. Analyses were conducted in Stata 17 [29].
Ethical Approval
Ethical approval was granted by the King’s College London Research Ethics Committee for Psychiatry, Nursing and Midwifery (HR-17/18-4629) and NHS Health Research Authority (18/HRA/0368).
Results
Sample Characteristics
A total of 35 healthcare practitioners across 14 NHS trusts participated. Twenty participants were qualified nurses, seven were medical doctors, and the remaining were allied health professionals, including therapists, psychologists, and pharmacists. Participants’ salary bands ranged from band 5 (newly qualified staff) to band 8 (managers with over five years’ experience). Twenty-seven participants were female, and the remainder were male. Ethnic backgrounds included White British or Irish (14 participants), Black African or Asian ethnic groups (12) and other White ethnic groups (9 participants). The mean age of participants was 35.3 years.
Realism, consistency with reality and immersion
As shown in Figure 3, most participants (88.6%) rated the VR simulation as at least moderately real, with 42.9% describing it as very or extremely real. Nearly half (45.7%) reported that the simulation was very or extremely consistent with their real-world experience, while 51.4% reported being only slightly or not at all aware of their real-world surroundings during the task, suggesting a high degree of immersion.
Open-ended feedback on technical aspects
Participants’ qualitative feedback emphasised usability and technical feasibility (Textbox 1). No participants reported motion sickness or physical discomfort. Many described the simulation as realistic, and most found the overall length appropriate for a training exercise. Some reported initial difficulty using the virtual dictaphone to record treatment recommendations, but these issues were quickly resolved with researcher assistance. This highlights the importance of brief, structured tutorials for maintaining usability in VR based studies.
Open-ended feedback on content
Feedback on the VR simulation content was broadly positive, with participants noting that consultation questions were clear, comprehensive, and clinically appropriate (Textbox). Nonetheless, several participants identified areas for improvement. Some reported that fixed, pre-determined questions felt restrictive, limiting opportunities for follow-up or elaboration and constraining more natural consultations. Others had difficulty navigating question sets, commenting that not all prompts were visible within each category, occasionally causing repetition and disrupting flow. A further challenge concerned treatment recommendations; participants found it difficult to decide without additional diagnostic information, such as test results. This was particularly evident among nurses in specialist roles, who felt independent recommendations were unrealistic compared with their usual practice.
Taken together, these points suggest future iterations could benefit from more complex dialogue trees, improved navigation to display all options, and supplementary clinical data to support decisions. Matching participants to patients relevant to their professional expertise may also enhance ecological validity.
Analysis of open-ended responses further highlighted several themes. Usability was prominent, with most participants finding the system intuitive, though a minority required clarification on tasks like the virtual dictaphone. Realism and immersion were frequently mentioned; participants consistently described the environment and avatars as realistic, though greater conversational flexibility would strengthen authenticity. Another theme concerned empathy and communication, as fixed response options limited opportunities to demonstrate sensitivity and rapport. Finally, participants emphasised the simulation’s potential as a training tool for raising awareness of bias and discrimination in healthcare decision-making.
These findings reinforce the feasibility of VR for examining bias in clinical contexts while pointing to refinements needed to balance ecological and cognitive validity [30].
Textbox 1. A subset of direct quotes from open-ended feedback provided by participants
Technical aspects of the VR simulation Positive
Negative
Content of the VR simulation Positive
Negative
Boxed Text
Discussion
VR remains underutilised in research examining socially undesirable behaviours, such as bias and discrimination, within clinical decision-making. Compared to self-report questionnaires, vignettes, and role-play, VR enables the creation of controlled yet realistic scenarios where patient behaviour and the clinical environment can be standardised. Unlike role-play with actors, VR is less resource-intensive and more scalable, while offering greater immersion and ecological validity than 2D video-based approaches [12]. VR occupies a middle ground between low-cost but limited modalities and high-fidelity but costly live simulations [7–10].
The aim of this study was to evaluate healthcare practitioners’ feedback on a novel VR prototype developed to explore bias and discrimination in clinical encounters. Thirty-five participants rated the simulation as realistic and immersive, supporting its feasibility for investigating bias. While not designed to directly measure bias, the study demonstrates that VR can elicit meaningful reflections on realism, usability, and empathy, essential precursors for developing tools for bias research and training.
A key contribution of this work lies in the development process. The simulation was designed with attention to realism in patient presentation and decision-making prompts, as well as intersectionality in patient identities and contexts. These design choices are critical, as the effectiveness of VR depends not only on technical fidelity but also on capturing the social and interpersonal dynamics of practice [25]. By foregrounding equity and intersectionality, VR can extend beyond technical skills training into complex domains such as bias, discrimination, and healthcare inequalities [14,15,17].
The findings also point to VR’s potential as an educational tool. Diagnostic errors in primary and secondary care can have harmful consequences for patients, practitioners, and health services [31,32]. Evidence suggests ethnic minority patients face disproportionate risks of complications and adverse events [33]. VR exposes practitioners to diverse, intersectional patient scenarios safely, helping build awareness of bias and supporting equitable clinical decision-making. Previous research shows implicit bias affects clinical interactions [34,35], but traditional studies often assess single social identities. VR allows interaction with patients possessing multiple intersecting identities, offering new insights into health inequalities.
Despite its promise, the study highlights limitations. Fixed response options constrained dialogue and empathetic engagement, and participants noted challenges navigating questions and making treatment recommendations without additional diagnostic information. These concerns reflect broader design challenges in VR healthcare training: ensuring ecological validity while preserving cognitive validity [26]. Such limitations provide guidance for refinement, including more flexible dialogue trees, supplementary clinical data, and scope-sensitive case assignments.
Strengths and limitations
This study demonstrated the feasibility of using VR to simulate complex clinical interactions while balancing methodological rigour with accessibility. By collecting both quantitative and qualitative data, we were able to evaluate immersion, realism, and user experience alongside practical considerations for clinical decision making. Although not within the scope of this paper, these data also provide opportunities for future analyses of how bias may influence treatment recommendations, as well as conversation analysis of clinician–patient interactions. A further strength was the accessibility of the prototype: the interactive features required minimal training, relying on natural speech and a simple point and click mechanism with the hand-held controller. This simplicity reduced participant burden, minimised the risk of motion sickness, and made the simulation suitable for diverse healthcare professionals.
Several limitations should also be noted. Participants were restricted to asking predetermined questions, since programming naturalistic responses to unrestricted input was not technically feasible within this prototype. This limitation reduced opportunities to observe potential indicators of bias in how clinicians select or phrase questions. Second, participants interacted with only three virtual patients to limit the risk of VR related discomfort with prolonged exposure, which constrained the range of scenarios available for analysis. Third, the rapid pace of VR hardware development posed challenges for implementation. Although the Oculus Quest was appropriate at the time of development, frequent software updates, including those associated with the transition from Facebook to Meta, created practical obstacles. Nonetheless, the simulation was designed with transferable software that remains compatible with newer headsets, supporting future adaptation and delivery.
Further research
Whilst participants in this study rated the VR simulation as highly immersive, future research should build on these findings by expanding both the technical capabilities and the scope of VR simulations. Advances in hardware now allow for greater interactivity, such as manipulating objects or conducting physical examinations, which could enhance ecological validity. Simulations situated in diverse clinical contexts, for example inpatient wards, outpatient clinics, or community services, would enable exploration of how organisational and environmental factors shape decision making. In addition, future studies should examine decision making with patients varying across a broader range of protected characteristics, such as disability, age, or sexual orientation, and presenting with both physical and mental health conditions. Such work would extend the utility of VR for investigating bias and inform the design of training interventions that prepare clinicians to deliver equitable care across diverse patient populations.
Conclusions
This proof of concept study demonstrates the feasibility and potential value of using VR to simulate complex clinical encounters for examining bias and discrimination in healthcare. Participants reported high levels of realism, immersion, and usability, indicating that VR can provide a controlled yet authentic platform for exploring clinician–patient interactions and eliciting meaningful feedback on decision making processes. By incorporating intersectional patient identities, the simulation also highlights how multiple social characteristics can influence clinical reasoning, offering a novel approach to studying bias that complements traditional methods such as vignettes, role-play, and 2D video.
The development process, grounded in collaboration between clinicians, psychologists, and VR specialists, ensured that both technical and social realism were central to the simulation. Feedback from participants further identified opportunities for refinement, such as greater flexibility in dialogue, enhanced access to clinical information, and better alignment of patient scenarios with practitioner expertise. These insights underscore how VR not only serves as a research tool but also has potential as an educational and training platform for healthcare professionals, supporting the development of skills in equitable decision making and empathy.
Future research can build on this foundation by expanding the scope and interactivity of VR simulations, diversifying clinical contexts, and including patients with a wider range of protected characteristics and health conditions. As VR technology becomes increasingly accessible and sophisticated, such applications have the potential to enhance evidence-based interventions, improve training in bias reduction, and ultimately contribute to more equitable healthcare delivery.
This study provides strong evidence that VR is a feasible and innovative method for investigating bias and discrimination in clinical decision making. By combining immersion, interactivity, and intersectional patient representation, VR offers a scalable, ecologically valid, and practical tool for both research and professional training in healthcare settings.
Data Availability
Data is available for research purposes upon request. All requests are reviewed by the study data committee. To apply for access to this data please contact care_hsc@kcl.ac.uk
Acknowledgements
This study was conducted by the TIDES team members and colleagues at the VR Research Lab, King’s College London. We thank Luke Connor and Ruth Mintah for their assistance with developing the study protocol and draft manuscripts; Kate Rimes, Sarah Dorrington; and Catherine Polling for their clinical expertise during the script writing phase; Dan Archer from Empathetic Media for building the simulation; and Jerome Di Pietro for technical support during the recruitment phase.