Educational considerations based on medical student use of polygenic risk information and apparent race in a simulated consultation

Abstract

Purpose:

In order to craft evidence-based educational approaches related to polygenic risk score (PRS) implementation, it is crucial to forecast issues and biases that may arise when PRS are introduced in clinical care.

Methods:

Medical students (n=84) were randomized to a simulated primary care encounter with a Black or White virtual reality-based patient and received either a direct-to-consumer-style PRS report for five common complex conditions or control information. The virtual patient inquired about two health concerns and her genetic report in the encounter. Data sources included participants’ verbalizations in the simulation, care plan recommendations, and self-report outcomes.

Results:

When medical students received PRS, they rated the patient as less healthy and requiring more strict advice. Patterns suggest that PRS influenced specific medical recommendations related to the patient’s concerns, despite student reports that participants did not use it for that purpose. We observed complex patterns regarding the effect of patient race on recommendations and behaviors.

Conclusion:

Educational approaches should consider potential unintentional influences of PRS on decision making and evaluate ways that they may be applied inconsistently across patients from different racial groups.

Introduction

Precision health aims to effectively prevent, screen for, and treat common complex diseases. There is a push toward developing polygenic risk scores (PRS) to quantify genomic risk for these diseases [1]. It is, however, unclear how PRS for common health conditions will function in the clinical sphere, as they are entangled with challenging issues related to health equity, racial identity, genetic ancestry, and clinical utility. Research to address these implementation questions is a high priority as it will pave the way for effective, equitable genomic medicine [2, 3]. To craft evidence-based educational approaches to PRS implementation, it is crucial to forecast what issues might arise in future clinical visits that integrate PRS.

PRS represent crucial progress in the ability to predict certain conditions. For example, previously identified individual genetic variants explain around 6% of the variance in body mass index (BMI), whereas an early PRS-based model predicted 23% of the variance [4, 5]. The highest risk PRS may be equivalent to having a rare high-impact variant and thus may be clinically actionable [5, 6]. While on a population level many individuals will receive high-risk feedback, the current ability of PRS to predict individual risk is generally weak [7].

Clinical use of PRS is expected to increase given continued scientific advancement and motivation to broaden usefulness through enhancing ancestral diversity in the genomic data [8, 9]. In advance of true clinical or public health utility, PRS for common complex diseases are conveyed by the direct-to-consumer (DTC) industry as ‘wellness’ information. For example, 23andme provides PRS for type 2 diabetes wherein individuals in the top 5% have about a 3-fold greater risk of developing type 2 diabetes [10]. Other companies have returned PRS for several conditions such as heart disease, hypertension, obesity, and some cancers, as well as for embryo screening [11, 12]. While it is uncertain if DTC companies will be a primary source of personalized genetic information in the future, market analysts predict their continued growth, especially in areas that overlap with disease prevention strategies such as precision nutrition [13]. Regardless, DTC services have already brought PRS to many individuals and understanding their use in these contexts can inform educational approaches to guide the future use of PRS in healthcare.

While these genomic advances have taken place in research and industry, healthcare has yet to implement PRS as standard of care. Healthcare providers report low levels of understanding about how genomic technologies can be used for predictive risk in common complex disease [14, 15]. Visions of PRS integration into clinical care include provider-facing education and instructions on their applications [16]. Creation of these educational materials should consider heuristics and biases that might arise when PRS are introduced. For example, it may be difficult for some providers to ignore PRS information presented in the clinic regardless of predictive ability or utility for the clinical task at hand [17].

Key potential moderators of whether and how providers consider PRS in clinical care are patient genetic ancestry and racial identity. The utility of race for PRS is part of a larger debate regarding the use of race in medicine [18]. Race is a social construct and not a biological one, so patients’ racial identification is not a good proxy for their ancestral or genetic makeup [19, 20]. Yet many providers report using race as a proxy for genetic factors relevant to health and disease [21].

Many of the conditions targeted by PRS are diseases with health disparities wherein racial and ethnic minoritized groups are at increased risk for disease and have worse outcomes once diagnosed [22, 23]. As such, early detection may be even more valuable. Furthermore, these processes occur within social environments and healthcare systems where structural barriers lead to worse health outcomes for racial and ethnic minoritized patients [24]. However, PRS are currently based on data from mostly European ancestry samples and are thought to be largely irrelevant to individuals from non-European ancestry groups at the present time [25, 26].

The only clear path forward is to bolster the genomic research efforts to represent all ancestral populations in the data that underlie PRS and other precision health approaches poised for clinical translation. At present, however, these intersecting issues of the use of race in clinical care, the ancestry-specific nature of PRS, and documented racial and ethnic health disparities hinder our ability to make a clear prediction on how clinical use of PRS might interact with patient race to influence provider behavior.

Given myriad possible effects of PRS on clinical encounters, particularly in intersection with patient racial/ethnic identity, it is important to explore the consequences of encountering such information in the clinic. Such understanding will enable health professionals to effectively counter any potential ill effects or enhance any positive effects of genomic translation preemptively, through education, so that these technologies will have maximal benefits and minimal harms upon implementation.

We conducted an experiment in a simulated clinical interaction wherein medical students interacted with a virtual patient; some were randomized to receive polygenic risk information similar to those provided by DTC companies (hereafter: PRS) with elevated genetic risk for the virtual patient. Given the crucial and complex questions related to race, ancestry, and health equity that surround clinical use of PRS, we randomized medical students to interact with a virtual patient who appeared as either Black or White. We were therefore able to assess differences that might be introduced in the patient’s care related to her apparent race.

We investigated three hypotheses and an exploratory research question.

Hypothesis 1: A majority of medical student participants who receive PRS-based genetic risk information will report considering and using the information to inform the patient’s care plan; they will also be more likely to report using the information in this way than those who receive control information.

Hypothesis 2: PRS-based genetic risk information (versus control) will influence medical student participants’ views of the patient, such that she is seen as more in need of preventive health measures and is perceived as less healthy.

Hypothesis 3: Medical student participants who receive PRS-based genetic risk information will incorporate this risk information into their consideration of the virtual patient’s specific medical concerns (knee pain and fatigue) such that recommendations differ from those made by participants who receive control information.

Research Question: What is the role of the patient’s apparent race in the above processes? We explore this both as a main effect and as a potential moderator of PRS-based genetic information provision effects.

Methods

Participants

We recruited third- and fourth-year medical students in the Washington, D.C., area. Recruitment strategies included flyers, word of mouth, social media, and email. The planned recruitment goal based on power analysis for primary outcomes was 200 medical students. Due to COVID-19 pandemic restrictions, this project was unable to meet the original goal for the full randomized control trial. Instead, data from the project were assessed for patterns that can guide formulation of educational approaches around PRS implementation. 84 participants were randomized; 3 were dropped from analysis due to technical failure.

Prospective participants were excluded if they had counterindications for virtual reality (VR) use in the current setting including seizure or vestibular disorders, known pregnancy, high proneness to motion sickness, low and uncorrected vision or hearing. They were also excluded if they received key information about the study purpose from a previous participant.

Recruitment ran from August 2019 to March 2020, resulting in a total accrual of 84 participants. Participants received a $100 gift card as compensation. This study was approved by the IRB of the National Human Genome Research Institute.

Procedure

After phone or online screening, enrolled participants scheduled a lab visit at the National Institutes of Health Clinical Center. We consented participants and randomly assigned them to condition using a random number generator. Participants were blindly randomized to one of four conditions based on a full-cross between two independent variables: whether the virtual patient was Black versus White, and whether the participant viewed results from genetic carrier screening alone versus results from carrier screening and PRS-based risk for five common diseases.

Medical student participants were seated in the VR environment where they practiced using the medical record located on a tablet within the virtual clinical exam room (Figure 1). This included the virtual patient’s history of present illness and examination findings. All participants could view a hypothetical genetic report for the virtual patient which contained carrier screening information for sickle cell disease, cystic fibrosis, and Tay-Sachs disease, all of which indicated the patient had no pathogenic variants. Participants randomized into the PRS conditions could additionally view the virtual patient’s PRS. This information indicated greater than average risk compared to the general population for obesity, type 2 diabetes, and kidney disease and average risk for hypertension and heart disease (Figure 1; supplement).

Figure 1.

Clinical environment elements.

The virtual patient spoke a prerecorded script prompted by the research assistant. Depending upon the randomly assigned condition, the virtual patient’s racial appearance was displayed as Black or White (Figure 1). All other aspects of the virtual patient including gender, clothing, voice, and nonverbal behavior were identical. Participants introduced themselves, after which the virtual patient began to explain her medical concerns. Participants could not ask the virtual patient questions. This ensured that every participant received the same information from the virtual patient.

The virtual patient asked four questions: the possible causes of her fatigue, the possible causes of her knee pain, an explanation of her genetic test results, and if there was anything else she should know. The participant responded after each question. See supplemental materials for the full script of virtual patient verbalizations.

Next, a voiceover prompted participants to verbally express any information they would have liked to know during their interaction and if they had any additional reflections. Finally, participants completed a series of computer-based questionnaires.

Materials

We employed a simulated VR paradigm as this method allows experimental control of patient race variables within a realistic clinical setting, holding all else constant [27]. VR approaches are quickly gaining popularity for training and student evaluation in medicine . The VR equipment for this study included an HTC Vive Pro headset and one hand controller. The virtual exam room and interaction (Figure 1) were developed on the Vizard software platform.

The medical record and the genetic reports reflected a female patient in good health with an elevated BMI (see supplement). The patient was chosen to present with a high BMI to put her at somewhat elevated risk for the five common diseases within the report, regardless of genetic predisposition.

Measures

Participants responded in writing to open-ended prompts regarding their use of the genetic information in their decision-making for the patient. Responses were coded by two coders (kappa statistics for agreement reached or exceeded 0.6) using a closed-ended codebook developed by the research team. Coders assessed whether participants reported considering the genetic information the patient provided, integrating the information into the patient’s care plan, and applying it to understand the patient’s specific symptoms.

Transcripts of the participants’ verbalizations during the virtual interaction were also coded by two coders using a closed-ended codebook (kappa levels reached or exceeded 0.6) in terms of whether and how they employed genetic information during the consultation. Coded items included the presence versus absence of discussion about the patient’s weight, weight loss, health behavior recommendations, type 2 diabetes, and kidney disease, as well as discussion of the uncertain, probabilistic nature of genetic risk.

Participants rated the virtual patient’s health on a scale from 1 = “not at all” to 7 = “extremely”. They additionally rated the extent to which they believed that the patient would require strict medical advice [28].

Participants gave open-ended recommendations for the virtual patient regarding her fatigue and knee pain. Recommendations were reviewed and categorized. Those recommended by at least 10% of participants were included in analysis . Those specifically related to elevated PRS reports (i.e., obesity, type 2 diabetes, and kidney disease) were noted. These included: weight loss, assessment of weight history, testing for diabetes markers such as hemoglobin A1C, and consideration of kidney function when recommending over-the-counter (OTC) pain medication.

Finally, participants completed the Genetic Variation Knowledge Assessment Index (GKAI). Participants’ scores were the number of correctly answered true/false items out of 8 [29]. Participants also completed the Racial Attributes in Clinical Evaluation (RACE) scale assessing reported use of race in clinical practice [29]. Responses to this assessment were assessed on a 5-point Likert-type scale from 1 = “all of the time” to 5 = “none of the time.” Self-reported demographics were collected at the conclusion of the questionnaire. These included race and ethnic background, which are required to be collected the National Institutes of Health.

Data analysis

Descriptive statistics were calculated for all variables, and demographic variables were compared between conditions using ANOVA or chi square depending upon whether they were continuous or categorical variables.

All binary outcome variables – including coded data from open-ended responses, conversation transcripts, and recommendations for the patient’s care – were analyzed via two-step binary logistic regression analysis with bootstrapping on 1000 samples. The main effects of each condition (patient race and presence of PRS) were both entered on the first step and the interaction between the two was entered on the second step. Main effects are reported from step 1 of the regression and interaction results reported from step 2. Some recommendation types exhibited data sparsity wherein some conditions had no instances of both observation types (i.e., a count of zero in some cells) and thus logistic regression could not be performed. In many cases this sparsity was expected given the lack of recommendation salience for participants who did not receive PRS. Here, we instead applied chi square analyses, first on each main effect (df=1). Because interaction between the two predictors could not be directly assessed in this case, we also assessed omnibus difference between all four condition groups (df=3) to denote where differences may occur among individual cells. Recommendations where chi square analyses were performed are denoted in Table 3. Ratings of the patient’s health and recommendation needs were assessed using ANOVA that included main effects and the interaction term in the model. Alpha was set at .05, two-tailed.

Table 3:

Recommendations for Knee Pain and Fatigue

Recommendation	Control White patient	Control Black patient	PRS White patient	PRS Black patient	Main effect of PRS provision	Main effect of patient race	Interaction of PRS provision by patient race5
Knee pain
	n (%)				Odds Ratios
Weight loss 1	1 (7)	2 (11)	10 (40)	5 (21)	0.24 *	1.97	0.24
Dietary modification4	0 (0)	2 (11)	2 (8)	5 (21)	X2=1.27	X2=2.73	X2=4.30
RICE	9 (64)	9 (50)	11 (44)	13 (54)	1.33	0.98	2.71
Physical therapy	6 (43)	13 (72)	17 (68)	20 (83)	0.42	0.35 *	0.68
Stretching	2 (14)	4 (22)	1 (4)	9 (38)	0.79	0.18 *	8.41 *
OTC medication4	14 (100)	17 (94)	23 (92)	24 (100)	X2=0.50	X2=0.43	X2=2.93
Medication depends on renal function 2, 4	0 (0)	0 (0)	6 (24)	2 (8)	X2=5.80 *	X2=2.57	X2=9.17 *
Inflammatory markers	3 (21)	7 (39)	9 (36)	2 (8)	1.60	1.70	0.69 *
Imaging	12 (86)	14 (78)	22 (88)	21 (88)	0.62	1.32	1.63
Fatigue
	n (%)				Odds Ratios
Weight history 1	5 (36)	1 (6)	2 (8)	1 (4)	4.26 *	5.21 *	4.71†
Dietary modification	3 (21)	10 (56)	9 (36)	10 (42)	1.03	0.49	0.28
Exercise	3 (21)	9 (50)	9 (36)	12 (50)	0.75	0.43	0.49
Sleep hygiene	8 (57)	10 (56)	10 (40)	12 (50)	1.55	0.80	1.60
Hydration	3 (21)	3 (17)	4 (16)	2 (8)	1.44	0.89	1.36
Caffeine use	3 (21)	3 (17)	4 (16)	2 (8)	1.76	1.69	0.65
Psychological health eval	4 (29)	2 (11)	10 (40)	9 (38)	0.37	1.51	2.88
Anemia eval	11 (79)	15 (83)	21 (84)	21 (88)	0.71	0.74	0.98
Metabolic panel	10 (71)	11 (61)	18 (72)	17 (71)	0.78	1.25	1.50
Micronutrient eval	0 (0)	5 (28)	4 (16)	5 (21)	X2=0.10	X2=2.60	X2=4.55
Autoimmune eval	5 (36)	6 (33)	2 (8)	3 (13)	4.57 *	0.88	0.55
Thyroid eval	8 (57)	9 (50)	16 (64)	13 (54)	0.80	1.43	0.89
Diabetes eval 3	1 (7)	2 (11)	3 (12)	11 (46)	0.21 *	0.21 *	0.26
Menopause eval	5 (36)	2 (11)	1 (4)	2 (8)	4.62 *	1.99	9.67 *

^*p<0.05;

¹related to genetic risk of obesity;

²related to genetic risk of kidney disease;

³related to genetic risk for diabetes;

⁴chi square performed due to data sparsity;

⁵chi square analyses in this column assess omnibus significance of four conditions rather than statistical interaction

RICE = rest, ice, compression, and elevation; OTC= over the counter

Results

Demographics, Genetic Knowledge, and Use of Race

See Table 1 for means and standard deviations. There were no differences in demographics, GKAI, or RACE scales by experimental condition.

Table 1:

Demographic factors and scale responses by condition, n(%) or M(SD)

Demographic Factor	Control White Patient	Control Black Patient	PRS Information White Patient	PRS Information Black Patient	Total
	n (%)
Gender (Female)	8 (57)	11 (61)	17 (68)	15 (63)	51 (64)
Race (Asian)	5 (36)	3 (17)	9 (36)	8 (33)	25 (31)
Race (Black)	2 (14)	5 (28)	3 (12)	5 (21)	15 (19)
Race (White)	5 (44)	8 (44)	8 (32)	10 (42)	31 (38)
Race (Other*)	0 (0)	0 (0)	1 (4)	1 (4)	2 (2)
Year in Medical School (3rd)	5 (36)	8 (44)	11 (46)	10 (42)	34 (42.5)
	Mean (SD)
Age	25.8 (1.5)	26.0 (0.9)	26.3 (2.0)	26.8 (3.2)	26.3 (2.2)
BMI	25.1 (3.6)	23.9 (4.4)	22.8 (4.2)	24.1 (4.5)	23.8 (4.2)
GKAI Scale	5.1 (1.4)	4.8 (1.2)	4.3 (1.6)	4.6 (1.4)	4.6 (1.4)
RACE Scale	23.8 (6.4)	24.5 (6.2)	24.5 (7.2)	21.6 (6.1)	23.5 (6.5)

BMI = body mass index; GKAI = Genetic Variation Knowledge Assessment Index; RACE = Racial Attributes in Clinical Evaluation;

^*includes one participant who self-identified as multiracial and one who self-identified as American Indian/Alaska Native

Reported Use of Genomic Information

Most participants (74%) reported that they considered the genetic information during the visit. Those that received the PRS were more likely to report considering the genetic information (Table 2). The interaction with perceived patient race did not reach statistical significance. Participants receiving PRS more frequently reported incorporating the genetic information into the patient’s care plan. A much smaller number (11%) reported using genetic information in relation to the patient’s symptoms; there were no significant differences by condition. There were no other significant effects of patient race or interactions for any of these outcomes.

Table 2:

Self-report responses by condition, n(%) or M(SD)

Self-Report Outcomes	Control White Patient	Control Black Patient	PRS Information White Patient	PRS Information Black Patient	Main effect of PRS provision	Main effect of patient race	Interaction of PRS provision by patient race
Coding of participant statements about use of genetics N(%)
Considered genetic info	8 (57)	11 (61)	19 (79)	21 (88)	.28 *	.69	1.56
Incorporated genetic info into care plan	3 (21)	1 (6)	12 (48)	15 (63)	.11 *	.94	7.73
Incorporated genetic info because related to symptoms	1 (7)	4 (22)	2 (8)	2 (8)	2.00	.52	.28
Coding of visit transcript N(%)
Transcript: discussed patient weight 1	3 (23)	4 (22)	12 (50)	11 (48)	.31 *	1.08	.96
Transcript: recommend weight loss 1	2 (15)	2 (11)	8 (33)	6 (26)	.36	1.43	1.03
Transcript: recommend lifestyle changes	2 (15)	4 (22)	20 (83)	18 (78)	.057 *	1.03	.46
Transcript: mentioned diabetes 3,4	0 (0)	0 (0)	7 (41)	8 (35)	X2=12.25 *	X2=.004	X2=12.49 *
Transcript: mentioned kidney disease 2 , 4	0 (0)	0 (0)	5 (29)	5 (22)	X2=7.57 *	X2=.030	X2=.030
Transcript: discussed risk uncertainty	1 (8)	4 (22)	21 (88)	19 (83)	.033 *	.81	.20
Closed-ended items M (SD)
Patient is healthy	5.0 (0.7)	4.3 (1.2)	3.8 (1.0)	4.2 (1.0)	F=7.98 *	F=.45	F=4.29 *
Patient requires strict medical advice	4.4 (1.3)	3.6 (1.2)	4.7 (1.4)	4.5 (1.4)	F=8.69 *	F=.089	F=.012

^*p<0.05;

¹related to genetic risk of obesity;

²related to genetic risk of kidney disease;

³related to genetic risk for diabetes;

⁴chi square performed due to data sparsity;

⁵chi square analyses in this column assess omnibus significance of four conditions rather than statistical interaction

PRS Influence on the Verbal Encounter

Coding of the verbal transcripts revealed that the patient’s weight was more likely to be mentioned when PRS was present. Participants who received PRS were no more likely to recommend weight loss to the patient or attribute her health problems to her weight. Participants were more likely to recommend lifestyle changes when they received PRS. Discussion of the probabilistic nature of the genetic feedback was prevalent in the PRS provision condition (discussed in 83% and 88% of interactions with the Black and White patient, respectively) and more common when PRS was given. There were no differences by patient race and no interactions for any of these outcomes.

Perceived Patient Health

Participant-rated patient health revealed a significant main effect of PRS provision, such that patients with PRS were viewed as less healthy than those without. There was a significant interaction, F(1,64)=4.29, p=0.042, such that participants viewed the White patient as less healthy when they received her PRS, whereas perceptions of the Black patient were relatively unchanged (Graph in supplement). Participants also indicated the patient would be more likely to require strict medical advice when she presented with PRS versus without, F(1,71)=8.69, p=0.004. There was no main effect of patient race and no interaction.

Healthcare Recommendations

See Table 3 for patient recommendations and their frequency (graphs in supplement). Participants were more likely to recommend weight loss in response to the patient’s knee pain when they received PRS. There were no significant differences by patient race. Recommendations to evaluate the patient’s weight history in response to her fatigue were more likely when the patient was White and when the participant did not receive PRS. The interactions were not significant.

The rate of recommendation for the kidney disease-related item (monitoring kidney function with respect to OTC drug use for knee pain) was significantly different depending upon whether participants received PRS, and there was a significant difference between the four conditions. This recommendation was most common for the White patient presenting with PRS versus other groups.

Recommending diabetes evaluation related to the patient’s fatigue was more common when she was Black and when participants received PRS. The interaction was not significant.

Discussion

In the present study, we assessed medical students’ decisions about what, if anything, to do with PRS vis-à-vis patient care in the absence of any accompanying training. Results support hypothesis 1 in that a majority of participants who received PRS for common health conditions reported considering and using genetic information, and this significantly exceeded the control condition who received non-pathogenic carrier screening information alone. Comparatively, very few participants reported using the information to understand the patient’s specific health concerns. This approach is sensible given the lack of clinical utility for PRS for this purpose and that the provided PRS report recommended only broad preventive health behaviors for risk reduction [8]. Indeed, less than half of participants who received PRS mentioned diabetes, kidney disease, or body weight in verbal conversation with the patient. In contrast, findings regarding attitudes about the patient and care recommendations suggest the PRS influenced medical students more than they realized.

In line with hypothesis 2, medical students rated the virtual patient as less healthy and in need of stricter medical advice when they received PRS versus in the absence of PRS. This suggests participants perceived the patient’s increased genomic susceptibility as a strain on her current health which would require active prevention. This is consistent with previous findings that individuals with a genetic predisposition for future health problems are often treated as if they are already unwell [30]. This interpretation is bolstered by the finding that participants addressed weight and lifestyle change more frequently in their virtual encounter when receiving PRS. As such, it appears that PRS influenced perceptions of the patient’s current health, causing participants to see her as in need of more care.

Results also suggest care plans can be influenced by PRS, even when participants believe this information is not or should not be influential. This fits with the larger literature showing that peripheral and even irrelevant factors can influence physicians’ care plans and medical decisions [31, 32]. In support of hypothesis 3, there were differences in rates of key recommendations that map to the PRS report, which reported elevated risk for obesity, type 2 diabetes, and kidney disease. Medical students who received PRS were more likely to recommend weight loss to address the patient’s knee pain. Overreliance on weight loss to address patients’ health conditions is a hallmark of weight stigma, which is typically reduced when genetic contributors to obesity are considered [33]. The current findings stand in contrast to this pattern and may be linked to participants’ perceptions that the patient with PRS required stricter advice. Interestingly, participants were less likely to plan to collect a weight history in conjunction with the patient’s fatigue when they had the PRS report. Although these results seem contradictory, participants may have focused on other potential explanatory factors for fatigue made salient by the PRS. Participants only mentioned monitoring renal function in conjunction with OTC medication for knee pain when presented with PRS. Finally, participants were far more likely to recommend screening for diabetes when they viewed the PRS. These patterns clearly suggest that medical students made recommendations in conjunction with PRS regardless of their awareness of it.

We began to parse the influence of patient race by assessing it as a potential moderator. The patient’s race influenced medical students’ beliefs about her health status. The association of PRS with beliefs that the patient is less healthy is driven by attitudes toward the White patient. Ratings of the Black patient’s health did not depend upon PRS provision. This could be the influence of negative biases among the participants wherein Black patients are treated as though they are more unhealthy than White patients [34], but this pattern requires further examination, especially given the complexity of the intersection of race, ancestry, genomics, and racial and ethnic health disparities.

The majority of the care recommendation rates did not differ by patient race. However, recommendations linked to heightened risk in the PRS report were particularly likely to demonstrate racial difference. Medical students were most likely to note the need to attend to renal function in conjunction with OTC medication when the patient was White and had PRS. This is unexpected as kidney disease is more prevalent among Black patients in the United States [35]. Participants were also more likely to recommend assessments related to diabetes when the patient was Black. This pattern is in line with disparities in diabetes rates between Black and White individuals in the United States [36]. Finally, participants were more likely to recommend taking a weight history when the virtual patient presented as White. These patterns are inconsistent, so replication and additional investigation is needed to interrogate their meaning. In general, however, results suggest that without educational intervention, PRS may play an outsized role in clinical decision-making, and may be applied inconsistently across patients from different racial groups.

Although additional research is needed, the current work has implications for the near future of genomic translation education in medical settings. In particular, medical students understand some limitations about genomic information, such as its probabilistic nature, and explicitly agree that it is unlikely to be a helpful contributor to medical decision-making for specific health concerns. However, it appears PRS affect medical students’ recommendations whether they realize it or not. Finally, these data suggest that apparent patient race influences how providers respond to PRS. These are areas that should be prioritized in further study and addressed by educational materials.

This study had limitations. Unlike real clinical encounters, the virtual patient could not answer questions. However, this strategy has the benefit of maintaining experimental control and has been implemented successfully . The virtual patient always appeared higher weight and had a BMI indicating obesity. The role of patient weight in PRS interpretation is unclear. The sample was comprised of 3^rd and 4^th year medical students as they would be a sensible target for educational materials. Practicing providers might respond differently to PRS; however, prior evidence suggests they are likewise new to PRS concepts, being rarely exposed to genomic information [37]. In addition, GKAI scores indicate that medical students had levels of genetic knowledge comparable to or higher than practicing providers [29, 38].

Conclusion

As use of PRS becomes more commonplace in DTC and research settings, it is critical to forecast the impact on clinical encounters and provider recommendations to form a bedrock for the development of educational materials. This study demonstrates medical students are largely unaware of the influence of PRS on their behaviors, recommendations, and perceptions. Furthermore, patients’ apparent race may influence interpretation and use of PRS in a complex manner. Bringing awareness to these areas through educational materials both in the training setting and accompanying PRS provision may be warranted. Likely, medical students are not alone in this, and such education could also benefit practicing providers. We anticipate future work will further interrogate these influences to develop educational interventions promoting equitable and appropriate use of PRS in clinical settings.

Data availability

Data will be made available upon request.