Polygenic Risk Scores in the Clinic: Health-System Leaders and Primary Care Providers Weigh In

Setting and Recruitment

We recruited two categories of key informants from UW Medicine, headquartered in Seattle, Washington: (1) leaders with decision-making authority for administrative and clinical practices, and (2) PCPs engaged in direct patient care. Leaders were identified through the UW Medicine website and snowball sampling. PCPs were recruited by e-mail, with an introductory message from the relevant medical director. To ensure that the sample included PCPs serving a representative range of demographic groups, we purposively selected clinics across the UW Medicine catchment area. Leaders and PCPs who indicated interest were e-mailed an information sheet. Recruitment ended when we reached theoretical saturation.¹³ All participants were offered a $100 electronic gift card.

Data Collection

Development of separate interview guides – one for leaders and one for PCPs (see Supplementary Material for details) – was informed by the empirical implementation science literature,^14,15 as well the Inner Setting and Individual domains of the Consolidated Framework for Implementation Research (CFIR).¹⁶ Because we also wished to understand how leaders and PCPs conceptualized and operationalized risk assessment elsewhere in their practice, we also looked to the literature on clinical reasoning and heuristics.^17,18 Self-reported demographic information was collected verbally at the end of each interview.

The leaders’ guide focused on their role as organizational decision-makers and included questions about the acceptability of dichotomous result reporting (i.e., high risk or not high risk, which was the form in which eMERGE results were to be returned), the potential impact of PRS on clinical decision-making, provider and patient education needs, and clinical decision-support priorities. The PCP guide included the same topics, additional questions about the clinical encounter, and a set of hypothetical cases aimed at eliciting how participants would respond to finding a research-generated PRS report in the EHR. We attempted to correct for any disease-specific anchoring effects by creating two sets of cases, one focused on type 2 diabetes (T2DM) and another on prostate cancer (Table 1). Each PCP participant was given either the T2DM (n=12) or the prostate cancer (n=8) cases. Each set contained three cases, all of which were presented to the participant: (1) one in which the patient has previously received a diagnosis of the condition, but their PRS indicates that they are below the study risk for that condition (C+/G−); one in which the patient appears clinically to have low risk for the condition, but the PRS result indicates increased risk (C−/G+); and one in which the patient appears clinically to have low risk for the condition, but the PRS report indicates wide confidence intervals for individuals of the patient’s population background (C−/G?). To avoid ordering effects, we varied the sequence of the cases per participant.

Two non-clinician team members (SBT, SMF) with extensive experience in qualitative research concerning genetics conducted the interviews via Zoom. Interviews lasted approximately 60 minutes. The interviewers drafted field notes following each interview and compared notes at regular intervals. Audio files of the interviews were professionally transcribed for analysis.

Analysis

We conducted a descriptive thematic analysis using the constant comparison approach.^19,20 The first interviewer performed several close readings of the transcripts and reviewed the field notes to develop an initial codebook, which was reviewed and revised by the second interviewer. The codebooks included both deductive codes (derived from the interview guide) and inductive codes (which emerged from the interviews but were not directly prompted by specific questions). The first interviewer coded the transcripts using ATLAS.ti, with approval or amendment from the second interviewer.

Using Risk Estimates

We invited interviewees to consider PRS in the context of other risk-assessment tools. PCPs and some leaders reported regular use of risk calculators for atherosclerotic cardiovascular disease (ASCVD), breast cancer, fracture, and kidney failure. Nearly all clinicians interviewed appreciated tools that were easy to use and well-integrated with the EHR; assessed both interim and lifetime risk; were endorsed by the relevant professional society; and produced graphical risk representations that could be shared with patients.

Clinicians further noted value in risk scores for patients who are not motivated by population-level screening and lifestyle recommendations, or who want more personalized guidance. PCP said that they often use risk calculators in the patient encounter:

With the [ASCVD] risk calculator, there are options for optimizing. So, [I might say], “What happens if you quit smoking? What happens if we get your [systolic] blood pressure down to 120? What happens if we can get your LDL down with a statin?” Sometimes it’s a motivator, like “Wow, I didn’t realize that if I quit smoking, that would really reduce my heart disease risk so much.” (PCP 33)

Whereas some described risk scores as an adjunct to shared decision-making, others used evidence-based risk estimates as a motivational tool, e.g., “If I wanted a patient to be on a statin and they were disagreeing with me, I might do [the ASCVD risk calculator] with them.” (PCP 74)

The majority of interviewees thought that a dichotomous reporting approach for PRS would be “cleaner” and easier for patients to understand. Others thought that reporting continuous results would offer providers and patients greater precision:

Just to make an analogy … Oncotype gives you a score and MammaPrint just says high/low. And I tend to use Oncotype more. I like the granularity of the score because we know that it is linear and I think that it does matter. I would want to know, if I got listed as high risk, if I was just barely high risk, or if I’m in that 99th percentile.” (Leader 15)

We also heard concerns about the potential for false reassurance with “below-threshold” results and the need for careful explanation from providers.

Clinical Utility

Clinicians in both groups correctly described PRS results as increasing the index of suspicion for a particular issue, but not necessarily definitive in their own right. PRS were seen as “a piece of the puzzle.” For example, one participant said:

PRS information is not the same as a lump on a mammogram. This is a piece of data in the midst of the clinical context for the patient …. I’m thinking like a 30-year-old who pops up high risk for cardiovascular disease, but otherwise has historically had normal blood pressure, exercises regularly, all that stuff. We have to take all that into account. That [PRS] doesn’t mean, start sucking down a statin and an aspirin every day. So, it is not diagnostic. It’s clinical data that aids, but doesn’t direct. (PCP 43)

Interviewees also described challenges in interpreting, acting upon, and explaining risk information to patients. One leader commented,

[A]s human beings and as physicians, we are very poor I think in general of knowing what to do with risk. Even if you had an objective [estimation] method, it’s not even clear what are the objective things you’re trying to maximize for. Are you trying to maximize survival? Are you trying to maximize cost-benefit ratio for things? Are you trying to maximize happiness? All of these things are very legitimate, and none of them immediately follow. (Leader 14)

Several interviewees noted the need to explain to patients that, as one put it, “low risk does not mean zero risk,” (PCP 93) and that genetics is only part – and potentially a small part – of what drives overall health risk. Participants also pointed to the relevance of non-clinical factors in determining the best course for a given patient, as in this comment:

You know, in primary care, … we cannot be so much just numbers driven. We bring in many factors: [the patient’s] personality or what kind of person that person is, are they compliant, what is their family history, many other things. What are the family situation, stressors? So, all those, combined, to help us make a decision. (PCP 67)

Participants said that, given that clinical screening is generally accurate, non-invasive, quick, inexpensive, and familiar to patients, PRS would need to generate actionable information that is not otherwise available. Both leaders and PCPs questioned the availability, burden, and cost of indicated follow-up and identified disease-specific considerations. For example, whereas a PRS result indicating increased T2DM risk might lead to more frequent HbA1c measurement with clear next steps, it is less obvious how clinicians should respond to other potential results:

If we say, “OK, you have an increased risk of pancreatic cancer,” the screening for that is not real straightforward . . . Not having really clear evidence-based guidelines and what to do with that information at the different levels of risk could put me as a physician in a hard position, because how do I counsel that person appropriately? (PCP 46)

Interviewees cited a number of potentially useful applications of PRS. Detecting substantial but clinically occult risk – and thereby identifying opportunities for primary prevention and more aggressive screening – was a strong theme. Examples included refining the timing, frequency, and modality of cancer and heart disease screening; identifying patients who could benefit from chemoprevention; deciding when to pursue more burdensome prevention or treatment strategies; and determining the appropriate blood pressure threshold for a given patient.

Potential Challenges

The main concern voiced by interviewees was the potential for PRS results to cause confusion and unnecessary worry for patients. The fact that, in most cases, results represent relative risk contributed to this perception. One interviewee put it this way:

[O]ne problem . . . is that the benefit of intervening is proportional to absolute risk, not relative risk. So, if I were to put someone on a therapy for which side effects are common, but the outcome itself is rare, even if they’re at high risk, the likelihood that they’re gonna have an adverse effect from the therapy may be higher than the outcome that I’m trying to prevent. (PCP 16)

In addition, interviewees noted the potential for patients to be falsely reassured by a “not at increased risk” PRS result, foregoing recommended screening or healthful habits. As one leader remarked, “You’re gonna be telling a lot of people who are going to die of cardiovascular disease that they’re low risk [but] I think what you’re really saying is, ‘You’re at low risk of dying of it early,’ not, ‘You’re not going to die of it.’ …. I do worry a little bit about the lifestyle effects of that.” (Leader 15)

Participants also raised the specter of opportunity costs, specifically the potential for PRS to “distract from broader population health, public health, societal health sort of focuses, in terms of where resources are allocated and just in terms of where we can get the most bang for our buck for the most people.” (PCP 19) Another dimension of the opportunity cost theme concerned access to care and the relative priority of PRS follow-up:

I frankly don’t have time for a patient to come see me saying, “Hey, I got a polygenic risk score and it says I’m at risk for [condition]. Can you screen me, please, because I’m worried about this?” You’re actually going to do indirect harm to the patients who actually do need to see me by glomming up access over something that I don’t know is actionable or at this point has no data to suggest that anyone’s outcome is gonna be improved by screening this [unselected] population. (Leader 63)

Primary Care Cases

As described in Table 1, we presented each PCP with three cases and asked them what they would make of the result; how they would use the result in the patient’s care; and what they would want the patient to understand about the result. In Case 1 (C+/G−), all participants said that the PRS result would have no impact on their clinical management. To explain the discordance to the patient, some said that they would note that not all disease has strong genetic contributors, or that other factors could outweigh the influence of genes. In the T2DM scenario, several providers said they would use the results to encourage or reinforce healthy behaviors, as in this example:

My hope would be maybe this could be a motivating factor of saying, “Hey, look, genetically, your body and the cosmos has you in low risk. So maybe if we just modify these other pieces – dietary, physical activity – sometimes that actually works. You blamed it on your genes all this time, and look, it’s not that.” (PCP 23)

For both the T2DM and prostate cancer versions of Case 2 (C−/G+), most participants said that the PRS would lead them to reinforce the patient’s healthy behaviors and consider screening earlier and/or more often. In the T2DM example, PCPs said they would likely use HbA1c as a medium-term indication of glucose control that is easy and inexpensive to order, if not always covered by insurance. In the prostate cancer example, PCPs said they would include the PRS information in a broader shared decision-making conversation, while pointing to a need for clear guidance about how to integrate this information into screening plans:

I would like to know, what are our next steps? Do we do PSAs? Do we do rectal exams? Is there more genetic testing that is more specific? Do we do ultrasounds? I don’t know, to be honest …. I’m really cautious to open that box, because the challenges of how aggressive or non-aggressive prostate cancer is, and the older they are, the more likely they are to get it anyway, but not necessarily have it affect their life. (PCP 72)

Case 3 (C−/G?) elicited the greatest uncertainty and showed the most divergent responses between the prostate cancer and T2DM scenarios. In the prostate cancer version, most interviewees said they would disregard the PRS information entirely, citing clinical guidelines and epidemiological data about a higher background risk among African American men. In the T2DM version, most said they would initiate or enhance screening and reinforce healthy behaviors, irrespective of race, and several said they would be inclined to start metformin earlier.

There was wide variability in which of the cases PCPs identified as most challenging. For both the T2DM and prostate cancer condition, several declined to identify a most challenging case, saying that it would depend on the patient (n=2), that two cases were tied for most difficult (n=2), or that none were difficult (n=2). In the prostate cancer condition, none of the PCPs found case C−/G+ most difficult, and equal numbers identified the other two cases (n=2 for each). In the T2DM condition, the C+/G− was rated most difficult by three participants, and the C−/G+ and C−/G? cases were each identified as most difficult by two participants.

Some PCPs expressed concern about unexpected, non-intuitive results harming patients’ trust and the therapeutic relationship. For example, a PCP who identified Case 3 (C−/G?) as most difficult said,

It can be hard when we feel like we are trying to educate a patient on something that is counterintuitive or if we feel like a patient is losing trust in the therapeutic relationship. Sometimes when [test results] don’t align with expected outcomes, that can create some conflict, and patients can feel maybe that they are not understanding it correctly or they might feel like the providers are mistaken or that there is some level of inaccuracy in the recommendations that they’re hearing from a provider. (PCP 99)

Operational / Organizational Impacts

Unprompted, both leaders and PCPs expressed concern regarding medicolegal liability. One leader compared the risk to “the incidental lung nodule that doesn’t get followed up on,” saying, “and we certainly have problems with those. I mean multimillion-dollar problems.” (Leader 73) For frontline clinicians, this concern was exacerbated if results were to be reported in dichotomous fashion and stored in the EHR as PDFs:

[Say] somebody heretofore not known to be high [risk] for cardiovascular disease.… then this result pops in, in a scanned media form. It’s what I call invisible data, because it’s an image. And then . . . the patient goes and has a stroke or heart attack. And then they say, “Hey, you knew that this patient is high risk. How come you didn’t do anything about it?” “‘Cause I didn’t know the patient was at high risk.” “But Doctor, see Exhibit A, a scanned report from the polygenic risk score study…” (PCP 43)

Clinical leaders indicated that training and coordination across departments would need to be established before PRS were routinely implemented. Procedures to coordinate care and referral, potentially across specialties, would be required; and if a panel approach were used (as is the case for eMERGE4), all potentially implicated departments would need to be trained and ready at once. This was described as entailing significant operational and administrative burden, which interviewees were uncertain was justified by the current clinical utility of these tests.

Leaders and PCPs also outlined the need for provider training, both on PRS proper and on how best to communicate with patients before and after such testing. They also saw a role for point-of-care resources. For example, one leader commented, “Sometimes with these scores, it’s a black box to folks. [Hyperlinks within the EHR to additional information], so that someone could educate themselves, I think are the bare minimum.” (Leader 14) . Participants also desired graphical tools and handouts for patients around interpreting risk results, similar to those produced by certain risk models built into the EHR. Although training needs were thought to be substantial, leaders felt that such an investment would support precision medicine more generally.

Finally, time considerations were a strong concern for both groups of interviewees. A recurring, unprompted theme concerned the possibility of patients messaging their providers through the EHR with questions about PRS results. PCPs were uniformly opposed to this approach, as exemplified by this comment:

What I would hate to see is a bunch of MyChart messages: “Hey, I got this thing. Can you tell me what I should do with this?” And for [the report] not just to say follow-up with their PCP – because then they think they can just e-mail us – but to make an appointment . . . . If it’s truly helpful information, it’s really worth the person sitting down and having a conversation. (PCP 64)

PCPs felt strongly that elevated risk results should be discussed with patients, preferably in person, both to deliver optimal patient counseling and to avoid “donating time.”

Study Limitations

This study was limited to a single healthcare system in the Pacific Northwest and may not represent the views of leaders and PCPs in other regions or health systems. UW Medicine is centered around a large academic medical center, and our findings may not generalize to smaller systems, rural settings, or standalone clinics. Although we attempted to recruit PCPs serving diverse communities, 12/15 leaders and 12/20 PCPs interviewed were White. Data were collected nearly four years ago, during a global pandemic; providers’ views may have changed in the interim. In addition, this study did not explore patient perspectives, which are an important area for future research.