Patient interviewsPatient understanding of and reaction to risk information

Patient baseline and demographic information are given in Additional file 1: Table S1. As per our recruitment strategy, five individuals self-identified as non-Hispanic Asian, five as non-Hispanic Black, five as non-Hispanic White, and ten as Hispanic White, five of whom were Spanish speakers. Five (20%) had low genetic literacy. For self-assessed numeracy, three (12%) were low, thirteen (52%) average, and nine (36%) high. Nine (36%) had inadequate health literacy.

Patients selectively engaged with the different types of information presented. The percentile risk rank included in the mock report was the number most engaged with (21/25), but the majority were confused by it (12/21), including two high numeracy patients. Many of these mistook percentile for percent chance. For example: “He‘s almost at a full whole risk. Ninety-nine percentile is almost at one hundred, so it’s like you’re one percent away from being completely at all risk of getting it. Doesn’t matter what age it is, you’re going to get it, that’s the thing.” There were other forms of confusion, for example: “That the person in the evaluation, those that have 95, have the least risk. And those with the least percentage, they have the most risk.” A minority (9/25) understood the percentile information, of which seven were high numeracy patients. Most patients referred to the odds ratio (15/25), the majority of whom were confused by it (9/15), including two with high numeracy. All of the patients who understood the odds ratios had high numeracy. Most patients interacted with the prevalence information (14/25), about half of whom were confused by it (6/14). For example, some thought the prevalence was their individual result. All of those who understood it had high numeracy. A common misinterpretation was to interpret “not high risk” as “low risk.” Also, in the context of the binary report, a common misinterpretation was to view the threshold as representing their own risk.

Additional information was desired; many wanted to see the inclusion of absolute risk information, and an integrated score: “I want the full assessment… and you’re only giving me the genetic score, which is necessary but not sufficient for a real assessment of my risk”. Two patients wanted to understand the genetic attributable risk, i.e., how much of the overall risk is captured by genetics, and by this score specifically: “I would want to say that it’s [patient’s risk] high, but not really, since there’s so many other factors that can contribute to getting a disease. And it’s not just your genetics. So it does come to this question in my head, how important is it really if it’s just like a small factor and it's not really like the only thing that means you're going to get the disease.”

When asked how they would interpret a fictional patient’s risk, many patients did not engage with the limitations section until explicitly prompted to do so. Most did not engage with the ancestry limitation. For those that did, many of them thought it should explain implications for patients: “it’s just it’s not really fully explaining, like, the implications”. Some did not understand: “Well, it says European ancestry. ... and I don’t understand when they say European, are they talking in Spain, Portugal?”

Of those patients asked whether they would “still like to receive this information if the results may be inexact for your ancestry?” (16/25) only a small handful (3/16) said they would not want it. There were no discernible patterns by the patients’ self-identified race/ethnicity.

Binary/continuous preference

Almost all (23/25) of the patients preferred the continuous report design. The few voices in favor of the binary report appreciated how direct the information was. In contrast to this, one patient expressed a common sentiment in reaction to the binary reports: the high binary “just made me feel like I have the disease, whatever it is” and the not high binary report “just doesn't give enough information.”

In their preference for the continuous report design, knowing their exact risk number was the most common reason given for their preferences. Patients expressed that not knowing their own risk number would leave them with questions, make them uncomfortable, and make them feel less empowered to alter their chances of developing a condition. They also wondered why the information was not being shared with them: “I would like to know my polygenic risk… Ignorance is bliss. Great. But knowledge is power. So I need to know my risk. If I know my risk, I can alter it, change it, and I can feel better about this thing. At least I can do something.” Another said, “I mean, the threshold here is like all I know is I’m not in the top 2 percent, it doesn’t really, what if I’m in the ninety seventh percent. ... it seems like a very narrow definition for having a high polygenic risk factor. But if you have that data, why not give it to me?”

Patients also appreciated use of the graph in the continuous report design, as an aid to understanding the risk information: “If I visually have a picture to match what I’m reading, it helps, just me personally, me, it helps me see visually what I would- what they’re trying to tell me…. seeing and knowing are two things that are always going to go hand in hand.”

Emotional reaction

A common theme that emerged was anticipated or imagined emotional reaction to the reports. A substantial minority (8/25) had an emotional reaction to the high binary report, and likewise (though fewer, 4/25) for the 99th percentile continuous report. This was often linked to a sense of genetic determinism, that they are going to develop the condition: “I mean, it gets to the point, but, me being the person that I am and not knowing too many big words, a lot of this would kind of scare me... just seeing the ‘high risk’, I probably would think that ‘Oh my God, I’m going to get cancer’.” And “it tells you for real, tells you to start getting ready. But I would die, it's very severe. Very alarming, very severe, It would make me worried.” We note that while both the educational materials and the reports themselves contained information designed to counter genetic determinism, our results suggest that this information was not effective at addressing these attitudes, suggesting that these reactions may be hard to dispel. Some had an emotional frame of being relaxed or not alarmed at the not high binary and the lower continuous report.

Questions for PCPs

When patients were asked “what questions would you have for your provider?,” all would ask what they could do to lower their risk. About half would ask for help interpreting the reports. Only a few other types of questions were raised, including what would happen if they develop the condition, for example, age of onset and disease trajectory.

Primary care provider (PCP) interviews

PCP baseline and demographic information are given in Additional file 1: Table S2. Eighteen of the PCPs interviewed were MDs, two were NPs, and one was PA. The small number of non-MDs prevented us from making comparisons between these types of PCPs.

PCP reactions to the reports PCP understanding of and preferences about risk information

Overall, most PCPs understood the information, though two of the PCPs read the percentile as an absolute risk, for example in responding to the 75th percentile report, “in this report it does say that out of a hundred people, 75 people will get the cancer.”

PCPs, like patients, expressed a strong desire for absolute risk information. Many PCPs were “doing the calculation,” multiplying the prevalence with the odds ratio to get an estimate of absolute risk. A handful of PCPs saw the percentile as possibly useful, but many commented that it was not useful: “I don’t actually care what the 70 percent is. Because at the end of the day, I’m treating an individual.” For considering the choice of risk metric (odds ratio or some measure of absolute risk), some PCPs mentioned that there are already standard ways to think about risk for some phenotypes. Some PCPs, like some of the patients, wanted an indication of genetic attributable risk, i.e., how much of an individual’s overall risk for a disease is accounted for by the information presented.

A small number (4/21) preferred the binary report design. This was mostly because they thought it would be easier for patients to understand. The “fear of missing something” also formed the basis for one PCP’s preference for a binary report: “I think that's what keeps people up at night: I missed something.” But the majority saw large downsides to not giving a continuous value. Many PCPs had concerns for those just below the threshold, including false reassurance for those just below the threshold, and the sense that it was paternalistic not to share the continuous result with the individual.

PCPs had diverse reactions regarding whether a threshold for high risk should be provided on a report. Many would not want a threshold; these PCPs were comfortable with fitting into a gray area, particularly because patients are different: “People do have different preferences. And I think given their preferences of an opportunity to use the score, I think that's more appropriate than the more paternalistic approach of picking high or low.” Many would want a suggested cutoff. Some PCPs expressed a desire for several risk categories.

One PCP displayed genetic determinism, though, as for the patients, this was coupled with a sense that they could do something to avert it: “I need to do something, whether it’s ninety nine percentile or ninety five percentile, it doesn’t matter to me. I know that the risk is high and that this person will develop prostate cancer in the future. So I’m going to take action.”

Several PCPs emphasized the importance of the design of the report not just to convey information, but to structure the patient-PCP interaction, to help the PCP “walk them through it.” This was connected with the sense that report design can make or break the patient’s understanding: “30 years of doing this, almost every patient can explain really complicated ideas if you present the information to them correctly.” And in connection with the short shrift usually given to the design of the report: “And how do you take something as complicated and make it simple? That is very challenging. That requires UI/UX experts… that is something that in health care is bizarrely nowhere on the list.”

PCPs emphasized that report design needed to work for a diversity of patient preferences and literacy: “I think what tends to happen is that you have some patients [that] will want very, very granular detail… And then there are going to be other patients who are just .. like well, I’m at high risk.” Additionally, “You’re going to have some patients who are even challenged to understand risk and benefit discussions …. certainly you don’t want to mislead somebody with a number… you really have to give them an idea of what the number actually represents.” Some PCPs stressed the need for easily understandable patient materials, in several languages.

Communicating differential performance by population groups

Many PCPs were confused about the relation of the differences in performance by ancestry group with different prevalence base rates by self-reported race/ethnicity. Some wanted prevalence by population group. Many thought that the relationship between the ancestry limitation and reporting the odds ratios by ancestry group was not clear. At a deeper level, there was some confusion over whether the differences reflected our current knowledge (as indicated on the report) or true differences in disease prevalence: “If I had an Asian patient, a person of Asian ancestry in front of me, I would say that the chances are higher for developing prostate cancer because of the ancestry.”

Some PCPs were comfortable with communicating this limitation, using “take this result with a grain of salt” language. Many drew analogies to other areas of medicine where data was similarly biased, for example in risk scores for certain conditions. On the other hand, many would struggle to communicate the ancestry limitation, and were unclear about what the implications were for the patient in front of them: “The last limitation says that score has been best validated in European ancestry... So does best validated mean that it’s the highest odds? I doubt it. I suspect that it means that there’s stronger research, but I wouldn’t know how to interpret that.” One PCP came up with their own interpretation: “As best as we can tell, it's underestimating your risk [of prostate cancer] as an African-American. And so I’m going to throw in another 10 or 20 percent.”

Some PCPs brought up the subject of the population categories used on the report. Some were comfortable with the categories: “We’re in a situation in our society where, like we’re so used to classifying groups of people this way.” Some PCPs expressed concern about what to do with an individual who didn’t neatly fit into a category, though a practical response was to look at the relevant ranges and say “something in there.” Some wanted an acknowledgement that this was difficult to interpret because few people fit neatly into a category. Some thought we did not need to break down by categories if the clinical action would not change.

Most PCPs used continental ancestry categories interchangeably with racial categories, for example “I think it is interesting to know that it’s better validated in European ancestry. I think we kind of tend to know that, but. So pretty much everything is validated in Caucasians.” This was particularly true when PCPs imagined how they would explain the report to patients: “And so specifically thinking of my Latino patients and my Black patients that, you know, ‘unfortunately, these scores historically have looked at greater white population than populations that look like you.’”

PCP perceived utility of PRS

In the examination of how PCPs would use PRS information, instead of a simple case of deciding whether or not the information was “actionable,” a more complicated picture emerged of three different use cases with very different perceived utility and engagement with the information (see Fig. 1).

Fig. 1

Three use cases for polygenic risk score (PRS) reports. These use cases for PRS emerged from the interviews with primary care providers (PCPs). PCPs also recognized the utility of PRS reports as an education tool, spanning these use cases. The relevance of lower than population-average-risk information varied by use case

Use Case 1: Clinical decision-making in the gray areas of patient care

PCPs saw value in personalized information informed by PRS for gray-area decision making: “Benefit, I think when there is a gray area and I feel like we can’t really tailor the recommendation as much as I want to. I think then it is definitely beneficial to have additional sort of data and scores to help guide both myself and the patient.” Some emphasized that this utility would be condition specific: “I think it depends on the condition and, you know, whether there really needs to be that kind of shared decision making and kind of risk calculation going on.” In general, PCPs were much less enthusiastic about using polygenic risk information for deciding whether or not to prescribe a medication; in this case, they would want to see proof of benefit.

For gray-area decision making, many PCPs perceived value in results below average population risk, ie. in the negative tail of the distribution. “If you had [a] report that says your patient is at low risk ...that would then probably guide a PCP like me to be more conservative in their recommendation for screening.” This was in the context of feeling that they might be “overdoing” certain tests. However, some would not use the lower risk in this way: “So I would use this basically to identify those who I might screen more regularly, but I wouldn’t use a low risk here to screen less regularly.”

PCPs used diverse heuristics for integrating a PRS into their holistic risk interpretation for the patient in front of them. Some PCPs operated with rules of thumb. Most used the odds ratio, for example, “I think that sort of one and a half increase cutoff is where it really helps to push me.” For a small handful their rule of thumb related to the percentile: “for anything more than 50 we should do a little more something.” Some were aware they did not have a good rule of thumb. A small handful of PCPs explicitly talked about the fact that PCPs were natural Bayesian integrators, “This is Bayes theorem sort of thing…. You actually have to have a fairly big increase in risk for a risk score to be useful... for most people it doesn’t make a difference because most people are either really low or really high to start with.” One PCP was aware that they should be operating in a Bayesian fashion, but was not willing to do this “naturally,” instead wanting explicit pre-test probabilities. In practice, this diversity of heuristics led to different PCP responses to the 75th percentile report, with some changing their recommendations based on this information and others refraining from doing so. Across the PCPs, the way they would use PRS information was consistent with how they would use other pieces of risk information such as family history or clinical risk factors.

Use Case 2: Encouraging patients to follow already recommended actions

PCPs perceived value in using PRS reports to encourage patients to do whatever they thought those patients should be doing anyway, in particular, to adopt recommended screening and a healthy lifestyle, using phrases such as “pushing” patients, providing “ammunition,” or “fuel for the conversation.” This use was noted in particular for overcoming the effects of a negative family history. “I mean many people won't do a colonoscopy. But if you’re presented with saying, ‘Hey, you're in the ninety-ninth percentile’ and you give this to them, they’re going to think very, very differently regarding it. So it could be a tool to sort of gently — or maybe not so gently — kind of push people to go get their screening updated.” Some expressed qualms about doing this, but would do it anyway: “The place where it would make a difference is if I really thought somebody should be taking a statin, for instance, or doing something with their blood sugars, it would give ammunition to do so, which is really not the right way to use these things, because it's not really doing shared decision making. It's using an argument that's unproven to get a patient to do something that you think they should do, even though they're hearing the data and don't want to do it.” For this use case, there was a concern about sharing below-population-average risk, on the basis that this might cause false reassurance.

Use Case 3: Identifying those at high risk who would otherwise be missed

PCPs saw value in PRS reports as a “hook” to have conversations that they might not otherwise have had. The foremost benefit of this was to detect disease early in those they might have otherwise missed: “But the benefits, I think, are huge. I think it’s going to catch people who would have been missed by traditional screening metrics. And I think that's the real benefit.” This was seen as particularly true when there is no perceived harm to screening, for example, A1C tests for diabetes.

There was also a perceived benefit to educating patients about their risk that is not achievable with other risk factors, “I think this has the potential to educate patients about themselves in a way that’s very hard in primary care…. the genetic scores indicate a really high risk independent of all these other things that we spend a lot of time thinking about, and that could be a huge actual time saver and actually be really efficient and really effective for patient education.”

How PRS reports could fit into PCP practice

Across these use cases, most PCPs saw using PRS reports as a natural extension of their practice. This use would fit into their existing ways of thinking about risk. Many emphasized that their practice has been trending in this direction over the last decade or so: “We all live in the risk score business now.”

Only a small handful would refer their patients to a specialist (in this case, a urologist). If asked, many said they would refer to a genetic counselor, but few brought this up before prompted. In the cases where they did bring it up, it was most often in connection with dealing with an anxious patient, and less often for help interpreting the results for a patient with many questions. PCPs frequently expressed wanting guidance on whether and when they should refer to a specialist or a genetic counselor.

Many saw the need for training and felt that with training, they would feel confident integrating this type of report into their practice: “And so it’s important that you arm those people with the right way to interpret it. So there has to be something that makes it so that I quickly come up to speed. So I understand the nuance, I understand the limitations.”

For gray-area decision-making, some PCPs were only willing to act on PRS info if there were evidence-based practice guidelines. Others wanted clarification on the relationship to professional guidelines via explicit statements on the report. For example, whether or not the relevant professional society approved or explicitly disapproved of the use of the PRS, and whether all patients were just being recommended standard guidelines.

Although most PCPs emphasized that PRS would be treated very similarly to other risk information they habitually deal with, a handful emphasized some differences. In addition to the quote above about the educational role of PRS information, one PCP described genetic information as different to other information because the underlying genetics does not change (unlike other tests), but interpretations do change more than other tests, concluding “I can say for certain that your potassium from five years ago is irrelevant to me. Where this may not be — I don't actually know in 10 years whether this will be as relevant, less relevant, more relevant.”

PCPs also expressed several concerns with the clinical implementation of PRS, which are given in Table 1. Chief amongst these were concerns about their own time, about the lack of an evidence base for the use of the scores, about potential adverse patient reactions, and about the potential implications of the differential performance by population group.

Table 1 Primary care provider (PCP) concerns about use of PRS information in their practice. The most cited concerns were overdiagnosis and overtreatment, lack of PCP time, and concerns over patient response

Overall, almost all PCPs saw the information as useful, though a small handful did heavily circumscribe this: “To be completely frank though, not a high priority for me in terms of if I had to choose what I got more of in primary care, in terms of resources. I just think there are a lot of other things before this that we really need.”

PCP responses were relevant to many reporting choices. We combine these with insights from the patient interviews in Points to Consider in the clinical reporting of PRS (see the “Discussion” section and Table 2).

Table 2 Recommendations in designing PRS reports. Integrating perspectives from both patients and PCPs, we offer the following points to consider for PRS clinical report design. Many of these recapitulate best practices from the risk communication literature; some are PRS-specific. See Discussion for the highlighting of certain points, in particular for a continuous (versus binary) sharing of risk information, the clear advantage of sharing absolute risk information, and the handling of differential performance of PRS by population group