Using HIV Risk Prediction Tools to Identify Candidates for Pre-Exposure Prophylaxis: Perspectives from Patients and Primary Care Providers

Abstract

Clinical guidelines for HIV pre-exposure prophylaxis (PrEP) include risk prediction tools to identify appropriate candidates. We conducted a qualitative interview study to explore the potential acceptability, interpretation, and anticipated impact of such tools from the perspectives of men who have sex with men (MSM) and primary care providers (PCPs). Our purposive sample of English-speaking participants included: (1) MSM reporting HIV risk behaviors (n = 32; median age = 38 years; 53% non-Hispanic white; 22% high school degree or less education); (2) PCPs specializing in health care for MSM (n = 12); and (3) PCPs in general practice (n = 19). MSM participants questioned the ability of risk tools to predict HIV acquisition, and their perceptions of what might constitute a high HIV risk score varied widely. Many MSM participants believed that receiving a high score would prompt them to consider PrEP or other risk reduction strategies. Some believed that the data would be useful, particularly if discussed with their providers, whereas others anticipated feeling fear, anxiety, or mistrust. PCPs expressed more confidence in HIV risk prediction and imagined integrating tools with medical histories and their clinical judgment to assess risk. PCPs were most enthusiastic about adopting HIV risk prediction as a teaching tool to help patients visualize and reduce risk, their concerns about time constraints notwithstanding. In conclusion, our findings suggest that PCPs' views of HIV risk prediction tools are generally positive, whereas MSM participants' are more mixed. Given that both groups emphasized the value of contextualizing risk, shared decision making may be needed to implement HIV risk prediction tools effectively.

Introduction

Nearly 40,000 people are newly diagnosed with HIV each year in the United States, with ∼70% being men who have sex with men (MSM).¹ Antiretroviral pre-exposure prophylaxis (PrEP) is up to 99% effective in preventing HIV infection among MSM and has the potential to eliminate HIV in this population if used with approaches such as antiretroviral treatment for HIV-infected persons.^2–8 Although the Centers for Disease Control and Prevention recommends PrEP as a key HIV-prevention strategy,⁹ few of the estimated 800,000 eligible MSM in the United States have initiated PrEP.^10,11 Strategies are needed to support PrEP prescribing for MSM, particularly in high-incidence underrepresented groups such as black and Latino MSM.^10,12–14

Primary care providers (PCPs) are well positioned to increase access to PrEP because they provide preventive health care to many MSM. However, few PCPs have prescribed PrEP,^15–18 in part because of challenges identifying suitable candidates.¹⁹ Providers may have difficulty in identifying MSM at risk of HIV acquisition because of time constraints, suboptimal patient–provider communication about sexual and substance use behaviors, and believing that few of their patients have indications for PrEP.^20–24 Similarly, from the patient perspective, a commonly cited barrier to PrEP use is the misperception of low HIV risk.^25–29 Notably, those at highest risk of HIV infection may be most likely to underestimate their risk.³⁰ Interventions are needed to support PCPs and MSM in more accurately and efficiently assessing HIV risk and suitability for PrEP.

PCPs rely on risk prediction tools to make decisions about provision of prophylactic medications in other areas of medicine, such as cardiovascular disease prevention,^31–33 and these tools hold promise for optimizing PrEP provision. Existing tools to estimate HIV risk and PrEP eligibility include risk indexes that require providers to manually gather and sum patient data^34–37 and, more recently, automated algorithms to identify at-risk patients using electronic health record data.³⁸ However, the usability of HIV risk prediction tools from the perspectives of intended audiences such as PCPs and MSM is largely unknown. Although previous studies have evaluated risk prediction tools in areas such as cancer, diabetes, and cardiovascular disease,^39–42 results from other areas of medicine may not generalize to HIV, for which risk is influenced by more stigmatized sexual and substance use behaviors. Thus, our objective was to explore perspectives among MSM who engage in HIV risk behaviors and PCPs on the acceptability, interpretation, and anticipated impact on PrEP use of HIV risk prediction tools.

Methods

Participants

We conducted semi-structured qualitative interviews with 31 PCPs and 32 MSM in Boston, Massachusetts, in 2013–2014. We used purposive sampling to recruit two types of PCPs: (1) “LGBTQ specialists” (n = 12), who practiced in a community health center specializing in health care for lesbian, gay, bisexual, transgender, and queer patients; and (2) “generalists” (n = 19), who practiced in an academic medical center. All prescribing clinicians were eligible. We recruited MSM participants from the same medical centers. MSM were eligible if they were aged 18 years or older, English-speaking, male at birth, HIV uninfected, and reported condomless anal sex with male partners with positive or unknown HIV status in the previous 3 months. Participants provided informed consent and received up to $100.

Data collection

Two interviewers (K.M.M. and D.S.K.) conducted 60- to 90-min in-person interviews as part of a larger study on patient–provider communication about PrEP.⁴³ For interview portions specific to HIV risk prediction tools, interviewers oriented participants with an introductory statement. For MSM, we said, “Let's say that you are informed of your risk for HIV infection in the form of a number. So in the next five years, your risk would be X%.” For providers, we said, “A risk prediction tool for HIV is a formula that can help healthcare providers figure out an individual's risk for becoming infected with HIV in a specific time frame, based on their personal characteristics, sexual behaviors, and other details.” Interviewers next asked participants to reflect on what they thought of such a tool, what they would consider to be a high and low risk score, and how they might use the tool. We recorded and transcribed interviews verbatim. The Institutional Review Boards of Beth Israel Deaconess Medical Center and Fenway Health approved the study protocol.

Analysis

We analyzed data using thematic content analysis.⁴⁴ In the first, inductive phase, two investigators (M.B.G. and J.L.M.) analyzed a subset of transcripts through a process of open coding to identify broad topics of discussion. Using a standardized codebook, one investigator (J.M.G.) then applied codes to the data systematically using NVivo version 12 (QSR International, Melbourne, Australia). A second investigator (M.B.G.) reviewed the coded transcripts to identify disagreements in coding, which the team resolved via discussion.

In the second, deductive phase of analysis, we considered data code-by-code and described emerging patterns thematically by participant type. Our resulting themes centered on participants' confidence in the concept of HIV risk prediction, their interpretation of high and low risk scores, how they anticipated using risk prediction tools, and the expected acceptability and feasibility of using these tools in clinical practice to inform counseling about PrEP and other HIV prevention strategies. We present themes by topic, alternating by participant type, to facilitate comparisons between provider and MSM perspectives.

Results

Participant characteristics

Of 32 MSM participants (Table 1), most were non-Hispanic white (n = 17), non-Hispanic black (n = 8), or Hispanic (n = 3). Most identified as homosexual or gay (n = 24), with smaller numbers identifying as bisexual (n = 6), heterosexual or straight (n = 1), or queer (n = 1). Of 31 provider participants, most were non-Hispanic white (n = 24), Asian (n = 3), or Hispanic (n = 2). Providers included physicians (n = 27), physician assistants (n = 2), and nurse practitioners (n = 2).

Table 1.

Participant Characteristics

	MSM (n = 32)	PCPs (n = 31)
Age, years, median (interquartile range)	37.5 (25.8, 47.0)	39.0 (34.0, 51.0)
Gender, n (%)
Male	32 (100)	17 (55)
Female	0 (0)	14 (45)
Race/ethnicity, n (%)
Non-Hispanic white	17 (53)	24 (77)
Non-Hispanic black	8 (25)	0 (0)
Latino/Hispanic	3 (9)	2 (6)
Non-Hispanic Asian	2 (6)	3 (10)
Other	2 (6)	2 (6)
Sexual orientation, n (%)
Homosexual/gay	24 (75)	8 (26)
Bisexual	6 (19)	0 (0)
Heterosexual/straight	1 (3)	20 (65)
Queer	1 (3)	1 (3)
Not reported	0 (0)	2 (6)
Recruitment location, n (%)
Academic medical center	7 (22)	19 (61)
LGBTQ community health center	25 (78)	12 (39)
Provider type, n (%)
Physician (MD, DO)	—	27 (87)
Physician assistant	—	2 (6)
Nurse practitioner	—	2 (6)
Educational attainment, n (%)
High school degree or less	7 (22)	—
Some college, no degree	11 (34)	—
College degree	9 (28)	—
Graduate/professional degree	5 (16)	—

LGBTQ, lesbian, gay, bisexual, transgender, and queer; MSM, men who have sex with men; PCPs, primary care providers.

Theme 1. Confidence in HIV risk prediction

Theme 1.1. MSM doubted that quantifying their risk at a single point in time could predict future HIV acquisition

Almost all MSM participants questioned the ability of a tool used at a single point in time to characterize the complex behavior of individuals or generate an accurate estimate of HIV risk. They noted that their sexual behavior could change dramatically depending on factors such as their relationship status, financial circumstances, substance use, and mental health. Furthermore, they perceived that their risk as closely linked to their partners' risk, which could be difficult for them to evaluate. MSM generally had low confidence in the ability of a numeric estimate to account for the multiple determinants of risk or variation in past and future behavior:

The problem with [a risk prediction tool] is it's all based on historical data, right? …. There's some ability to make some jackass prediction. But I know, just from my own personal experience, that my risk level has changed with my mental status. Meaning that, you know, there have been low points in my life, where I felt like I was more willing to put myself at risk than, say, right now.—MSM Participant #25

Some MSM conceived of HIV infection as an inherently unknowable or unpredictable event, ultimately determined by luck or fate more than any measurable risk factor. Such perceptions further undermined confidence that HIV risk could be summarized into a single number with any predictive value:

[Risk prediction] just seems pretty generic …. Anything can happen in the next five years. You can be very lucky or you can be very unlucky. … you can meet someone and you can have unprotected sex, they can lie to you, and you can be a receptive partner, and be very unlucky …. So it's a very random thing.—MSM Participant #1

Theme 1.2. Providers were more confident than MSM about quantifying HIV risk

Providers in our sample were familiar with the concept of risk prediction and often referenced risk prediction tools they already used, such as the Framingham risk score for cardiovascular disease. In contrast to MSM participants, few providers rejected the idea of HIV risk prediction outright, but rather predicated their confidence in the tool on validation:

I would welcome [an HIV risk prediction tool] as long as it's gone through some rigorous testing. And I think it would just be another tool to use … like a depression scale, a PHQ9 [Patient Health Questionnaire 9] or something.—Provider #9, LGBTQ Specialist

Despite being generally comfortable with the concept of risk prediction, some providers did share MSM participants' concern that using the tool at a single point in time would fail to capture changes in behavior and that population data were not always applicable to individuals. Several providers additionally worried that their patients' tendency to underreport HIV risk behaviors could compromise the tool. However, in only a few cases were these concerns serious enough to undermine providers' confidence entirely.

Theme 2. Interpretation of risk scores

Theme 2.1. Perceptions among MSM of what would constitute “high” and “low” HIV risk prediction scores varied widely

When asked what they would view as a “high” risk prediction score, MSM participants gave answers ranging from 0% (or “any”) to 100%. Within this range, about half of responses clustered at the low end of the scale (0–15%), and another one-third clustered around 50%, suggesting a tendency to mean-center risk:

I feel like 51% up and above [is high risk]. Because 50% is, like, the median, it's like average … You're in the middle. One hundred percent is, like, you're fully high risk. You're doing everything in your possible way to get HIV ….—MSM Participant #15

MSM participants' perceptions of “low” risk prediction scores ranged from 0% to 50%, with most specifying 10% or less. The exercise of naming high and low scores was difficult; many expressed uncertainty about their answers, and some named multiple scores or resisted answering:

[What is] a low risk number? Like, for instance? I'm not sure …. I think you should always be worried about it. I don't think there is an acceptable percentage because you might be that unlucky one. And I don't want to be the unlucky one.—MSM Participant #16

Theme 2.2. Providers most often placed “high” HIV risk at 10% or lower

Like MSM participants, providers were uncertain about what would constitute a high or low HIV risk prediction score. Those who responded most often defined high 5-year risk as 5–10%, and several mentioned that their experience with the Framingham risk score informed their answer. Providers noted the need for research to inform such classifications, as well as a preference for gradients or ranges of risk rather than single-point thresholds.

Theme 3. Anticipated impact

Theme 3.1. Many, but not all, MSM believed that receiving a high HIV risk prediction score would prompt them to consider using PrEP or changing their sexual behavior

Despite expressing skepticism about the predictive power of HIV risk prediction tools, most MSM reported that, if presented with a high score, they would seek to reduce their risk. Some imagined changing their sexual behavior, for example, by using condoms more frequently or having fewer partners, whereas others anticipated contemplating PrEP use:

If my healthcare provider says that I'm at more risk, then I may go back to making myself less risky in behavior. Or I might look at options like PrEP.—MSM Participant #8

A smaller proportion of MSM did not believe that risk scores would prompt protective behaviors. Some felt unable to change or that they were comfortable accepting their current level of risk. Others feared that a high risk score would be so demoralizing or anxiety provoking as to cause them to “give up” trying to reduce their risk. Several noted that their response would likely be commensurate with the magnitude of the score:

I think it depends on what percentage they are giving you. If a healthcare provider said [my 5-year risk of HIV infection was] 25–50%, I think I would probably think about PrEP, maybe start using it, maybe change some of my sex practices to be safer. But if, you know, if it's a really high number … I think that would make me very despondent or not willing to try anymore. Like, you have a 90% chance of getting HIV, so you might as well get it.—MSM Participant #24

At the other end of the risk spectrum, some MSM expressed the concern that receiving a low score could encourage HIV risk behaviors through behavioral disinhibition:

[If I received a low risk score] I would just be, like, “Oh, thank goodness.” But that is not helpful because then I would feel like I could do [high risk behaviors] with more people.—MSM Participant #12

Theme 3.2. Providers anticipated using HIV risk prediction as one of an array of strategies to inform their counseling and prescribing practices

Most providers imagined that they would use an HIV risk prediction tool in some capacity. Some believed that the tool would have value in identifying high-risk patients and triggering further discussion about PrEP and other HIV prevention strategies but noted that they would not rely on the tool alone. Rather, they would consider risk scores alongside other information, including their discussions with patients, their “gut feeling” about the patients' risk, and patients' history of sexually transmitted infections:

I think that if in that moment I feel like what [my patient is] doing is riskier than what [their] percentage is showing, I think I would go with my feeling. If I felt like they were doing something less risky than the percentage, I would go with the percentage.—Provider #2, LGBTQ Specialist

While providers believed a high risk score would prompt them to counsel on safer sex practices or to discuss PrEP, few saw a low risk score as providing sufficient reason to rule out preventive counseling and services. Compared with the generalists, LGBTQ specialists expressed more confidence in their existing approaches to assessing HIV risk, and perhaps for this reason, more often prioritized their intuition or assessments over a risk prediction score.

In addition to informing their own clinical decisions, many providers anticipated using a risk prediction tool to inform patients' decisions. Providers described risk prediction as a potential “teaching tool” that they could use to help patients' visualize their risk and motivate behavior change:

[An HIV risk prediction tool] could be educational for the patient in terms of what their risk is for becoming HIV positive …. I like risk prediction tools to the extent that you can modify and show patients who have different levels of risk …. It sort of makes it meaningful to look at their risk and have an estimate. There's a discussion about how one may modify their risk.—Provider #15, Generalist

Theme 4. Acceptability and feasibility

Theme 4.1. MSM differed on their views of the acceptability of HIV risk prediction tools and emphasized the need to contextualize risk scores

MSM participants expressed at least three views about using an HIV risk prediction tool. One group was amenable to receiving risk scores, which they viewed in dispassionate terms as comparable to information received from other routine screening tests or even fitness trackers. For this group, a risk score was simply another data point that might help guide their decision making, and they perceived the “objective” nature of the score as a benefit:

I would think [an HIV risk prediction score] actually could be helpful, in a way. I mean, obviously, I would take it with a grain of salt …. But I think it would help me understand where I am, you know how I've been doing, you know where I am. It would be an objective way of looking at my behavior, more than a subjective way. And I think that would be helpful in a way. ‘Ok, do I continue on the course that I'm on? Or do I need to readjust?’—MSM Participant #23

A second group also believed that risk scores could be useful for guiding their behavior but imagined that receiving the score would be fraught with fear, anxiety, or guilt. This group viewed risk scores as possibly motivating, but unpleasant and always on their minds:

I'd get depressed, I'd worry. When's [HIV infection] going to happen? Is it going to happen? Can I beat the odds? … And if there are things that can lower the percentage, then obviously I would do them.—MSM Participant #16

A third group rejected the risk scores as being unacceptable or unhelpful. Some believed that the numeric description would be “marginalizing” or dehumanizing. Others referenced their lack of confidence in risk prediction, as described in Theme 1:

I probably would blow it off. [I] wouldn't give [an HIV risk prediction score] much thought, because no one knows stuff like that …. Every time you have sex with someone, you do run the risk of getting infected. But I don't think like that …. It ain't easy, but I would try to ignore it.—MSM Participant #11

MSM who offered advice for increasing the acceptability and usefulness of risk scores focused on opportunities to overcome the limitations of presenting risk as a percentage. For example, they recommended using natural frequencies (e.g., one of five people) to make the scores more “human” or presenting risk in relative terms as lower or higher than average. Others noted that they preferred having health care providers or others interpret and contextualize risk, rather than receiving a numeric risk score.

Theme 4.2. Most providers believed that HIV risk prediction tools could be useful but noted that their feasibility would depend on careful implementation

Providers' views on the acceptability of HIV risk prediction tools were positive on the whole, with many seeing value in the opportunity to make risk assessments more routine, streamlined, evidence-based, and comfortable:

I think [an HIV risk prediction tool] would be useful. I think it would increase my own level of comfort, knowing that I was using a validated tool. As opposed to me kind of floundering through questions.—Provider #18, Generalist

Consistent with their confidence in assessing HIV risk, some LGBTQ specialists expressed the view that risk prediction tools were most likely to be helpful to generalists with less experience prescribing PrEP or providing care for MSM.

At the same time, providers noted potential feasibility challenges. Time constraints were a primary concern, and providers emphasized the need for tools to be very brief and easy to interpret. Some providers expressed a preference for a self-administered tool that patients could complete in the waiting room or even before the office visit, whereas others favored in-person administration because of concerns about literacy and privacy. In either case, providers noted the need for attention to the sensitivity of the task:

Time in the office is always the thing that we're thinking about …. Because already we're doing Framingham Risk and PHQ9 for depression, so there's a lot of scores and calculators going around …. A lot of those ones we have in the medical record to give patients at check in. But [an HIV risk prediction tool] might be one that is not good to give someone right at check in because of the embarrassing questions on it. And patients may not be that honest with these kinds of things.—Provider #22, Generalist

Both generalists and LGBTQ specialists expressed a variety of views about how to integrate HIV risk prediction tools into clinical workflows, suggesting that when, how, and how often they could be administered would vary by practice or provider.

Providers also noted the need for care in conveying HIV risk prediction scores to patients. Some were concerned about the possibility that their patients could underestimate or overestimate their risk based on their scores and expressed uncertainty about how best to meet patients' communication needs around receiving a risk score.

I think you're really having to think very clearly of how are you going to express that risk in a way that is supportive and affirming of the person who is getting the score.—Provider #20, Generalist

Discussion

This qualitative study found support among PCPs, including both generalists and LGBTQ specialist providers, for using HIV risk prediction tools to inform clinical counseling about HIV prevention strategies, including PrEP. In our sample, providers saw value in using a validated quantitative tool, alongside their clinical judgment, to assess risk. PCPs were particularly enthusiastic about using such a tool to engage patients in counseling about behavioral risk reduction. Compared with providers, MSM participants were more mixed as to the acceptability of HIV risk prediction tools, due in part to a lack of familiarity with or confidence in risk prediction. Nevertheless, many believed that receiving a risk score could be helpful. Our study suggests that HIV risk prediction tools show promise for use among providers and some patient populations and that their usability and clinical impact warrant further evaluation.

Many participants in our study discussed challenges to implementing HIV risk prediction tools in clinical practice, with a primary concern being risk communication. Both providers and MSM anticipated that patients would interpret risk scores differently and expressed concern that misinterpretation of risk could lead to false reassurance in the case of underestimation or needless worry and intervention in the case of overestimation. Our exploration of how participants conceived of high or low risk scores similarly suggests the potential for variation in interpretation. In the absence of contextual information, participants struggled to locate “high” risk, at times anchoring their answers on the midpoint of 50% or their experience using other risk prediction tools. These findings speak to the need to carefully define and contextualize HIV risk prediction scores and are consistent with the broader literature on health numeracy and personalized risk assessment.^45,46 This literature finds that providing a numeric score does have value and is preferable to the presentation of qualitative descriptors, such as “low risk” or “high risk,” alone.⁴⁷ Other best practices include the use of absolute versus relative risk, pictographs or other graphical displays of risk, and both positive and negative frames for describing the estimated chance of getting and not getting HIV.⁴⁷

Our findings lay the foundation for piloting validated HIV risk prediction tools in clinical practice. Next steps include the development of formats and displays for communicating HIV risk scores to patients, trainings and discussion guides to prepare providers for using such tools in a culturally competent way, and processes for integrating tools into clinical workflows. Our findings suggest that, in addition to decision making, such evaluations should carefully assess unintended consequences of using HIV risk prediction tools. Some MSM in our sample believed that receiving their score would elicit anxiety or fear, and a few did not want to receive scores. Given these differing views, implementation of HIV risk prediction tools may benefit from a shared decision-making approach in which patients are given a choice in whether and how to engage with such tools.

This study provides novel data on the usability of HIV risk prediction tools from the perspective of three key groups of intended users: PCPs in general practice, LGBTQ specialists, and MSM at higher risk of HIV acquisition. Limitations include recruitment of English-speaking adult participants from a single Northeastern urban area. To maximize variation in our sample, we recruited participants from both an LGBTQ health center with relatively high PrEP adoption and an academic medical center with more typically low adoption. However, our findings may be less transferable to suburban or rural areas, to private practices or nonacademic medical centers, or to other populations where PrEP acceptability, accessibility, and uptake are lower, such as people of color and those living in the South.^10,48 With regard to patients, our findings are most relevant to understanding the perspectives of adult MSM who report HIV risk behaviors and may be less transferable to patients who are younger, not MSM, non-English speaking, or who do not engage in HIV risk behaviors. Although our qualitative approach was well suited to the study's aim of understanding the range of ways that intended users conceive of and anticipate adopting HIV risk prediction tools, our findings cannot quantify the prevalence of those views or speak to how those views have changed over the past several years with modest increases in PrEP adoption in general practice. Future research can build on the present study by assessing the acceptability of HIV risk prediction tools quantitatively in larger and more diverse samples, with a particular focus on understanding the perspectives of MSM who underestimate their risk of HIV acquisition, perceive stigma related to PrEP use, or lack access to primary care.^49,50 Finally, given that some MSM in our study indicated that receiving a low risk score could encourage HIV risk behaviors, future implementation studies should assess the impact of HIV risk prediction tools on sexual behavior.

In conclusion, PCPs can play a central role in expanding access to PrEP, but to do so, they will need tools for efficiently identifying MSM who are good candidates. Our findings indicate that PCPs see value in using HIV risk prediction tools to increase their capacity to assess risk and counsel patients on risk reduction. Alongside testing of tools' predictive validity, future research is needed to guide the implementation of risk tools in a supportive and culturally competent way to help patients more accurately understand and manage their HIV risk.

Author Disclosure Statement

J.L.M. has received research grant support from Merck and has consulted for Kaiser Permanente Northern California on a research grant sponsored by Gilead Sciences. D.S.K. has received funding to complete educational content relating to HIV prevention for Medscape, MED-IQ, DKBmed, and UptoDate, Inc., and has been a consultant to Fenway Health on a research grant sponsored by Gilead. All other authors (M.B.G., J.M.G., V.E.P., and K.M.M.) declare no conflicts.