HEalth Record Optimization for Identifying Candidates for HIV PRe-Exposure Prophylaxis: A Community-Informed Approach to Model Development

Abstract

Electronic health record (EHR)-based models to identify individuals who may benefit from pre-exposure prophylaxis (PrEP) outperform traditional risk scores and may alleviate challenges associated with PrEP initiation. Pre-implementation work is critical to ensure algorithms are optimized for the local context, particularly given regional differences in the US HIV epidemic. To inform the derivation and implementation of EHR-based models within health systems in New Orleans and Baton Rouge, Louisiana, we conducted focus group discussions (FGDs) with community advocates and in-depth interviews (IDIs) with emergency department, primary care, and HIV-trained clinicians. We asked about their perspectives on HIV epidemiology and PrEP uptake and sought suggestions for locally relevant variables to optimize model performance. FGDs and IDIs were audio-recorded and analyzed using thematic analysis. From January to March 2023, FGDs were conducted with 18 community advocates and IDIs with 12 clinicians. Community advocates did not believe that PrEP had reduced local HIV incidence, primarily due to a lack of inclusive marketing. Clinicians noted that improving PrEP uptake would require better access to education, PrEP providers, and affordable medication. Community advocates suggested adding sexual assault history and number of pregnancies to the model; clinicians suggested adding hepatitis B, more sexually transmitted infection treatment modalities, incarceration history, and opiate use. To optimize model implementation, community advocates emphasized the need to convey model output respectfully and compassionately, and clinicians suggested involving ancillary staff in PrEP discussions. Although evidence supports the use of EHR-based models to identify PrEP candidates, local stakeholders can provide unique insight into optimizing model performance and implementation.

Introduction

While HIV pre-exposure prophylaxis (PrEP) is remarkably effective at preventing HIV acquisition, its uptake has not been distributed equitably.¹ In the United States, more than half of new HIV infections occur in the Southeastern part of the country (“the South”); meanwhile, the region accounted for only 39% of PrEP users in 2022.² In the South, HIV diagnosis rates are significantly higher than in other regions, even among PrEP users.³ Additionally, in 2022, there were only five Black PrEP users for every new HIV diagnosis among Black persons compared with 27 PrEP users per new HIV diagnosis among White persons in the region.²

Many factors contribute to the slow and uneven uptake of PrEP on a population level. Systems-level barriers such as cost and insurance challenges, lack of awareness and education, and logistical hurdles have been repeatedly identified.^4–8 Additionally, stigma, culture, risk perception, and discomfort discussing sexual practices are barriers on both the provider and client levels.^5,8 Providers may also have biases that hinder equitable prescription of PrEP among clients.⁹ Machine learning models that use electronic health record (EHR) data to identify individuals who may benefit from PrEP can alleviate challenges associated with PrEP initiation, as they have potential to reduce bias in identification of PrEP candidates and mitigate discomfort associated with discussions about PrEP.^10–12 Further, such models have been shown to outperform traditional risk scores, thereby further promoting equity.^13,14

Use of these risk scores, however, must be considered in the context of concerns about privacy and sensitivities around the use of EHR-extracted sexual health data. While potential patients see value in receiving a risk score, they also acknowledge that such scores may engender feelings of fear, anxiety, and mistrust.¹⁵ Engagement of local community leaders and clinicians early in the development phase is thus critical to ensure algorithm development and deployment is acceptable and useful to clients and clinicians. Further, local voices can inform variable input based on the local context, which is a necessary step given regional differences in the US HIV epidemic. HEalth Record Optimization for Identifying Candidates for HIV PRe-Exposure Prophylaxis (HEROIC-PrEP) is a multi-stage study to derive, validate, and implement an EHR-based model to identify PrEP candidates within two large health systems in Southeastern Louisiana. Uniquely, the project involves extensive formative work to determine best practices for model development and implementation.

Methods

Study design

Two focus group discussions (FGDs) of nine participants each were conducted in-person in January 2023 in New Orleans and Baton Rouge. A leader of a local community-based organization focusing in HIV testing, linkage to care, and/or prevention recruited community advocates at his or her discretion, aiming for participants with relevant expertise (e.g., person living with HIV, community health worker, and linkage coordinator). Additionally, 12 in-depth interviews (IDIs) with clinicians were conducted virtually from January to March 2023. To ensure a range of perspectives, clinicians were purposively selected from two large health systems in Baton Rouge and New Orleans. FGDs and IDIs were recorded on an encrypted audio device, and files were professionally transcribed by Rev.com. The research activities were reviewed and approved as an exempt protocol by the Louisiana State University Health Sciences Center-New Orleans and Duke University Institutional Review Boards.

IDI and FGD content

Participants were first given an overview of our previously validated HIV risk prediction model.¹⁶ The overview provided a brief evidence base around PrEP, the rationale for the use of machine learning models to identify persons who could benefit from PrEP, and concluded with the variables from our previously validated model.

Analysis

IDIs and FGDs were analyzed separately, although results were integrated in the interpretation phase and are presented together. Structural codes were created based on the interview topics and questions. A second round of thematic coding allowed for similar information and recurrent themes to be identified and grouped across transcripts. To ensure reliability, two analysts conducted intercoder agreement assessments on both the structural and thematic codes. Disagreements and coding inconsistencies were discussed, and codebook changes were made as needed. Once coding was complete, analysts wrote narrative summaries of the data within each domain to provide further context for the thematic codes and overall findings.

Results

In total, 18 community advocates participated in the FGDs. Eleven participants identified as women, six identified as men, and one identified as agender. Sixteen participants identified as Black, one identified as American Indian or Alaskan Native, and one identified as “Other.” No participants identified as Hispanic/Latinx (Table 1).

Table 1.

Demographic Characteristics of Community Advocates and Clinicians

Variable	Community advocates n = 18	Clinicians n = 12
Race
White	0 (0%)	11 (91.67%)
Black or African American	16 (88.89%)	1 (8.33%)
American Indian or Alaskan Native	1 (5.56%)	0 (0%)
Other	1 (5.56%)	0 (0%)
Gender
Woman	11 (61.11%)	6 (50%)
Man	6 (33.33%)	6 (50%)
Agender	1 (5.56%)	0 (0%)
Age
21–30	1 (5.56%)	0 (0%)
31–40	3 (16.67%)	5 (41.67%)
41–50	8 (44.44%)	4 (33.33%)
51–60	3 (16.67%)	2 (16.67%)
61–70	3 (16.67%)	1 (8.33%)
Hispanic/Latinx
No	18 (100%)	12 (100%)
Yes	0 (0%)	0 (0%)
Years working as provider
More than 10 years		8 (66.67%)
5–10 years		3 (25%)
3–4 years		1 (8.33%)
1–2 years		0 (0%)
Less than 1 year		0 (0%)

Twelve clinicians participated in interviews, six from each health system; clinicians were from primary care (n = 3), emergency department (n = 4), infectious disease (n = 4), and one hospitalist (Table 1). Half identified as men and half as women. Eleven clinicians identified as White, and one identified as Black. No participants identified as Hispanic/Latinx. The majority of participants (n = 8) reported having worked as a clinician for more than 10 years.

Community advocates’ perspectives on local HIV epidemiology and PrEP

Community advocates in New Orleans and Baton Rouge reported that recent increases in HIV awareness and education have had a beneficial impact on attitudes toward testing and prevention services in their communities. They reported noticing an overall positive change in the HIV landscape over the years, and a few participants attributed this change to scientific advancements in prevention and treatment. One participant reported,

“I think that’s basically because a lot of education has been given to the community and not only the education, but people have seen the diseases in different people. But another thing, it’s not only the knowledge part that has changed, also the… different types of treatment and prevention methods are available now.”—Participant 13, Community Advocate, Baton Rouge

Despite these improvements, community advocates noted a perceived increase in HIV diagnoses, attributed to a lack of financial resources and inadequate adolescent sex education. Community advocates emphasized that adolescents, Black women, and Black men who have sex with men (MSM) are disproportionately impacted by HIV. Participants also noted several other groups at elevated risk, including individuals engaging in survival sex, individuals who are sex trafficked, and those with low health care access. Participants expressed that individuals do not seek preventative care, often because of past experiences in which they were met with judgment or lack of respect.

Community advocates reported a lack of PrEP education and inclusivity around PrEP access as the two biggest barriers to PrEP uptake. They noted that PrEP pharmaceutical commercials have historically marketed PrEP toward White, gay men. One community advocate reported,

“I think it’s a lot of miseducation as well around PrEP. And I say this because I have experience testing in the local jail and I always ask the question, “Have you heard of PrEP? Do you know anything about PrEP?” …and they’ll be like, “Oh yeah, yeah, I saw that. That’s the gay commercial. I don’t need that.” …it was rolled out totally geared toward one group of people when it should have been opened up to, ‘this is for everybody.’”—Participant 19, Community advocate, Baton Rouge

Community advocates emphasized how PrEP commercials are not inclusive of the Black community, specifically Black women, and expressed feeling “left behind” and “not relevant” in the PrEP conversation. They discussed hardships that are unique to Black women, such as health care access, and how PrEP education and advertisements should be cognizant of their needs.

Community advocates discussed the importance of community relationships and meeting people where they are to provide PrEP information. Many participants agreed that providers need to earn the trust of community members to facilitate health care access and thus PrEP access. Some community advocates felt that the increasing willingness of adolescents and young adults to discuss “risky” sexual behaviors could serve as an avenue to discuss PrEP.

Clinicians’ perspectives on local HIV epidemiology and PrEP

Clinicians discussed a wide range of social determinants of health that are associated with HIV incidence and lack of PrEP usage, including low socioeconomic status, area of residence (particularly rural locations), drug use including injection drugs, limited health care access, and lack of education. One clinician noted that many have more immediate needs compared with HIV prevention, stating,

“If you’re hungry and you don’t pick up your food, you’ll feel that. But if you’re using IV drugs or a sex worker and don’t pick up your prophylactic medications, it’s not going to impact you at the moment.”—Participant 8, Emergency Medicine Clinician, New Orleans

Clinicians reported seeing HIV cases among all age groups but especially among individuals in their teens to their 30s–40s. Clinicians reported seeing HIV diagnosed among Black and Hispanic/Latinx patients more than among White patients. Black MSM was the demographic group mentioned most frequently (by all clinicians), followed by Black heterosexual and cisgender patients. Hispanic and Latinx MSM and transgender individuals were mentioned less frequently.

Clinicians noted that stigma associated with HIV and PrEP is a barrier to PrEP access as it contributes to the avoidance of conversations around HIV risk and prevention. One clinician noted,

“I wonder if the climate here is such that a lot of people who could be facilitators…maybe have their own biases and stigma about becoming a clinic for folks who have certain behaviors and they don’t want to be associated, they just want to be a regular clinic, so to speak.”—Participant 11, Infectious Diseases/HIV Clinician, New Orleans

Additionally, clinicians noted insufficient health care infrastructure to support PrEP access and usage. Clinicians reported a lack of training around PrEP prescribing and identifying patients at risk of HIV. They perceived that the current workflow does not build in opportunities to discuss HIV prevention, which is exacerbated further by workload. As one clinician shared,

“I think the biggest barrier is always when you have a list of sometimes dozens of things that you have to ask and check off in terms of healthcare maintenance… .”—Participant 2, Primary Care provider, New Orleans

Clinicians felt that patients who could benefit from PrEP may not have established routine health care or be able to afford health care costs, particularly those with low socioeconomic status. Providers also reported a lack of clinics providing comprehensive HIV prevention services, particularly in rural areas. Lack of transportation and proximity to providers were seen as barriers to PrEP. Lastly, clinicians noted that many patients do not know what PrEP is, and if they do, they have little knowledge about who could benefit. As one provider said while referencing a PrEP billboard,

“The average person riding down the interstate, that doesn’t mean anything to them. They don’t know what PrEP is.”—Participant 4, Emergency Medicine clinician, Baton Rouge

The exception to this is in young, White, gay communities, in which PrEP usage is far more common, and where it was felt to be having an impact.

Community advocates’ perspectives on model implementation

There was some optimism from community advocates around the idea of a prediction model implemented at local facilities to facilitate conversations around HIV that might not otherwise happen. They discussed the opportunity to ask about HIV risk at the doctor’s office or urgent care in a routine manner, in the same way questions about blood pressure or other chronic diseases might be discussed.

Community advocates were, however, concerned about how HIV risk and model output would be communicated. They felt that there would be a variety of reactions to delivery of HIV risk information, ranging from acceptance to denial and from confusion to offense. They also highlighted the importance of thoughtfully selecting the health care professional who delivers the model output. One community advocate reported,

“I just think a risk calculator can be misconstrued. It can be taken the wrong way… Because if you’re a regular nurse office assistant behind the window and, I don’t relate to you, and you just like, ‘Okay, this is what’s going on with you because this is what the computer shows.’ I’m going to be highly offended because you don’t know what’s going on with me.”—Participant 25, Community advocate, New Orleans

Another advocate shared,

“You’re going to have definite mixed reactions to something like that. Some people are going to be like, okay, yeah, well maybe I am at high risk because I am doing a little stuff. Then they’re going to have other people that say I’m not at high risk. Like she just said, I’m married, I didn’t do nothing.”—Participant 30, Community advocate, New Orleans

Community advocates felt that younger populations would be more receptive to receiving this information than older populations. Participants wanted to focus on prevention options as opposed to “risk.” Many community advocates felt that model output should be communicated by someone that individuals respect and trust, and ideally someone with whom they have an established relationship. This professional should be culturally competent and practice excellent bedside manner. Community advocates discussed medical assistants, nurses, and community health care workers as having more cultural competency than doctors and that they might be more appropriate to discuss HIV risk and prevention.

Community advocates also stressed the importance of how the HIV risk conversation is approached. Respect and dignity were emphasized, as well as clear explanations of what EHR variables contributed to their elevated HIV risk, why HIV prevention services would help them, and options on how to move forward (i.e., providing condoms and not just PrEP prescribing). Because HIV stigma is still prevalent in the medical field, compassion and empathy in communicating increased risk were thought to be essential to successful model implementation.

Clinicians’ perspectives on model implementation

Some providers had enthusiasm for the model and emphasized the feasibility of model implementation if the right health care team members were selected to act upon the model’s output. One clinician shared,

“I think this is fantastic. And if you can always know the audience of who you’re asking to do the work, then I think it could be successful.”—Participant 7, Emergency Medicine Clinician, Baton Rouge

Another was optimistic about utilizing the EHR for identifying PrEP candidates:

“If you do end up with a measure that gets validated and is widely applicable, gosh, I think putting it into an EHR just really has a lot of potential and a lot of possible fruit.”—Participant 11, HIV/Infectious Diseases Clinician, New Orleans

However, this provider and others voiced concern about the existing clinician burden.

“The whole concept of one more thing as a physician is exhausting… There’s so much stuff that just piles up…There’s a lot of ‘one more things.’”—Participant 11, HIV/Infectious Diseases clinician, New Orleans

Clinician suggested having a designated nonprovider staff member such as an integrated PrEP care coordinator or social worker who could initiate PrEP conversations and facilitate follow-up care.

“If you’re going to rely on the provider to do [it], it’s just not going to get done.”—Participant 1, Emergency Medicine clinician, New Orleans

Many providers emphasized the importance of making implementation as simple and easy as possible to facilitate adoption. An automatic EHR alert was suggested as a mode to relay patient risk information. Other suggestions included basic training for providers on how to effectively prescribe PrEP. Encouragingly, one provider noted,

“I think it would be enhanced if there was some education to the providers as far as how effective the medications actually are, because… I didn’t realize how truly effective they were. So I think that educating the providers and then putting them in the chart, so it’s easier to do, I think would help.”—Participant 8, Emergency Medicine clinician, New Orleans

Providers also suggested developing order sets for the EHR and having a simple paper handout for patients to facilitate PrEP discussions.

HIV predictive EHR model variables

Community advocates suggested adding history of sexual assault/abuse (e.g., rape, incest, and sex trafficking) and number of pregnancies to the model. Number of pregnancies was felt to be a direct and indirect reflection of HIV risk, for example, a woman with children might be in a sexual relationship in order to provide financially for her family. One participant shared,

“And you have some women that will do… what they have to do, in order to survive, especially if they have children…”—Participant 13, Community advocate, Baton Rouge

The existing model variables that received the most support from clinicians (mentioned by four or more) were age, race, sexual orientation, neighborhood deprivation index, number of urine toxicology tests, and substance abuse disorder. Variables commonly suggested to add to the model were hepatitis B, broader sexually transmitted infection treatment such as antibiotics, incarceration (active or historical), opiate use or history, and a history of sexual assault. One provider noted,

“Cause I think that we will probably encounter a lot of people who potentially have experience with risky behaviors, participating in sex trade, participating in IV drug use or other drug use that might increase risk to the extent that incarceration status might be worth. And notably, we may have been overtaken now by New Mexico maybe, but I think we may have claimed it back. Not something for anyone to be proud of, but Louisiana's the most incarcerated state in the most incarcerated country in the world, right.”—Participant 10, Primary care provider, New Orleans

Clinicians reported observing cisgender women in heterosexual long-term relationships getting HIV from their partners, and suggested that variables particularly relevant to cisgender women were unintended pregnancies, number of pregnancies, substance abuse, age, and zip code.

Discussion

The analysis presented here is the first step in the model development for HEROIC-PrEP and includes formative work to understand the HIV/PrEP landscape in Southeastern Louisiana and gather perspectives on model derivation and implementation. Overall, both community advocates and clinicians shared the belief that PrEP is not reaching people who need it most. Both groups of participants generally supported a model to identify people likely to benefit from PrEP. However, there was hesitation regarding model implementation, both the “who” and the “how.” Uniquely, these stakeholder voices will inform both model derivation, including new variables to evaluate, and implementation, including strategies to pursue in hopes of optimizing reach and impact, in an approach that is truly community and clinician informed.

Community advocates and clinicians alike shared the perspective that PrEP has not mitigated HIV incidence in Louisiana in any significant way, consistent with statewide epidemiological data demonstrating little meaningful change in new HIV cases over the last several years (e.g., 999 cases newly diagnosed in 2017 compared with 858 cases diagnosed in 2022).¹⁷ Both community advocates and clinicians felt that PrEP was not reaching Black communities, and there was a prevalent sentiment among both groups that PrEP is only reaching White MSM. Community advocates additionally noted exclusion from PrEP marketing and advertising, which has been previously described in the literature.^18–20

Skepticism was expressed by both community advocates and clinicians on the ability of medical providers to meet the standard of delivering model output consistently and with the patient-centered approach required. Although early in the implementation process, understanding community and clinician perceptions of the model can help our team anticipate challenges and adapt strategies. While it may be difficult to change all clinicians’ attitudes and behaviors, incorporating cultural competency training and PrEP education into model rollout can be feasibly accomplished. As suggested, involvement of ancillary staff, who can both speak about HIV risk compassionately and offload clinicians, may be useful or even required for successful model implementation. Additionally, stakeholders recommended that PrEP conversations can be augmented by hand-outs, and that impact on clinical workflow can be minimized with order sets, suggestions which can be incorporated in the future. These identified barriers are the subject of ongoing discussions with local partners as a larger part of efforts to identify solutions to facilitate model implementation.²¹

Participants additionally shared valuable input on new variables to improve model performance. Notably, prior models have focused on or been able to identify PrEP candidates who are cisgender women—who comprise 19% of new HIV diagnoses but only account for 8% of PrEP users—although with some limitations.^2,22,23 Given regional differences in the HIV epidemic, our team previously sought to derive a model in a Southern cohort and incorporated additional variables such as trichomoniasis and pelvic inflammatory disease, which have been shown to be associated with HIV diagnosis in women.¹⁶ We demonstrated improved predictive performance of our model among women, and other models have recently done the same.^24,25 Adding newly suggested variables on the number of pregnancies or history of sexual assault has promise for improving model performance further; in fact, recent work in Chicago to develop a risk prediction model in women demonstrated that pregnancy was associated with new HIV diagnosis.²⁶

Our study has limitations that impact generalized interpretation. We engaged participants only in New Orleans and Baton Rouge, Louisiana, where our model will be implemented, and thus our findings are specific to these areas. A different group of community advocates and clinicians from outside of this region may have shared different perspectives. Additionally, because our model is still in the development phase, participants gave hypothetical feedback about acceptability. Other barriers, such as lack of perceived model accuracy and model mistrust from clinicians, were demonstrated through real-world experience with such a model,27 but such challenges would be difficult for participants to anticipate and thus comment on without first-hand experience with model implementation.

In summary, development of new models suited to the epidemiology of HIV in the South is critical. Our experience strongly supports the need for local community and clinician input to better inform context-specific models for determining PrEP candidacy in priority jurisdictions. As our team pursues model derivation, we will analyze how the newly recommended variables impact model performance. Next steps include a multi-tiered assessment to further develop and refine our implementation strategies with stakeholder input in preparation for a pilot implementation trial.

Author Disclosure Statement

M.E.C. has served on Scientific Advisory Boards for ViiV Healthcare and received grants to institution for research from Gilead Sciences and ViiV Healthcare. N.W.L. has served on Scientific Advisory Boards for ViiV Healthcare. For the remaining authors, none were declared.

Funding Information

Funding was provided by the National Institute of Allergy and Infectious Diseases (R01AI169641) and the Duke University Center for AIDS Research (CFAR), an NIH-funded program (5P30 AI064518).