Leveraging Electronic Health Record Data in a Risk Prediction Model to Increase HIV PrEP Intake in Primary Care

Overview

Kaiser Permanente Northern California (KPNC) has been a national leader in implementation of HIV preexposure prophylaxis (PrEP), a daily antiretroviral pill that is more than 99% effective in preventing HIV acquisition. However, many patients at risk for HIV at KPNC are not yet using PrEP, resulting in preventable infections each year. In prior work, the study team developed and validated a prediction model that used electronic health record (EHR) data from 3.7 million KPNC members to identify patients at high risk of HIV acquisition but not using PrEP. The model substantially outperformed models based only on CDC criteria for PrEP use, particularly for Black patients, a population with high HIV incidence and lower PrEP uptake. The objective of this proposal is to evaluate the feasibility of implementing this EHR-based model at the point of care to increase PrEP referrals and uptake at KPNC. The specific aims of this project are to 1) conduct provider focus groups to identify barriers and facilitators to PrEP referrals and to optimize the delivery mechanism of our clinical decision support intervention, 2) evaluate the feasibility and acceptability of a clinical decision support intervention for primary care providers (PCPs) to increase PrEP referrals and uptake among high-risk patients, and 3) assess patient- and provider-based characteristics associated with PrEP referrals and uptake. To accomplish these aims, the study team proposes a randomized controlled trial of a clinical decision support intervention for PrEP, which involves alerting KP San Francisco (KPSF) Adult and Family Medicine PCPs about patients identified by our prediction model as being at high risk for HIV acquisition prior to in-person clinic visits. We will compare PrEP referrals and uptake among patients who are seen by PCPs randomized to intervention and usual care study arms using an intention-to-treat analysis.

Study Design This is a randomized controlled trial among PCPs at KPSF to evaluate the feasibility of implementing a decision support intervention for PrEP and its impact on PrEP referrals and uptake among patients predicted to be at high risk for incident HIV infections. Aim 1 will consist of focus groups with PCPs to identify barriers and facilitators to PrEP referrals and to optimize the intervention. After first testing the proposed study intervention with two PCPs in each study arm, Aim 2 will compare PrEP referrals and uptake between members who are seen by PCPs randomized to intervention and usual care study arms. Aim 3 will evaluate provider- and patient-based factors associated with PrEP referrals and uptake among patients in the intervention arm. The trial will randomly allocate PCPs at KP San Francisco to one of two arms: 1) intervention arm that includes clinical decision support for PrEP referrals prior to visits scheduled with patients flagged as high-risk by the prediction model, or 2) usual care, with an estimated 50 PCPs in each arm. Eligible PCPs will be contacted by the study team prior to enrollment and given an opportunity to opt out of participating in this study. PCPs randomized to the usual care arm will not receive the clinical decision support intervention. The intervention will be conducted over 8 months. Setting. KPNC is a large healthcare delivery system that provides integrated pharmacy, laboratory, and medical care to more than a third of insured individuals in California. KPSF has been chosen as the study site given the disproportionate burden of HIV in this KPNC service area. At KPSF, there are 121 Adult and Family Medicine (AFM) PCPs serving approximately 189,000 patients. Subjects. The study team will recruit 10-15 PCPs each from the two KPSF clinical sites in San Francisco prior to randomization. Providers will be recruited at routine staff meetings and through email invitations for a one-time, 60- minute, audio-recorded focus group. The study team will assess what information PCPs would like to have included in the clinical decision support tool. Recognizing that time constraints are a barrier to all healthcare interventions, we will seek feedback from PCPs regarding their preferences for intervention delivery and timing, as well as their perceptions of facilitators and barriers to acting on the intervention. Open-ended questions will minimize bias and allow for unanticipated themes to emerge. Randomization. PCPs who meet study criteria and do not opt-out will be randomized to one of the two study arms in a 1:1 ratio. There are 12 providers who care for both HIV-infected and uninfected patients, and allocation of this subset of providers will be balanced between the trial arms. To ensure this, the study team will use simple randomization to assign the 13 PCPs who care for both HIV-infected and uninfected patients to intervention or usual care and then use simple randomization again to assign the remaining 109 PCPs. Specificity of the prediction model. The prediction model acts as a screening test to be followed by a provider's clinical evaluation. The study team expects that some patients with scores indicating an elevated risk for HIV acquisition will not have clinical indications for PrEP. The relationship between PCPs and patients is critical given the sensitive nature of sexual health discussions. In addition, prior to initiation of treatment with PrEP, patients will attend visits where a member of the dedicated KPSF PrEP team (NP or MD) will assess HIV risk and discuss PrEP indications, follow-up, side effects, and contraindications. Scalability of the intervention outside KPNC. Even if the intervention is successful within KPNC, other health settings outside KPNC will need to adopt technology solutions to update risk scores, automate lists of patients with elevated risk scores, identify eligible appointments, and successfully deliver interventions to providers. While the study team's robust model that included 44 EHR variables provided the greatest predictive value for incident HIV, simpler models that included only a handful of key variables also improved identification of patients who may benefit from PrEP. The study results will provide a roadmap for successful implementation at other institutions, even if technology and EHR resources are more limited.