ABSTRACT
With the highest HIV incidence rates observed in the US, Florida strives to develop effective and sustainable
HIV prevention strategies. HIV screening and pre-exposure prophylaxis (PrEP) are proven interventions to
prevent transmission and reduce new HIV infections. However, uptake of these HIV prevention services is low
among women relative to their needs and male counterparts. One of the key barriers to promoting HIV testing
and PrEP among women is the challenge of identifying women at risk of HIV acquisition by healthcare
providers. Researchers demonstrated potential HIV risk prediction models using electronic health records
(EHRs) among men. Unfortunately, they failed to identify HIV risk among women due to having fewer women
HIV incident cases in their datasets and a lack of risk factors tailored for women. To fill the gap, we propose to
develop an HIV risk prediction model specifically tailored for women and integrate the prediction model into
an EHR system as a clinical decision support prototype to assess its feasibility, acceptability, and usability with
primary care providers. In Aim 1, we will develop an HIV risk prediction model specific for women to identify
potential candidates for HIV testing and PrEP. Leveraging patients’ structured EHRs, ZIP code-linked
community-level factors and social determinants of health, and factors extracted from clinical notes via a state-
of-the-art natural language processing (NLP) algorithm, we will use AI/machine learning to develop and
validate an HIV risk prediction model specifically developed for women (ACTION-HIV algorithm). The results
will be used to design a clinical decision support (CDS) prototype to help providers better identify women in
need of HIV testing and PrEP. In Aim 2, using a user-centered design approach guided by the five “rights” of
the CDS intervention framework, we will conduct 6 focus groups with providers to design and prototype a
women-specific HIV risk prediction CDS tool (ACTION-HIV CDS). In Aim 3, using think-aloud protocols and
surveys, we will assess the feasibility, acceptability, and usability of the ACTION-HIV algorithm and CDS in a
simulated EHR environment with 20 primary care providers at the UF Health outpatient clinics. We will
integrate our ACTION-HIV algorithm into UF Health’s EHR system (Epic) to produce the CDS alerts for HIV
testing and pilot test ACTION-HIV CDS in a simulated Epic environment presenting synthetic patient data.
Our proposed research is highly innovative as it expands past HIV risk prediction models and pioneers in
designing and prototyping HIV testing and PrEP-related CDS in primary care settings. This project will provide
valuable insights into a future clinical trial which we will investigate the efficacy (including patient outcomes
such as rates of HIV testing and rates of PrEP prescription) of a women-specific HIV risk prediction CDS tool
within the real-time EHR production in real-world settings.