The Parkland Center for Clinical Innovation (PCCI) is using AI to tackle a persistent problem: HIV infections. PCCI's Vice President of Clinical and Social Health explains how it works.
Editor’s Note: Jacqueline Naeem, MD, is the Vice President of Clinical and Social Health for the Parkland Center for Clinical Innovation (PCCI) in Dallas. She was also the program leader for the U.S. Centers for Medicare & Medicaid Services (CMS) Accountable Health Communities (AHC) Model in Dallas County.
During the last decade, we have seen major breakthroughs in preventing HIV infection. However, even with these advances, that infection rate has not appreciably dropped. Despite improvements in morbidity and mortality associated with HIV due to antiretroviral therapy and the availability of an effective preventative medication, the incidence of HIV has only modestly decreased, with a 9% decrease between 2015 and 2019, to a total of 36,136 cases in 2021.
Pre-exposure prophylaxis (PrEP) has emerged as a highly effective preventive strategy for HIV, reducing the risk of HIV infection by up to 99% when taken consistently. Due to its effectiveness, the CDC recommends that medical providers counsel and prescribe PrEP to all sexually active patients if they are at risk for HIV infection. However, despite its efficacy, PrEP remains underutilized, in large part due to lack of awareness.
This is where AI has stepped in to significantly advance our HIV prevention efforts.
Jacqueline Naeem, MD, Vice President of Clinical and Social Health at the Parkland Center for Clinical Innovation (PCCI). Photo courtesy PCCI.
With its position as North Texas' largest safety-net hospital system, Parkland Health serves an extensive population of at-risk patients, creating a vital opportunity to address this problem and enhance HIV testing and facilitate connections to PrEP programs. In Dallas County, we found an opportunity where we could apply our AI-driven modeling to solve a difficult situation. According to HIV.gov, the HIV incidence rate in Dallas County was 41.4 per 100,000 people in 2022, which is significantly higher than the state average of 17.1 per 100,000 people.
Although we knew the mission was clear, the challenge was also great. We have an effective preventive treatment— PrEP, and opportunities to reach PrEP candidates—through Parkland, but what we were lacking was a way to identify candidates for referral in a simple way that could be incorporated into Parkland’s workflows and be paired with provider tools to guide discussion and assessment of indications and eligibility criteria for PrEP.
To address this critical gap, we developed and implemented a predictive model, PCCI’s HIV Detection AI/ML Model, informed by EHR data and paired with provider tools to guide discussions on PrEP eligibility criteria, to efficiently identify (and target for outreach) individuals who stand to benefit most from PrEP.
PCCI’s HIV Detection AI/ML Model project work began in the latter part of 2020. Once underway, we then worked with Parkland’s IT department to integrate the developed model for provider alert-based, risk-stratified interventions, in silent mode. We then automated PrEP Model load to the Parkland test table for piloting and testing the workflow. We also identified the patient population cohort eligible for HIV risk scoring.
In late 2022, the model went live, using information from the EHR to predict the individuals at increased likelihood of acquiring HIV and who may be candidates for HIV PrEP. Once identified, the patients can be offered HIV testing, and if negative can be offered PrEP.
So far, the HIV Detection AI/ML Model has risk stratified hundreds of thousands of patients, demonstrating that machine learning models can be used for predicting and classifying the risk of HIV using available EHR data.
We see this as a breakthrough for identifying candidates who are at risk for HIV infection. PCCI’s HIV Detection AI/ML Model has been shown to effectively address the needs of vulnerable populations and can be implemented in hospital settings with limited resources. There are opportunities to expand this model to reach even more patients in Dallas County, through an additional project underway with Dallas County Health and Human Services.
Leveraging predictive models within Parkland and Dallas County allows providers to identify individuals at high risk for HIV acquisition and those who are prime candidates for PrEP. By doing so, we can implement proactive interventions that can bridge critical gaps in the HIV prevention cascade, thereby contributing to the broader goal of reducing HIV incidence in Dallas County.
