Two new studies develop algorithms to identify patients at risk of acquiring HIV

Two new studies developed algorithms that can identify patients who are at risk of acquiring HIV and may benefit from preventive care. Both studies appear in the July 5 issue of The Lancet HIV.

Preexposure prophylaxis (PrEP) is an antiretroviral pill that is over 90% effective in preventing HIV acquisition when taken as prescribed. PrEP was recently given a Grade A recommendation from the U.S. Preventive Services Task Force but is vastly underutilized. There are nearly 40,000 new HIV infections annually in the United States, yet the Centers for Disease Control and Prevention estimates that only 7% of the 1.1 million individuals at substantial risk for HIV infection used the antiretroviral pill in 2016.

One barrier to use is the difficulty for providers in identifying patients who are at high risk of HIV acquisition. Providers often have limited time, may have limited knowledge about PrEP, and may lack training in how to talk to patients about sex or substance use. Risk prediction tools, a form of electronic clinical decision support using the data in patients' electronic health records (EHRs), are often used in other areas of medicine. Researchers from both studies, one using a patient population in California and the other in Massachusetts, built HIV risk prediction models that could be used in EHRs as automated screening tools for PrEP.

The two studies looked back at the medical records of millions of patients who were HIV-uninfected and had not yet used PrEP. Researchers extracted demographic and clinical data from these patients' EHRs on numerous potential predictors of HIV risk. A machine-learning algorithm automatically selected important HIV risk-related variables for the final models.

In the California-based study, which used medical record data of 3.7 million patients at Kaiser Permanente Northern California, the final risk prediction model included such variables as sex, race, living in a neighborhood with high HIV incidence, use of medications for erectile dysfunction, and sexually-transmitted infection (STI) testing and positivity. The model flagged 2% of the general patient population as potential PrEP candidates and identified 46% of male HIV cases, but none among females.

Although risk prediction tools are imperfect and cannot replace the clinical judgement of skilled providers, our algorithms can help prompt discussions about PrEP with the patients who are most likely to benefit from it."

Julia Marcus, PhD, MPH, lead author of the California-based study, Assistant Professor of Population Medicine at the Harvard Pilgrim Health Care Institute and Harvard Medical School

The Massachusetts-based study used a patient population of 1.1 million patients at Atrius Health as well as the population of Fenway Health, an independent community health center in Boston specializing in sexual health care, to test performance in a new setting with higher rates of new HIV infection. The final risk prediction model included sex, race, primary language, as well as diagnoses, tests, or prescriptions for STIs. The model flagged 1.8% of the general patient population at Atrius Health and 15.3% of the population at Fenway Health as potential PrEP candidates. The model also identified 37.5% of new HIV cases at Atrius Health and 46.3% at Fenway Health.

According to Douglas Krakower, MD, lead author of the Massachusetts-based study and Assistant Professor at Beth Israel Deaconess Medical Center, the Harvard Pilgrim Health Care Institute, and Harvard Medical School, "integrating these prediction models into primary care with routine, comprehensive HIV risk assessments by clinicians could play an important role in increasing the prescription of PrEP and preventing new HIV infections."

Jonathan Volk, MD, senior author of the California-based study and an infectious disease physician at Kaiser Permanente San Francisco Medical Center added that "a recent publication by the U.S. Preventive Services Task Force in JAMA cites the lack of effective prediction models as a major gap in research that is critical to improving PrEP delivery. Our model helps fill that gap."