Douglas S. Krakower
An automated algorithm that used electronic health record data to identify individuals at increased risk for HIV infection could have averted about 40% of new HIV infections at a U.S.-based organization serving a general primary care population.
A second tool, a prediction model that also utilized EHR information, identified nearly half of male HIV cases at a U.S. health care system.
Both tools were recently described in The Lancet HIV.
The CDC estimates that 1.1 million individuals in the United States have indications
for HIV pre-exposure prophylaxis, or PrEP, but only 100,000 were prescribed the therapy in 2017, according to Douglas S. Krakower, MD, assistant professor of medicine at Harvard Medical School, and colleagues.
“There is consequently a need for tools to help providers to identify people at high risk of HIV acquisition who might benefit from PrEP. Electronic clinical decision support using data embedded in patients’ EHRs could address this need and might empower more primary care providers to prescribe PrEP,” they wrote.
Krakower and colleagues developed a machine learning algorithm that could identify those at increased risk for HIV based on EHR data from about 1.1 million patients. The algorithm, which included 180 predictors of HIV risk, was validated in a prospective cohort from Atrius Health (n = 586,384) and in an external cohort treated at Fenway Health (n = 28,702).
The algorithm identified 1.8% patients at Atrius Health and 15.3% of Fenway Health patients as potential candidates for PrEP.
“If clinicians had discussed PrEP with all patients with risk scores above our cutoff and prescribed PrEP to those who indicated current high-risk behaviors, our model could have helped to avert nearly 40% of the new HIV infections [at Atrius Health] in 2016,” Krakower and colleagues wrote.
In the second study, Julia L. Marcus, PhD, MPH, assistant professor in the department of population medicine at Harvard Medical School, and colleagues used EHR data from adult patients at Kaiser Permanente Northern California to predict incident HIV diagnoses within 3 years. Their model was developed using data from about 3.1 million patients and validated in a cohort of more than 600,000. The researchers said they used 81 EHR variables to predict new HIV infections in patients.
According to the researchers, the model flagged 2.2% of the general patient population as potential PrEP candidates, and it identified 38.6% of 83 incident HIV cases, including nearly half of the 69 male HIV cases but none of the 14 female cases.
“Our study suggests that HIV prediction models could be embedded in EHRs as an automated screening tool to help to identify the subset of patients most likely to benefit from discussions about PrEP,” Marcus and colleagues added.
The model could be used in any clinical setting with EHRs, the researchers noted. – by Janel Miller
Disclosures : Krakower reports conducting research supported by Gilead Sciences and receiving honoraria for authoring or presenting CME content outside of the submitted work for DKBmed, MED-IQ, Medscape and UpToDate. Marcus reports receiving research grant support from Merck and consulting on a research grant to Kaiser Permanente Northern California from Gilead Sciences, both outside of the submitted work. Please see the studies for all other authors’ relevant financial disclosures.