Studies funded by the National Institute of Mental Health (NIMH) and the National Institute of Allergy and Infectious Diseases (NIAID) assess a novel method that may help clinicians identify individuals most in need of PrEP. This is based on studies that show the effectiveness of using algorithms that analyze electronic health records (EHRs) to help physicians identify patients at risk for HIV who may benefit from preexposure prophylaxis (PrEP), which significantly reduces the risk of getting HIV.
What is PrEP?
PrEP is a strategy in which healthy people routinely take one or more antiretroviral drugs to reduce their risk of getting HIV. It is highly effective for reducing the risk of HIV acquisition, yet it remains greatly underutilized. The Centers for Disease Control and Prevention estimates that as many as 1.1 million Americans may be candidates for PrEP use, but in 2016, only an estimated 78,360 (about 7%) were prescribed PrEP medication.
In two large-scale studies, which used EHRs from large health systems in Massachusetts and California, researchers created and tested algorithms that analyze a rich array of health data and patient information to help clinicians automatically identify those at highest risk for HIV infection and therefore most likely to benefit from PrEP medications.
In the first study, Krakower and colleagues used machine learning to create an HIV prediction algorithm using 2007-2015 EHR data from more than 1 million patients attending Atrius Health, a large healthcare system in Massachusetts. The model used variables in the EHRs, such as diagnosis codes for HIV counseling or sexually transmitted infections (STIs), laboratory tests for HIV or STIs, and prescriptions for medications related to treating STIs. The model was subsequently validated using data from 537,257 patients seen by Atrius Health in 2016, as well as 33,404 patients seen by Fenway Health, a community health center in Boston that specializes in providing healthcare for sexual and gender minorities, between 2011 and 2016. In these validation studies, the prediction algorithm was able to be successfully distinguished between patients who did or did not acquire HIV, and between patients who did or did not receive a PrEP prescription with high precision.
According to Krakower, “A striking outcome is that our analysis suggests nearly 40% of new HIV cases could potentially have been averted had clinicians received alerts to discuss and offer PrEP to their patients with the highest 2% of risk scores.”
The second study, led by Julia Marcus, Ph.D., of Harvard Medical School and Harvard Pilgrim Health Care Institute, with Krakower and colleagues, scaled-up this prediction approach by using the EHRs of more than 3.7 million patients receiving outpatient services from Kaiser Permanente Northern California. They developed a model to predict HIV incidence, using data from patients who entered the Kaiser Permanente system between 2007 and 2014, and they validated the model on data from patients who had entered the Kaiser Permanente system between 2015 and 2017. The model used variables in the EHRs, such as high-risk sexual behavior indications, HIV and STI testing frequency, STI diagnoses and treatments.
“Our model was able to identify nearly half of the incident HIV cases among males by flagging only 2% of the general patient population,” Marcus said. “Embedding our algorithm into the Kaiser Permanente EHR could prompt providers to discuss PrEP with patients who are most likely to benefit.”
Both studies are among the first to demonstrate that EHR-based prediction algorithms can effectively identify individuals in general populations who are at high risk for HIV and potential candidates for PrEP. These models offer clinicians an important new tool to reduce new HIV infections. Future research will continue the development of these predictive models and discover the best ways to integrate them with healthcare systems to improve PrEP use and prevent HIV infections.
Douglas Krakower, M.D., of Beth Israel Deaconess Medical Center and Harvard Medical SchoolSource: NIH