Using EHR data, a study team identified 6076 potential PrEP candidates in a large health system, of whom 5577 (92%) were not on PrEP at the time.
It is well known that HIV pre-exposure prophylaxis (PrEP) is effective in preventing HIV acquisition but, despite this fact, PrEP has not been widely prescribed to individuals who are at an increased risk of HIV exposure.
When considering ways to increase the uptake of PrEP, a team of investigators hypothesized that using data from electronic health records (EHR) could be a method to identify patients who are at a higher risk for HIV and could benefit from PrEP.
The findings of the study were presented in an oral abstract session at the Annual Conference on Retroviruses and Opportunistic Infections (CROI 2019) on Wednesday, March 6, 2019.
For the study, the investigators developed and validated a prediction model to identify appropriate PrEP candidates in a cohort of individuals at Kaiser Permanente Northern California who did not have HIV, had at least 2 years of enrollment, and at least 1 outpatient visit between 2007-2017.
Using EHR data on 68 variables including demographic, clinical, and behavioral elements that were potentially predictive of HIV risk, the study team applied logistic regression and machine learning methods to create predictions of incident HIV cases in patients entering the cohort between 2007-2014.
"For this study we developed a prediction model using electronic health record data that were collected during routine clinical care at Kaiser Permanente," Julia Marcus, PhD, MPH, assistant professor at Harvard Medical School and Harvard Pilgrim Health Care Institute, and an author of the study, told Contagion® in an exclusive video interview. "Just like any other prediction model that's been developed for other areas of medicine, like cardiovascular disease or sepsis, we were trying to figure out whether we could identify patients who were at an increased risk of a particular outcome and in this case, we were looking at HIV diagnosis."
According to the study of more than 3.7 million eligible patients in 2007-2017, there were 1422 incident HIV cases documented.
The candidate models were evaluated by cross-validating the area under the curve (AUC, range 0-1). From there, the investigators determined the prospective performance of the best-performing model through validation among members who entered the cohort between 2015 and 2017. The best-performing model for predicting incident HIV was least absolute shrinkage and selection operator (Lasso) with an AUC of 0.90 in 2007-2014.
Next, the full model was compared with simpler models that incorporated only variables that are considered traditional risk factors, including the detection of sexually transmitted infections (STIs).
The final model included 41 predictors of incident HIV cases, which the investigators note included black race, home zip code, urine positivity for methadone, and use of medication for erectile dysfunction.
When the model was evaluated prospectively in the 2015-2017 data, it performed well (AUC 0.89). Model performance continued to remain high when the MSM variable (AUC 0.87) or STI variables (AUC 0.90) were excluded, but performance was reduced when including only MSM (AUC 0.74), STIs, (AUC 0.61) or both (AUC 0.78).
“Patients in the top 1% of risk scores included 45/68 (66%) male HIV cases but 0/13 female HIV cases among those entering the cohort in 2015-2017,” write the investigators.
Through using the top 1% of risk scores to define potential PrEP candidates in 2015-2017, the study team identified 6076 potential PrEP candidates, of whom 5577 (92%) were not on PrEP at the time.
The study investigators report that prediction models using EHR data can be used to identify patients who are at a high risk of HIV acquisition and could benefit from PrEP and should be tested as a strategy to improve PrEP use.
Additional EHR variables or other data are needed to identify females who may benefit from PrEP, the study investigators conclude.
The study, “Using EHR Data to Identify Potential PrEP Candidates in a Large Health Care System,” was presented at CROI 2019 on Wednesday, March 6, 2019 in Seattle, Washington.