Krakower DS, et al. - Researchers sought to develop and validate an automated prediction algorithm using electronic health record (EHR) data that may help clinicians to identify individuals at increased risk for HIV acquisition. Using machine learning algorithms, incident HIV infections were predicted with 180 potential predictors of HIV risk drawn from EHR data from 2007–15 at Atrius Health, an ambulatory group practice in Massachusetts, USA. The ten-fold cross-validated area under the receiver operating characteristic curve (cv-AUC) with 95% CIs was used to evaluate the model's performance at distinguishing individuals with incident HIV and patients independently prescribed PrEP by clinicians. Validation of the best-performing model was then done. From 2007–15, 1,155,966 Atrius Health patients (150 [< 0·1%] patients with incident HIV) were included in the development cohort; in 2016, 537,257 Atrius Health patients (16 [< 0·1%] with incident HIV) were included in the prospective validation cohort; and from 2011–16, 33,404 Fenway Health patients (423 [1·3%] with incident HIV) were included in the external validation cohort. Using least absolute shrinkage and selection operator (LASSO), the best-performing algorithm was obtained; the algorithm exhibited a cv-AUC of 0·86 (95% CI 0·82–0·90) for identification of incident HIV infections in the development cohort, 0·91 (0·81–1·00) on prospective validation, and 0·77 (0·74–0·79) on external validation. Findings support the efficacy of automated algorithms in identifying patients at increased risk for HIV acquisition. Combining these models into EHRs may assist providers in recognizing patients who might benefit from PrEP, in improving prescribing and in preventing new HIV infections.