CDC STACKS serves as an archival repository of CDC-published products including scientific findings, journal articles, guidelines, recommendations, or other public health information authored or co-authored by CDC or funded partners.
As a repository, CDC STACKS retains documents in their original published format to ensure public access to scientific information.
i
Development and validation of an automated HIV prediction algorithm to identify candidates for pre-exposure prophylaxis: a modelling study
-
10 2019
-
-
Source: Lancet HIV. 6(10):e696-e704
Details:
-
Alternative Title:Lancet HIV
-
Personal Author:
-
Description:Background:
HIV preexposure prophylaxis (PrEP) is effective but underutilized, in part because clinicians lack tools to identify PrEP candidates. We developed and validated an automated prediction algorithm using electronic health records (EHR) data to identify individuals at increased risk for HIV acquisition.
Methods:
We used machine learning algorithms to predict incident HIV infections using 180 potential predictors of HIV risk drawn from EHR data from 2007-2015 at Atrius Health, an ambulatory group practice in Massachusetts, USA. The best-performing model was validated prospectively using 2016 data from Atrius Health and externally using 2011-2016 data from Fenway Health, a community health center specializing in sexual healthcare in Boston, Massachusetts. We assessed the model’s performance at identifying individuals with incident HIV and patients independently prescribed PrEP by clinicians using cross-validated area under the curve (cv-AUC).
Findings:
Cohorts included 1,155,966 Atrius Health patients from 2007-2015 (including 150 [<0·1%] patients with incident HIV), 537,257 patients in 2016 (16 [<0·1%] with incident HIV), and 33,404 Fenway Health patients from 2011-2016 (423 [1·3%] with incident HIV). The best-performing algorithm had a cv-AUC of 0·86 (95% CI 0·82-0·90) for identifying incident HIV infections in the development cohort, 0·91 (95% CI 0·81-1·00) on prospective validation, and 0·77 (95% CI 0·74-0·79) on external validation. The model successfully identified patients independently prescribed PrEP by clinicians at Atrius Health (cv-AUC 0·94, 95% CI 0·90-0·97) or Fenway Health (cv-AUC 0·79, 95% CI 0·78-0·80). HIV risk scores increased steeply at the 98th percentile. We designated patients with scores above this threshold as potential PrEP candidates and prospectively identified 9,515/536,384 (1·8%) new PrEP candidates at Atrius Health in 2016.
Interpretation:
Automated algorithms can efficiently identify patients at increased risk for HIV acquisition. Integrating these models into EHRs to alert providers about patients who may benefit from PrEP could improve PrEP prescribing and prevent new HIV infections.
Funding:
The Harvard University Center for AIDS Research, the Providence/Boston Center for AIDS Research, the Rhode Island IDeA-CTR [U54GM11567], and the US Centers for Disease Control and Prevention.
-
Subjects:
-
Source:
-
Pubmed ID:31285182
-
Pubmed Central ID:PMC7522919
-
Document Type:
-
Funding:
-
Volume:6
-
Issue:10
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type: