Using Clinical Notes and Natural Language Processing for Automated HIV Risk Assessment.

Area of Science:

• Medical Informatics
• Public Health
• Computational Linguistics

Background:

• Universal HIV screening is resource-intensive and may miss high-risk individuals.
• Electronic Health Records (EHRs) contain valuable social and behavioral determinants of health in unstructured notes.
• Automated risk assessment using EHRs can optimize targeted HIV screening programs.

Purpose of the Study:

• To evaluate the effectiveness of Natural Language Processing (NLP) in improving predictive models for HIV diagnosis.
• To determine if NLP analysis of unstructured EHR data can enhance the identification of individuals at high risk for HIV.

Main Methods:

• Developed three machine-learning predictive algorithms using EHR data.
• Model 1: Structured EHR data only.
• Models 2 & 3: Included NLP-derived features (topics and keywords) from clinical notes.

Main Results:

• The NLP keyword model achieved the highest predictive performance (F-measure of 0.74).
• Key terms like "msm," "unprotected," "hiv," and "methamphetamine" significantly improved prediction.
• NLP enhanced the model's ability to identify high-risk behaviors from clinical text.

Conclusions:

• NLP significantly improves automated HIV risk assessment by extracting critical behavioral indicators from clinical text.
• Future research should focus on advanced NLP techniques for social and behavioral determinant extraction.
• This approach can lead to more targeted and effective HIV screening strategies.