HIV Data Quality Score (DQS) Project
Description
Currently, 60 health departments and 150 community-based organizations submit National HIV Monitoring and Evaluation (NHM&E) program data using a standard online form. However, this data often contains errors that require significant manual cleanup by CDC's HIV data managers. To address this issue, our project aims to create a large language model (LLM)-based data quality score capable of detecting errors in datasets and measuring dataset cleanliness levels. LLMs can also be utilized to automatically fix some detected errors.
Detailed example
The outputs include both a List of identified erroneous data fields in a dataset and a Dataset with some errors automatically corrected. This will be available for future evaluation efforts.
AI / analytics pattern
Natural Language Processing: AI that processes, interprets, and shares information in human language.
Automation level / stage
a) Pre-deployment – The use case is in a development or acquisition status.
Expected benefit
This project intends to enable HIV data managers to quickly identify errors, track trends in data quality by site, provide targeted technical assistance (TA), and automate some error corrections.
Controls / human review
ATO: Not reported; PIA: Not published