Detecting, evaluating, and redacting PII in NAMCS HC Component
Description
How effective are open-source PII detection models in identifying and redacting PII? The National Center for Health Statistics’ (NCHS) Division of Health Care Statistics has collected millions of health records with laboratory results from encounters at health centers via the National Ambulatory Medical Care Survey (NAMCS), Health Center (HC) Component. Due to inadvertent errors during data entry and processing, some records contain identifiers (e.g. names, locations, etc.) in non-PII fields. Due to the PII, certain fields cannot be made available for restricted or public use, but reviewing millions of records for PII is not practical.
Detailed example
A semi-automated process to conduct a quality control review of health data records, including potential PII records flagged for manual review.
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
If the process is feasible, it will significantly increase the healthcare lab data available to researchers for analysis. Additionally, the process could be applied to additional tables and years of data, increasing overall data availability.
Controls / human review
ATO: Not reported; PIA: Not published