Detecting PPE concerns in OSHA complaints using machine learning to support infectious disease outbreak response.

Area of Science:

• Occupational Health
• Public Health Surveillance
• Machine Learning Applications

Background:

• Workers face challenges with personal protective equipment (PPE) during infectious disease outbreaks.
• Existing public health systems lack real-time data on workforce PPE challenges.
• Effective PPE interventions require timely characterization of these issues.

Purpose of the Study:

• To develop and assess a machine learning (ML) approach for detecting PPE concerns in workplace safety complaints.
• To evaluate the feasibility of repurposing Occupational Safety and Health Administration (OSHA) complaints for public health surveillance.
• To generate timely data on worker PPE concerns during infectious disease outbreaks.

Main Methods:

• Utilized a dataset of 78,770 OSHA complaints from the COVID-19 pandemic.
• Developed and trained a machine learning model to detect specific PPE concerns within complaint text.
• Assessed model performance using precision and recall metrics.

Main Results:

• OSHA complaints revealed a significant variety and volume of PPE concerns.
• The ML model achieved over 90% precision and recall for detecting unavailable PPE, lack of PPE use, and inadequate enforcement.
• ML-facilitated analysis identified national and industry-specific trends in worker PPE concerns.

Conclusions:

• Machine learning can effectively repurpose OSHA complaints to generate real-time data on worker PPE concerns.
• This method offers a valuable tool for public health surveillance during future outbreaks.
• Further development is needed to expand the range of detectable PPE concerns.