AgISM: A Novel Automated Tool for Monitoring Trends of Agricultural Waste Storage and Handling-Related Injuries and Fatalities Data in Real-Time

Availability of summarized occupational injury data is essential for establishing complete incident surveillance systems, targeting incident preventative efforts, assessing the efficacy of prevention programs, and enhancing workplace safety. There are currently limited automated injury monitoring systems for summarizing occupational injuries obtained from electronic news and other sources, or for visualizing real-time data through an output platform. A "near" real-time surveillance tool could enable researchers to visualize data as it is being collected and provide a more rapid monitoring method to identify patterns in injury data. An automated data pipeline method could provide more current, consistent, and reliable information for injury surveillance systems and injury prevention purposes. Such a system could help public policy makers, epidemiologists, and injury prevention professionals spend less time and effort on classifying cases, increase confidence in the data, and respond quicker to "patterns" of specific types of incidents. Currently, injury surveillance approaches generally rely on manual coding of injury data, resulting in inconsistencies in classification of incident, and contributing factors and considerable delays in publishing results. This study focused on developing and testing a more automated coding methodology for use with incident narratives for further data mining, analysis, and interpretation. The concept was tested on 491 documented fatalities or serious injuries involving agricultural waste storage, handling, and transport operations. The approach provided current and real-time summarization of incident data along with data analysis and visualization by using a standard questionnaire for record-keeping, Python data frames, and the MySQL database. Findings in this study provided evidence for the reliability of classifying injury news clipping narratives into external real-time incident categories. Results showed a very encouraging performance for the chosen model to monitor injury and fatality incidents with efficiency, simplicity, data quality, timeliness, and a consistent coding process.