Using text summarization for surveillance: A case study involving OSHA fatality and catastrophe reports

Background: OSHA and NIOSH recently released a Hazard Alert on methylene chloride and bathroom refinishing, following the deaths of at least 14 workers since 2000 in related circumstances. Although OSHA maintains a database of narrative reports describing fatalities and catastrophes, similar patterns of preventable exposure to occupational hazards may be difficult to identify efficiently through manual review alone, given the large number of narratives in this database. Using methylene chloride as a case study, we consider whether text mining techniques can help identify important patterns in circumstances of hazardous exposures. Methods: Using keywords entered by compliance officers to categorize narrative reports, we apply our text algorithms to generate summaries for each category. By manually examining the resulting words and phrases associated with methylene chloride, we consider whether our text summarization tool effectively characterizes the circumstances of the bathroom refinishing fatalities. Initial and Expected Results: Based on initial results and earlier work using the same algorithms, we expect that our text summarization tools will accurately portray the differences and unique characteristics of each keyword category. This presentation will explore whether these automated summaries are useful for surveillance purposes in particular. Preliminary Conclusions: While text mining algorithms are unlikely to identify all meaningful patterns in narrative reports, the computational advantages of text summarization tools may complement manual review. The summaries themselves, for example, can serve as a launching point for inquiry by raising potential flags that can direct further investigation.