Background ICD-10 codes provide limited information on the location of drowning deaths, particularly those occurring in natural water. Application of natural language processing (NLP) to literal text on death certificates can provide greater detail.

Methods Records that include literal text were selected from 2011–2017 U.S. National Vital Statistics System mortality data using ICD-10 multiple cause codes W65-W74, V90, V92, X71, X92, Y21 and T75.1. SAS Contextual Analysis software was applied to literal text data from six fields on the U.S. standard death certificate to create a list of possible terms describing drowning locations. Terms were categorized as: Domestic (e.g., swimming pool, bathtub, spa), Non-Domestic Freshwater (e.g., lake, pond, river, reservoir, well), Saltwater (e.g., ocean), Other, and Unspecified. For Non-Domestic Freshwater, subcategories included Natural and Man-made.

Results Of 26,638 drowning deaths, 32% were categorized to Domestic, 52% to Non-Domestic Freshwater, 10% to Saltwater, and 6% to Other or Unspecified. For Non-Domestic Freshwater, 90% occurred in natural settings and 10% involved a man-made source (e.g. well, reservoir, cistern, tank). Among the Natural Non-Domestic Freshwater drownings, about half (52%) occurred in still water (e.g., lake, pond, wetlands, marsh) and half (47%) occurred in moving water (e.g., river, creek, rapids, flooding).

Conclusion Application of NLP methodology can provide more detailed information on location of drowning deaths. This information can be used to target prevention efforts to high risk populations by location.

Learning Outcomes To learn about the application of NLP processing to death certificate literal text to identify and categorize locations of drowning deaths.