Research utility and limitations of textual data in the National Violent Death Reporting System: A scoping review and recommendations
Dang, L.N., Kahsay, E.T., James, L.N., Johns, L.J., Rios, I.E., & Mezuk, B.
Background Many studies of injury deaths rely on mortality data that contain limited contextual information about decedents. The National Violent Death Reporting System (NVDRS) is unique among such data systems in that each observation includes both quantitative variables and qualitative texts (called “narratives”) abstracted from original source documents. These narratives provide rich data regarding salient circumstances that can be used to inform prevention efforts. This review provides a comprehensive summary of peer-reviewed research using NVDRS narratives over the past 20 years, including the limitations of these texts and provides recommendations on utilizing and improving narrative quality for researchers and practitioners. Main body Studies that used narratives to examine deaths related to suicide, homicide, undetermined intent, accidental firearm, or legal intervention were identified by a title/abstract screening, followed by a full-text review. The search was conducted on English-language, peer-reviewed literature and government reports published from 2002 to 2022 in PubMed, PsycInfo, Scopus, and Google Scholar. Abstracted elements focused on the methodologies used to analyze the narratives, including approaches to explore potential biases in these texts. Articles were abstracted independently by two reviewers, with disagreements resolved through consensus discussion. During the 20-year period, 111 articles used narratives. Two-thirds studied suicide (n = 48, 43%) and homicides (n = 25, 23%). Most studies analyzed the narratives using manual review (n = 81, 73%) and keyword searches (n = 9, 8%), with only 6 (5%) using machine learning tools. Narratives were mainly used for case finding (n = 49, 44%) and characterization of circumstances around deaths (n = 38, 34%). Common challenges included variability in the narratives and lack of relevant circumstantial details for case characterization. Conclusion Although the use of narratives has increased over time, these efforts would be enhanced by detailed abstraction of circumstances with greater salience to injury research and prevention. Moreover, researchers and practitioners would benefit from guidance on integrating narratives with quantitative variables and standardized approaches to address variability in the completeness and length of narratives. Such efforts will increase the reliability of findings and set the stage for more widespread applications of data science methods to these texts.