The Detection of Suicide Data Availability for Potential Analyses through Text Mining

Background: Suicide is the second leading cause of death for youth and adults aged 15-34, and the third highest causes of death for youth 10-14 (CDC WISQARS 2013). Despite the significant public health and interpersonal health impact, the phenomena has relatively little resource into the effects of interventions on suicide ideation or completion due to ethical concerns, comparatively low base rate of suicide occurrence, and frequent exclusion of potential randomized controlled trial participants with a history of suicidal ideation or attempts. Further, even when suicide related information is collected, it often remains unanalyzed or reported as a subgroup analysis instead of a primary publication, limiting the detection through traditional searches or review methodology.  This project aimed to identify suicide related data and sub-analyses acquired through depression prevention trials in order to estimate whether these trials had acquired sufficient data to detect whether the tested depression prevention interventions could be useful towards not only depression prevention, but suicide prevention as well.  Methods: First, a comprehensive list of depression measures utilized in depression prevention trials was compiled from a review of 19 adolescent depression prevent trials (Brown et al., in press).  Each measure was reviewed in detail to determine whether suicide related items or subscales were included in each measure.  Next, we created a comprehensive list of suicide related keywords.  This list, as well as the list of the names and common abbreviations of the identified depression measures that included suicide items and subscales were used to create a text analytic pipeline that searched PDFs of published depression prevention programs (N=249).  In this presentation, we describe the computational process in greater detail. Results: The computational search yielded 2427 hits of suicide related terms, and 1979 mentions of psychiatric measures that include suicide items or subscales. The high incidence of measures and terms related to suicide reported in trials with potential suicide related information points to an untapped potential of information related to the prevention of suicide and warrants further analysis.  Finally, the feasibility of similar searches and potential public health impact are discussed.