Prediction of the mechanism of suicide among Minnesota residents using data from the Minnesota violent death reporting system (MNVDRS).

Abstract

Background: Suicide remains a major public health problem, and firearms are used in approximately half of all such incidents. This study sought to predict the occurrence of suicide specifically by firearm, as opposed to any other means of suicide, in order to help inform possible life-saving interventions.

Methods: This study involved data from the Minnesota Violent Death Reporting System. Models evaluated whether data beyond basic demographics generated increased prediction accuracy. Models were built using random forests, logistic regression and data imputation. Models were evaluated for prediction accuracy using the area under the curve analysis and for proper calibration.

Results: Results showed that models constructed with social determinants and personal history data led to increased prediction accuracy in comparison to models constructed with basic demographic information only. The study identified an optimised 'top 20' variables model with a 73% chance of correctly discerning relative incident risk for a pair of individuals. Age, height/weight, employment industry/occupation, sex and education level were found to be most highly predictive of firearm suicide in the study's 'top 20' model.

Conclusions: The study demonstrated that the use of a firearm in a death by suicide, as opposed to any other means of suicide, can be reasonably well predicted when an individual's social determinants and personal history are considered. These predictive models could help inform many prevention strategies, such as safe storage practices, background checks for firearm purchases or red flag laws.