Design and Application of an Area-Level Suicide Risk Index with Spatial Correlation

In this study, we design a novel model-based Suicide Risk Index to assess and identify area-level suicide risk. We construct a Bayesian Spatial Factor Analysis model, treating suicide risk as an underlying latent factor that manifests through multiple observable variables. Our method is applied to county-level data from multiple sources in Florida and Georgia. We utilize 14 manifest variables classified into three dimensions: “suicidal behavior”, “mental illness”, and “substance abuse.” The posterior means and 95% credible intervals of the model-based SRI ranks are estimated. Our results show substantial disagreement between the SRI rankings and age-adjusted suicide rate which only captures reported suicides. Furthermore, we find strong evidence of spatial spillovers in suicide risk across counties. The “mental illness” dimension of our model represents the greatest contribution to county suicide risk in Florida while the “suicidal behavior” dimension accounts for the most variation in suicide risk in Georgia. We also test the sensitivity of our model-based SRI ranks to an alternative spatial correlation specification and different methods for imputing missing data. Finally, we show that greater deprivation and social fragmentation, each estimated using the same SFA model, are positively associated with suicide risk. Our findings suggest that existing suicide prevention guidelines used by policymakers to identify high-risk counties based on suicide death rates may be misleading. The model-based SRI identifies counties with both high suicide risks and greater likelihoods of transferring their risks across county borders. Policy may benefit from singling out these counties for aid and targeted interventions.