Accessing the impact of youth suicide prevention program in small communities

点击翻译按钮获取中文摘要

Project Summary/Abstract Suicide is a leading cause of death in the United States, particularly among youth aged 10–24 years. Youth in smaller, more rural communities, particularly American Indian/Native American youth, face heightened risk. Developing and delivering effective suicide prevention programs requires the means to rigorously evaluate their impact. Conducting impact evaluations of suicide prevention programs faces many challenges common to other fields, such as the difficulty in estimating the counterfactual when randomization is not feasible. These challenges are further compounded by the difficulty of obtaining reliable estimates of suicide rates in communities and segments of the population with heightened risk due to the small size of such communities. Building on small area estimation techniques, Bayesian disease-mapping models were developed to improve estimation of disease or health outcome rates in small areas by borrowing information from neighboring areas. Despite their potential, these models have not been applied to program impact evaluation. This study has three aims: (1) examine a method to assess suicide prevention impact in small areas with heightened suicide risk; (2) identify particular conditions under which our proposed method can outperform alternative methods for impact evaluation; and (3) extend earlier assessments of the impact of the Garrett Lee Smith (GLS) Memorial youth suicide prevention program administered by the Substance Abuse and Mental Health Services Administration by focusing on small, high-risk communities, such as tribal communities, that were underrepresented in previous study samples. This study will use both simulated and real data to empirically assess the accuracy of the proposed approach vis-à-vis three alternative methods to estimate impact: synthetic control, elastic net, and matrix completion. The simulated data will allow us to gauge how different levels of spatial and temporal dependence, as well as variations in other relevant parameters, affect the relative performance of approaches. Finally, we will use a sample of micropolitan and noncore communities exposed to the GLS program between 2006 and 2018 to estimate the impact of the GLS program on both suicide and self-harm hospitalization rates among youth. In sum, by applying Bayesian spatiotemporal models to suicide prevention impact evaluation, we expect to both extend the utility of the approach beyond disease mapping and advance the ability to understand suicide prevention programming impact in some of the highest-risk populations.