Prediction of suicide death using EHR and polygenic risk scores

Utah State Higher Education System--University Of Utah, Salt Lake City UT

Investigators

Linked publications & trials

Abstract

ABSTRACT Suicide death is at epidemic levels in the US, but remains extremely difficult to predict and prevent. For individuals with evident suicidality, <10% die by suicide; accurate prediction of who makes up this 10% is not currently possible. For the ~50% of suicide deaths with no prior evidence of suicidality, prediction is even more elusive. Effective prevention will involve more work to accurately identify and characterize subsets at specific risk of mortality. Our research program, the Utah Suicide Mortality Risk Study (USMRS) has unique resources to address these knowledge gaps, including unprecedented demographic, clinical, genetic, geocoding for urban/rural and other spatial studies, and exposure data from the world’s largest genetically-informative, population-ascertained cohort of suicide deaths. Currently, >12,000 suicide deaths are linked to statewide electronic health records; >7,200 of these suicides have genotyping for computation of polygenic scores, and >1,000 have whole genome sequencing, enabling discovery of more rare risk-associated changes in gene pathways. Clinical and demographic comparisons are made to 10:1 population-matched data, and genetic data can also be compared to samples from population-ascertained deaths where cause of death was not suicide (>34,000 with health data and biosamples for DNA). Studies in our previous award period engaged in developing and using a natural language processing (NLP) algorithm to more accurately define suicide deaths with vs. without prior nonfatal suicidality. This NLP greatly increased the specificity of detection of suicidality, yielding far more precisely defined groups. These groups demonstrated strikingly significant clinical differences driven in part by significant differences in underlying polygenic liability such that the SD-S group showed highly increased risk of a broad range of psychopathology in both clinical and polygenic data, but SD-N group had low clinical co-occurring psychopathology and greatly attenuated polygenic risk of psychopathology. In this renewal, it is critical that we extend these findings to refine the characteristics of these broad, fascinatingly different groups of suicide deaths that span psychiatric and physical health. We will double our sample sizes and will use data elements including clinical data, extended familial risks, environmental factors, polygenic scores, and targeted gene pathways to ascertain and characterize groups where risks lead specifically to mortality. We will use knowledge of group differences to ascertain and characterize more precise subgroups of suicide mortality. Our previous work also identified clusters of diagnoses with putative protective associations with suicide behaviors and death. We will characterize these clusters to determine factors associated with resilience. Finally, we will pursue additional external replications of our results in the Mount Sinai Medical School BioME data and in the large PsychEMERGE Consortium.