Geospatial suicide risk prediction in the State of Maryland and the limits of discernible intent.

Background: The rate of suicide death is known to be unequally distributed by geography and associated with multiple sociocultural and clinical risk factors. It is less clear whether the same patterns exist for deaths where the manner is undetermined, acknowledging that it is not always possible for a medical examiner to know whether death is intentional. Methods: Cumulative counts of suicide and undetermined death were aggregated to the Census tract level of the State of Maryland, between 2012-2020 and used as primary outcomes of interest. We evaluated death rate using two statistical approaches: 1) negative binomial models and 2) SaTScan geospatial clustering of pooled cases. Results: Model effects for suicide reveal a marginal association with older age, and greater area deprivation index (ADI), but an inverse association with proportion of minority race population. Undetermined death was found to be significantly associated with ADI. SaTScan clusters for suicide revealed a small number of large clusters covering rural areas of Maryland, but many small clusters around metropolitan areas for undetermined manner of death. Conclusions: The two techniques identified distinct regions of interest for public health surveillance and intervention planning, and with respect to each manner of death. Suicide deaths were found to be elevated in non-metropolitan areas. Limited access to care and increased social burden are found to be associated with undetermined death. There is more uncertainty around intent for these cases, often due to a lack of diagnoses, treatment or other clinical context.