PREDICTING SUICIDE: UTILITY OF SOCIAL DETERMINANTS OF HEALTH AND ACCESS DATA
Johns Hopkins University
By the numbers, 45,979 people died by suicide in 2020, making suicide the twelfth leading cause of death overall in the United States and the second-leading cause of death for ages 10-34.1 Suicide remains a particularly difficult condition upon which to intervene. In this respect, it perhaps resembles another high-mortality disease, cancer. A commonly held belief about cancer is that it is a single disease, while the reality is that it is a hundred – or hundreds – of different diseases each with its distinct etiology and clinical course.4 While suicides bear a superficial resemblance in their result, they are the result of a hundred, or hundreds, different pathways. While several risk factors of death by suicide, such as a diagnosis of depression or firearm ownership, stand above the rest, it is largely a murky picture. Suicide Prediction Models (SPM) are one way to understand and trace those pathways. These statistical models use data about the facts of a person’s life to estimate their risk of suicidal behavior. SPMs are a particularly useful tool as they can consider a broad range of potential risk factors and boil that down to a single measure of risk. A common use case for these would be to identify patients at risk for death by suicide and prompt some kind of intervention. Models like this could also be used to guide interventions to reduce the population-level risk of suicide. Through this thesis, I attempt to illustrate the role of Social Determinants of Health (SDoH) and Access to Care play in improving the performance of SPMs. The first manuscript within the thesis explores the extant literature on this subject. The second manuscript describes the development of a SPM to test the value of SDoH data. The third manuscript details an ecological model to determine the impact of SDoH and Access data on predictive ability by SPMs.
Suicide, Suicide Prediction Model, Social Determinants of Health, Access, Predictive Analytics