Suicide is one of the most serious public health problems facing the United States. Recent evidence indicates that many if not most of individuals who die by suicide have been in contact with the healthcare system in the months prior to their death, providing data that can be used to identify patients at risk prior to an attempt. The proposed project will develop an innovative method for identifying patients at risk of suicidal behavior using data from a large multistate health information exchange, integrated with data from the State of Connecticut’s CHIME hospital database. We will develop and test suicide risk algorithms using principles associated with transfer learning, in which information from a comprehensive external data source is used to improve prediction in a more limited dataset. Specifically, we will use multimodal data fusion techniques to develop and test algorithms that can identify patients at risk of suicidal behavior by clinicians in hospitals with limited numbers of patients, select patient populations, and lack of access to outpatient data. This approach is not only generalizable to hospitals throughout the US but can be extended to very diverse clinical settings, e.g., primary and specialty care practices, community health centers, urgent care clinics. The potential public health significance of this study is substantial. The fragmentation of the healthcare system, particularly in relation to patients’ behavioral health needs, highlights the critical need to cultivate comprehensive, system-wide approaches to identifying and managing at patients at risk of suicide.

Narrative Suicide is one of the most serious public health problems facing the United States. The proposed project will demonstrate an innovative method for identifying patients at risk of suicidal behavior by integrating data across disparate healthcare settings.