Search

Patient representation learning methods create rich representations of complex data and have potential to further advance the development of computational phenotypes (CP). Currently, these methods are either applied to small predefined concept sets or all available patient data, limiting the potential for novel discovery and reducing the explainability of the resulting representations. We report on an extensive, data-driven characterization of the utility of patient representation learning methods for the purpose of CP development or automatization. We conducted ablation studies to examine the impact of patient representations, built using data from different combinations of data types and sampling windows on rare disease classification. We demonstrated that the data type and sampling window directly impact classification and clustering performance, and these results differ by rare disease group. Our results, although preliminary, exemplify the importance of and need for data-driven characterization in patient representation-based CP development pipelines.

Objective To evaluate risk for suicide among veterans with a history of stroke, seeking care within the Veterans Health Administration (VHA), we analyzed existing clinical data. Design This retrospective cohort study was approved and performed in accordance with the local Institutional Review Board. Veterans were identified via the VHA's Corporate Data Warehouse. Initial eligibility criteria included confirmed veteran status and at least 90 days of VHA utilization between fiscal years 2001-2015. Cox proportional hazards models were used to assess the association between history of stroke and suicide. Among those veterans who died by suicide, the association between history of stroke and method of suicide was also investigated. Setting VHA. Participants Veterans with at least 90 days of VHA utilization between fiscal years 2001-2015 (N=1,647,671). Data from these 1,647,671 veterans were analyzed (1,405,762 without stroke and 241,909 with stroke). Interventions Not applicable. Main Outcome Measures Suicide and method of suicide. Results The fully adjusted model, which controlled for age, sex, mental health diagnoses, mild traumatic brain injury, and modified Charlson/Deyo Index (stroke-related diagnoses excluded), demonstrated a hazard ratio of 1.13 (95% confidence interval, 1.02-1.25; P=.02). The majority of suicides in both cohorts was by firearm, and a significantly larger proportion of suicides occurred by firearm in the group with stroke than the cohort without (81.2% vs 76.6%). Conclusions Findings suggest that veterans with a history of stroke are at increased risk for suicide, specifically by firearm, compared with veterans without a history of stroke. Increased efforts are needed to address the mental health needs and lethal means safety of veterans with a history of stroke, with the goal of improving function and decreasing negative psychiatric outcomes, such as suicide.