SDOH Associated with Increased Risk of Suicide Among US Veterans

In a recent study published in JAMA Network Open, researchers found that nine social determinants of health (SDOH) extracted from structured and unstructured EHR data using natural language processing (NLP) were associated with an increased risk of suicide-related death among US veterans.

The impact of social factors on health has been the subject of much recent research, including ongoing work by the US Department of Veterans Affairs to leverage SDOH for veteran suicide prevention. However, fewer studies are using unstructured data and clinical notes from EHRs to help determine suicide risk.

To close this gap, researchers conducted a nested case-control study comprised of 6.1 million veterans who received care under the US Veterans Health Administration (VHA) from Oct. 1, 2010, to Sept. 30, 2015. Using an NLP tool, which allows researchers to sort through unstructured data, the research team pulled SDOH data from free-text EHR notes while relying on a combination of ICD-10 codes and VHA stop codes to extract SDOH information from structured data.

For each study participant, the research team included multiple types of clinical notes for analysis, including emergency department notes, nursing assessments, primary care notes, hospital admission notes, inpatient progress notes, pain management, and discharge summaries.

Patients with prior suicide attempts, missing or incorrect demographic information, or no EHR notes were excluded from the cohort.

The NLP-extracted SDOH data included eight factors: social isolation, job or financial insecurity, housing instability, legal problems, violence, barriers to care, transition of care, and food insecurity. The structured SDOH data included six factors: social or familial problems, employment or financial problems, housing instability, legal problems, violence, and nonspecific psychosocial needs.

Five of the same SDOH factors were present in both groups. But barriers to care, transition of care, and food insecurity were found only in the NLP-extracted group, and nonspecific psychosocial needs were found only in the structured group. Thus, the researchers combined these data into nine distinct SDOH factors.

Occurrences of SDOH factors over the two-year study period were then evaluated across follow-up outcomes within the cohort. Follow-up outcomes were defined as the earliest of four outcomes: suicide, death from other causes, end of last record for the patient, or end of the study period. Suicide incidence information was gathered from the National Death Index.

Of the veterans in the cohort, 8,821 committed suicide. The researchers found that SDOH factors from both structured and unstructured data were significantly associated with an increased risk of suicide within the cohort. Legal problems, violence, and nonspecific psychosocial needs were the three SDOH factors with the most significant impact on suicide risk.

The researchers also discovered that NLP-extracted SDOH factors, with and without structured SDOH data, were associated with increased suicide risk among veterans, indicating NLP’s potential utility for future public health and suicide prevention studies.

Other studies have also taken advantage of artificial intelligence (AI)-driven approaches to address veteran suicide risk.

In 2019, researchers at Lawrence Berkeley National Laboratory teamed up with the VA and the Department of Energy to use deep learning to comb through the VA’s EHR data assets and Million Veteran Program genomic databank, identify patterns in clinical data that may indicate a likely future suicide attempt, and connect patients with mental health resources necessary to help them avoid a crisis event.