Using classification models that identify mentions of Social Determinants of Health (SDoH) within clinical notes to develop open-source tools for identifying SDoH categories
Social determinants of health (SDoH) play a significant role in health and health care systems. SDoH refers to the conditions in which people are born, grow, live, work, and age, which include social and environment factors that impact the health of individuals. According to a study cited by the researchers, although more than 80% of physicians in the U.S. document SDoH, SDoH often only appear in the notes of electronic health records (EHRs). Being able to accurately identify, and in turn act upon, SDoH information remains a significant challenge for healthcare professionals.
Recently published research supported by NIMH utilized EHR notes from the Veterans Health Administration (VHA) to develop a 3-step framework to identify SDoH concepts in the Unified Medical Language System (UMLS), then map those concepts onto one of the six SDoH categories identified by the Kaiser Family Foundation (economic stability, physical environment, education, food, community and social context, healthcare system).
The team randomly selected 3,176 EHR notes from VHA patients who had been prescribed an opioid in FY2016. Three domain experts reviewed the notes, identified the SDoH mentioned in the notes, and mapped them to UMLS concepts. They then developed a hybrid classification system to assign each SDoH concept to a high-level SDoH category. The manual annotation process resulted in a total of 901 unique SDoH concepts from UMLS, while the keyword-based search extracted 117,089 unique SDoH concepts across UMLS. However, combining both sources resulted in a final concept set that could be used to build the classification model. The Logistic Regression model performed the best, with the greatest degree of accuracy. To help ensure that this classification model could have broad application, the research team developed an open-source tool called Extract SDoH from EHRs (EASE) that can automatically identify SDoH and their SDoH categories from clinical notes.
The study findings have some limitations. The classification model was developed using a relatively small subset of VHA patients who had been prescribed opioids during FY2016. It is not clear the extent to which this model would accurately categorize SDoH among other populations. Additionally, although EASE is an open-source tool, it does require some programming language skills and may not be accessible to less technologically savvy healthcare professionals.
In summary, classification models can be used to extract and categorize SDoH-related information from EHR notes. This methodology holds great promise for building clinical decision support systems that empower healthcare professionals to address the needs of their patients more comprehensively. This work also highlights the need for research in diverse healthcare settings to ensure that tools and models that are developed can correctly identify and categorize SDoH among diverse populations.
Citation:
Rawat, B. P. S., Keating, H., Goodwin, R., Druhl, E., & Yu, H. (2023). An Investigation of the Representation of Social Determinants of Health in the UMLS. AMIA Annu Symp Proc. 2023 Apr 29; 2022:912-921. PMID: 37128364; PMCID: PMC10148271.
Income–associated disparities in children’s brain structure can be alleviated with state-level anti-poverty programs
Prior research has indicated an association between family income levels and structural differences in the developing brain, which in turn contributes to disparities in health outcomes later life. Lower income has been found to be associated with several environmental processes, including resource constraints that impact how caregivers provide for their children, higher exposure to stressful life events (e.g., violence) and chronic stressors (e.g., food and housing insecurity), and an unpredictable environment. Adults who spent their childhood in lower income households have been found to have lower educational attainment, increased dependence on public assistance, and more mental and physical health challenges than those raised in families with higher incomes. Government sponsored anti-poverty programs such as the Earned Income Tax Credit (EITC), Temporary Assistance for Needy Families (TANF), and Medicaid expansion are designed to assist low-income families by providing income relief, temporary cash benefits, and health coverage, respectively. The degree of availability and generosity of anti-poverty policies varies by state; the more substantial programs are associated with improved health outcomes, further academic achievement, child wellbeing, and better functioning of families.
Recently published research supported by NIMH and others utilized data from the large, multi-site Adolescent Brain and Cognitive Development (ABCD) study that included data of over 10,000 9 to 11-year-old children across 17 states, to explore the effect of state-level macro-economic factors on early adolescent brain development and mental health. Higher costs of living were hypothesized to place additional stress on low-income families leading to brain structure disparities and mental health symptoms in children.
Replicating findings from previous studies, researchers found lower hippocampal volume and higher internalizing psychopathology in children of families with lower socioeconomic status. Additionally, bivariate correlations of study variables also found differences in brain structure were reduced by 34%–48% in high cost of living states with more generous cash benefits versus states with less generous anti-poverty programs. The variance of hippocampal volume was 195 millimeters between high- and low- income study participants in states with high costs of living and low cash benefits, but only 129 millimeters for participants in states where cost of living and cash benefits were more sizeable. Similarly, the developing brain disparity between high- and low-income participants was 43% less in states with Medicaid expansion. The income-associated disparity in mental health conditions was 48% lower in higher cost of living states with larger cash benefits. This study demonstrates that the strength of these associations varies as a function of state level policies, including the impact of cost of living and the inclusiveness of anti-poverty programs.
The study results have some limitations, including that the study design is cross-sectional and observational, however the sensitivity analyses conducted confirmed that no other state level social, economic, educational, or political characteristic could explain the associations found. Also, study estimates are likely conservative, given that ABCD study sites are limited to only 17 out of 50 states. Overall, the results of this study support that the developing brain is influenced by its’ environment and indicate that structural policy interventions that allow for more generous state social safety nets could potentially reduce the negative impacts driven by socioeconomic and health inequities on mental health and brain development.
Citation:
Weissman DG, Hatzenbuehler ML, Cikara M, Barch DM, McLaughlin KA. State-level macro-economic factors moderate the association of low income with brain structure and mental health in U.S. children. Nat Commun. 2023 May 2;14(1):2085. doi: 10.1038/s41467-023-37778-1. PMID: 37130880; PMCID: PMC10154403.
Factors associated with improvements in physical health and functioning of women during midlife
Studies have shown that functional limitations impacting women’s health often arise in midlife and become exacerbated into late adulthood. While some studies have found associations between these limitations and the development of chronic diseases in later life, other studies have shown that this association is not always a negative one; in other words, functional status and physical health are dynamic and capable of being improved. Interest in this area is also peaking as researchers continue to investigate factors associated with “successful aging” and personal longevity.
A recent study supported by NIA, NINR, and ORWH aimed to identify factors associated with changes in physical health and function among women during midlife. Researchers used data from Study of Women’s Health Across the Nation (SWAN) sites, which collect information from multiethnic, multiracial community-based cohorts regarding menopause transition and midlife. This study used data from 2004 to 2017, although the SWAN study contains baseline data from 1996. The primary outcome variable was clinically important improvement in the physical component score (PCS) of the SF-36, a 36-item health survey. The PCS is a validated score containing components from 8 domains covering physical and emotional health, personal limitations, and health functioning. Researchers examined changes in the PCS among the study population over a ten-year period. Other study variables included demographic characteristics (i.e., age, race, ethnicity), socioeconomic characteristics (i.e., educational level, financial strain), and health/behavioral characteristics (i.e., substance use, financial presence of chronic conditions, medication use, BMI, and sleep quality). Study analyses examined significant associations between covariates and the primary outcome variable, as well as the association of changes in variables over time.
A total of 1,807 women were included in the analytic sample, with a mean age of 54.5 years (White = 50%, Black = 27%, Chinese = 11%, Japanese = 12%). About half of the sample reported having less than a college education and one-quarter reported financial strain. Among the entire analytic sample, 15% of women had a “clinically important PCS improvement” (considered to be at least a 5-point improvement) over a median 11 years. At baseline, the most prevalent problems were reported to be sleep disturbances (44%), hyperlipidemia (44%), and osteoarthritis (41%). Study results found that several factors were significantly associated with clinically important PCS improvements, including a lower BMI, no reported financial strain, no sleep disturbance, fewer medications, no osteoarthritis, and having a higher physical activity score. In addition, those who had a lower baseline PCS were more likely to show improvements in health and function.
This study confirmed the work of several past studies on positive aging and physical health and function. It also went a step further to suggest specific factors that are associated with these improvements over time, as well as show that health and function are actually dynamic and modifiable from midlife into late adulthood. The research team was able to demonstrate these significant results among a racially, ethnically, and geographically diverse population, testing a broad range of health- and quality-of-life-related factors. Further work can build on this study to develop targeted interventions for women in midlife aimed at improving physical and functional health for successful aging.
Citation:
Santacroce LM, Avis NE, Colvin AB, Ruppert K, Karvonen-Gutierrez C, Solomon DH. Physical and Behavioral Factors Associated with Improvement in Physical Health and Function Among US Women During Midlife. JAMA Netw Open. 2023 May 1;6(5): e2311012. doi: 10.1001/jamanetworkopen.2023.11012. PMID: 37126345; PMCID: PMC10152304.
