Abstract: Parsing the Effects of Culture, Language, and Insurance: An Agent-Based Model of Access to Care Among Hispanics in the United States (Society for Social Work and Research 22nd Annual Conference - Achieving Equal Opportunity, Equity, and Justice)

Parsing the Effects of Culture, Language, and Insurance: An Agent-Based Model of Access to Care Among Hispanics in the United States

Schedule:
Friday, January 12, 2018: 3:52 PM
Archives (ML 4) (Marriott Marquis Washington DC)
* noted as presenting author
Hyunsung Oh, PhD, Assistant Professor, Arizona State University, Phoenix, AZ
Mai Trinh, PhD, Assistant professor, Arizona State University, Tempe, AZ
Cindy Vang, MSW, Doctoral student, Arizona State University, Phoenix, AZ
BACKGROUND: Disparities in access to care among Hispanics in the United States are correlated with lower English proficiency, lower health insurance coverage, and cultural barriers when visiting an English-speaking physician. However, these findings primarily come from observational studies, making it difficult to examine interactions between these risk factors over time. To provide an alternative approach to explaining the disparities in Hispanics’ access to care, we employed agent-based modeling (ABM) to describe how Hispanics seek care based on a combination of determinants such as English proficiency, insurance coverage, and geographical distance between a patient and a provider. ABM is a powerful tool integrating what has been known from existing research to simulate what-is, what-might-be, and what-should-be scenarios.

METHODS: We used NetLogo to build an ABM simulating the process in which Hispanics in a community of 10,000 seek primary care within a six-month period.[1] Each round of the model represents two weeks in real life. Sample input parameters include physicians’ capacity, travel distance, insurance rate, English proficiency among patients, and Spanish proficiency among physicians. Most of these parameters were drawn from the National Health Interview Survey (NHIS) 2015. The model uses simple behavioral rules in which Hispanics’ decisions to seek or delay care depends on four logical algorithms involving their insurance status, English proficiency, travel distance, and physicians’ availability. To examine disparities in access to care, we focus on two outcomes: the proportion of Hispanics who delayed care at the six-month mark, and the average number of times a Hispanic delayed care during this six-month period. We ran six controlled experiments with different levels of likelihood to seek care, English proficiency, health insurance coverage, mobility limit, number of providers, and proportion of Spanish-speaking doctors. Each experiment was iterated for 1,000 times.

RESULTS: Preliminary results confirmed that known risk factors associated with lower access to care indeed led to higher risk of delayed care. Patients with no insurance or no English proficiency delayed care more than their counterparts. Surprisingly, when we examined a hypothetical model with Hispanics’ rate of insurance increased to the national average, no substantial difference in average delayed care was observed. Post-hoc analysis showed that English proficiency but not insurance explained the difference in the likelihood to delay care between the two scenarios. Lastly, when mobility limitation was removed, health care outcomes became substantially improved, suggesting that spatial distributions of patients and providers are important.

IMPLICATIONS: We experimented different what-might-be hypothetical scenarios to generate policy implications and provide recommendations to improve access to care for Hispanics. Models considering interactions between agents and attributes of both patients and providers showed that population health may not be dramatically improved when a single patient’s barriers to care, such as English proficiency and insurance coverage, is addressed. Our models suggested that roles of culturally competent care and spatial distribution of culturally competent providers deserve further scholarly research.



[1] Model shows stability after 13 rounds, or six months in real time. More details are available upon request.