Abstract
Introduction: This study utilizes the Virginia all-payer claims database (APCD) to examine the relationship between primary care utilization and emergency department (ER) use and to explore geographic variation in primary care utilization and ER use. We hypothesize that higher rates of primary care utilization will be associated with lower ER use rates, with maps showing clear geographic patterns.
Methods: This retrospective observational analysis utilized Bayesian smoothing techniques, regression analysis, and geographic information system (GIS) mapping to explore the association of ER use with primary care utilization. Our analysis included 866 ZIP Code Tabulation Areas (ZCTAs) in Virginia.
Results: Primary care utilization was significantly associated with ER usage rates. The results show that for every increase of 10 primary care visit rates per 1,000 population, ER use rates decline by 7 per 1,000. The maps show clusters of higher rates of PC utilization throughout central and eastern Virginia, with lower rates in many parts of southern Virginia. Higher rates of ER use are observed in western Virginia, particularly along the border with West Virginia, with clusters of lower rates in northern Virginia near Washington, DC.
Conclusions: Utilizing the Virginia APCD and GIS mapping, this study finds that primary care utilization is associated with lower rates of ER use. The maps show clear geographic patterns for both ER use and primary care utilization. Important next steps include identifying priority areas, exploring their characteristics, and conducting qualitative research to better understand local factors contributing to their high or low rates of ER use.
- Database Management Systems
- Emergency Room Visits
- Geographic Information Systems
- Primary Health Care
- Regression Analysis
- Retrospective Studies
- Secondary Data Analysis
- Virginia
Introduction
Emergency department (ER) visits in the US have increased by more than 10% over the past 2 decades,1 which coincides with a 30% decrease in primary care acute visits1 and concerns over shortages of the primary care workforce.2 While ER visits are costly and often preventable,3,4 there is substantial evidence that better access to and utilization of primary care is associated with fewer ER visits and preventable hospitalizations.5⇓⇓⇓⇓–10
Several studies have examined the geographic access to primary care, as well as the association of hospitalizations for ambulatory care sensitive conditions (ACSCs) with small-area geographies.11⇓–13 Frequency of ACSC hospitalizations is a commonly used measure for preventable hospitalizations and includes conditions (such as such as diabetes, hypertension, and bacterial pneumonia) that can be effectively treated in outpatient care settings.14 Using Medicare data, Chang et al found that primary care service areas (PCSAs) in the highest quintile for adult primary care workforce had significantly fewer ACSC hospitalizations.11 Basu et al. similarly identified primary care availability as inversely associated with ACSC hospitalizations.12 Daly et al. utilized more advanced geographic approaches at the ZIP Code Tabulation Area (ZCTA) in the State of Virginia, finding that better geographic access to primary care physicians was associated with fewer ACSC hospitalizations.13
These studies highlight the potential of using subcounty (small-area) data to help elucidate relationships between primary care access and avoidable hospitalizations. However, because data on actual primary care utilization is not usually available for small areas, little research has explored the relationship between ER use and primary care utilization. Data sources that could be utilized for exploring primary care utilization for small-area analyses are all-payer claims databases (APCDs), which have “the potential to provide a deeper understanding of patterns, quality, and cost of care across entire populations.”15 Researchers in several states have utilized APCDs to explore important questions around health care utilization, outcomes, and workforce, including in the state of Virginia.16⇓–18
This study utilizes the Virginia APCD to examine the relationship between primary care utilization and ER use and to explore geographic variation in primary care utilization and ER use. We hypothesize that higher rates of primary care utilization will be associated with lower ER use rates, with maps showing clear geographic patterns.
Methods
Study Design
This retrospective observational analysis utilized regression analysis and GIS mapping to explore the association of ER use with primary care utilization.
Data Sources
We combined data from multiple sources for this study, including: the Virginia APCD (2016 to 2020),19 containing about 80% of all claims in Virginia, the American Community Survey (ACS, 2016 to 2020),20 and the Center for Disease Control and Prevention (CDC) PLACES,21 which are small-area estimates for a variety of health measures. All data utilized in this study are publicly available and can be downloaded from HealthLandscape Virginia: an interactive mapping tool with data from the Virginia APCD and several other sources, including those used here.22 All data are aggregated to the Zip Code Tabulation Area (ZCTA) level. After removing 30 ZCTAs that were missing APCD data for all 5 years (2016 to 2020), our final model included 866 ZCTAs.
Measures
Our main outcome measure was ER use rates, which was defined as the number of people having visited the emergency department 1 or more times in the past year per 1000 population (2016 to 2020) per ZCTA. Our independent variable was primary care utilization, which was defined as the number of people with a primary care visit in the past year per 1000 population (2016 to 2020) per ZCTA. Covariates were selected based on previous literature on factors impacting ER utilization and included 2 measures obtainable from the Virginia APCD: mental health diagnoses per 1000 population per ZCTA, and type 2 diabetes diagnoses per 1000 population (2016 to 2020) per ZCTA. Other covariates included the percentage of adults ages 18 to 64 without health insurance (2020) from CDC PLACES, the percentage of population that is Black (2016 to 2020) from the American Community Survey, and a rural dichotomous variable from the USDA Rural-Urban Continuum Areas (defined as ZCTAs in nonmetropolitan Rural-Urban Continuum Areas [RUCA 4 to 10]23). Additional measures in the analysis included the primary care physician rate (derived from the Virginia APCD)18 and several measures from the American Community Survey (2016 to 2020): total population, the percentage of people ages 65 and older, the percentage of Non-White populations, the percentage of Hispanic populations, the percentage of population below 100% of the federal poverty limit, the percentage of population on public insurance.
Statistical Analysis and Mapping
All analyses were completed using an open-source geospatial software program, GeoDa 1.20.0.22.24 Our cross-sectional approach included several methods, including using empirical Bayes (EB) smoothing, GIS mapping, and regression analysis. Given the wide range of patients at the ZCTA level, we first utilized an EB approach for smoothing rates from the Virginia APCD (ER Use per 1000, Primary Care Use per 1000, type 2 Diabetes Diagnoses per 1000, Mental Health Diagnoses per 1000). Numerators and denominators for each of the APCD measures were averaged over 5 years (2016 to 2020) and used to calculate the EB adjusted rates. The EB approach smooths rates toward the overall average based on the total number of patients in each ZCTA, where ZCTAs with smaller numbers are smoothed more toward the average, while ZCTAs with larger numbers are adjusted less.25 Next, a GIS was used to create maps of primary care utilization and ER use and visually explore geographic patterns, and to stratify ZCTAs by ER use quartiles to examine the characteristics of areas by high ER use rates.
Finally, we modeled the EB-adjusted ER use rates on the number of people with a primary care visit per 1000 (EB adjusted) while controlling for mental health and type 2 diabetes prevalence (EB adjusted), the percentage of adults without health insurance, the percentage of population that is Black, and a rural dichotomous variable.
Results
Figure 1 displays PC utilization and ER use by quartiles. The maps show clusters of higher rates of PC utilization throughout central and eastern Virginia, with lower rates in many parts of southern Virginia. Higher rates of ER use are observed in western Virginia, particularly along the border with West Virginia, with clusters of lower rates in northern Virginia near Washington, DC.
Primary care utilization and emergency department (ER) use by quartile.
Table 1 shows clear patterns regarding the characteristics of areas with high or low ER use. In general, areas with higher ER use rates are more rural, have higher percentages of Black or older populations, and have higher rates of poverty. Further, these areas have higher rates of uninsured, higher percentages on public insurance, and higher rates of disease prevalence (Diabetes and Mental Health). In addition, high ER use areas have significantly higher rates of primary care utilization, while having lower rates of primary care clinicians (though not statistically significant).
Characteristics of Areas by Emergency Department (ER) Use Quartile
As displayed in Table 2, primary care utilization and most covariates (except for % uninsured) were significantly associated with ER usage rates. In contrast to Table 1, the regression results show that primary care utilization and ER use have an inverse relationship - for every increase of 10 primary care visit rates per 1000 population, ER use rates decline by 7 per 1000. Consistent with Table 1, Table 2 also shows that higher rates of type 2 Diabetes, mental health prevalence, and Black residents, along with rural areas, were significantly associated with higher rates of ER use.
Factors Associated with Increased emergency department (ER) Use per 1000
Discussion
Our findings show increased ER use is more likely in communities with decreased primary care use, increased disease prevalence, higher proportions of Black residents, and being a rural area. These findings are aligned with prior literature that find increased ER visits are associated with 4 primary factors – prevalence of illness (including mental illness), social needs, access to care primary, and preference for care. Research has found that high utilizers of ER care tend to be those with chronic illness, particularly with mental illnesses, or high social needs, such as food or housing insecurity.26⇓–28 In addition, work at the individual level has focused on the relationship between neighborhood socioeconomic conditions and children’s use of the ER.
Similar to our work, ER visits were more likely to be among children from areas of low childhood opportunity index, an indicator of neighborhood context, or from low socioeconomic communities.29,30 Prior work examining the role of primary care and acute care utilization has primarily focused on hospitalizations, finding that primary care access is associated with reductions in probability of hospitalizations for ambulatory sensitive conditions in older individuals.9,13,14 Preferences may also be critical in understanding emergency department use as some individuals may prefer hospital-based care, such as the emergency department, if it is perceived to be less expensive, more accessible, or of higher quality.31,32
We also found substantial geographic variation in primary care utilization and ER use in Virginia, though areas with low primary care utilization do not always indicate higher ER use. This is not unexpected given that areas in the top quartiles for ER use also had significantly higher rates of primary care utilization compared with areas in the bottom quartile, suggesting that while primary care utilization may reduce ER use rates, the factors mentioned above (percentage of Black populations, rurality, disease prevalence) are driving higher ER use. Moreover, Daly et al. found that having more primary care access (in terms of the location of primary care clinicians) in Virginia is associated with decreased avoidable hospitalizations – suggesting that having better access to primary care does not necessarily mean higher rates of overall health care utilization. The maps show that some parts of Virginia, including south-central and eastern Virginia, have higher rates of both PC utilization and ER use, indicating higher rates of overall health care utilization in these areas. Northern Virginia outside of Washington DC have low rates of PC utilization and low rates of ER use. These inconsistent geographic patterns suggest the need for further study of the characteristics contributing to higher or lower rates of PC utilization and ER use in these areas. Our research collaborative is working on research next steps using more advanced geospatial analysis to identify “bright spot” ER use areas – defined as those with lower-then-expected ER use rates.33 Future research will target these areas for in-depth, qualitative research to better understand factors contributing to their success.
Limitations
Though ER use has been associated with a lack of primary care access, hospitalizations for ambulatory sensitive conditions (ASCs) is a more commonly used measure. We plan on using the APCD to further explore the differences between ER visit rates and ASC hospitalizations once the data become available. While HealthLandscape Virginia is a rich data source that contains a growing number of small area measures from the Virginia APCD, it currently has few measures related to morbidity and mortality. Future research will incorporate morbidity and mortality measures to better understand the association of primary care utilization, ER use, and health outcomes. Despite this limitation our work is unique in that it uses publicly available data, focuses on small areas, combines APCD data with community social determinants of health, and includes populations under 65. One additional limitation is that since the data are at the community level and show correlations; caution should be used when inferring causality.
Conclusions
Using the Virginia APCD and GIS mapping, this study finds that primary care utilization is associated with lower rates of ER use. The maps show clear geographic patterns for both ER use and primary care utilization. Important next steps include identifying priority areas, exploring their characteristics, and conducting qualitative research to better understand local factors contributing to their high or low rates of ER use.
Notes
This article was externally peer reviewed.
Funding: No funding was received to complete this research.
Conflict of interest: The authors declare no conflicts of interest.
To see this article online, please go to: http://jabfm.org/content/38/2/247.full.
- Received for publication July 19, 2024.
- Revision received October 14, 2024.
- Accepted for publication October 16, 2024.
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.