Introduction
In this report, we will examine how high blood pressure prevalence is spatially clustered in Portland, OR. An important association in demography is how spatial relationships between places are formed and how places are influenced by other nearby geographies and populations. These associations impact neighborhood level public health, and demonstration how different environmental characteristics have an effect on our well-being. These two concepts are related in the sense that nearby, bordering neighborhoods or Census tracts likely deal with many of the same health concerns. Using spatial analysis techniques, we can study where and why certain health outcomes are clustered in certain areas of a city.
Neighborhoods don’t exist in a vacuum, they are part of a larger region in which environmental factors can either promote or discourage good health practices. As a result, nearby neighborhoods may also suffer or benefit from these attributes due to havening demographically similar residents, or due to where the neighborhoods are located within a city. Portland, OR is a city that has experienced a lot of change in the last decade. Increasing gentrification has caused longtime residents to be displaced from their neighborhoods as the racial and economic diversity of the city has shifted. Measuring a single health outcome such as high blood pressure at the Census tract level is one way to look at how neighborhoods differ and how certain areas of the city may suffer from underinvestment or environmental hazards.
The numerous characteristics and attributes that make up a neighborhood can be determinants in specific health outcomes. Hypertension is a unique health outcome to study because it is highly impacted by both an individual’s physical and social environment. We know that diet and exercise play a critical role in blood pressure levels, as well as individually experienced stress levels. We also know that the built environment and social health are associated with one another, and we can determine why certain geographic clusters of this outcome may exist. There may be environmental reasons why a certain area experiences greater prevalence of high blood pressure, such has having limited access to outdoor recreation or low food access. In neighborhoods with less community organizations, individual level social health can suffer. All of these factors are important in ensuring positive public health outcomes.
Literature Review
There is a significant amount of existing literature that discusses where and why different measures of health are clustered across cities. There are a range of categories that researchers can study with regard to public health including health outcomes, preventative care, and unhealthy behaviors. This report will ultimately focus on high blood pressure prevalence in Portland, OR, but reviewing the studies done on other classifications of health may help us better understand the motivating factors in outcomes as we determine independent variables to include in a model. Since there is literature on prevention and health behaviors and how they are spatially distributed, it will be useful to see what methodologies were used in the research process. We’ll also discuss studies done specifically on neighborhood level determinants of health outcomes in Portland, OR.
It is a common research topic throughout public health and demography journals that the built environment has an effect on our health. The neighborhood environment influences many of our individual behaviors. Differences in the physical environment impact what types of transportation are available, the prevalence of grocery stores, and our social relationships with neighbors. All of these factors can vary by neighborhood and be clustered in different parts of a metro area. In “Neighborhood built environment and income: Examining multiple health outcomes”, Fallis et al. study outcomes including physical activity, body mass index, and quality of life. This article is particularly useful to this review because the study was conducted on 32 neighborhoods in the Seattle, WA and Baltimore, MD regions (Fallis et al., 2009). Seattle is similar to Portland in the socioeconomic characteristics of the population while Baltimore and Portland are similarly sized in their overall population.
This research cites the neighborhood quality of life study (NQLS) as method of identifying neighborhoods as having low or high walkability ratings. Their findings showed that
living in neighborhoods with a high walkability rating was associated with higher physical activity and lower levels of obesity, while these patterns were sustained across income levels. However, individuals in high-income areas did experience lower body mass index (Fallis et al., 2009). Therefore, they conclude that better walkability leads to healthier behaviors, but our research wants to answer how behavior led outcomes are clustered. Living in a pedestrian friendly neighborhood with a high density of available resources could be a factor that leads to clustering of adults with lower BMI, and therefore potentially lowers levels of high blood pressure or diabetes.
Air pollution is another commonly researched environmental risk that is associated with a wide range of health outcomes. A study conducted in Barcelona found that asthma cases related to air pollution were worse among tracts that were lower on the MEDEA social deprivation index that was used in the study. This index includes parameters such as share of manual workers, education levels and unemployment rates. The research states that the wealthiest groups in Barcelona are more likely to live in the city center where traffic density is high leading to increased air pollution (Pierangeli, 2020). They found that this correlation differs among other countries within Europe, but the reverse trend is found throughout cities in North America.
We can also look at research that has been done on changes to the neighborhood environment, and how those changes affect the health of its residents. An article in Health & Place studied how diabetes incidence is associated with neighborhood level social and economic changes in Madrid. This study is valuable because they used a large sample size of 199,621 adults over six years of socio-economic change. They state that their research was unique from studies that had been done on the health effects of refugees relocating to wealthier areas, with a change of residential mobility (Bilal et al., 2019). Here, their focus was on the individuals who do not move which is a much higher share of the population.
To do our analysis in this report, we will use data from the CDC 500 Cities Project. While a longitudinal study like the one done by Bilal et al. could be informative, it appears that the 500 Cities Project only has data for 2017. Therefore, with a health outcome of high blood pressure as the dependent variable in the model, we won’t be able to see how tract levels have changed over a certain amount of time. While this study will be different from the study done in Madrid, it will be helpful as a framework as they focused on shifts in socioeconomic status at the tract level. Its also noted that there are few studies on neighborhood dynamics and diabetes incidence at the this large of a scale.
Another issue that is widely research in neighborhood level health is food access and food insecurity, the latter with specific focus typically on children and their development. Access to fresh produce and healthy foods is a major barrier to positive health outcomes for many Americans. In a study on food accessibility in the Detroit metro area, proximity to the nearest grocery store is similar across Census tracts that experience very low levels of poverty, irrespective of the racial / ethnic characteristics of the neighborhood. In the most impoverished neighborhoods, there was a racial disparity as African Americans lived on average 1.1 miles further from a grocery store than White neighborhoods (Zenk et al., 2005). If this research was done specifically on children, it could provide information into health disparities over time and the health risks that are involved with low food accessibility.
Research by Breyer and Voss-Andreae look at the issue of food access in Portland through the lens of geographic and economic barriers. The geographic component to this study is critical because it addresses the reality that food prices differ at different grocery stores. Due to increased levels of gentrification, some low-income individuals live in a neighborhood with access to a high-end grocery stores, but not a low-cost store where they want to shop because it is what is affordable (Breyer & Voss-Andreae, 2013). With this realization the researchers concluded that while food deserts are less common in Portland than in other American cities, what they identified as “food mirages” are much more prevalent as “81% of low income households in Portland reside in food mirages” (Breyer & Voss-Andreae, 2013, p. 134).
Due to the findings of Breyer & Voss-Andreae, the use of data involving the USDA declaration of a food desert will be omitted as a predictive variable of health outcomes in this study. If low-income Portland residents are traveling an additional 1.9 miles to reach a low-cost grocery store, they are functionally in the same position as those who live in food deserts. The Mapping Food Deserts in the United States page on the USDA site mentions the definition of a food desert to be “lacking access to healthy and affordable food” but doesn’t detail the process of how they determine affordability. At least for low-income Portland residents, this is an important distinction.
While we won’t be using food insecurity as a variable in this research, it still has a significant effect on neighborhood level health outcomes. Food access is associated with other socio-economic determinants of health, and neighborhoods with low food access may still be where we find clusters of highly concentrated negative health outcomes. Leonard et al. discuss this idea in their research on overlapping geographic clusters of poor health and food insecurity. There are two keys results from their research; the first states that the incidence of overlapping geographies of unfavorable outcomes was more likely than those of favorable outcomes. This is to say that a county that suffers from low food access is more likely to also face negative health outcomes, while positive determinants of health don’t overlap as often. Also, geographies with higher black populations and higher levels of poverty were more likely to see an increase in overlapping negative outcomes (Leonard et al. 2018).
From this research we can see that counties with high shares of black residents, poverty and food insecurity are most likely to be at risk of experiencing unfavorable health outcomes. While Portland overall has a low share of black residents, there are historically black neighborhoods throughout north and northeast Portland (Bodenner, 2016). This specific area has also seen some of the highest levels of gentrification in the city over the last ten years and may be prone to the “food mirage” issue of food accessibility. In looking at the spatial clustering of health outcomes, from the existing literature we would expect north and northeast Portland to experience different outcomes than other parts of the city.
Most of the research that has been discussed thus far is focused on neighborhood environments or specific health behaviors in outcomes. In this review it is also critical to study the methods of how these topics have been spatially analyzed in previously published literature.
The paper “A spatial analysis of health status in Britain, 1991–2011” examines disparities across the country. The research utilized spatial analysis techniques to conclude that health outcomes are spatially clustered in Britain, and it is has become more segregated by long-term illness (Dearden et al. 2019). They also used what was called Supergroup Area Classification to identify regions of Britain including identifiers such as “Scottish Countryside”, “Prosperous England”, and “Business & Education Centres”. A similar classification could be used for the city of Portland that is commonly divided into six different sections. They are known as the four ordinal directions (NE, SE, NW and SW) plus North and East Portland. Tract boundaries follow lines that these four sections can be divided by including the Willamette River, Burnside St and N Williams Ave.
Another report uses the CDC’s variation of a social vulnerability measure that was discussed previously in the research that was done in Barcelona. Here, researchers are studying the spatial relationship between heat-related illness throughout Georgia and social vulnerability. The CDC’s SVI is found by using “15 U.S. census variables at tract level to help local officials identify communities that may need support in preparing for hazards; or recovering from disaster” (Center for Disease Control and Prevention, 2016). While the study in Georgia uses the SVI at the county level, the measure is also available down to the Census tract level which could be a useful variable in determining how health outcomes are clustered. The rest of the study mostly discusses the issues faced by large rural areas of poverty throughout the state (Lehnert et el. 2020). This isn’t applicable to our research, but the methods and analysis used for spatial exploration are useful to review.
By reviewing the existing literature on spatial clustering of health outcomes, we’ve seen some methods that will be used in our analysis as well as some gaps in areas that haven’t been studied. It should be noted that this isn’t an exhaustive and comprehensive review of all of the published research on this topic. However, there doesn’t seem to be extensive research on specific outcomes at the Census tract level in Portland. Most of the research we’ve seen in the review is done at the county level of observation for a state or a larger city. Another takeaway is that we see unfavorable health outcomes in much higher numbers in areas with both more black and more low-income residents. In Portland specifically, low income groups face a unique challenge to food access that isn’t captured in traditional data on food deserts. Due to these factors and Portland’s low racial / ethnic diversity, we may see at risk groups more highly concentrated than we would in other cities.
Data and Methods
To examine the prevalence of high blood pressure in Portland, we will perform exploratory spatial data analysis (ESDA) to better understand how this health outcome is spatially clustered. We also use a regression model to look at the relationship between high pressure, and two independent variables of educational attainment and median household income. This provides a view of how socioeconomic indictors relate to health outcomes in Portland. The data used for high blood pressure prevalence is from the CDC 500 Cities Project. This dataset is a collection of information on health outcomes, prevention, and unhealthy behaviors. The data is available at the tract level as well as aggregated data for each state. For our socio-economic variables, we used American Community Survey data and analyzed the information in R.
First, we’ll look at maps of our variables. This allows us to see how the tracts are dispersed throughout Portland and provides some insight on how each variable is clustered with neighboring tracts. The map below shows high blood pressure prevalence by tract.
This is the map we want to explain in our analysis. We can see that there are some tracts clustered in the center of the map, on the inner east side of Portland where high blood pressure prevalence is below 20%. The western area of the map shows fairly low prevalence values, while the highest area is in outer east Portland. Next, we’ll look at maps of median family income by tract and share of the population twenty-five years and older who have a bachelor’s degree.
Median Family Income by Tract in Portland, OR
Percent of the Population over 25 with a Bachelor’s Degree by Tract in Portland, OR
In these maps, we can see some similarities in the independent variables as well as education and median income’s association with high blood pressure. We could hypothesize that education and income would have a negative relationship with high blood pressure from what we see here. Outer east Portland has the lowest levels in educational attainment, while median income is more scattered by tract throughout the city. In both maps we see dark blue tracts on the west side of the city in the West Hills neighborhoods that have an older, higher income population than the rest of Portland.
Now we’ll look at some of the ESDA methods that we’ve done to examine these relationships. First, the Moran’s I statistic is a very common measure of spatial autocorrelation. This statistic provides a correlation coefficient between -1 and +1 that is a measure of the association of tracts that are nearer to one another. The Moran’s I statistic for high blood pressure prevalence is 0.527, which indicates that tract clustering is present since the value is above 0. Perfect clustering would be a value of +1, while -1 would imply perfect dispersion of the variable. We can also see this affirmation of spatial autocorrelation in the Moran’s scatterplot that indicates the relationship between observed values and spatially lagged values.
Scatterplot of Blood Pressure Prevalence by Tract
The other component of ESDA to examine is the LISA clusters map. The LISA map shows where different clusters of tracts with high values of the variable border other high tracts, as well as the same for bordering low value tracts. It will also provide observations of bordering high/low and low/high tracts which would indicate more randomness among tracts and less correlation. In our map below, we have 15 instances of tracts with low high blood pressure prevalence that border each other. This includes the area of inner east Portland that we’ve pointed out previously, as well as a cluster in northwest Portland. The low/low tract cluster in the map are highlighted in red.
In addition to the ESDA components of this study, we’ll also discuss a regression model that compares the relationship of education and income to high blood pressure. For this model, we’ve merged the CDC 500 Cities data set with American Community Survey data from the Census in R. The regression results from the linear model are shown below. Educational attainment for the population 25 years and over did have a highly significant negative relationship with high blood pressure at the 0.001 level. Even though it is significant, it has a low correlation coefficient of -2.668e-01 which indicates it isn’t highly associated. The adjusted R squared denotes that 24.61% of the variation in high blood pressure among tracts is explained by the two variables in the model.
Regression Results Table | ||||
Estimate | Std. Error | t value | Pr(>|t|) | |
(Intercept) | 4.968e+01 | 3.208e+00 | 15.489 | <2e-16 |
Education (pop >25 years) | -2.668e-01 | 3.983e-02 | -6.700 | 1.7e-10 |
Med. Household Income | -4.008e-06 | 1.241e-05 | -0.323 | 0.747 |
Summary of Results
The findings from this report signify that there is spatial autocorrelation present among high blood pressure in Portland, but there are still other variables that could better explain the relationship than the two we’ve examined here. We had a fairly high Moran’s I statistic that tells us that tracts are clustered, and this was supported in the LISA map where 15 tracts of low observed prevalence were found to border each other. In future steps for this research, it would be ideal to test more independent variables for their association with high blood pressure. If several variables with high significance are found, we could then discuss whether or not the independent variables are endogenous. If the independent variables are correlated with one another we could then research if there are built environment effects on the variables or other reasons why certain characteristics predict high blood pressure prevalence. Overall, this project has identified spatial clustering and discussed the relationship between education, income and high blood pressure prevalence.
References
Bilal, U., Glass, T. A., Cura-Gonzalez, I. D., Sanchez-Perruca, L., Celentano, D. D., & Franco, M. (2019). Neighborhood social and economic change and diabetes incidence: The HeartHealthyHoods study. Health & Place, 58, 102149. doi: 10.1016/j.healthplace.2019.102149
Bodenner, C. (2016, August 23). Gentrification in Portland: Residents and Readers Debate. Retrieved April 11, 2020, from https://www.theatlantic.com/notes/2016/08/albina/493793/
Breyer, B., & Voss-Andreae, A. (2013). Food mirages: Geographic and economic barriers to healthful food access in Portland, Oregon. Health & Place, 24, 131–139. doi: 10.1016/j.healthplace.2013.07.008
CDC 500 Cities Project (2017). Portland, OR High Blood Pressure Prevalence. [Data File]. Retrieved from https://chronicdata.cdc.gov/browse?category=500+Cities
CDC Social Vulnerability Index (SVI). (n.d.). Retrieved April 11, 2020, from https://svi.cdc.gov/
Dearden, E. K., Lloyd, C. D., & Catney, G. (2019). A spatial analysis of health status in Britain, 1991–2011. Social Science & Medicine, 220, 340–352. doi: 10.1016/j.socscimed.2018.11.014
Lehnert, E. A., Wilt, G., Flanagan, B., & Hallisey, E. (2020). Spatial exploration of the CDCs Social Vulnerability Index and heat-related health outcomes in Georgia. International Journal of Disaster Risk Reduction, 46, 101517. doi: 10.1016/j.ijdrr.2020.101517
Leonard, T., Hughes, A. E., Donegan, C., Santillan, A., & Pruitt, S. L. (2018). Overlapping geographic clusters of food security and health: Where do social determinants and health outcomes converge in the U.S? SSM – Population Health, 5, 160–170. doi: 10.1016/j.ssmph.2018.06.006
Mapping Food Deserts in the United States. (n.d.). Retrieved April 11, 2020, from https://www.ers.usda.gov/amber-waves/2011/december/data-feature-mapping-food-deserts-in-the-us/
Pierangeli, I., Nieuwenhuijsen, M., Cirach, M., & Rojas-Rueda, D. (2020). Health equity and burden of childhood asthma – related to air pollution in Barcelona. Environmental Research, 109067. doi: 10.1016/j.envres.2019.109067
Sallis, J. F., Saelens, B. E., Frank, L. D., Conway, T. L., Slymen, D. J., Cain, K. L., … Kerr, J. (2009). Neighborhood built environment and income: Examining multiple health outcomes. Social Science & Medicine, 68(7), 1285–1293. doi: 10.1016/j.socscimed.2009.01.017
Zenk, S. N., Schulz, A. J., Israel, B. A., James, S. A., Bao, S. L., & Wilson, M. undefined. (2005). Neighborhood Racial Composition, Neighborhood Poverty, and the Spatial Accessibility of Supermarkets in Metropolitan Detroit. American Journal of Public Health, 95(4), 660–667. doi: 10.2105/ajph.2004.042150
Reflection #2
Throughout the research for this project, I wasn’t sure exactly how I was going to perform the ESDA and which program I was going to use. I’m more comfortable using GeoDa as I’ve used it for two other classes before, but I was having trouble getting the shapefile to load in that program. I’m glad that I did end up doing everything in R because it allowed me to reflect on the course and see how much I’ve learned about the program. I was able to successfully create the maps I wanted and was able to merge the 500 Cities Data with the ACS data to perform a linear regression model.
The research itself does inform us about the spatial distribution of high blood pressure in Portland, but there are many ways it could be expanded in a future project. I thought about looking at a regression model that had other health outcomes as independent variables rather than socioeconomic characteristics. It could be an interesting way to examine where the highest and lowest risk tracts are with regard to health, and why outcomes are clustered in certain neighborhoods. A more in-depth project would then look at neighborhood characteristics and attempt to determine if there are built and social environment factors that highly influence certain behaviors that lead to bad health. Looking at our map of Portland, we can generally see that higher income, more educated neighborhoods have lower prevalence of high blood pressure, even if income isn’t significantly associated. What is also important to discuss is not just the socioeconomic predictors, but how the environment of the neighborhood effects our health.