This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Public Health and Surveillance, is properly cited. The complete bibliographic information, a link to the original publication on https://publichealth.jmir.org, as well as this copyright and license information must be included.
With the increased availability of data, a growing number of studies have been conducted to address the impact of social determinants of health (SDOH) factors on population health outcomes. However, such an impact is either examined at the county level or the state level in the United States. The results of analysis at lower administrative levels would be useful for local policy makers to make informed health policy decisions.
This study aimed to investigate the ecological association between SDOH factors and population health outcomes at the census tract level and the city level. The findings of this study can be applied to support local policy makers in efforts to improve population health, enhance the quality of care, and reduce health inequity.
This ecological analysis was conducted based on 29,126 census tracts in 499 cities across all 50 states in the United States. These cities were grouped into 5 categories based on their population density and political affiliation. Feature selection was applied to reduce the number of SDOH variables from 148 to 9. A linear mixed-effects model was then applied to account for the fixed effect and random effects of SDOH variables at both the census tract level and the city level.
The finding reveals that all 9 selected SDOH variables had a statistically significant impact on population health outcomes for ≥2 city groups classified by population density and political affiliation; however, the magnitude of the impact varied among the different groups. The results also show that 4 SDOH risk factors, namely, asthma, kidney disease, smoking, and food stamps, significantly affect population health outcomes in all groups (
The analysis reveals that population density and political affiliation are effective delineations for separating how the SDOH affects health outcomes. In addition, different SDOH risk factors have varied effects on health outcomes among different city groups but similar effects within city groups. Our study has 2 policy implications. First, cities in different groups should prioritize different resources for SDOH risk mitigation to maximize health outcomes. Second, cities in the same group can share knowledge and enable more effective SDOH-enabled policy transfers for population health.
Social determinants of health (SDOH), defined by the World Health Organization, encompass economic policies, social and physical environments, and access to health services and shapes the conditions “in which people are born, grow, work, live, and age” [
It is well established that SDOH factors, such as health behaviors, clinical care, social and economic status, and physical environment, account for 30% to 55% of health outcomes [
Understanding the ecological association between SDOH factors and population-level health outcomes is vital. Such an understanding is particularly relevant for researchers, clinicians, and policy makers in assessing new SDOH-enabled programs or policies to improve population health, enhance quality of care, and reduce health inequities [
The CHR model uses >30 ranking criteria to measure the impact of SDOH factors on the current and future population health outcomes at the county level [
This study investigates the ecological associations between the SDOH and population health outcomes at the city level to support health policy decisions using the CHR as the foundation. For this purpose, we curated data from 5 different sources and integrated them at the census tract level from 29,126 census tracts within 499 cities across all 50 states in the United States. With such smaller geographical delineations, researchers, policy makers, and other relevant parties can aggregate data into larger administrative divisions to make more impactful and effective decisions.
After grouping cities through 2 factors, namely, population density and political affiliation, we formalized measures for SDOH factors and population health outcomes and used a voting-based feature selection approach to reduce the original 148 sociodemographic and SDOH-related variables to 9. We then used a linear mixed-effects model to examine the ecological associations between the SDOH factors and population health outcomes. At the census tract level, the goodness of fit ranges from 0.65 to 0.75. At the city level, the total variance explained by the model was high, ranging from 0.86 to 0.90. The effect size of variables is different across groups. Noticeably, asthma, kidney disease, smoking, and food stamp variables were significant in all groups. Post hoc analysis was later conducted using predictive margin to assess group differences in health outcomes for the 4 behavioral health indicators that majorly affect health outcomes across all groups.
SDOH encompass a wide set of dimensions such as socioeconomics, education, physical environment, food access, health care system condition, health behaviors, community status, and politics [
In the United States, the CHR model is a widely applied model that describes how SDOH factors contribute to population health [
Although the CHR model makes health outcomes and factors easy to calculate and be understood by the general public and policy makers, it has 2 major drawbacks. First, the weights proposed by experts in the CHR may not be applicable to the entire population because different locations may have various population characteristics and dissimilar social and policy environments. Two recent studies that empirically tested the CHR [
Generally, SDOH research has both breadth and depth. Research can be found using study sites worldwide, with different takes on what SDOH are, and using a diverse range of research methods. The same perspective can be gleamed when focusing on the relationship between health outcomes and the SDOH. Owing to the divergent and seemingly agglomeration research that explores the relationship between health outcomes and SDOH factors, researchers need to be careful when defining (1) the health outcomes measurement, (2) the SDOH factors, (3) the population and study sites, and (4) the research methods. These definitions set up the phenomenon for research while providing concrete findings to help promote population health, which in turn affects policy changes.
Health outcomes and SDOH factors are interlinked. For instance, poverty, education, and income are known to be closely related to health outcomes [
Health behavior is another essential component of SDOH [
At any location, the population can be delineated into living in 3 areas: urban, suburban, and rural. Although the general public could easily discern the differences between the 3, in research, this has been proven as not as clear-cut as it may seem [
Indeed, population density is closely related to health service delivery and further influences health outcomes. On the one hand, health care institutions located in low population–density places may need to manage small-scale operations and handle financial losses owing to low volume, whereas people who live in low population–density locations may have difficulties accessing health care facilities and services because of human service and resource deficiency. For example, in less populated areas, ambulance response time is likely to be longer [
Furthermore, these low population–density locations usually have an increasingly aging population, demanding more health services and resources [
Previous studies have found that population density is associated with various health outcomes, such as mortality rate [
Local governments and their political leaning greatly affect health policy. The US political system is a constant wrestle between 2 major political parties: the Democratic Party (often colorized as blue) and the Republican Party (often colorized as red). Contention exists throughout the United States and at various levels of the government. At the state level, the differences between the parties encompass both health policy and social issues (eg, attitude and policy regarding abortion and substance use) and the preferred role of the government (eg, big vs small government) in addressing health-related problems [
At the individual level, partisan polarization in public attitudes shapes individuals’ health behaviors. For example, studies found that there are diverging attitudes between Republicans and Democrats toward influenza and the COVID-19 vaccine, in which Republicans displayed a negative attitude and intention toward vaccine, whereas attitudes and intentions of Democrats remained largely stable [
Despite the research outcomes suggesting linkages between health outcomes and SDOH, policy makers do not view SDOH as a priority, as is evident in the absence of SDOH in the general government policy agenda, despite earlier emphasis [
Furthermore, Embrett and Randall [
This study makes the following contributions to research and practice. First, the study divulges additional insights into how different SDOH factors affect health outcomes by curating related data in census tracts. In contrast to the county level, targeted but limited SDOH data are available at the census tract level. Our study presents a novel data curation process that creates additional SDOH variables that are otherwise not readily available. At the census tract level, data could be aggregated to the city level, allowing policy makers to devise policy more effectively. Second, city is a living and emergent ecosystem, and each city presents its own opportunities and challenges in terms of population health [
The study design includes 4 considerations. First, to perform the ecological analysis at the city level, we need to identify data sources for SDOH factors and health outcomes at the group level. Second, we need to determine the unit of analysis, that is, how to define the study population and the method of grouping. Third, we need to formalize the measures for SDOH factors and population health outcomes. Finally, we need to determine the appropriate data analysis method. In the subsequent sections, we describe each consideration in detail.
We curated and integrated a data set from 5 different sources: the PLACES program [
The results of our data collection process yielded the census tract as the default unit of analysis. We further examined whether this unit of analysis was sufficient or whether an additional grouping mechanism was warranted. As the census tracts resided within the city boundary, a natural grouping was to coalesce tracts based on the city itself. However, this type of grouping did not help explain the relationship between cities. Rather, additional grouping of cities was required. Therefore, we used political affiliation and population density, as discussed in the previous sections with the same name.
To determine the political affiliation of a city, we revisited the political affiliation of each census tract, which was determined using the abovementioned description. With a simple majority rule, if a city has more census tracts that are red, we assign red as the city’s political affiliation. Similarly, a city will be coded blue as its political affiliation if a majority of the census tracts residing within it are blue. Correspondingly, the population density of a tract was calculated by dividing the total population by its area in square miles [
After integrating and cleaning the data, we then formulated the health outcomes measurements for our research based on the CHR model [
Similar to the CHR model, our proposed health outcome measure assigns equal weightage to the length of life (through life expectancy) and quality of life (through poor physical health days and poor mental health days). The subtraction signs signify the negative effects of having poor physical health days and poor mental health days on health outcomes.
To formalize the measurements for SDOH factors, we first standardized the different scales such as rates, percentages, and averages of the survey responses. We also followed the CHR model by standardizing all measures with
Metadata of the final data seta.
Variable name | Type | Description |
Health outcome | DVb | Health outcome of the population (refer to equation 1) |
Asthma | IVc | Percentage of the population aged >18 years and with asthma |
Kidney disease | IV | Percentage of the population aged >18 years and with chronic kidney disease |
Smoking | IV | Percentage of the population aged >18 years who has smoked >100 cigarettes in their lifetime and currently smoke |
Teeth lost | IV | Percentage of the population aged ≥65 years who has lost all their natural teeth |
Annual checkup | IV | Percentage of the population aged >18 years who has visited a physician for a routine checkup |
Lack of sleep | IV | Percentage of the population aged >18 years who sleeps <7 hours over a 24-hour period |
Lack of health insurance | IV | Percentage of the population aged between 18 and 64 years who does not have health insurance |
Below poverty | IV | Percentage of the population below the federal poverty level |
Food stamps | IV | Percentage of households using food stamps or other cash public assistance programs |
Group | Grouping | Grouping of tracts based on the city’s population density and political affiliation (5 groups in total) |
aThe final data set included 1 dependent variable, 9 independent variables, and 1 grouping variable. Metadata applies for all 29,126 census tracts within 499 cities in the United States in 2021.
bDV: dependent variable.
cIV: independent variable.
For data analysis, a linear mixed-effects model [
The data used in this study are secondary data, which do not involve human participants, and are collected and aggregated by the respective organizations mentioned in the
All the linear mixed-effects models converged. We present the results of the linear mixed-effect models for 5 groups: blue-low (model 1), blue-mid (model 2), blue-high (model 3), red-low (model 4), and red-high (model 5) in
Our results suggest that all the groups examined in the study have an equipotential baseline, as evidenced by the comparable intercepts across all 5 city groups classified by population density and political affiliation. Furthermore, the results showed that all SDOH variables had a statistically significant impact on population health outcomes for ≥2 or city groups, but the magnitude of the impact varied among the different groups. For example, an increase in the proportion of the population with asthma, kidney disease, smoking, or using cash assistance programs (such as food stamps) is associated with a decline in population health outcomes across all groups, consistent with the SDOH literature [
Some SDOH variables were statistically significant for a subset of the city groups. For instance, the percentage of the population having an annual checkup does not have a statistically significant impact on population health outcomes for red cities with low population density (model 4: red-low), whereas the impact of the percentage of the population below the federal poverty level is not statistically significant only for red cities with high population density (model 5: red-high). Similarly, the impact of the percentage of the population without health insurance is not statistically significant for blue cities with low density (model 1: blue-low). Finally, the impact of lack of sleep is only statistically significant for rural cities, and the impact of teeth loss is statistically significant for cities with a high population density (model 3: blue-high and model 5: red-high), regardless of their political affiliation.
To further investigate these findings, a post hoc analysis was conducted using predictive margins to assess group differences in health outcomes for the 4 SDOH variables, namely, asthma, kidney disease, smoking, and food stamps, that substantially affected health outcomes across all groups (as shown in
Census tract level–fixed effect results of the linear mixed-effect model.
|
Model 1 (blue-low) | Model 2 (blue-mid) | Model 3 (blue-high) | Model 4 (red-low) | Model 5 (red-high) | |||||||||
|
Coefficient (SE) | Coefficient (SE) | Coefficient (SE) | Coefficient (SE) | Coefficient (SE) | |||||||||
Lack of health insurance | −0.124 (0.095) | .19 | −0.284 (0.042) | <.001 | −0.697 (0.038) | <.001 | −0.792 (0.093) | <.001 | −0.808 (0.138) | <.001 | ||||
Asthmaa | −0.848 (0.158) | <.001 | −1.006 (0.109) | <.001 | −1.248 (0.075) | <.001 | −1.266 (0.228) | <.001 | −1.731 (0.256) | <.001 | ||||
Kidney diseasea | −1.247 (0.130) | <.001 | −1.491 (0.096) | <.001 | −1.973 (0.092) | <.001 | −0.738 (0.193) | <.001 | −1.348 (0.284) | <.001 | ||||
Smokinga | −3.203 (0.123) | <.001 | −1.925 (0.094) | <.001 | −2.259 (0.098) | <.001 | −1.955 (0.167) | <.001 | −1.136 (0.187) | <.001 | ||||
Teeth lost | 0.011 (0.112) | .92 | −0.049 (0.081) | .54 | 0.342 (0.081) | <.001 | −0.301 (0.158) | .06 | −0.656 (0.228) | .004 | ||||
Below poverty | −0.158 (0.064) | .01 | −0.256 (0.044) | <.001 | −0.328 (0.042) | <.001 | −0.184 (0.088) | .04 | −0.079 (0.113) | .48 | ||||
Food stampsa | −0.281 (0.072) | <.001 | −0.346 (0.047) | <.001 | −0.156 (0.039) | <.001 | −0.279 (0.103) | .007 | −0.377 (0.121) | .002 | ||||
Annual checkup | 0.292 (0.110) | .008 | 0.520 (0.079) | <.001 | 1.022 (0.065) | <.001 | −0.056 (0.162) | .73 | 0.563 (0.222) | .01 | ||||
Lack of sleep | 0.955 (0.158) | <.001 | 0.193 (0.108) | .07 | 0.048 (0.078) | .07 | 1.346 (0.232) | <.001 | 0.565 (0.305) | .01 | ||||
Intercept | 1.344 (0.044) | <.001 | 1.261 (0.025) | <.001 | 1.248 (0.037) | <.001 | 1.314 (0.034) | <.001 | 1.383 (0.042) | <.001 |
aVariables are statistically significant in all groups.
City level–random effect results of the linear mixed-effect model.
|
Model 1 (blue-low) | Model 2 (blue-mid) | Model 3 (blue-high) | Model 4 (red-low) | Model 5 (red-high) |
City groups–intercept, variance (SD) | 0.116 (0.340) | 0.094 (0.307) | 0.143 (0.378) | 0.358 (0.189) | 0.068 (0.260) |
Residual, variance (SD) | 0.050 (0.223) | 0.064 (0.252) | 0.055 (0.235) | 0.054 (0.231) | 0.059 (0.243) |
Random effects | 0.2225 | 0.1936 | 0.2861 | 0.0926 | 0.1416 |
Overall model statistics and performance of the linear mixed-effect model.
|
Model 1 (blue-low) | Model 2 (blue-mid) | Model 3 (blue-high) | Model 4 (red-low) | Model 5 (red-high) |
Number of census tracts | 3350 | 9803 | 9144 | 1991 | 1728 |
Number of city | 73 | 198 | 120 | 53 | 55 |
0.6816 | 0.6759 | 0.6034 | 0.7690 | 0.7351 | |
0.9041 | 0.8695 | 0.8895 | 0.8616 | 0.8767 |
Post hoc analysis via predictive margins for (A) asthma, (B) smoking, (C) kidney disease, and (D) food stamps. Each colored line represents each city grouping. The x-axis shows whether the variable is high or low, where low means 1 SD below the mean and high means 1 SD above the mean. The y-axis displays the value of health outcome. This figure applies to all 29,126 census tracts within 499 cities in the United States in 2021.
In this study, we have demonstrated that, with the use of census tracts as the unit of analysis, more insights are extracted regarding how SDOH factors would affect the health outcomes of a diverse population. Furthermore, aggregating data into city jurisdictions enables a more targeted examination of how each city operates and influences health outcomes with an eye on policy generation and implementation at the city level.
Using a unique grouping method for cities based on population density and political affiliation, we reveal that cities are similar in many ways yet exhibit remarkable differences in public health. Specifically, we observed a major divide in public health and the different impacts of SDOH factors on population health outcomes between different political support in the United States, which contains only 2 main political parties. The results reinforce the current political climate in the United States, namely, the polarization between the 2 parties and how it affects health care outcome [
Our research yielded several interesting findings related to different SDOH factors. For example, we found that the number of adults who lost all their natural teeth before age 65 contributed to health outcomes in an oscillating manner. Although there is an established link between oral health and quality of life, there seems to be a disconnect between oral health and population health outcomes in the literature [
This study has several implications for public health. First, we highlight that public health policies should differ among cities with different population densities and political affiliations. For instance, policies targeting the population with asthma, such as promoting environmental cleanliness, reducing pollution particles, and reducing the costs of asthma treatment and medication, would have a much stronger effect in red-leaning cities with a high population density than in cities in other groups. Likewise, smoking affects the blue-leaning cities the most, so a policy to discourage smoking and promote quitting would be more beneficial to population health outcomes in those cities. As resources are scarce, cities should allocate their resources according to the effectiveness of the proposed solutions. Therefore, each devised policy could target different SDOH factors for policy interventions to optimize health outcomes based on the categorization of each city. Second, our findings provide additional variables of interest for invigorating public health policy transfer possibilities among cities. Policy transfer is a well-studied phenomenon worldwide, especially in the European Union, but it has been much less studied in the United States. This study indicates that policy transfer between cities in the same group is possible. For example, a blue-leaning city with low population density might try to focus more on reducing smoking by perusing policies from health care offices residing in other blue-leaning rural cities. Similarly, cities may be able to pool resources together and procure a repository containing all related health policies. This repository could help facilitate faster and more efficient knowledge transfer between cities across the United States.
This study has several limitations. First, the curated data were not equivalent to those of the CHR. Only a handful of equivalent variables are available at the census tract level. Second, data normalization in the CHR encompasses all counties in the United States, although this study limits the data set to 499 cities, potentially skewing the results. Third, although the cities may represent many populations, sparse suburban areas and rural areas are largely neglected owing to data availability. Fourth, as an ecological study, there are possible confounders that existed outside our data set. Finally, the ecological nature of the study prohibits generalization of conclusions such as giving more food stamps to an individual, which could result in a change in personal health outcomes. Future research should continue to explore the relationship between SDOH and health outcomes in other delineations suitable for policy decision making.
This study aimed to investigate the impact of SDOH factors on population health outcomes using a large data set comprising 29,126 census tracts within 499 cities across all 50 states in the United States. Our results identified 4 SDOH factors, namely, asthma, kidney disease, smoking, and food stamps, that have major effects across cities with different population densities and political affiliations. In addition, this study highlights the need for differentiated public health policies among cities with different population densities and political affiliations. The analysis of data at the city level, in which policies and decisions directly affect its citizens, promotes an understanding of how SDOH factors affect population health outcomes. The grouping mechanism, based on the combination of population density and political affiliation, provides a useful framework for separating and comparing different census tracts in different cities. To that end, this study adds to the existing literature on various ways to improve health equity among geographic areas or demographic and socioeconomic groups.
List of 499 cities, breaking down by 5 groups: blue-low, blue-mid, blue-high, red-low, and red-high.
List of the initial 148 independent variables.
County Health Ranking
social determinants of health
The data set will be available upon reasonable request to the corresponding author.
This work was partially funded by the Le Family Endowed Faculty Opportunity Fund at the College of Business Administration, Loyola Marymount University, California, United States. The Le Family Endowed Faculty Opportunity Fund had no direct or indirect influence on the results or direction of this research. AV and YT thank the Le Family for their support.
None declared.