Improving the Environmental Quality Component of the County Health Rankings Model

Abstract

Objectives. We examined the association between environmental quality measures and health outcomes by using the County Health Rankings data, and tested whether a revised environmental quality measure for 1 state could improve the models.

Methods. We conducted state-by-state, county-level linear regression analyses to determine how often the model’s 4 health determinants (social and economic factors, health behaviors, clinical care, and physical environment) were associated with mortality and morbidity outcomes. We then developed a revised measure of environmental quality for West Virginia, and tested whether the revised measure was superior to the original measure.

Results. Measures of social and economic conditions, and health behaviors, were related to health outcomes in 58% to 88% of state models; measures of environmental quality were related to outcomes in 0% to 8% of models. In West Virginia, the original measure of environmental quality was unrelated to any of the 8 health outcome measures, but the revised measure was significantly related to all 8.

Conclusions. The County Health Rankings model underestimates the impact of the physical environment on public health outcomes. Suggestions for other data sources that may contribute to improved measurement of the physical environment are provided.

A recent significant effort to characterize population health across counties in the United States is that of the County Health Rankings model developed by the University of Wisconsin’s Population Health Institute¹ based on the United Health Foundation America’s Health Rankings for states.² The model equally weights 2 health outcomes–morbidity and mortality–to form a total county health outcome score and identifies and weights 4 primary determinants of health. These determinants include social and economic factors (measured by education, employment, poverty, family and social support, single-parent households, and community safety) at 40% weight, health behaviors (tobacco use, obesity, physical inactivity, alcohol use, and unsafe intercourse) at 30% weight, clinical care (quality of care and access to care) at 20% weight, and physical environment (environmental quality and the built environment) at 10% weight. Considerations for choosing health determinants and assigning weights included existing studies, potential community modifiability of determinants, county availability and reliability of measures, and expert analysis paired with feedback.¹

Not included in the model are genetic determinants of health. In an often-cited model, McGinnis et al.³ assigned 30% weight to genetic influences on health. The Working Paper for the County Health Rankings team recognized that genetics influence health outcomes, but the paper indicated that the contribution of genetic factors was excluded from the model because they were considered to be nonmodifiable and nonmeasurable.⁴

Our study summarizes evidence regarding the interdependence of genetic structure and physical environment, along with evidence of increased exposure to environmental toxicants that has occurred during recent decades. We then conducted an analysis of the 2012 County Health Rankings data to test the hypothesis that the impact of environmental quality as measured in the County Health Rankings model is significantly underestimated. We tested the hypothesis by conducting a case study of 1 state to incorporate additional state-specific environmental quality indicators into a revised environmental quality measure.

THE PHYSICAL ENVIRONMENT

A stream of new technologies in recent decades has increased the quantities and potencies of chemicals that are released into the air, soil, and water, thus damaging ecosystems and increasing human disease.^5,6 More than 2.8 billion kilograms of toxic chemicals are being released into the US environment annually.⁷ Humans and other species are frequently exposed to such chemicals as mercury, benzene, and pesticides, all shown to be related to a variety of diseases and disease-related birth defects.⁸ Americans now carry a burden of at least 116 chemicals in their bodies from external sources⁹; almost every person carries lead, mercury, dioxins, and polychlorinated biphenyls.¹⁰ About 40% of world deaths are related to environmental degradation from pollution of water, air, and soil.⁸

Chemical exposures have bearing on cancer rates, birth defects, immune system defects, and developmental and behavioral problems, and contribute to reduced fertility, altered sex hormones and metabolism, and organ dysfunction.¹⁰ Between 1970 and 2002, US cancer-related deaths increased from 331 000 to 563 000,¹¹ a per capita increase from 1.63 to 1.96 per 1000 population despite improvements in cancer treatment over time. The lifetime risk of breast cancer in the United States has risen from 1 in 22 women in 1940, to 1 in 8 women currently, ¹² and although this reflects multiple causes including dietary changes and women delaying childbirth until later in life, environmental exposures may also be a factor. Cardiopulmonary and lung cancer mortality in the United States is caused by particulate matter from vehicle exhaust.¹³ The toxicity of pesticides has increased from 10-fold to 100-fold in recent years compared with those in use 4 to 5 decades ago.¹⁴ The number of pesticide poisonings in the United States has risen significantly during the past 2 decades.^15,16

GENETIC AND EPIGENETIC EFFECTS OF CHEMICAL EXPOSURES

It is understood that exposures to toxins can produce genetic mutations (i.e., changes in DNA sequence), and efforts have been undertaken to categorize chemical agents according to their ability to alter the DNA sequence in the interest of better regulation to reduce exposures.^17,18 More recently, however, the study of epigenetics has shown that heritable changes in gene expression can occur without changes in the DNA sequence.^19,20 Instead, toxic exposures can initiate epigenetic mechanisms such as DNA methylation and histone acetylation that can alter gene expression (rather than DNA sequencing) resulting in heritable changes.²¹

Regarding metals, several studies have shown an association between DNA methylation and environmental metals including nickel, cadmium, lead, and arsenic.^22–24 Exposure to particulate matter has been associated with greater cardiorespiratory and lung cancer morbidity and mortality,^25–27 and a recent study revealed that such exposure has resulted in gene-specific methylation.²⁸ Environmental exposures to particulate matter and gaseous pollutants in childhood affect susceptibility to disease later in life consistent with the long-term nature of epigenetic effects.^29,30

Thus, evidence shows an interaction between environmental toxicants and our genetic structure and functioning, changes that are heritable and that affect short- and long-term health. In view of the nature of this interaction, it is possible that our physical environment explains a significantly greater proportion of health dysfunction than previously recognized. Toxic physical environments can be modeled in county-level ecological studies, helping public and private health providers as well as health policymakers understand more about the sources of health problems.

The purpose of this study was to use the County Health Rankings data to test on a state-by-state basis the significance of the model’s 4 health determinant factors in understanding the model’s health outcome measures. The particular focus was on the contribution of the physical environment measure to the outcomes. Then, we conducted a case study of West Virginia to show how tailoring the model to state-specific environmental conditions may improve the strength of the model. A more accurate assessment of factors that influence population health may be useful to state and county public health officials in prioritizing public health needs.

METHODS

The 2012 County Health Rankings data were downloaded from the County Health Rankings Web site.¹ Data were converted for analysis using SAS Software version 9.2 (SAS Institute, Inc., Cary, NC.) The data included the state-specific county rankings for the 3 outcome measures (morbidity, mortality, and the combined health outcome rank resulting from equally weighted morbidity and mortality scores), the state-specific county ranking for the 4 factors that hypothetically contribute to the outcome measures in the health rankings model (social and economic factors, health behaviors, clinical care, and physical environment), the raw county data for the 5 measures that contribute to the mortality and morbidity outcomes, and the 23 measures that contribute to the 4 health determinant factors. The submeasures that were combined to form the primary dependent and independent variables were:

• Mortality: years of potential life lost rate.

• Morbidity: percentage of population in fair or poor self-rated health, number of poor physical health days in the last 30 days, number of poor mental health days in the last 30 days, and percentage of low birth weight deliveries.

• Social and economic: high school graduation rate, percentage of adult population with at least some college, unemployment rate, percentage of children in poverty, percentage of children in single-parent households, and rate of violent crime.

• Health behaviors: tobacco use, adult obesity rate, percentage physically inactive, percentage engaged in excessive drinking, motor vehicle crash death rate, sexually transmitted infection rate, and adolescent birth rates.

• Clinical care: percentage without health insurance, primary care physicians per capita, rate of preventable hospital stays, rate of diabetes screening, and rate of mammography screening.

• Physical environment: days when particulate matter air pollution (PM2.5) exceeded the US Environmental Protection Agency (EPA) standard, days when ozone level exceeded the standard, per capita supply of recreational facilities, percentage of population with limited access to healthy foods, and fast food restaurants as a percentage of total restaurants. For the current study our focus was on the first 2 measures of environmental quality. Analyses were therefore conducted using the full physical environment score from all 5 measures, and then again using only the 2 air quality measures.

Further details about specific measures may be found on the County Health Rankings Web site.¹

Analysis

Two sets of 3 multiple linear regression models were run separately for each state (6 models per state). The models used the morbidity, mortality, and combined health outcome county ranks as the dependent variables. In the first set, the 4 county rank scores measuring social and economic conditions, health behaviors, clinical care, and physical environment (including all 5 measures of physical environment) were the independent variables. In the second set, the physical environment measure was limited to the 2 air quality indicators, which were measured from the raw data as days when ozone or PM2.5 were out of EPA compliance. A count was made of the number of states in which each of the 4 factors were significantly (P < .05) related to the 3 outcome measures. For state analyses we conducted models only for 40 states with at least 20 counties with complete data, providing for a ratio of observations to measures of at least 5:1. State-by-state models were run, rather than a national analysis, because the intent of the rankings data was to provide local and state officials with information on priority areas for health improvement based on county ranks within states.

Case Study

For the case study, we created 2 new indicators of physical environmental quality for the state of West Virginia. One was an indicator of Stream Integrity, which is a widely used measure in ecological studies.^31–33 Stream integrity is a measure that ranges from 0 to 100 based on assessments of the population size, number of taxonomic groups (taxa), and diversity of benthic macro-invertebrate taxa living in streams. Some species are more sensitive to stream pollution than others; healthier streams will have more species diversity, greater populations, and greater relative presence of pollution-sensitive species. Taxa used in the assessment of stream integrity included Chironomidae (midges), Ephemeroptera (mayflies), Plecoptera (stoneflies), and Tricoptera (caddisflies). Stream integrity is impaired by a variety of anthropogenic features including roads, housing developments, agricultural runoff, mining, industry, atmospheric deposition, and other developmental features; as such, the quality of stream life may reflect the burden of environmental exposures that people experience.³⁴ The measure of stream integrity for this study was based on 4718 samples from West Virginia streams between 1996 and 2006. The number of stream samples per county varied from a low of 15 to a high of 277 (mean = 85.8). The number of samples per county was unrelated to the county’s stream integrity score (Pearson's r = 0.07). A mean stream integrity score was found for each county.

The second indicator was a measure of mountaintop and other coal mining in each county. Each county received a score of 1 for no mining, 2 for mining but not mountaintop mining, and 3 for the presence of mountaintop coal mining. Mountaintop mining in West Virginia and other parts of Appalachia creates serious and persistent air and water quality impairments^35,36 and has been linked to human health consequences.^37,38 Other forms of surface and underground mining are also environmentally damaging, but evidence suggests that their health impacts are not as severe as the repercussions from mountaintop mining; thus, the 3-point mining measure was used as an ordinal scale to estimate coal mining’s environmental health effects. The 2 new measures were expressed as z scores for each county relative to the state mean and summed so that each measure had equal weight.

We then tested whether this state-tailored modification to the physical environment measure would improve its association to morbidity and mortality outcomes over the existing measures of the physical environment. We used the 3 dependent measures used in the state-by-state analysis (county ranks of morbidity, mortality and the combined outcome rank) as dependent measures in regression analyses. To these 3 we added the 5 actual measures that constituted these outcomes to test for the robustness of the new environmental measure, conducting a total of 24 regression models: 8 models using the original physical environment measures, 8 using the original air quality measures, and 8 using the revised measure.

RESULTS

The behavior factor was significantly associated with the morbidity, mortality and combined outcome measures in 58% to 88% of the state-by-state multiple regression models (Table 1). The social and economic factor was significantly associated with the outcomes in 70% to 88% of the models. The clinical care and physical environment measures, by contrast, were related to health outcomes less often. In particular, the original physical environment measure was related to outcomes in 5% to 18% of the state analyses; and the measure that included only the 2 air quality indicators was related to outcomes in 0% to 8% of models.

TABLE 1—

States With Significant Determinants in Relation to Health Outcome Measures

	Behavior, No. (%)	Clinical Care, No. (%)	Social and Economic, No. (%)	Physical Environment, No. (%)
Set 1 (n = 40)a
Health outcome (morbidity + mortality)	35 (88)	8 (20)	35 (88)	4 (10)
Morbidity	23 (58)	4 (10)	31 (78)	7 (18)
Mortality	28 (70)	11 (28)	28 (70)	2 (5)
Set 2 (n = 39)b
Health outcome (morbidity + mortality)	33 (85)	6 (15)	34 (87)	2 (5)
Morbidity	24 (62)	4 (10)	31 (79)	3 (8)
Mortality	25 (64)	8 (21)	28 (72)	0 (0)

Note. Set 1 and set 2 are the same except that set 1 includes 5 measures of the physical environment, and set 2 includes only the ozone and particulate matter air pollution (PM 2.5) measures of environmental quality.

^aInsufficient sample size to test AZ, CT, DE, DC, HI, MA, ME, NV, NH, RI, and VT.

^bData missing for AK. Insufficient sample size to test AZ, CT, DE, DC, HI, MA, ME, NV, NH, RI, and VT

Using the original rankings data, West Virginia presented a typical case. Across the 8 outcome measures, the adjusted R² value for the regression models varied from 0.16 to 0.48 (mean = 0.35). The original physical environment measure was not significantly associated with any of the 8 outcomes. When the physical environment measure was reduced to the 2 measures of air quality, results for West Virginia remained essentially unchanged. The environmental quality measure was unrelated to outcomes in all 8 models (Table 2).

TABLE 2—

Summary Across Models for the Original and Revised Environmental Quality Measures

	Originala			Reviseda
Dependent Variable	R2	b (SE)	P	R2	b (SE)	P
Combined outcome	0.41	0.01 (0.009)	.16	0.58	0.08 (0.02)	< .001
Mortality	0.41	0.01 (0.01)	.14	0.49	0.06 (0.02)	.002
Morbidity	0.30	0.01 (0.01)	.28	0.53	0.10 (0.02)	.001
Years of potential life lost	0.43	108.6 (77.1)	.17	0.57	647.7 (147.0)	< .001
Physically unhealthy days	0.41	0.06 (0.04)	.13	0.61	0.39 (0.07)	< .001
In fair or poor health, %	0.48	0.07 (0.19)	.72	0.64	1.61 (0.35)	< .001
Mentally unhealthy days	0.18	0.06 (0.04)	.17	0.37	0.34 (0.08)	< .001
Low birth weight deliveries, %	0.16	0.02 (0.06)	.75	0.35	0.44 (0.11)	< .001

Note. Adjusted R² values are for the full models containing 4 independent variables. Estimates, SEs, and P values are specific to the original or revised environmental quality measure.

^aThe original environmental quality measure is based on air quality measured by ozone and particulate matter air pollution (PM2.5) US Environmental Protection Agency data. The revised environmental quality measure is based on stream integrity and mining data.

When we replaced the environmental quality measure with the new measure based on stream integrity and mining, the environmental quality measure was significantly related to all 8 of the health outcomes. The adjusted R² values across models ranged from 0.35 to 0.64 (mean = 0.52). These results illustrate how the model may be improved by consideration of state-specific indicators of the physical environment. Table 2 summarizes the parameter estimates, standard errors, and P values for the original and revised environmental quality measure across all 8 models. Table 3 provides an illustration of 1 of the 8 full models using the original and revised environmental quality measures.

TABLE 3—

Original and Revised Environmental Quality Linear Regression Model Results for West Virginia for Years of Potential Life Lost, by Health Determinant (n = 55)

	Original (Adjusted R2 = 0.43)		Revised (Adjusted R2 = 0.57)
Determinants	b (SE)	P	b (SE)	P
Behavior	3498.6 (1119.8)	.003	2445.8 (999.6)	.02
Clinical care	2275.0 (1133.7)	.06	1710.4 (871.5)	.06
Social/economic	1410.8 (1142.2)	.23	1569.2 (956.9)	.11
Environmental quality	108.6 (77.1)	.17	647.7 (147.0)	< .001

DISCUSSION

The results of the study show that for most states, morbidity and mortality outcomes at the county level are related to measures of social and economic conditions and health behaviors, but are infrequently related to clinical care as assessed by access and quality measures. The results also show that the physical environment as it is measured in the rankings model is usually unrelated to health outcomes, but that a state-specific modification of this environmental measure changes the measure from nonsignificant to strongly significant.

Social and economic conditions included in the rankings model include poverty, education, employment, social support, and community safety. There is a growing and frankly overwhelming body of evidence to support the critical importance of basic social and economic conditions in determining population health.^39–41 For example, recent research has estimated that

each year, an average of 195 619 deaths would have been averted if mortality rates among adults with an inadequate education had been the same as mortality rates among college-educated adults.^40(p680)

This is compared with an average of 25 456 lives saved each year through medical advances (e.g., drugs and other medical devices). Lalonde’s 1974 observations³⁹ regarding the key aspects of lifestyle and environmental factors continue to be underappreciated. On one hand, the current results are not surprising, but on the other, the disparity that exists between the determinants of population health and health policy is difficult to reconcile. Health policy in the United States continues to be dominated by medical care issues, and conversely basic population health drivers receive scant attention from the policy community. To cite 1 example of the relative importance of medical care and primary prevention in US health policy, the fiscal year 2011 budget for the Medicare program was $468.6 billion, ⁴² whereas the fiscal year 2011 budget for the Centers for Disease Control and Prevention and the EPA combined was $15.0 billion.^{42, 43}

The current measures of environmental quality in the rankings model include 2 measures of air quality: days out of compliance with EPA standards for levels of ozone and PM2.5. These measures are deeply flawed. They are flawed because the existing cut-points to determine safe from unsafe levels of air pollution are known to be inadequate^44,45 and because the standards inadequately address the role of ultrafine particles in impacting health.⁴⁶ But they are most seriously flawed because the majority of counties around the country contain zero monitoring stations for the assessment of ozone and PM2.5,⁴⁷ and yet the EPA database does not distinguish between counties where air quality is known to be in compliance and counties where there are no data. The unmonitored counties tend to be in more rural places. In West Virginia, only 15 of the state’s 55 counties monitor ozone and PM2.5,⁴⁷ and according to the rankings data, the best air quality in the state exists where there are no monitors. The Pearson correlation between better air quality and the absence of air monitoring stations in West Virginia is 0.68 (P < .001). None of the West Virginia air monitoring stations is located in southern portions of the state characterized by mountaintop coal mining; air quality in mining areas is suspect,^{36, 48,49} and we cannot assume that air quality is good in the absence of data.

Limitations

Our study is limited by a number of features. First, all factors may be limited by the measures available and chosen to represent them, and any factor could hypothetically be improved by adding or revising indicators. The 2012 County Health Rankings model, relative to earlier years, has provided additional indicators in efforts to improve the utility of the model. Second, the ecological design of the study is such that we have only county-level associations and not person-level links between health determinants and health outcomes. The associations may be useful to guide county or state-level planning but not to establish causal links for individuals. Third, sample sizes for the state-level regression models were small for some states and models must be interpreted with caution; rather than relying on results for a single state, the attempt was made to discern patterns in variables that were significant in multiple models across states. Those general patterns would not alter if models were restricted to fewer states with greater numbers of counties. However, that the analyses were not conducted for states with few counties may limit the generalizability of results to the full country. Finally, the health determinant “factors” may not be factors in a psychometric sense; they are conceptual labels more than true factors, and state public health officials wishing to use the rankings data may consider focusing on the individual measures to identify priority needs, rather than the more general factors.

Conclusions

The County Health Rankings model can make valuable contributions to helping state and local public health officials identify priority needs as well as identify areas where performance is strong.^50–52 The results confirm the significant contributions of social and economic conditions, and health behaviors, to differences in health outcomes that have been demonstrated often in prior research. However, the model should not be adopted uncritically, and there are ways in which it may be improved. County and state health officials who use the model in its current form to identify priority areas for public health improvement will overlook key factors that determine population health, in particular the impact of the physical environment.

There is a clear need to reconsider and improve the assessment of environmental quality in the rankings model. Exposure to environmental pollutants has a larger impact on population health than the model indicates. In addition to measures of mining activity and stream integrity examined in this study, which are applicable to some but not necessarily all states, other possible national measures for inclusion in the model may be developed from Toxics Release Inventory data,⁵³ satellite imagery data on land use,⁵⁴ and data on the locations of power plants and industrial facilities.⁵⁵