Regional Risk: Mapping Single and Multiple Chronic Conditions in the United States

Abstract

Prevalence of single and multiple chronic conditions continues to increase in the United States. Chronic conditions predict significant morbidity and health care costs, especially when complicated by additional conditions. Likewise, many conditions are linked to health risk behaviors, and thus amenable to prevention. We examine regional differences in prevalence of single and multiple chronic conditions. In the process, we examine the ability of health risk behaviors to predict condition prevalence in each region. We recommend national prevention strategies with targeted content for specific geographic regions. We used 2009 Behavioral Risk Factor Surveillance System (BRFSS) data (N = 432,607) for all analyses. After grouping states into nine U.S. Census divisions, we fitted generalized linear mixed regression models and compared regional odds ratios with national averages. Analyses controlled for helpful and harmful behaviors, health insurance coverage, and demographic characteristics. Odds ratios for single and multiple chronic conditions deviated significantly from national averages in all nine regions. Health behaviors significantly predicted prevalence for both single and multiple conditions within regions, but differences in behaviors between regions did not fully account for observed disparities in prevalence. Significant regional differences in disease prevalence suggest priority areas for prevention efforts. Promoting healthy behaviors and mitigating harmful behaviors in high-risk regions may help to reduce overall chronic condition prevalence, but is unlikely to obviate disparities between regions. Targeted needs assessment should be conducted within each region with higher-than-average risk to determine intervention strategies with the greatest likelihood of near-term impact.

Keywords

adult Behavioral Risk Factor Surveillance System multiple chronic conditions hierarchical logistic modeling population surveillance

Introduction

In the United States, national and regional prevalence estimates for most chronic diseases have increased dramatically in recent decades (Perrin, Bloom, & Gortmaker, 2007). Health conditions that cause persistent problems (Goodman, Posner, Huang, Parekh, & Koh, 2013) are increasing in both incidence and prevalence. These conditions include chronicities present from early life and those that begin with acute health events later in life (Cooper et al., 2000), as well as those complicated by other health issues. Increasing prevalence of diverse and complex chronic conditions strongly predicts both significant morbidity and rising health care costs (Thorpe & Howard, 2006).

Care costs increase exponentially as disease prevalence and complexity increase (Thorpe & Howard, 2006). This is of particular concern in the United States, where latest estimates suggest that multiple chronic conditions affect one in four adults (Ward, Schiller, & Goodman, 2014). Annual costs for treating multiple chronic conditions were higher among young adults aged between 18 and 49 years than among older adults aged between 50 and 64 years (Naessens et al., 2011). Costs for treatment of single and multiple chronic conditions can vary by several factors, including geographic region (Wennberg & McAndrew Cooper, 1996) and siloization of care services for specific diseases (Parekh, Goodman, Gordon, Koh, & HHS Interagency Workgroup on Multiple Chronic Conditions, 2011). Researchers have thus sought to examine the role of region in determining health outcomes—and resultant differences in costs—among U.S. populations (McGlynn et al., 2003).

Data from the Behavioral Risk Factor Surveillance System (BRFSS) have previously been used to examine differences in condition prevalence, and to inform development and evaluation of programs and policies to mitigate geographic disparities in morbidity (Holtzman, Powell-Griner, Bolen, & Rhodes, 2000). However, most of this work has focused on state-level differences. Previous studies have revealed that particular clusters of chronic conditions are common within data captured by the BRFSS for each U.S. state (Chen, Baumgardner, & Rice, 2011). We hypothesized that the same might be true for multistate regions as defined by the Census. Racial/ethnic and socioeconomic disparities exist in health-related quality of life among people with specific chronic conditions (Hayes et al., 2006), and states within specific U.S. regions often share broad population demographics. We thus sought to explore multiple chronic condition prevalence and etiology at the regional level.

Our study uses data from the 2009 BRFSS to examine regional differences in the prevalence of single and multiple chronic conditions. Specifically, we compare risk for single and multiple chronic conditions within each region to the national average, and attempt to isolate the unique predictive value of regional residence for multiple chronic condition risk when other health determinants are accounted for. We conclude with recommendations for translating these findings into evidence-based mitigation.

Materials and Methods

We used the SAS V.9.3 procedure, PROC GLIMMIX, to compute generalized linear mixed models of 2009 BRFSS data to explore region-level risks for single and multiple conditions. Outcome prevalence values for specific regions were compared and contrasted using odds ratios (ORs). We then generated hypotheses about the influence that regional variance in health behaviors may exert on observed disparities in single and multiple chronic condition prevalence.

The BRFSS collects data on health behaviors that are linked to chronic diseases via telephone surveys of community-dwelling U.S. adults. In 2009, cell phone interviews were also included in a pilot study to reach people in households with only cellular phone service (Centers for Disease Control and Prevention, 2014). These data were included in both the full 2009 BRFSS dataset and the specific sample used in this study. Interviews were administered to a total of 432,607 respondents. The median response rate was 52%, and the median cooperation rate was 75% (Centers for Disease Control and Prevention, 2014). BRFSS data are weighted to adjust for selection, nonresponses, and noncoverage differences in probability (Centers for Disease Control and Prevention, 2014). BRFSS person-level weights (_finalwt) and state strata indicators were used for estimation of unadjusted regional prevalence using SAS procedure PROC SURVEYFREQ (Centers for Disease Control and Prevention, 2014). We tested the sensitivity of our logistic mixed model findings using person-level weights by substituting rescaled weights following Carle (2009). This study used BRFSS data from the 50 states and the District of Columbia with valid responses (N = 304,982). We used modeling frameworks accounting for the multilevel structure of the BRFSS data, allowing us to compare region-level results with national averages. We further adjusted our analyses for sampling design using standard survey weights, and performed a variety of sensitivity analyses using rescaled weights.

We assessed nine conditions: diabetes, high blood pressure, heart attack, angina or coronary heart disease, stroke, high cholesterol, asthma, arthritis, and cancer. Inclusion in these categories was based on participant self-reports of ever having received a formal diagnosis of a given condition from a health professional. First, we fitted logistic regression models predicting the prevalence of one or more chronic conditions versus none in our full study population (N = 304,982). We also generated prevalence estimates for the subset of respondents with at least one chronic condition and modeled the likelihood of multiple chronic conditions versus only one condition (N = 215,715). To control for potential confounding of net effects, each of our models included variables indicating health insurance coverage status, race, age, sex, income, and education.

To facilitate exploration of single and multiple chronic condition etiology and thus recommend potential preventive strategies, we focused intensively on behavioral determinants of health. Specifically, we incorporated robust information on both harmful and helpful health behaviors. Harmful behaviors included smoking, caregiving for one or more persons with illness/injury, and heavy alcohol consumption. Helpful behaviors such as number of days of adequate rest or sleep during the past 30 days; participation in nonwork-related physical activity such as running, calisthenics, golf, gardening, or walking during the past 30 days; meeting recommendations for moderate and vigorous physical activity; degree of emotional support and life satisfaction; daily fruit and vegetable consumption; receipt of injection or spray influenza vaccine during the past 12 months; and receipt of any pneumonia/pneumococcus vaccine during the past 12 months were included in the model. We included two different measures of physical activity to facilitate capture of BRFSS’ participants’ total amount of regular exercise, that is, exercise both within and outside of work settings. Although these measures may overlap somewhat—we computed a Pearson correlation coefficient of .27 between our exercise predictors—our concerns about potential collinearity were low for two reasons. First, we used only logistic modeling frameworks, which are not sensitive to collinearity. Second, we saw value in incorporating a measure capturing work-related physical activity rather than only accounting for leisure time exercise.

To examine regional differences, we grouped states into nine divisions using U.S. Census’s geographic definitions: New England (Maine, New Hampshire, Vermont, Massachusetts, Rhode Island, and Connecticut), Middle Atlantic (New York, New Jersey, and Pennsylvania), East North Central (Ohio, Indiana, Illinois, Michigan, and Wisconsin), West North Central (Minnesota, Iowa, Missouri, North Dakota, South Dakota, Nebraska, and Kansas), South Atlantic (Delaware, Maryland, District of Columbia, Virginia, West Virginia, North Carolina, South Carolina, Georgia, and Florida), East South Central (Kentucky, Tennessee, Alabama, and Mississippi), West South Central (Arkansas, Louisiana, Oklahoma, and Texas), Mountain (Montana, Idaho, Wyoming, Colorado, New Mexico, Arizona, Utah, and Nevada), and Pacific (Washington, Oregon, California, Alaska, and Hawaii) (U.S. Census Bureau, 2014).

To accommodate the nested data structure of individuals living within broader geographic areas, we generated hierarchical logistic models (Dai, Li, & Rocke, 2006). We used a two-level modeling framework to assess differences in chronic condition prevalence between regions, with national averages as our reference category for dummy variables capturing specific regions. Our first model predicted prevalence of at least one chronic condition in our full study population. Our second model predicted multiple chronic conditions among people with at least one diagnosed condition. We set our p value for determining statistical significance level at .05.

In response to feedback from peer reviewers, we subsequently ran two sets of sensitivity analyses to assess any needed revisions to our original models. The first set of these analyses reproduced the original models with adjustments for multiple comparisons. The second set of these analyses created “two-step” versions of the original models—an alternate approach to estimating the significance of parameters more rigorously. In both cases, our original results remained substantively unchanged. We thus present the original models in our tables, but encourage conservatism in interpreting results despite the corroboration of these models’ significance estimates via sensitivity analyses. In all cases, results are presented using ORs for ease of interpretation by public health professionals. These metrics remain the standard in the field for communicating relative levels of risk to health in a relatively lay-oriented format.

We also received peer review feedback suggesting that we present coefficients of variation. However, coefficients of variation are for the comparison of mean values. The number of chronic conditions across the regions are not normally distributed; we therefore did not think that the mean number of chronic conditions was suitable for summarizing the data used to compute our two regression models. Instead, we present weighted percentages in our tables to illustrate the distribution of chronic condition frequencies across regions. Significant differences between regions were illustrated by chi-square test statistics from the regression models themselves.

Ethics Statement

This study was approved by the Florida State University Human Subjects Review Committee. Approval was granted effective May 3, 2011. Participants did not interface directly with our research team. Rather, this study was conducted using secondary data from the BRFSS. Verbal consent was obtained by Centers for Disease Control and Prevention’s employees for each interviewee in the 2009 BRFSS sample. These consent procedures were approved by the appropriate internal committees at the Centers for Disease Control and Prevention prior to collection of data for the 2009 BRFSS.

Results

Among U.S. adults (age 18 or older), nationally, about 59% reported they had at least one of the following conditions: diabetes, hypertension, heart attack, angina or coronary heart disease, stroke, high cholesterol, asthma, arthritis, or cancer. Approximately 56% of these individuals reported having more than one diagnosed condition (Table 1). Regional differences were observed in these prevalences. Sensitivity analyses performed with rescaled survey weights did not substantively alter our original findings for prevalence of either at least one chronic condition or multiple conditions among those with at least one condition. We thus present estimates based on standard weights for ease of interpretation in Tables 1 and 2.

Table 1.

Weighted Percentage of Population Affected by Chronic Conditions Among Adults Aged 18 Years or Older by U.S. Census Division Compared With National Average, 2009 Behavioral Risk Factor Surveillance System.

U.S. Census division	Number of diagnosed conditions among all adults with valid responses in Census division 1 to 9 (Excludes PR, VI, and GU)(N = 304,982)					≥2^a (95% CI)
U.S. Census division	0(95% CI)	1(95% CI)	2(95% CI)	3(95% CI)	≥4(95% CI)	≥2^a (95% CI)
New England	39.89 [38.93, 40.86]	27.90 [27.05, 28.75]	16.62 [16.02, 17.22]	8.83 [8.43, 9.23]	6.76 [6.42, 7.09]	53.58 [52.48, 54.67]
Middle Atlantic	39.32 [38.11, 40.53]	26.94 [25.88, 27.99]	16.77 [15.99, 17.56]	9.70 [9.13, 10.26]	7.27 [6.80, 7.74]	55.61 [54.21, 57.00]
East North Central	40.09 [39.16, 41.03]	26.11 [25.32, 26.89]	15.99 [15.42, 16.57]	9.57 [9.15, 10.00]	8.23 [7.85, 8.62]	56.42 [55.38, 57.47]
West North Central	42.77 [41.77, 43.76]	26.54 [25.68, 27.39]	14.96 [14.35, 15.57]	9.02 [8.57, 9.46]	6.72 [6.38, 7.06]	53.64 [52.48, 54.79]
South Atlantic	39.66 [38.69, 40.62]	26.41 [25.57, 27.25]	16.46 [15.85, 17.06]	9.48 [9.04, 9.92]	8.00 [7.62, 8.38]	56.23 [55.13, 57.34]
East South Central	37.12 [35.90, 38.33]	25.45 [24.37, 26.52]	17.06 [16.23, 17.89]	10.68 [10.09, 11.28]	9.69 [9.15, 10.23]	59.53 [58.10, 60.96]
West South Central	41.53 [40.11, 42.95]	25.26 [24.07, 26.45]	16.22 [15.35, 17.09]	9.34 [8.72, 9.95]	7.66 [7.16, 8.15]	56.80 [55.19, 58.41]
Mountain	43.31 [42.07, 44.56]	26.12 [25.09, 27.15]	15.98 [15.18, 16.79]	8.13 [7.68, 8.59]	6.45 [6.05, 6.85]	53.92 [52.52, 55.33]
Pacific	44.23 [43.25, 45.22]	25.60 [24.77, 26.44]	15.29 [14.68, 15.89]	8.67 [8.24, 9.10]	6.21 [5.87, 6.55]	54.09 [52.93, 55.25]
Total	40.99 [40.60, 41.38]	26.17 [25.84, 26.50]	16.11 [15.86, 16.35]	9.28 [9.11, 9.46]	7.45 [7.30, 7.60]	55.65 [55.21, 56.10]

Note. PR = Puerto Rico; VI = Virgin Islands; GU = Guam; CI = confidence interval.

Multiple chronic conditions among those who have at least one chronic condition (N = 215,715).

Table 2.

Weighted Percentage of Population Affected by Particular Chronicities Among Adults Aged 18 Years or Older by U.S. Census Division 1 to 9 (Excludes PR, VI, and GU), 2009 Behavioral Risk Factor Surveillance System (N = 304,982).

U.S. Census division	Diabetes(95% CI)	Hypertension(95% CI)	Heart attack(95% CI)	Angina or coronary heart disease(95% CI)	Stroke(95% CI)	High cholesterol(95% CI)	Asthma(95% CI)	Arthritis(95% CI)	Cancer(95% CI)
New England	6.93^a [6.56, 7.30]	26.23^a [25.51, 26.96]	3.37 [3.13, 3.60]	3.64 [3.4, 3.89]	1.76^a [1.59, 1.92]	31.48 [30.68, 32.28]	10.21^b [9.60, 10.81]	25.06 [24.37, 25.74]	9.78 [9.36, 10.21]
Middle Atlantic	8.41 [7.85, 8.96]	28.58 [27.58, 29.58]	3.55 [3.20, 3.89]	3.98 [3.62, 4.35]	2.12 [1.81, 2.44]	32.22^b [31.19, 33.24]	9.25^b [8.55, 9.95]	25.68 [24.81, 26.55]	9.04 [8.51, 9.57]
East North Central	8.67 [8.25, 9.10]	28.87 [28.13, 29.62]	3.77 [3.51, 4.03]	4.10 [3.82, 4.38]	2.34 [2.12, 2.55]	30.67 [29.92, 31.43]	9.23^b [8.69, 9.76]	27.84^b [27.13, 28.55]	9.11 [8.71, 9.51]
West North Central	7.11^a [6.73, 7.50]	26.24^a [25.49, 26.99]	3.44 [3.20, 3.69]	3.79 [3.53, 4.06]	2.39 [2.17, 2.61]	28.82^a [28.04, 29.61]	7.86 [7.27, 8.45]	24.98 [24.23, 25.72]	9.44 [8.99, 9.89]
South Atlantic	9.06 [8.57, 9.56]	29.82^b [29.04, 30.61]	3.96 [3.66, 4.27]	4.30 [3.97, 4.63]	2.33 [2.13, 2.54]	31.86^b [31.04, 32.68]	7.35^a [6.88, 7.83]	25.54 [24.85, 26.23]	10.15^b [9.71, 10.58]
East South Central	10.43^b [9.82, 11.04]	34.69^b [33.63, 35.75]	4.70^b [4.31, 5.10]	4.48^b [4.13, 4.83]	3.02 [2.71, 3.32]	30.78 [29.74, 31.83]	8.05 [7.44, 8.67]	30.37^b [29.36, 31.38]	9.79 [9.22, 10.36]
West South Central	9.79^b [9.06, 10.52]	30.94^b [29.79, 32.08]	3.41 [3.1, 3.73]	4.00 [3.66, 4.33]	2.41 [2.15, 2.68]	30.51 [29.36, 31.66]	6.90^a [6.26, 7.53]	24.27 [23.28, 25.26]	8.73 [8.16, 9.30]
Mountain	6.92^a [6.45, 7.39]	24.55^a [23.65, 25.45]	3.41 [3.08, 3.74]	3.15^a [2.89, 3.41]	2.08 [1.87, 2.29]	28.84^a [27.83, 29.84]	9.15 [8.38, 9.92]	23.78^a [22.96, 24.60]	9.68 [9.19, 10.17]
Pacific	8.39 [7.91, 8.87]	26.07^a [25.30, 26.84]	3.17^a [2.90, 3.44]	3.16^a [2.91, 3.4]	2.18 [1.96, 2.39]	29.07^a [28.27, 29.88]	8.24 [7.73, 8.75]	21.80^a [21.14, 22.47]	8.61^a [8.21, 9.00]
Total	8.58 [8.39, 8.76]	28.50 [28.19, 28.81]	3.63 [3.52, 3.73]	3.87 [3.76, 3.98]	2.29 [2.20, 2.38]	30.60 [30.28, 30.92]	8.33 [8.12, 8.53]	25.26 [24.98, 25.54]	9.31 [9.14, 9.47]

Note. PR = Puerto Rico; VI = Virgin Islands; GU = Guam; CI = confidence interval.

Significantly lower than national average.

Significantly higher than national average.

The weighted percentages of adults reporting none of the nine chronic conditions ranged from 37.12% (confidence interval [CI] = [35.90%, 38.33%]) in the East South Central Division to 44.23% (CI = [43.25%, 45.22%]) in the Pacific Division. The absence of chronic conditions in the East South Central division was statistically lower than all divisions except the Middle Atlantic (39.32%, CI = [38.11, 40.53]) indicating an increased burden of chronic disease. The Pacific division (44.23%, CI = [43.25%, 45.22%]) appeared healthiest, with a significantly higher weighted percentage of adults without any reported chronic conditions than any other division except for the Mountain division. Among adults with at least one chronic condition, the percentage of people with multiple conditions was significantly higher in the East South Central division (59.53%, [CI = 58.10%, 60.96%]) than in any other except for the West South Central division (Table 1).

The most prevalent conditions were high cholesterol, hypertension, and arthritis; these conditions showed prevalences of 30.60%, 28.50%, and 25.26%, respectively. Stroke and heart attack were least prevalent, with less than 5% prevalence each in the overall U.S. population. Weighted percentages of prevalence for each condition by division varied from the respective national average in many cases as noted. However, most of the unadjusted point estimates were tightly clustered around the national average, with the exception of arthritis in the East South Central Division (Table 2). Both the Pacific and Mountain divisions showed significantly lower weighted percentages of hypertension, angina or coronary heart disease, high cholesterol, and arthritis. The Pacific and Mountain divisions had lower weighted percentages of heart attack, cancer, and diabetes. The East South Central division showed the largest number of significantly higher weighted percentages (compared with national averages) of diabetes, hypertension, heart attack, angina or coronary heart disease, and arthritis.

Only the South Atlantic division showed significantly higher weighted percentages of cancer, hypertension, and high cholesterol compared with national averages. The West North Central division showed significantly lower than average weighted percentages for diabetes, hypertension, and high cholesterol. The Middle Atlantic and East North Central divisions had significantly higher than average weighted percentages for high cholesterol and arthritis, respectively. Asthma was highly prevalent in the New England, Middle Atlantic, and East North Central divisions. Conversely, asthma was less prevalent in the South Atlantic division than the national average. The New England division also showed lower prevalences of diabetes, hypertension, and stroke.

Estimated ORs from hierarchical logistic models are presented in Table 3. The first model estimates the ORs for at least one chronic condition among all adults. The second model estimates the ORs for having multiple chronic conditions among those with at least one diagnosed condition. Here, we likewise present results from models using standard weights because sensitivity analyses performed with rescaled survey weights did not substantively alter the findings.

Table 3.

Estimated Odds Ratios for Prevalence of Single and Multiple Conditions Among Adults Aged 18 Years or Older, 2009 Behavioral Risk Factor Surveillance System.

Outcome variable	Model 1: At least one diagnosed condition among all adults (N = 304,982)		Model 2: Multiple diagnosed conditions among those who have at least one chronic condition (N = 215,715)
Outcome variable	OR (95% CI)	Pr > \|t\|	OR (95% CI)	Pr >\|t\|
Intercept	0.074 [0.067, 0.081]	<.001	0.059 [0.050, 0.069]	<.001
Behaviors that enhance health
Sleep days	1.022 [1.021, 1.023]	<.001	1.016 [1.015, 1.017]	<.001
Exercise
No	1.186 [1.157, 1.217]	<.001	1.213 [1.183, 1.244]	<.001
Fruit and vegetable consumption
Less than once per day or never	0.997 [0.949, 1.048]	0.91	0.878 [0.832, 0.927]	<.001
Once but less than 3×/day	1.042 [1.017, 1.069]	< .01	0.976 [0.951, 1.003]	0.08
3×, 5×/day	1.028 [1.004, 1.052]	0.02	0.993 [0.968, 1.018]	0.58
5+×/day	[Reference]	[Reference]	[Reference]	[Reference]
Physical activity
No activity	1.437 [1.381, 1.495]	<.001	1.446 [1.388, 1.507]	<.001
Insufficient	1.291 [1.258, 1.326]	<.001	1.277 [1.239, 1.316]	<.001
Vigorous	1.194 [1.161, 1.229]	<.001	1.187 [1.149, 1.226]	<.001
Moderate	1.062 [1.027, 1.098]	0.001	1.020 [0.980, 1.062]	0.33
Moderate and vigorous	[Reference]	[Reference]	[Reference]	[Reference]
Life satisfaction
Very dissatisfied	2.252 [2.010, 2.524]	<.001	1.667 [1.508, 1.843]	<.001
Dissatisfied	1.667 [1.582, 1.758]	<.001	1.423 [1.351, 1.499]	<.001
Satisfied	1.244 [1.219, 1.270]	<.001	1.191 [1.165, 1.217]	<.001
Very satisfied	[Reference]	[Reference]	[Reference]	[Reference]
Emotional support
No support	0.827 [0.784, 0.871]	<.001	0.881 [0.835, 0.930]	<.001
Rarely	1.136 [1.077, 1.199]	<.001	1.108 [1.051, 1.169]	<.001
Sometimes	1.082 [1.049, 1.117]	<.001	0.982 [0.951, 1.015]	0.28
Usually	0.986 [0.965, 1.007]	0.20	0.955 [0.933, 0.977]	<.001
Always	[Reference]	[Reference]	[Reference]	[Reference]
Immunization
Yes	1.844 [1.810, 1.879]	<.001	1.714 [1.679, 1.750]	<.001
Behaviors that are detrimental to health
Smoking
Current smoker—smokes everyday	1.099 [1.067, 1.132]	<.001	1.060 [1.026, 1.094]	<.001
Current smoker—smokes some days	1.145 [1.096, 1.197]	<.001	1.082 [1.030, 1.136]	< .01
Former smoker	1.295 [1.267, 1.323]	<.001	1.243 [1.216, 1.271]	<.001
Never smoked	[Reference]	[Reference]	[Reference]	[Reference]
Caregiver
Yes	1.215 [1.190, 1.240]	<.001	1.114 [1.090, 1.139]	<.001
Heavy drinker
Yes	1.040 [0.999, 1.082]	0.06	0.922 [0.883, 0.963]	<.001
Health insurance coverage
Yes	1.431 [1.388, 1.475]	<.001	1.291 [1.244, 1.339]	<.001
Demographic attributes
Ethnoracial background
Black—non-Hispanic	1.210 [1.166, 1.255]	<.001	1.180 [1.134, 1.227]	<.001
Multiracial—non-Hispanic	1.194 [1.112, 1.282]	<.001	1.269 [1.175, 1.370]	<.001
Other race—non-Hispanic	0.921 [0.879, 0.965]	< .01	0.969 [0.918, 1.023]	0.26
Hispanic	0.820 [0.789, 0.853]	<.001	0.903 [0.861, 0.947]	<.001
White—non-Hispanic	[Reference]	[Reference]	[Reference]	[Reference]
Age
65+	25.806 [24.298, 27.407]	<.001	13.934 [12.417, 15.636]	<.001
55-64	13.872 [13.095, 14.695]	<.001	9.257 [8.254, 10.381]	<.001
45-54	6.415 [6.064, 6.787]	<.001	4.953 [4.417, 5.555]	<.001
35-44	3.098 [2.927, 3.280]	<.001	2.835 [2.524, 3.184]	<.001
25-34	1.601 [1.510, 1.699]	<.001	1.654 [1.464, 1.867]	<.001
18-24	[Reference]	[Reference]	[Reference]	[Reference]
Sex
Male	1.110 [1.090, 1.132]	<.001	1.136 [1.113, 1.159]	<.001
Annual household income
<US$15,000	1.492 [1.434, 1.553]	<.001	1.785 [1.714, 1.858]	<.001
US$15,000-US$25,000	1.207 [1.169, 1.245]	<.001	1.389 [1.345, 1.434]	<.001
US$25,000-US$35,000	1.092 [1.057, 1.127]	<.001	1.223 [1.183, 1.264]	<.001
US$35,000-US$50,000	1.066 [1.038, 1.095]	<.001	1.161 [1.128, 1.194]	<.001
US$50,000+	[Reference]	[Reference]	[Reference]	[Reference]
Education level
No high school	1.242 [1.189, 1.299]	<.001	1.323 [1.265, 1.384]	<.001
High school	1.189 [1.160, 1.219]	<.001	1.190 [1.159, 1.222]	<.001
College or tech school	1.200 [1.173, 1.228]	<.001	1.182 [1.153, 1.212]	<.001
College graduate	[Reference]	[Reference]	[Reference]	[Reference]

Note. OR = odds ratio; CI = confidence interval.

Behavioral factors were significant in both models; this effect persisted across models for all demographic groups. Health enhancing behaviors such as sleeping well, exercising, engaging in physical activity, having strong life satisfaction, getting good emotional support, and receiving vaccines were all associated with protective effects for both models. Mid-level consumption of fruit and vegetables as compared with the highest level were associated with somewhat higher prevalence in the first model but not the second. Anomalously, the lowest levels of consumption were associated with decreased prevalence of multiple conditions. This may have owed to over-reporting of fruit and vegetable consumption among people whose daily diets did not include substantial amounts of fresh produce.

Detrimental behaviors generally predicted increased prevalence of at least one chronic condition in Model 1, and multiple chronic conditions in Model 2. Both current and former smokers were more likely than nonsmokers to have both single and multiple conditions. However, excessive drinking behavior did not significantly predict prevalence of single conditions and, in fact, associated negatively with multiple conditions. Demographic factors were also related to chronicity prevalence after controlling for health behaviors. Compared with non-Hispanic Whites, non-Hispanic Blacks and people of more than one race were significantly more likely to have single and multiple conditions. Older males with health care access deficiencies, lower income, and lower levels of education also experienced higher risk of both single and multiple conditions compared with their peers in other demographic categories.

Significant regional differences also appeared in prevalences of chronic conditions in our hierarchical logistic models (Figures 1 and 2). Compared with the national average, adults living in the East North Central (OR = 1.052 and 1.048), South Atlantic (OR = 1.041 and 1.040), East South Central (OR = 1.028 and 1.077), and West South Central (OR = 1.024 and 1.120) divisions had higher than average odds of at least one condition (first number) and multiple conditions (second number). By contrast, respondents from the West North Central (OR = 0.845 and 0.839) and Pacific (OR = 0.975 and 0.988) divisions had lower than average odds of having either single or multiple conditions. Adults living in the New England (OR = 1.015) and Middle Atlantic (OR = 1.045) divisions had higher odds of having single but lower odds of having multiple conditions (OR = 0.916 and 0.990, respectively). Adults living in the Mountain division had lower odds of having single conditions (OR = 0.994) but higher odds of having multiple conditions (OR = 1.013). All observed differences were statistically significant, with a majority significant at the 99.9% confidence level.

Figure 1.

Estimated odds ratios for having at least one chronic condition among adults aged 18 years or older by U.S. Census division compared with the national average, 2009 Behavioral Risk Factor Surveillance System.

Figure 2.

Estimated odds ratios for having multiple chronic conditions among adults aged 18 years or older by U.S. Census division compared with national average, 2009 Behavioral Risk Factor Surveillance System.

Across U.S. regions, the reported prevalence of one or more chronic conditions appeared to be exacerbated by detrimental health behaviors including heavy drinking, caregiving, and smoking and mitigated by health enhancing behaviors after controlling for demographic differences. Indeed, our results provide additional evidence of a statistically significant overall relationship between health behaviors and chronic condition prevalence. We did not test for regional differences in the prevalence of behaviors but did observe that regional populations in the United States showed fairly consistent levels of both harmful and helpful health behaviors.

Discussion and Conclusion

Results from our study are consistent with previous studies on diabetes and cardiovascular disease using BRFSS data from 2007, suggesting that health behaviors may play a pivotal role in the observed prevalence of single and multiple chronic conditions within each Census region (Tsai et al., 2010). We build on these findings by assessing regional influences on risk of other single and multiple chronic conditions. In doing so, we also isolate the unique predictive value of geographic location in contrast to health behaviors. We find that while health behaviors within regions do significantly predict chronic condition prevalence, regional differences in single and multiple chronic condition prevalence cannot be explained by corresponding differences in either health behaviors or demographics. Instead, we present overall differences in condition prevalence by region, and assess the implications of these contrasts for prevention efforts.

Our results reveal important disparities in chronic condition prevalence between geographic regions in the United States. Overall, 59% of adults reported having at least one of the specified conditions. Residents of the East South Central and Middle Atlantic divisions experienced highest risk for having at least one condition, while persons living in the Pacific and Mountain divisions experienced lowest risk. In addition, certain regions experienced elevated risk for single and multiple conditions. Risk of both single and multiple conditions was highest in the East North Central, South Atlantic, West South Central, and East South Central divisions. These risks were lowest in the West North Central and Mountain divisions; risk of multiple conditions was also very low among New England residents.

Many researchers have also studied socioeconomic disparities in health-related outcomes. Indeed, our findings suggest that race, income, education, and access to health care can significantly affect chronic condition prevalence within specific regions, both in concert with and net of health behaviors. These observed differences echo findings from literatures on geographic disparities in health at the neighborhood, city, county, and state levels. Broadly, these literatures suggest that social disadvantage across multiple domains can fundamentally cause poor health outcomes (Link & Phelan, 1995). Region-level disparities observed in our own results may thus stem from similar structural inequalities that hold substantive predictive value for population health outcomes.

Yet our findings also show persistence of geographic disparities in condition prevalence even after controlling for sociodemographic factors. Moreover, these disparities mirror established trends in BRFSS and other data for overall health disparities in the United States. For example, analysis of trends in obesity and other nutritional health-related conditions breaks down along similar geographic lines (Ford, Giles, & Dietz, 2002). These trends may help to explain observed prevalence patterns for other conditions that we explored in our analyses (Flegal, Carroll, Ogden, & Curtin, 2010). A number of factors that we were not able to measure directly in this study may have played a role in creating these disparities. These factors may include regional culture, population-level differences in life stress, government institutions and programs, unequal access to social resources not captured by BRFSS variables, religiosity, and other contextual forces linked in the literature to chronic condition outcomes.

Our findings mirror prior evidence that health risk behaviors play a role in causing both single and multiple chronic conditions for U.S. residents. Behavioral risk factors such as smoking and physical inactivity were strongly linked to single and multiple chronic conditions across U.S. regions. National chronic disease prevention programs may benefit from focusing on these specific domains of health (Thun et al., 2013). Such broad-ranging efforts can also address the general relationship between health behaviors and chronic conditions by adopting person-centered approaches that address the convergence of multiple harmful and helpful behaviors within specific individuals. Our results suggest that in individual regions strongly affected by single and multiple chronic conditions, behavioral interventions may exert a positive impact.

However, interventions focused solely on health behaviors may not translate to mitigation of observed disparities in chronic condition prevalence between different regions. Indeed, our findings suggest that regional differences in single and multiple condition prevalence may not owe to regional differences in health behaviors. Effective interventions must thus incorporate sophisticated consciousness of the intersectional relationships between geographic area of residence and cultural differences in shaping not only health behaviors (Diez Roux, 2001) but also the long-term impacts of these behaviors.

With respect to potential interactions between social and environmental factors and health behaviors in determining chronic condition prevalence, our results lend additional weight to robust evidence from biosocial literature on the relationship between demographic characteristics and health. Most important in the context of this study are the striking regional differences in racial composition of residential populations historically observed in the United States (Krivo, Peterson, Rizzo, & Reynolds, 1998). The literature suggests that racial background is associated not only with health behaviors and cultural values related to wellness but also with environmental and community exposures that affect health outcomes (Kwate, 2008). Social exposures can play just as important a role as physical ones in shaping population health outcomes, including chronic condition prevalence (Phelan, Link, & Tehranifar, 2010). Research shows that adverse social exposures can negatively affect health over time, and even compound other types of disadvantage in health already experienced by particular individuals and populations (Geronimus, 1992).

Large sample sizes and diverse sampling frames afforded by the use of BRFSS data help to contextualize and generalize our findings in the broader landscape of epidemiologic surveillance on health behaviors and outcomes. Although the BRFSS collects data using random sampling methods, it does so using specific algorithms that actually oversample specific populations (Centers for Disease Control and Prevention, 2014). Capture of robust samples for minority populations via BRFSS also facilitates identification of possible programs and interventions to improve the health resources and profiles of disadvantaged groups.

Although incorporating survey weights generally produces little difference in final parameter estimates for hierarchical models using BRFSS data (Carle, 2009), this remains an important consideration. We addressed this concern as robustly as possible by conducting our initial analyses using standard weights, and subsequently by comparing these results to a variety of sensitivity analyses using rescaled weights. Despite our rigorous approach to accounting for the BRFSS data design, we firmly acknowledge limitations on our ability to capture the magnitude of observed regional differences with complete accuracy and precision. Results should thus be interpreted as general sentinels of major disparities between different U.S. regions, as well as potential social and environmental factors contributing to those disparities. This is especially important to remember in light of the fact that the large area represented by different Census divisions likely misses pockets of higher disease prevalence at the county or municipality level, possibly attenuating the size of the computed ORs.

The low overall response rate for the 2009 BRFSS presents another important limitation. Although that year was the first to include cell phone only houses in sampling, a variety of socioeconomic constraints may still have influenced which particular household answered the phone and took the survey. Differences in response may have introduced selection bias. Specifically, householders who did not answer the phone may not have been home because they worked longer hours, had child care obligations, or experienced other constraints on leisure time. These constraints could also affect lifestyle and thus chronic condition risk.

Yet even with a middling overall response rate, the 2009 BRFSS still contains data on over 300,000 people—a highly powered sample. We thus encourage caution in interpreting our findings due to the immense size of incorporated samples. The large quantity and aggregated nature of BRFSS data offer superb advantages with respect to statistical power and generalizability of findings. Yet these same qualities may also lead to oversampling and thus introduce Type I error. Likewise, large samples may mask the importance of individual-level factors in determining behavior. Individual influences on “health lifestyles” stem from multiple demographic, social, economic, behavioral, and medical circumstances (Cockerham, 2005). As previous studies have indicated substantial clustering in behavioral risk factors by region, failure to incorporate these characteristics into modeling of regional differences can substantially confound results (Fine, Philogene, Gramling, Coups, & Sinha, 2004).

Changes in BRFSS data design circa 2011 prevented us from comparing our 2009 data directly with data from more recent years. Consequently, we could not assess whether or not health behaviors were changing in regions that had higher levels of multiple chronic condition prevalence. Such changes in health behaviors would likely also take time to exert positive impacts on population prevalence statistics. In considering potential implications of our study results, we thus attend to the possibility that we did not see differences in health behaviors between regions because populations in regions with higher prevalences of single and multiple chronic conditions had already made significant behavioral changes. We also reiterate our encouragement of conservatism in interpreting apparent significant differences that did appear in our results—not only because of the large sample sizes we used but also because of potential additional issues introduced by multiple comparisons that were not captured by our existing sensitivity analyses.

Efforts to develop new public health interventions aimed at reducing regional disparities identified by this study must take into account social justice and stratification concerns faced by population subgroups with high residence in particular geographic areas. For example, the high prevalence of type 2 diabetes among African Americans in the Southeastern United States has spurred a mass movement to adapt traditional “soul food” dishes for a healthful diet (Campbell et al., 2007). This movement originated mainly via American Methodist Episcopal (AME) churches and, indeed, has succeeded in helping Black Americans to prevent and manage type 2 diabetes while continuing to embrace traditional foods and food ways (Resnicow et al., 2001).

Realistic, culturally affirming interventions such as the “healthy soul food” movement have been shown to make substantial positive impacts on health status and behaviors for traditionally disadvantaged groups. Yet, such interventions cannot be developed without first conducting sophisticated and thorough needs assessments for each target community (Williams & Yanoshik, 2001) that inform an intersectional awareness of how health behaviors influence chronic condition outcomes. The fact that relatively identical health behaviors across regions did not translate to similar prevalences for single and multiple chronic conditions suggests that social contextual factors may shape the impact of interventions that target health behaviors without addressing underlying inequalities.

We also note that interactions between different types of social contextual factors, such as class and race, can further influence health behavior. Indeed, the dynamics of intersection between different social characteristics and processes constitute an important part of what we refer to as “culture” in the interdisciplinary sociomedical sciences literature. Differences in culture can affect perceptions of the health consequences of particular behavioral activities. However, we cannot comment meaningfully on these dynamics using the BRFSS data themselves because the survey instrument does not include questions that address cultural perceptions of health behavior. This topic should be addressed in greater detail with further research.

Along with other region-level studies of health status and behavior, our study can serve as a guiding framework for strategies to mitigate disadvantage in wellness. We recommend two key priorities for future research. First, researchers should continue to analyze regional disparities using secondary data from the BRFSS and other instruments. Comparing data across multiple collection years may prove particularly useful for illuminating changes in health behaviors at the regional level over time, and any corresponding changes in chronic condition prevalence. Second, researchers should use a combination of qualitative and quantitative methods to conduct detailed, participatory needs assessments for regional communities identified by this intervention as high-priority areas for interventions that target underlying structural inequalities rather than merely focusing on health behaviors.

Our findings demonstrate stark regional differences in chronic condition prevalence among U.S. adults. Although health lifestyles may vary somewhat between different regions in the United States (Cockerham, 2005), our findings suggest that different parts of the country are remarkably similar in how they approach health promotion and disease prevention behaviorally. Consequently, differences in health risk behaviors may not play a substantial role in accounting for these regional disparities. Keen awareness of intersectional influences thus represents a crucial priority for preventing and mitigating chronic disease. Applying findings from this research using integrative, multidisciplinary frameworks can enhance access to effective, empowering health promotion and disease prevention interventions across the United States. We identified the New England, Middle Atlantic, South Atlantic, East South Central, and East North Central regions as having elevated risk for single conditions. By contrast, we identified the South Atlantic, East South Central, East North Central, West South Central, and Mountain regions as having elevated risk for multiple conditions. Public policy should thus draw on this distinction in crafting population-level approaches to morbidity prevention. For example, understanding the differences between groups whose chronicities generally trace back to a single underlying condition and those with clusters of less strongly related conditions can help to shape effective interventions. In regions with higher odds of single conditions, primary prevention using culturally affirming strategies may suffice to address much of the burden of chronic disease. Regions with higher odds of multimorbidity among people with single conditions may require a different approach—one that focuses more robustly on secondary prevention in a culturally affirming way for people already living with chronic disease.

At a general level, we recommend that policy makers implement our findings by legislating and funding targeted needs assessment activities to inform development of primary and secondary chronic disease prevention efforts. Regions with higher-than-average risk for single and multiple chronic conditions should be prioritized for assessment and intervention development. Given the weak relationships we observed between differences in health behaviors and differences in chronic condition prevalence between regions, we recommend that needs assessment and program planning efforts in high-risk regions focus on underlying social inequalities that can fundamentally cause chronic conditions (Phelan et al., 2010).

Footnotes

Acknowledgements

The authors are grateful to Dr. Leslie Beitsch for his careful review and insightful guidance regarding this manuscript. They are also grateful to their peer reviewers and editors for suggesting a variety of changes to improve their original draft.

Authors’ Note

The research discussed in this article was originally presented at the Annual Meeting of the American Public Health Association on November 3, 2015. An abstract for the poster was published in the conference proceedings from this APHA meeting.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Author Biographies

Alexandra C. H. Nowakowski is a medical sociologist and public health program evaluator who studies aging with chronic disease. Their scholarship addresses questions of equity in access to and outcomes from chronic disease management services across the life course. They currently serve as an assistant professor in the Geriatrics and Behavioral Sciences and Social Medicine departments at FSU College of Medicine.

Jihyung Shin is a statistician and data analyst who provides statistical expertise for a variety of research and evaluation projects. Her scholarship engages interdisciplinary health datasets with complex sampling designs. She currently serves as a research fellow at the Korea Information Society Development Institute.

Henry J. Carretta is a health services researcher and public health program evaluator who studies the epidemiology and management of multiple chronic conditions. His scholarship addresses how health systems identify and respond to people with multiple intersecting health challenges. He currently serves as an assistant professor in the Behavioral Sciences and Social Medicine department at FSU College of Medicine.

References

Campbell

M. K.

Hudson

M. A.

Resnicow

Blakeney

Paxton

Baskin

(2007). Church-based health promotion interventions: Evidence and lessons learned. Annual Review of Public Health, 28, 213-234.

Carle

A. C.

(2009). Fitting multilevel models in complex survey data with design weights: Recommendations. BMC Medical Research Methodology, 9(1), Article 49.

Centers for Disease Control and Prevention. (2014). 2009 Behavioral Risk Factor Surveillance Survey. Atlanta, GA: Author. Retrieved from http://www.cdc.gov/brfss/annual_data/annual_data.htm

Chen

H. Y.

Baumgardner

D. J.

Rice

J. P.

(2011). Health-related quality of life among adults with multiple chronic conditions in the United States, Behavioral Risk Factor Surveillance System, 2007. Preventing Chronic Disease, 8(1), Article A09.

Cockerham

W. C.

(2005). Health lifestyle theory and the convergence of agency and structure. Journal of Health and Social Behavior, 46, 51-67.

Cooper

Cutler

Desvigne-Nickens

Fortmann

S. P.

Friedman

Havlik

. . . Thom

(2000). Trends and disparities in coronary heart disease, stroke, and other cardiovascular diseases in the United States: Findings of the national conference on cardiovascular disease prevention. Circulation, 102, 3137-3147.

Dai

Rocke

(Eds.). (2006). Hierarchical logistic regression modeling with SAS GLIMMIX. Irvine, CA: Conference proceedings of Western Users of SAS Software.

Diez Roux

A. V.

(2001). Investigating neighborhood and area effects on health. American Journal of Public Health, 91, 1783-1789.

Fine

L. J.

Philogene

G. S.

Gramling

Coups

E. J.

Sinha

(2004). Prevalence of multiple chronic disease risk factors: 2001 National Health Interview Survey. American Journal of Preventive Medicine, 27(2), 18-24.

10.

Flegal

K. M.

Carroll

M. D.

Ogden

C. L.

Curtin

L. R.

(2010). Prevalence and trends in obesity among US adults, 1999-2008. Journal of the American Medical Association, 303, 235-241.

11.

Ford

E. S.

Giles

W. H.

Dietz

W. H.

(2002). Prevalence of the metabolic syndrome among US adults: Findings from the third National Health and Nutrition Examination Survey. Journal of the American Medical Association, 287, 356-359.

12.

Geronimus

A. T.

(1992). The weathering hypothesis and the health of African-American women and infants: Evidence and speculations. Ethnicity & Disease, 2, 207-221.

13.

Goodman

R. A.

Posner

S. F.

Huang

E. S.

Parekh

A. K.

Koh

H. K.

(2013). Defining and measuring chronic conditions: Imperatives for research, policy, program, and practice. Preventing Chronic Disease, 10, Article 120239.

14.

Hayes

D. K.

Denny

C. H.

Keenan

N. L.

Croft

J. B.

Sundaram

A. A.

Greenlund

K. J.

(2006). Racial/ethnic and socioeconomic differences in multiple risk factors for heart disease and stroke in women: Behavioral Risk Factor Surveillance System, 2003. Journal of Women’s Health, 15, 1000-1008.

15.

Holtzman

Powell-Griner

Bolen

J. C.

Rhodes

(2000). State-and sex-specific prevalence of selected characteristics–Behavioral Risk Factor Surveillance System, 1996 and 1997. MMWR. CDC Surveillance Summaries: Morbidity and Mortality Weekly Report. CDC Surveillance Summaries, 49(6), 1-39.

16.

Krivo

L. J.

Peterson

R. D.

Rizzo

Reynolds

J. R.

(1998). Race, segregation, and the concentration of disadvantage: 1980–1990. Social Problems, 45, 61-80.

17.

Kwate

N. O. A.

(2008). Fried chicken and fresh apples: Racial segregation as a fundamental cause of fast food density in black neighborhoods. Health & Place, 14, 32-44.

18.

Link

B. G.

Phelan

(1995). Social conditions as fundamental causes of disease. Journal of Health and Social Behavior, 80-94.

19.

McGlynn

E. A.

Asch

S. M.

Adams

Keesey

Hicks

DeCristofaro

Kerr

E. A.

(2003). The quality of health care delivered to adults in the United States. The New England Journal of Medicine, 348, 2635-2645.

20.

Naessens

J. M.

Stroebel

R. J.

Finnie

D. M.

Shah

N. D.

Wagie

A. E.

Litchy

W. J.

. . . Nesse

R. E.

(2011). Effect of multiple chronic conditions among working-age adults-page. American Journal of Managed Care, 17, 118-122.

21.

Parekh

A. K.

Goodman

R. A.

Gordon

Koh

H. K.

& HHS Interagency Workgroup on Multiple Chronic Conditions. (2011). Managing multiple chronic conditions: A strategic framework for improving health outcomes and quality of life. Public Health Reports, 126, 460-471.

22.

Perrin

J. M.

Bloom

S. R.

Gortmaker

S. L.

(2007). The increase of childhood chronic conditions in the United States. Journal of the American Medical Association, 297, 2755-2759.

23.

Phelan

J. C.

Link

B. G.

Tehranifar

(2010). Social conditions as fundamental causes of health inequalities: Theory, evidence, and policy implications. Journal of Health and Social Behavior, 51(Suppl. 1), S28-S40.

24.

Resnicow

Jackson

Wang

A. K.

McCarty

Dudley

W. N.

Baranowski

(2001). A motivational interviewing intervention to increase fruit and vegetable intake through Black churches: Results of the Eat for Life trial. American Journal of Public Health, 91, 1686-1693.

25.

Thorpe

K. E.

Howard

D. H.

(2006). The rise in spending among Medicare beneficiaries: The role of chronic disease prevalence and changes in treatment intensity. Health Affairs, 25(5), w378-w388.

26.

Thun

M. J.

Carter

B. D.

Feskanich

Freedman

N. D.

Prentice

Lopez

A. D.

. . . Gapstur

S. M.

(2013). 50-year trends in smoking-related mortality in the United States. The New England Journal of Medicine, 368, 351-364.

27.

Tsai

Ford

E. S.

Zhao

Pearson

W. S.

Balluz

L. S.

(2010). Multiple healthy behaviors and optimal self-rated health: Findings from the 2007 Behavioral Risk Factor Surveillance System Survey. Preventive Medicine, 51, 268-274.

28.

U. S. Census Bureau. (2014). Geographic area reference manual. Bethesda, MD: Author. Retrieved from http://www.census.gov/geo/reference/garm.html

29.

Ward

B. W.

Schiller

J. S.

Goodman

R. A.

(2014). Peer reviewed: Multiple chronic conditions among US adults: A 2012 update. Preventing Chronic Disease, 11, Article 130389.

30.

Wennberg

J. E.

McAndrew Cooper

(1996). The Dartmouth atlas of health care in the United States. The trustees of Dartmouth College. Chicago, IL: AHA Press.

31.

Williams

R. L.

Yanoshik

(2001). Can you do a community assessment without talking to the community? Journal of Community Health, 26, 233-247.