Health Disparities in Later Life

Abstract

Objective: This article examines health disparities between older Blacks/Whites by recognizing the importance of health services utilization. Although previous studies have examined health and utilization independently, this is among the first to (a) model its endogenous relation with utilization, and (b) use a continuous measure for health. Data: Household Component files from Medical Expenditures Panel Survey (MEPS) from 2004 and 2005, with 1,369 observations (1,169 for White and 200 for Black) between the ages of 61-69. Methods: The methods employed are two-equation modeling where Medicare eligibility functions as the identification criterion and also as an exogenous shock. Results: The results show older Blacks continue to remain in poorer health despite access to care and insurance status. The author shows underutilization accounts for some of this observed disparity and offers novel approaches to overcome this issue. Conclusion: With the baby-boom cohort approaching retirement, this area of research is timely. This work is also practical because the Department of Health and Human Services has launched various projects examining health care issues for Americans. One major project, Healthy People 2010, provides a “framework for prevention for the Nation [and is] designed to identify the most significant preventable threats to health and to establish national goals to reduce these threats” (http://www.healthypeople.gov/About/). Of the 28 areas, this article complements objectives relating to (a) disability and secondary conditions, and (b) access to quality health services. This article also supports Healthy People 2020, which sets a high priority on access to health care (one of 12 topic areas) and categorizes health disparities as part of the Leading Health Indicators Framework.

Keywords

health services policy well-being

Introduction

Although researchers have examined later-life health disparities by race for many years, a clear consensus as to what factors play a critical role remains unresolved. Health scholars, policy analysts, and social scientists have discussed the importance of access to care, comorbidities, diet/exercise/ nutrition, economic resources, environmental factors, genetic predisposition, insurance, and utilization. Although such studies contribute to the body of literature examining aging and health, the racial health gap still remains. This article aims to address why this gap persists by investigating the endogenous role of utilization. Another scholarly contribution relates to the modeling of health as a continuous variable, where most previous studies employ a categorical measure.

The Role of Socioeconomic Status (SES)

Researchers have utilized various data sets and econometric specifications to unravel the impact of social and economic factors. As an early example, consider the work by Mutchler and Burr (1991). These scholars address health disparities between Blacks and Whites by claiming SES “conditions many factors that relate to health, ranging from knowledge of health care practices and nutrition [as well as the] the ability to purchase medical care” (p. 344). Using the Survey of Income and Program Participation or SIPP from 1984, they use logit and Tobit regressions to show Black older adults are in worse health due to being “less advantaged in terms of income, wealth, access to health care [and] even after controlling for many of these differences, race retains a significant effect” (p. 350). A more recent article by Warner and Hayward (2006) reiterates some of these key points. Using the National Survey of Older Men (1966-1990), these researchers employ nested discrete-time hazard models to show early-life conditions relating to SES have health/mortality consequences in later life. Their work suggests not only adult socioeconomic factors but also childhood conditions relating to family structure and parental occupation account for the race gap in men’s health and mortality.

Consistency of Care and Access to Care

A recent article by Howard, Carson, Holmes, and Kaufman (2009) highlights the importance of consistency of care. Consistency of care is defined as a measure of access to care and signifies a relationship between patient and physician. Using the Piedmont Health Survey of the Elderly (PHSE) from 1987, 1990, 1994, and 1998, these scholars show older Whites and African Americans maintain better control of blood pressure (BP) with some level of consistent care. Given the aforementioned articles by Mutchler and Burr (1991) and Warner and Hayward (2006), Howard et al. (2009) find controlling for sociodemographic factors decreases the association between consistent care and BP control. If insurance, satisfaction with care, and medical history are added, then the association further decreases. Specifically, the odds ratio reduces from 1.51 to 1.06, suggesting individuals with some level of consistent care are still more likely to maintain BP. Certainly, controlling for SES and insurance are critical given their significant impact on BP management. Although insightful, this research does not fully account for why the differences in outcomes still exist and what can be done to close the gap.

The Importance of Race

Kawachi, Daniels, and Robinson (2005) provide a good framework for causal interpretations of health disparities by race and class. They suggest race and class need to be considered separately when examining health disparities and biological differences should be evaluated with skepticism. These scholars make an interesting point by arguing that hypertension and diabetes are 2 to 3 times higher among Blacks than Whites in the United States. However, “representative surveys of populations in West Africa and African-origin populations in the Caribbean have revealed prevalence rates of hypertension and diabetes that are two to five times lower than those of Black Americans or Black Britons” (p. 345). In conjunction with the above findings, consider the work by Ferraro (1987), a leading scholar in later-life health. Using the Survey of Low-Income Aged and Disabled (SLIAD), Ferraro found Black older adults to be in poorer health than White older adults. However, the groups did not “differ substantially in the number of reported chronic conditions” or “the likelihood of having a serious illness, but the illnesses are more functionally debilitating to elderly Blacks” (p. 530). Ferraro and Farmer (1996) reexamined health disparities using the National Health and Nutrition Examination Survey I: Epidemiological Follow-up Survey (NHEFS). Using longitudinal regressions on this 15-year panel data set, Ferraro and Farmer contend being Black and old does not account for higher morbidity and mortality rates. Nevertheless, these authors agree older Blacks maintain more serious illnesses and are more likely to have a faster decline in health. However, their work does not offer concrete explanations as to what accounts for this.

This article hopes to move the field of aging and health forward not only by appreciating the important contributions to these scholars and building upon their work but also by recognizing that health disparities research needs to recognize the importance of utilization and its endogenous relation with health.

Conceptual Framework

My approach builds on the work by Andersen (1968) and, more specifically, Wolinsky, Mosely, and Coe (1986), which examines older adults’ health utilization from a behavioral model. Mathematically, this can be depicted as: U = f(P,E,N).

Wolinsky et al. (1986) assert utilization is a function of Predisposing factors, Enabling factors, and Need characteristics. Predisposing factors refer to sex, marital status, race, education, and labor-force participation; Enabling refers to income, type of residence, Census region; Need refers to perceived health status and limited activity. My analysis closely resembles this with the key difference being I model both health and utilization simultaneously (see Figure 1). More specifically, what accounts for the persistent health disparity between Blacks and Whites, even after controlling for key demographic variables, socioeconomic status, insurance status, obesity, and smoking? I argue utilization accounts for some of this disparity and this variable also serves as a proximate determinant for health. In other words, utilization has a profound effect on health, particularly during later life.

Figure 1.

General determinants of health with endogenous utilization

Mathematically, my approach can be modeled as:

Health = a + b (Utilization) + \dots + e

Utilization = a * + b * (Health) + \dots + e *

In terms of a hypothesis, this research tests three, interrelated suppositions:

Hypothesis H1₀: Utilization has no association with health.

Hypothesis H1_A: Utilization has a positive association with health.

Hypothesis H2₀: Older Blacks and Whites do not differ in health.

Hypothesis H2_A: Older Blacks maintain a lower health profile (lower PCS).

Hypothesis H3₀: Older Blacks and Whites do not differ in utilization.

Hypothesis H3_A: Older Blacks have lower levels of utilization.

Finally, some readers may benefit from an illustration. The following figure shows predisposing, enabling, and need factors are baseline attributes that indirectly impact an individual’s health. In turn, these three foundational attributes play a role in influencing the next level, utilization. For older adults, health utilization becomes more important in determining health outcomes and particularly for those in poorer health: slower immune response, loss of muscle, weakening of bones, and onset of various age-related diseases and conditions. As such, health utilization serves a larger role and recognizing it (in addition to age, race, SES, nutrition, marital status, insurance) can uncover important relationships.

Data

This analysis uses Household Component (HC) files from the Medical Expenditures Panel Survey or MEPS. MEPS data series, which are projects at the Agency for Healthcare Research and Quality or AHRQ, consist of a “set of large-scale surveys of families and individuals, their medical providers, and employers across the United States. MEPS is the most complete source of data on the cost and use of health care and health insurance coverage” (MEPS survey background homepage at http://www.meps.ahrq.gov/mepsweb/). The HC includes demographic, economic, and health characteristics, as well as use and cost of medical services, for each person in the household.

Besides extensive information on health costs, services, and utilization, MEPS serves as a good data set because different years can be appended (as early as 1995), the variables definitions are consistent and comparable across years, the overall sample size generally exceeds 30,000 observations for any time period, and response rates approximate nearly 70%. I use HC files from 2004 and 2005, which correspond to data files H89 and H97, respectively. Since MEPS is structured as a changing panel in which some of the respondents are replenished/replaced at the beginning of each year, the 2004 wave also contains some respondents from 2003 and the 2005 wave also contains some respondents from 2004. Since I append H89 and H97 to create a stacked data set, I have cross-sectional observations from all three time periods.

Analytic Sample

I extract demographic variables (age, gender, marital status, race, MSA [metropolitan statistical area], Census region), socioeconomic characteristics (education, income), health variables (BMI, smoking, usual source of care, Mental Component Summary, Physical Component Summary), and several variables relating to health behaviors (valuing the need for insurance, seeking medical attention).

Dependent Variables

The main variables of interest are Physical Component Summary (PCS), which serves as a proxy for health, and Provider Visits (Visits), which serves as a proxy for utilization. MEPS uses the Short-Form 12 or SF-12 to provide two distinct measures of self-rated health—Physical Component Summary (PCS) and Mental Component Summary (MCS; see Appendix and Ware et al., 1996). The PCS weights responses to the first five items more heavily and MCS weights responses to Items 6 to 9 more heavily. The advantage of using PCS and MCS relates to not only the broad dimensions of health encompassed in the SF-12, but these measures are continuous. In other words, typical self-rated health variables generally fall into four categories: excellent, good, fair, or poor. As such, the researcher is limited in selecting regression models when self-rated health is the dependent variable. MEPS is among the very few nationally representative, panel data sets with a continuous self-rated health measure. In addition, studies by Gandek et al. (1998) and Jenkinson et al. (1997) show the adequacy of using PCS and MCS. Gandek et al. cross-validate the SF-12 selection and scoring against the SF-36 in nine countries. Since their analysis shows a high degree of correspondence between the two self-rated health instruments, they suggest the SF-12 is a practical alternative “for purposes of large group comparisons in which the focus is on overall physical and mental health outcomes” (p. 1171). Jenkinson et al. arrive at a similar conclusion, stating “PCS and MCS scores calculated from the SF-36 or a subset of 12 items (the ‘SF-12’) were virtually identical, and indicated the same magnitude of ill-health and degree of change overtime” (p. 179).

I operationalize utilization as the number of visits to access medical services. Specifically, these are “encounters that took place primarily in office-based settings and clinics. Care provided in other settings such as a hospital, nursing home, or a person’s home are not included in this category” (MEPS codebook). I use Office-based Provider Visits to measure utilization and this variable encompasses visits to physician, nonphysicians, and unknown. This is a reasonable approach since such a measure is commonly used and reported in health disparities research (see Gornick, 2008). Equally important, Lubeck and Hubert (2005) do not find inconsistencies with under- or over-reporting of self-reported physician visits when measuring health care utilization with older adults (with the exception of dermatological or ophthalmological specialties). A recent study by Zuvekas and Olin (2009) also finds a high degree of concordance when comparing self-reported in-patient visits against Medicare claims. However, there exists greater variability with office visits. At most, households underreport such visits by as much as 19%. My research attempts to overcome some of this measurement error by taking nonphysician and unknown visits into account. This broader measure may capture a portion of the unaccounted interactions with less common specialties.

Independent Variables for Health Behaviors

The Usual Source of Care, Currently Smoke, and Obesity are critical variables, particularly given Marmot’s (2007) work. Marmot highlights contemporary public health interventions have “often given primary emphasis to the role of individuals and their behaviours” (p. 1155). This is an important consideration because health disparities fundamentally derive from unhealthy lifestyles. Usual Source of Care (USOC) refers to “a particular doctor’s office, clinic, health center, or other place” a person visits for advice or when sick (MEPS Codebook). This variable is coded 1 = yes and 0 = no. Having a usual source of care functions as an additional measure of access and suggests the individual can seek continuous care. Several studies show the importance of USOC in maintaining health (see Blewett, Johnson, Lee, & Scal, 2008; Doescher, Saver, Fiscell, & Franks, 2004). A recent article by Winter et al. (2010) uses the National Health and Nutrition Examination Survey (1999-2006) to examine cholesterol management among adults aged 21 to 79. Their results suggest USOC is positively associated with statin use. As Ridker (2003) shows, statins can decrease cholesterol levels and also reduce inflammation. This is noteworthy since heart disease is the leading causes of death in the United States. More specifically, for those 65 years and older, 476,519 deaths from a total of 1,796,620 (nearly 26.5%) can be attributed to diseases of the heart for 2010 (Murphy, Xu, & Kochanek, 2012). This exceeds malignant neoplasms, chronic lower respiratory diseases, cerebrovascular diseases, Alzheimer’s, and diabetes. As Winter et al. claim, “Having a USOC is often a prerequisite for initiating and continuing chronic disease management” (p. 184).

Smoking is another dummy variable and coded 0 = does not smoke and 1 = does smoke. The deleterious effects of smoking have long been examined from an economic, social, and environmental perspective. Peruzza et al. (2003) show the adverse impact of smoking on chronic obstructive pulmonary disease (COPD) for older adults. Almeida et al. (2008) investigate its effect on brain density. Their results suggest smoking reduces regional gray matter related to early stages of Alzheimer disease. An interesting article by Chiolero et al. (Chiolero, Faeh, Paccaud, & Cornuz, 2008) suggests smoking “increases insulin resistance and is associated with central fat accumulation” (p. 801). This could exacerbate diabetes and obesity. A recent article by Hubbard et al. (Hubbard, Searle, Mitnitski, & Rockwood, 2009) adds to this enormous body of literature by assessing frailty. Utilizing the Canadian Study of Health and Aging, these scholars generate a modified Frailty Index (FI) and then show heavy smokers are most frail. Results suggest the oldest old have the highest mortality rate and there is “no emergence of ‘survivors’ with fitness or longevity advantages” (p. 468).

Finally, the Obesity measure provides some insight into respondents’ general nutrition and functional ability. Since this measure is continuous, I develop four categories for Body Mass Index (BMI): Underweight for BMI < 18.5, Normal weight for 25 > BMI > 18.4, Overweight for 30 > BMI > 24.9, and Obese for BMI ≥ 30 with the modal or omitted or reference category as Overweight. This classification derives from the World Health Organization (WHO) Technical Report #854 (World Health Organization, 1995), which is consistent with the National Institutes of Health guidelines. Similar to that of smoking, the adverse effects of obesity are well-documented across a wide range of age groups. Koster et al. (2008) find older men and women with higher levels of BMI exhibit greater functional limitations. Adams et al. (2006) find older, obese individuals to be at greater risk for cardiovascular diseases. Obese individuals are more likely to suffer from diabetes.

Independent Variables for Health Attitudes

I also measure attitudes regarding insurance, an important consideration lacking in previous studies. Specifically, I examine whether Insurance Status, Do Not Need Health Insurance, and Overcome Illnesses without Medical Help play a role in explaining later-life health disparities by race. Insurance Status refers to the respondent being covered for medical services (such as hospitalization, surgery, prescriptions) and belonging to HMO, Medicare, Private Plan, or any other type of provider. This dummy variable is listed as No Insurance and coded 0 = have insurance and 1 = do not have insurance.

For the Do Not Need Health Insurance variable, respondents are asked if they “do not need insurance” and replies are coded as disagree strongly, disagree somewhat, uncertain, agree somewhat, and agree strongly. In general, more than 80% of respondents answer disagree strongly or disagree. I recode the five categories into a dummy variable, which is listed as Not Need Insurance, with those responding uncertain or disagree as 0 = need insurance and those responding agree as 1 = do not need insurance.

Overcome Illness without Medical Help is another important attitude/behavior measure. Respondents are asked if they “can overcome an illness without seeking medical services” and replies are coded similar to Not Need Insurance question. In general, more than 70% of respondents answer disagree strongly or disagree. I recode the five categories into a dummy variable, which is listed as No Med. Help, with those responding uncertain or disagree as 0 = need medical help and those responding agree as 1 = do not need medical help.

Other Independent Variables

Other covariates include marital status, race, education, Census regions, and age. The marital status corresponds to a categorical variable for single, separated, divorced, and widowed with married as the reference group. Race corresponds to Black with White as the reference group. Education is another categorical variable for less than high school, some college, college, and graduate degree with high school as the reference group. The four Census regions refer to Northeast, Midwest, and West with South as the reference group. Finally, since this analysis focuses on older adults, I retain observations of individuals between 61 and 69 years of age. In the Methods section, I explain why this age range is critical for the analysis.

Methods

One of the most important statistical/methodological factors to consider when using two-equation modeling of health and utilization relates to the identification criteria. Below is the full-model:

Two-equation model:

PCS = a + b (Visits) + d (Pop .) + n (Marital) + v (Region) + t (Health Behaviors) + e

Visits = a * + b * (PCS) + d * (Pop .) + n * (Marital) + t * (Health Attitudes) + e *

Equation 1 shows PCS (health) is explained by visits (utilization), matrix of population characteristics, marital status, Census region, and health behaviors (population characteristics refers to age, education, ethnicity/race, gender). Equation 2 shows visits, in turn, are explained by PCS, matrix of population characteristics, marital status, insurance, and health attitudes. For the visits variable, the importance of functional form cannot be overstated. Since its impact on health does not exhibit a linear relation for the entire range of the function, a transformation may be used. Specifically, I employ both visits and square root of visits. This additional variable, which represents a decrease along the ladder of powers, also corrects for overdispersion in visits since this is a count variable.

To use this analysis, both Equation 1 and Equation 2 need to be solved simultaneously to produce the reduced-form equation. This can only be done if the two equations are properly identified. In other words, simultaneous equations with full information not only require order and rank but also identification (see Baum, Schaffer, & Stillman, 2007; Greene, 2003). To fulfill this necessary condition, I use the Age of eligibility into Medicare as the exogenous shock in Equation 2. This variable is predetermined by factors outside the model. That is, there are causal factors that determine Medicare eligibility but are not part of the model used to explain health and utilization. This variable, denoted MC age, is a dummy variable and coded 0 = not Medicare eligible and 1 = Medicare eligible. I use 65 years of age to determine eligibility into Medicare. Using age of eligibility is appropriate because it explains utilization (visits) without directly influencing health (PCS) outcomes. Using Stata: Release 10 (2007), I formally conduct an identification test, and the results indicate the system is properly identified.¹ Age of eligibility also serves another critical function. By using 65 years of age as the marker, I am able to extend the age range ± 4 years and those just below it (ages 61-64) can be compared with those just above it (ages 66-69).

In order to better understand this two-equation approach, I am compelled to show separate equations modeling health (PCS) and utilization (visits). Using ordinary least squares (OLS) regression, this comparison model also shows single equations produce biased estimates due to the endogeneity present between health ↔ utilization. Below are the formal models:

PCS = a + b (Visits) + d (Pop .) + n (Marital) + v (Region) + t (Health Behaviors) + e

Visits = a * + b * (PCS) + d * (Pop .) + n * (Marital) + t * (Health Attitudes) + e *

As compared with the full-model, there is no difference except each equation is determined independently.²

Results

For the entire sample of 1,369 observations, the mean PCS (or health) approximates 47 with minimum of 17 and maximum of 63, where higher PCS values indicate better health (see Table 1). The mean visits correspond to nearly 8, the average age approximates just below 65 with a family-size of two and income of nearly US$12,100. More than 29% of the sample is obese and nearly 17% smoke. In terms of characteristics by race, 1,169 observations are White and 200 represent Black. For Whites, the average PCS equals 47, visits are nearly 8, income exceeds US$12,750, 29% are obese, and 16% smoke. For Blacks, the average PCS equals 44, visits approximate 6, income is just above US$8,500, more than 36% are obese, and nearly 20% smoke. Additional descriptive statistics are provided in Table 2.

Table 1.

Descriptive Statistics for Entire Sample

	M	SD	Min.	Max.
Visits	8.15	13.54	0.00	187.00
PCS	47.01	10.05	17.00	62.87
Black	0.14	0.35	—	—
Age	64.66	2.58	61.00	69.00
Male	0.47	0.50	—	—
Less than high school	0.25	0.43	—	—
Some college	0.17	0.38	—	—
College	0.13	0.33	—	—
Graduate degree	0.10	0.31	—	—
Family size (log)	0.63	0.45	0.00	2.45
Income (log)	9.39	2.39	0.00	12.10
Single	0.05	0.21	—	—
Separated	0.02	0.13	—	—
Divorced	0.13	0.33	—	—
Widowed	0.11	0.32	—	—
Northeast	0.16	0.37	—	—
Midwest	0.23	0.42	—	—
West	0.21	0.41	—	—
Non-MSA	0.23	0.42	—	—
USC	0.89	0.34	—	—
MCS	50.89	9.67	13.08	66.44
Smoke	0.17	0.38	—	—
Underweight	0.01	0.10	—	—
Normal weight	0.28	0.45	—	—
Obese	0.30	0.46	—	—
MC age	0.50	0.50	—	—
No insurance	0.10	0.30	—	—
Not need insurance	0.07	0.25	—	—
No medical help	0.17	0.37	—	—
N	1,369

Note: PCS = Physical Component Summary; MCS = Mental Component Summary; MSA = metropolitan statistical area; USC = U.S. Census; MC = Medicare; modal categories include white, female, high school, married, South, overweight

Table 2.

Descriptive Statistics by Race

	White mean	Black mean
Visits	8.39	6.78
PCS	47.44	44.53
Age	64.60	64.62
Male	0.46	0.49
Less than high school	0.22	0.42
Some college	0.18	0.13
College	0.14	0.07
Grad degree	0.11	0.07
Family size (log)	0.64	0.60
Income (log)	9.45	9.05
Single	0.04	0.11
Separated	0.01	0.06
Divorced	0.12	0.15
Widowed	0.11	0.18
Northeast	0.16	0.15
Midwest	0.24	0.21
West	0.22	0.14
Non-MSA	0.24	0.15
USC	0.90	0.87
MCS	51.14	49.43
Smoke	0.16	0.20
Underweight	0.02	0.00
Normal weight	0.29	0.25
Obese	0.29	0.36
MC age	0.50	0.47
No insurance	0.10	0.13
Not need ins	0.07	0.07
No medical help	0.17	0.12
N	1,169	200

I construct the complete two-equation, full-model by adding different sets of variables for Equation 1 and Equation 2. The base model begins with:

Model 1:

PCS = a + b (Visits) + d (Pop .) + n (Marital) + v (Region) + t (Health Behaviors) + e

Visits = a * + b * (PCS) + d * (Pop .) + n * (Marital) + e *

*In this model, only USC and MCS are added as Health Behaviors

The intermediate model adds smoking and obesity.

Model 2:

PCS = a + b (Visits) + d (Pop .) + n (Marital) + v (Region) + t (Health Behaviors) + e

Visits = a * + b * (PCS) + d * (Pop .) + n * (Marital) + e *

The full-model is presented as follows.

Model 3:

PCS = a + b (Visits) + d (Pop .) + n (Marital) + v (Region) + t (Health Behaviors) + e

Visits = a * + b * (PCS) + d * (Pop .) + n * (Marital) + t * (Health Attitudes) + e *

In the full-model, there exists a positive and significant association between health and visits (Equation 1, see Table 3). This result is expected because seeking health services results in treatment and medical awareness. More formally, the Equation 1 full-model rejects the null hypothesis stating utilization has a no association with health. However, the full impact of visits must also consider the square root of visits estimate. When both are calculated, one visit is associated with a decrease in PCS. This result may seem striking at first. More specifically, one would expect visits to increase PCS.³ This outcome does occur after 20 visits or the inflection point, where each additional visit increases PCS.⁴

Table 3.

Two Equation Models with Equation 1 as PCS, Equation 2 as Visits, and Single-Equation OLS Regression Results

	Base model 1		Intermediate model 2		Full model 3		Comparison 1	Comparison 2
	Equation 1	Equation 2	Equation 1	Equation 2	Equation 1	Equation 2	OLS	OLS
	PCS	Visits	PCS	Visits	PCS	Visits	PCS	Visits
	B (SE)	B (SE)	B (SE)	B (SE)	B (SE)	B (SE)	B (SE)	B (SE)
Visits	2.88		2.73		0.62*		0.04
	(3.32)		(2.87)		(0.26)		(0.03)
Sq. root visits	−22.60		−21.56		−5.71**		−1.34***
	(24.84)		(21.54)		(1.95)		(0.27)
PCS		−0.91***		−0.88***		−0.84***		−0.33***
		(0.06)		(0.06)		(0.06)		(0.04)
Race/age/gender
Black	−4.36	−2.79**	−4.23	−2.76**	−2.01**	−2.80**	−1.40**	−2.37**
	(3.72)	(1.12)	(3.36)	(1.11)	(0.71)	(1.10)	(0.59)	(1.03)
Age	0.09	−0.63*	0.06	−0.62*	−0.15*	−0.61*	−0.21**	−0.41
	(0.41)	(0.30)	(0.36)	(0.30)	(0.09)	(0.30)	(0.08)	(0.28)
Male	−3.25	−0.96	−3.24	−0.98	−1.05*	−0.95	−0.44	−1.26*
	(3.51)	(0.79)	(3.17)	(0.79)	(0.55)	(0.78)	(0.42)	(0.73)
Education/family/income
Less than high school	−1.65	−3.27**	−1.74	−3.16**	−1.50*	−2.91**	−1.43**	−0.64
	(1.49)	(1.05)	(1.44)	(1.05)	(0.61)	(1.04)	(0.55)	(0.96)
Some college	2.21	1.19	2.28	1.14	1.48*	1.00	1.26*	0.02
	(1.96)	(1.13)	(1.85)	(1.12)	(0.66)	(1.11)	(0.58)	(1.04)
College	5.51	3.35**	5.32	3.27*	2.95***	3.06*	2.30***	1.37
	(4.05)	(1.28)	(3.64)	(1.27)	(0.80)	(1.27)	(0.66)	(1.18)
Grad degree	5.22	4.28**	5.07	4.19**	2.93***	3.82**	2.34**	1.75
	(3.69)	(1.38)	(3.45)	(1.38)	(0.85)	(1.37)	(0.72)	(1.27)
Family size (log)	−1.86	−1.86*	−1.78	−1.89*	0.13	−1.75*	0.65	−2.25*
	(3.20)	(1.00)	(2.93)	(1.00)	(0.64)	(0.99)	(0.53)	(0.93)
Income (log)	0.66		0.63		0.25*		0.15*
	(0.64)		(0.56)		(0.11)		(0.09)
Marital status
Single	−4.97	−2.34	−4.97	−2.30	−1.52	−1.80	−0.57	−0.71
	(5.62)	(1.92)	(5.35)	(1.91)	(1.20)	(1.90)	(1.00)	(1.78)
Separated	−2.27	−4.84*	−2.24	−4.67	−4.12*	−4.15	−4.64**	−0.57
	(4.75)	(2.88)	(4.60)	(2.86)	(1.69)	(2.84)	(1.48)	(2.64)
Divorced	−1.22	−2.47*	−0.94	−2.45*	−0.28	−2.11	−0.10	−1.50
	(2.21)	(1.30)	(1.98)	(1.29)	(0.77)	(1.29)	(0.68)	(1.21)
Widowed	−4.44	0.61	−4.26	0.62	−0.56	1.06	0.46	1.38
	(5.85)	(1.33)	(5.34)	(1.32)	(0.90)	(1.32)	(0.70)	(1.24)
Region
Northeast	−0.00		−0.01		0.83		1.06*
	(1.99)		(1.90)		(0.67)		(0.59)
Midwest	1.94		1.80		0.83		0.57
	(2.12)		(1.88)		(0.60)		(0.52)
West	−2.40		−2.42		−0.21		0.40
	(3.47)		(3.31)		(0.67)		(0.54)
Non-MSA	−2.00	−1.54*	−1.96	−1.53*	−1.08*	−1.43	−0.83*	−1.23
	(1.84)	(0.92)	(1.73)	(0.92)	(0.56)	(0.91)	(0.48)	(0.85)
Health behaviors
USC	10.99		10.17		1.90		−0.38
	(13.49)		(11.36)		(1.26)		(0.68)
MCS	0.63***		0.63***		0.59***		0.58***
	(0.07)		(0.07)		(0.03)		(0.02)
Panel	−0.00	0.00	−0.00	0.00	0.00	0.00	0.00	0.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
Smoke			−1.30		−0.65		−0.47
			(1.67)		(0.62)		(0.55)
Underweight			4.14		−0.74		−2.09
			(8.39)		(2.32)		(2.00)
Normal weight			−1.99		−0.52		−0.11
			(2.37)		(0.58)		(0.49)
Obese			0.08		−1.33**		−1.72***
			(2.29)		(0.57)		(0.48)
Medicare eligibility		3.67**		3.63**		3.05**		1.96
		(1.54)		(1.54)		(1.54)		(1.44)
Health attitudes
No insurance						−2.98*		−3.75**
						(1.34)		(1.25)
Not need insurance						−0.30		−2.07
						(1.59)		(1.49)
No medical help						−0.85		−2.10**
						(1.05)		(0.98)
Constant	26.38**	91.55***	28.54**	89.80***	31.78***	88.37***	32.67***	52.81**
	(15.28)	(19.26)	(14.17)	(19.13)	(5.89)	(18.96)	(5.22)	(17.51)
Observations	1369		1369		1369		1369	1369
Adjusted R²							0.477	0.082

Note: PCS = Physical Component Summary; MCS = Mental Component Summary; MSA = metropolitan statistical area; USC = U.S. Census; MC = Medicare; OLS = ordinary least squares; modal categories include white, female, high school, married, South, overweight. First three columns show estimates for Equation 1 or PCS/health and Equation 2 or visits. Last two columns show estimates for single-equation OLS for PCS/health and single-equation OLS for utilization/visits.

p < .10. **p < .05. ***p < .01.

The other main independent variable corresponds to Black. In all models, Blacks are associated with lower PCS (or worse health) than Whites, but this is significant only in the full model. This difference becomes more evident when comparing two-equation modeling with single-equation OLS. As can be seen in Table 3, the estimate for Black is biased downward without accounting for endogeneity between health and utilization. Specifically, the full model approximates −2.01 and OLS estimates only −1.40. In summary, the Equation 1 full-model rejects the null hypothesis stating older Blacks and Whites do not differ in health.

In all three models and OLS, lower levels of education are large and highly significant on health or Equation 1. Specifically, individuals with less than high school, as compared to the reference group of high school, have nearly a 2-point penalty. Individuals with a college or graduate degree maintain a 3-point health advantage. A 4-point difference is associated for separated individuals in the full model and the single-equation OLS model, with respect to married. As expected, those with more income maintain better health. Obese individuals, as compared to the reference group of overweight, maintain a 1.33 lower PCS score in the full model.

As for estimates for provider visits (Equation 2), higher PCS is associated with fewer visits, where a 1-unit increase in PCS results in almost one fewer visit. This outcome is expected because individuals in better health are less likely to seek health services. The other main independent variable of interest corresponds to Black. All three models and OLS show Blacks are associated with fewer visits than Whites, ceteris paribus. However, the full model reveals a more significant and larger difference than the biased OLS model. Even after accounting for unhealthy behaviors, the full-model estimates Blacks as having nearly three fewer visits than Whites and the OLS model estimates slightly above two. More formally, the Equation 2 full-model rejects the null hypothesis stating older Blacks and Whites do not differ in utilization.

Once again, those with lower education levels have substantially fewer visits. Individuals with less than high school maintain three fewer visits and those with a graduate degree maintain three more visits, as compared to the high school reference group. As for the Medicare eligibility variable, a very important variable since it satisfies the identification condition, the estimate is significant in the full model and shows eligibility has a large and positive impact on visits. Specifically, individuals eligible for Medicare have three more visits. In contrast, those without insurance of any type have three fewer visits.

Discussion

In order to better understand later-life health, researchers must recognize differences in utilization account for some of the observed health disparity among Blacks and Whites. A stark illustration of this can be found in “an ounce of prevention is worth a pound of cure.” A special report by the Partnership for Prevention (2007) found

If the 42 percent of African Americans age 50 and older up-to-date with any recommended screening for colorectal cancer increased to 90 percent, 1,800 additional lives would be saved annually. This is a rate of 26 per 100,000 African Americans age 50 and older, substantially more than the corresponding rates of 17, 15, and 15 per 100,000 additional lives saved for Whites, Hispanics, and Asians, respectively.

(This study was sponsored by Centers for Disease Control and Prevention, Robert Wood Johnson Foundation, and WellPoint Foundation)

This same report highlights influenza immunization for Whites aged 50+ at 40.3% and only 26.2% for Blacks. For pneumococcal immunization aged 65+, Whites average 58.5% and Blacks average 38.9%. In short, utilization functions much more as a proximate determinant for health in later life.

Demand Side: Educate to Increase Utilization

Given this, what can policy analysts propose as feasible recommendations? One alternative relates to education. As all models show, those with less than high school education have markedly lower health scores. Consider the recent article on health literacy by Oates and Paasche-Orlow (2009). These scholars present an interesting point by stating, “Patients with the largest disease burdens are often those with the least ability to understand and use health information” (p. 1049). Citing Baker, Wolf, Feinglass, Thompson, and Gazmararian (2009), they offer evidence of higher hazard ratios for mortality for older adults with limited health literacy. Specifically, individuals lacking literacy maintain a 52% greater likelihood of mortality in a 5-year period.

This expected result can be traced to Omran (1971), Kitagawa and Hauser (1973), and, more recently, with Ross and Wu (1996) and Mirowsky and Ross (2003). Mirowsky and Ross claim education/schooling results in greater human capital, which in turn influences health behaviors. Recognizing this is noteworthy because education and learning do not stop as one ages—in other words, individuals can still learn and develop skills. Although individuals with lower education levels may have difficulty pursuing more schooling due to income constraints or mobility restrictions, technological advancements have virtually eliminated such limitations. As a case in point, consider the internet and online learning. With computers becoming more affordable and the internet reaching more communities, providing educational software to older adults is not cost-prohibitive or inefficient. Such software, in the form of instructional videos, point-and-click tutorials, and interactive games, can provide this group with abundant health information. Even though gains in human capital may be marginal, the multiplier effect due to increased knowledge can positively influence several health attributes. As an example, consider the work by Bennett et al. (Bennett, Chen, Soroui, & White, 2009) in the Annals of Family Medicine. Utilizing the 2003 National Assessment of Adult Literacy, these scholars find “health literacy significantly mediated both racial/ethnic disparities and education-related disparities in self-rated health status” among older adults (p. 204). Employing both the NHANES for the periods 1999-2000, 2001-2002, 2003-2004, 2005-2006 and NHIS for the period 1997-2008, a recent and comprehensive analysis by Martin et al. (Martin, Schoeni, & Andreski, 2010) suggests increased education is associated with a better health profile for older adults in terms of disability, obesity, and smoking.

As a final example, consider the work by Leung et al. (Leung, Chi, & Lui, 2006). These scholars examine learning preferences among older adults in the United States, the United Kingdom, Finland, and China. Although their survey does not yield a high response rate, the analysis does provide some insight into learning choices in later life. Specifically, older adults prefer courses related to hobbies/interests and ones that are knowledge driven (e.g., language, health). In terms of barriers to learning, most U.S. older adults indicated a lack of time. This work suggests older adults value education and may be interested in programs that only require a small time investment.

Supply Side: Enhanced Medical Training to Increase Utilization

Another cost-effective and practical solution focuses on supply. For example, training medical students and interns to reach out to minority groups during patient consultations can develop a relationship that enhances communication and builds trust. Work by Musa, Schulz, Harris, Silverman, and Thomas (2009) examines the importance of trust in medical relationships. Using the Medicine Enrollment File (MEF) for Allegheny County, Pennsylvania, these scholars randomly selected adults aged 65 years for telephone interviews. Their exploratory factor analysis shows Blacks maintain lower levels of trust with doctors; Blacks are also much more willing to rely on church or religious leaders for information.

Given a large body of research suggesting “trust between patients and physicians is an important way to increase preventive service use,” doctors must learn important relationship skills (see Musa et al., 2009, p. 1297). In addition, doctors should be taught about cultural and generational values. Reconsider the work by Oates and Paasche-Orlow (2009). These scholars offer numerous communication strategies for clinicians/physicians: (a) speak slowly and use plain language, (b) show pictures and videos, (c) help patients ask questions, and (d) stress action steps for patients (p. 1050). All of these topics can be introduced in any first- or second-year medical program and reinforced during professional retraining. As a case in point, consider the Division of Geriatric Medicine and the Center for Aging and Health at University of North Carolina at Chapel Hill. With the Geriatrics Practice and Teaching program and Healthy Aging Partnership initiative, medical professionals receive up-to-date training on various issues involving older adults. Another example relates to the Resource Center for Minority Aging Research (RCMAR) at the University of California at Los Angeles. The center conducts research on health disparities and also examines ways to improve research measures and methodology. What is notable about this center is the Dissemination and Translation core. This unit, with the aid of a central coordinating center, recognizes the critical roles of policy makers, consumers, and providers. As such, research findings are channeled to nonacademic markets.

As a final example, consider the training module provided by the College of Nursing at University of Iowa (Smith, 2006). This program, which was developed and evaluated from 1989 to 1994 and updated starting in 2003, instructs on effective communication with older adults. The objectives relate to recognizing different types of communication, explaining how attitudes and beliefs shape communication, and how disabilities can interfere with effective communication. The manual is rich with examples, case studies, and discussion questions, all of which can be freely used by individuals for personal development or continuing education. In fact, one of the project goals is to “facilitate the widest dissemination and use of the training modules” (p. 1).

Limitations

Although this research provides another explanation for health disparities among older adults and recognizes the endogenous relationship with utilization, I am unable to identify the underlying causes. That is, why do Blacks have fewer visits, particularly given significant t tests showing this group is in worse health and lower PCS? If anything, one would expect Blacks to have more visits, particularly given nearly 90% have insurance and less than 12% indicate medical help is unnecessary, as compared to more than 17% for White. Although MEPS serves as a rich data set, quantitative research suffers from limitations. For one, understanding psychological factors that drive an individual to seek or not seek health services is very difficult. Simply, most of this is not available in the data and in those few instances where such questions are asked, measurement error can be pronounced. This research compels a qualitative or mixed-methods approach to better understand health disparities. By conducting detailed interviews and ethnographic studies, researchers gain insight into human behavior at an interpersonal or micro level.

Gornick (2003) also highlights the importance of utilization in reducing health disparities for the older adults and cites historical discharge rates by race to claim Medicare has improved access to care for Blacks. However, Gornick emphasizes: (a) Medicare utilization does not ensure equal use of services, (b) SES impacts utilization and explains some but not all racial disparities, and (c) differences by race in need for care do not explain disparities (p. 164). Again, a qualitative approach is needed and it should be viewed as complement, as opposed to a substitute, in further developing this line of research.

Finally, this analysis only examines utilization with respect to office visits. Older adults also utilize specialty practices and emergency rooms. These types of visits are not captured and adding this dimension would provide further insight into racial disparities.

Conclusion

Utilization serves a critical function in determining later-life health, and this is partly due to endogeneity. In addition to this work, more research is needed to better understand the intricate relationships between health and utilization. Although finding a data set as rich as MEPS is difficult, a logical starting point may be to conduct cross-national health comparisons to determine the extent of health disparities by population subgroups using the Integrated Public Use Microdata Series (IPUMS). Given that Blacks maintain poorer health than Whites and Blacks can gain from utilization, finding ways to increase use of health services is a logical starting point. This focus may improve both the quality and quantity of life. As discussed at the beginning, this goal is also practical given the significant demographic changes due to aging, retirement, increased life expectancy, and spiraling health care costs. As Koster et al. (2008) and Adams et al. (2006) suggest, obesity is another important consideration, and this research confirms the adverse impact of obesity on health. Perhaps, measures that incentivize or punish healthy food choices may be good policy. A recent article by Vastag and Aizenman (2012) discusses New York Mayor Bloomberg’s ban on sodas. Although contentious, these are necessary discussions and compel us to act collectively to find ways to improve population health.

Finally, this research complements the goals of Healthy People 2020. Access to health services is one of 12 topic areas and eliminating health disparities is an important goal. By focusing on demand- and supply-side interventions, we can fulfill the vision of health equity and improved health for all groups.

Footnotes

Appendix

Acknowledgements

I would like to acknowledge the Public Policy and Sociology faculty and the Carolina Population Center at the University of North Carolina at Chapel Hill for providing the training necessary to undertake this research.

Declaration of Conflicting Interests

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

Funding

The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: A part of this work was funded by NIA Grant 2T32 AG000155-20.

Notes

References

Adams

K. F.

Schatzkin

Harris

T. B.

Kipnis

Mouw

Ballard-Barbash

. (2006). Overweight, obesity, and mortality in a large prospective cohort of persons 50 to 71 years old. New England Journal of Medicine, 335, 763-778.

Almeida

O. P.

Garrido

G. J.

Lautenschlager

N. T.

Hulse

G. K.

Jamrozik

Ficker

(2008). Smoking is associated with reduced cortical regional gray matter density in brain regions associated with incipient Alzheimer disease. American Journal of Geriatric Psychiatry, 16, 92-98.

Andersen

(1968). A behavioral model of families’ use of health services (Research Series No. 25). Chicago, IL: Center for Health Administration Studies, University of Chicago.

Baum

C. F.

Schaffer

M. E.

Stillman

(2007). Enhanced routines for instrumental variables/GMM estimation and testing (Boston College Department of Economics Working Paper No. 667). Retrieved from http://fmwww.bc.edu/ec-p/WP667.pdf

Bennett

I. M.

Chen

Soroui

J. S.

White

(2009). The contribution of health literacy to disparities in self-rated health status and preventive health behaviors in older adults. Annals of Family Medicine, 7(3), 204-211.

Blewett

L. A.

Johnson

P. J.

Lee

Scal

P. B.

(2008). When a usual source of care and usual provider matter: Adult prevention and screening services. Journal of General Internal Medicine, 23, 1354-1360.

Chiolero

Faeh

Paccaud

Cornuz

(2008). Consequences of smoking for body weight, body fat distribution, and insulin resistance. American Journal of Clinical Nutrition, 87, 801-809.

Doescher

M. P.

Saver

B. G.

Fiscell

Franks

(2004). Preventive care. Journal of General Internal Medicine, 19, 632-637.

Ferraro

K. F.

(1987). Double jeopardy to health for Black older adults? Journal of Gerontology, 42(5), 528-533.

10.

Ferraro

K. F.

Farmer

M. M.

(1996). Double jeopardy to health hypothesis for African Americans: Analysis and critique. Journal of Health & Social Behavior, 37(1), 27.

11.

Gandek

Ware

J. E.

Aaronson

N. K.

Apolone

Bjorner

J. B.

Brazier

J. E.

Sullivan

(1998). Cross-validation of item selection and scoring for the SF-12 Health Survey in nine countries: Results from the IQOLA Project. International Quality of Life Assessment. Journal of Clinical Epidemiology, 51(11), 1171-1178.

12.

Gornick

M. E.

(2008). A decade of research on disparities in Medicare utilization: Lessons for the health and health care of vulnerable men. American Journal of Public Health, 98(9), 162-168.

13.

Greene

(2003). Econometric analysis (5th ed.). Upper Saddle River, NJ: Prentice Hall.

14.

Howard

D. L.

Carson

A. P.

Holmes

D. N.

Kaufman

J. S.

(2009). Consistency of care and blood pressure control among elderly African Americans and whites with hypertension. Journal of the American Board of Family Medicine, 22(2), 307-315.

15.

Hubbard

R. E.

Searle

S. D.

Mitnitski

Rockwood

(2009). Effect of smoking on the accumulation of deficits, frailty and survival in older adults: A secondary analysis from the Canadian Study of Health and Aging. The Journal of Nutrition, Health & Aging, 13(5), 468-472.

16.

Jenkinson

Layte

Jenkinson

Lawrence

Petersen

Paice

Stradling

(1997). A shorter-form health survey: Can the SF-12 replicate results from the SF-36 in longitudinal studies? Journal of Public Health Medicine, 19(2), 179-186.

17.

Kawachi

Daniels

Robinson

(2005). Health disparities by race and class: Why both matter. Health Affairs, 24(2), 343-352.

18.

Kitagawa

E. M.

Hauser

P. M.

(1973). Differential mortality in the United States: A study of socioeconomic epidemiology. Cambridge, MA: Harvard University Press.

19.

Koster

Patel

K. V.

Visser

van Eijk

J. T. M.

Kanaya

A. M.

de Rekeneire

. (2008). Joint effects of adiposity and physical activity on incident mobility limitation in older adults. Journal of the American Geriatrics Society, 56, 636-643.

20.

Leung

Chi

Lui

Y. H.

(2006). A cross-cultural study in older adults’ learning experience. Asian Journal of Gerontology & Geriatrics, 1(2), 78-83.

21.

Lubeck

D. P.

Hubert

H. B.

(2005). Self-report was a viable method for obtaining health care utilization data in community-dwelling seniors. Journal of Clinical Epidemiology, 58(3), 286-290.

22.

Marmot

(2007). Achieving health equity: From root causes to fair outcomes. The Lancet, 370(9593), 1153-1163.

23.

Martin

L. G.

Schoeni

R. F.

Andreski

P. M.

(2010). Trends in health of older adults in the United States: Past, present, future. Demography, 47, 17-40.

24.

Mirowsky

Ross

C. E.

(2003). Education, social status and health. New York, NY: Aldine de Gruyter.

25.

Murphy

S. L.

Kochanek

(2012). Deaths: Preliminary data for 2010 (National Vital Statistics Report, Vol. 60, No. 4). Atlanta, GA: Centers for Disease Control and Prevention.

26.

Musa

Schulz

Harris

Silverman

Thomas

S. B.

(2009). Trust in the health care system and the use of preventive health services by older Black and White adults. American Journal of Public Health, 99(7), 1293.

27.

Mutchler

J. E.

Burr

J. R.

(1991). Racial differences in health and health care service utilization in later life: The effect of socioeconomic status. Journal of Health and Social Behavior, 32(4), 342-356.

28.

Nie

J. X.

Wang

Tracy

C. S.

Moineddin

Upshur

R. E.

(2008). Health care service utilization among the elderly: Findings from the Study to Understand the Chronic Condition Experience of the Elderly and the Disabled (SUCCEED project). Journal of Evaluation in Clinical Practice, 14(6), 1044-1049.

29.

Oates

D. J.

Paasche-Orlow

M. K.

(2009). Health literacy: Communication strategies to improve patient comprehension of cardiovascular health. Circulation, 119(7), 1049-1051.

30.

Omran

A. R.

(1971). The epidemiologic transition: A theory of the epidemiology of population change. Milbank Memorial Fund Quarterly, 49, 509-538.

31.

Partnership for Prevention. (2007). Preventive care: A national profile on use, disparities, and health benefits. Washington, DC: Author.

32.

Peruzza

Sergei

Vianello

Pisent

Tiozzo

Manzan

Enzi

(2003). Chronic obstructive pulmonary disease (COPD) in elderly subjects: Impact on functional status and quality of life. Respiratory Medicine, 97, 612-617.

33.

Ridker

P. M.

(2003). C-reactive protein: A simple test to help predict risk of heart attack and stroke. Circulation, 108, 81-85.

34.

Ross

C. E.

(1996). Education, age, and the cumulative advantage in health. Journal of Health and Social Behavior, 37, 104-120.

35.

Schreiber

(2008). The Hausman test statistic can be negative even asymptotically. Journal of Economics and Statistics, 228(4), 394-405.

36.

Smith

(2006). Getting the facts: Communicating with the elderly (The Geriatric Mental Health Training Series). Iowa City: the Hartford Center of Geriatric Nursing Excellence, College of Nursing, University of Iowa.

37.

StataCorp. (2007). Stata Statistical Software: Release 10. College Station, TX: Author.

38.

Vastag

Aizenman

N. C.

(2012, June 2). New tactic in war on obesity: Attack portion size. The Washington Post. Retrieved from http://www.washingtonpost.com/national/health-science/new-tactic-in-war-on-obesity-attack-portion-size/2012/06/02/gJQA4wzi9U_story.html

39.

Ware

J. E.

Kosinski

Keller

S. D.

(1996). A 12-item short-form health survey: Construction of scales and preliminary tests of reliability and validity. Medical Care, 34(3), 220.

40.

Warner

D. F.

Hayward

M. D.

(2006). Early-life origins of the race gap in men’s mortality. Journal of Health and Social Behavior, 47, 3.

41.

World Health Organization. (1995). Physical status: The use and interpretation of anthropometry (Technical Report Series, #854, p. 9). Geneva, Switzerland: Author.

42.

Wolinsky

F. D.

Mosely

R. R.

Coe

R. M.

(1986). A cohort analysis of the use of health services by elderly Americans. Journal of Health and Social Behavior, 27(3), 209-219.

43.

Zuvekas

S. H.

Olin

G. L.

(2009). Validating household reports of health care use in the medical expenditure panel survey. Health Services Research, 44, 1679-1700.