Understanding Racial/Ethnic Disparities in Physical Performance in Midlife Women: Findings From SWAN (Study of Women’s Health Across the Nation)

Abstract

Objectives

Evaluate degree to which racial/ethnic differences in physical performance are mediated by sociodemographic, health, behavioral, and psychosocial factors.

Methods

Physical performance was evaluated using a decile score derived from grip strength, timed 4 m walk, and timed repeat chair stand in 1,855 African American, Caucasian, Chinese, Hispanic, and Japanese women, mean age = 61.8 (SD = 2.7) in the Study of Women’s Health Across the Nation. Mediators included education, financial strain, comorbidities, pain, body mass index (BMI), physical activity, and perceived stress. Structural equation models provided estimates of the total difference in physical performance between Caucasians and each race/ethnic groups and differences due to direct effects of race/ethnicity and indirect effects through mediators.

Results

The mean decile score for Caucasian women was 16.9 (SD = 5.6), 1.8, 2.6, and 2.1 points higher than the model-estimated scores in African Americans, Hispanics and Chinese, respectively, and 1.3 points lower than the Japanese. Differences between Caucasians and the Chinese and Japanese were direct effects of race/ethnicity whereas in African Americans and Hispanics 75% or more of that disparity was through mediators, particularly education, financial strain, BMI, physical activity, and pain.

Discussion

Addressing issues of poverty, racial inequality, pain, and obesity could reduce some racial/ethnic disparity in functional limitations as women age.

Between 1982 and 2004, the prevalence of self-reported disabilities in activities of daily living (ADLs) and instrumental activities of daily living (IADLs) in the U.S. population declined from 26.5% to 19.0% (Manton, 2008). However, this encouraging trend was not distributed equally throughout the population. The greatest decline was in those aged 85 years and older, whereas the prevalence of disability among those 65–84 years old remained relatively constant (Freedman et al., 2013). A recent analysis of trends between 1998 and 2012 found an increase in the prevalence of limitations in IADL among 55–64 year olds (Choi, Schoeni, & Martin, 2016).

Moreover, racial/ethnic disparities in activity limitations still appear to exist, with African Americans and Hispanics experiencing more disability in later life than Caucasians or Asians (Freedman, Martin, Schoeni, & Cornman, 2008; Schoeni, Martin, Andreski, & Freedman, 2005). This is an issue of considerable public health concern given the increasing diversity of the aging population, and recent increases in life expectancy in African Americans (Cunningham et al., 2017). Various factors may contribute to these disparities, including differences in neighborhood environment (e.g., safety, walkability), socioeconomic status, education, family structure, number and type of comorbidities, smoking, body size, and other lifestyle behaviors (Barnes et al., 2011; Choi et al., 2016; Dong, Chang, & Simon, 2014; Haas & Rohlfsen, 2010; Kirkness & Ren, 2015; Louie & Ward, 2011; Quiben & Hazuda, 2015). However, in most analyses (Barnes et al., 2011; Clay et al., 2015; Fuller-Thomson, Nuru-Jeter, Minkler, & Guralnik, 2009; Haas & Rohlfsen, 2010; Kirkness & Ren, 2015; Mendes de Leon, Barnes, Bienias, Skarupski, & Evans, 2005), although not all (Louie & Ward, 2011; Quiben & Hazuda, 2015; Thorpe et al., 2008; Thorpe et al., 2014), the racial/ethnic disparity in disability, particularly between African Americans and Caucasians, persists, even after accounting for confounding by these factors. As race/ethnicity is at least as much of a social construct as a biological reality, the inability to “explain” observed differences in physical function and disability entirely in terms of social and behavioral factors may reflect methodological limitations that fail to capture the multidimensional context in which individuals and groups function. Few studies have considered multiple factors simultaneously with a formal mediation analysis that considers how social-, behavioral-, and disease-related factors interrelate within racial/ethnic groups to affect functional limitation and disability.

In addition, most studies of disparities in physical performance have been conducted in older adults (≥65), but disability in ADLs and IADLs is a distal stage in the disablement process that may begin decades earlier with functional limitations resulting from disease, injury, and disuse (Jette & Branch, 1985; Verbrugge & Jette, 1994). Many physiological, behavioral, and social factors that differ by race/ethnicity (Bowen, 2009) influence the disablement process and are present in midlife, and even earlier, and functional limitations often first appear during midlife. In the Health and Retirement Study, for instance, the prevalence of at least one physical function limitation in those 55–69 years old ranged between 55.7% and 59.9% during the period 1998–2012 (Choi et al., 2016). In the Study of Women’s Health Across the Nation (SWAN), 29.6% of the cohort, aged 46–56 years (fourth follow-up examination, 2000–2002), reported moderate limitations and 11.0% reported severe limitations (Pope, Sowers, Welch, & Albrecht, 2001), and over the following 10 years, 26% reported a substantial limitation at least once, 42% reported having some limitation, and only 33% reported never having a limitation (Ylitalo et al., 2013).

Finally, the U.S. Asian population, which grew by 72% between 2000 and 2015, has the fastest growth rate of any major racial/ethnic group (https://www.pewresearch.org/fact-tank/2017/09/08/key-facts-about-asian-americans/ (last accessed August 22, 2019)), but little knowledge exists regarding the prevalence of functional limitations in these groups or whether they experience disparities in functioning relative to Caucasian Americans. Although life expectancy among all Asian Americans exceeds that of Caucasian-Americans by nearly 8 years (Acciai, Noah, & Firebaugh, 2015), this advantage is not distributed equally among the Asian population subgroups (Mui, Bowie, Juon, & Thorpe, 2017), or by immigration status (Hastings et al., 2016), underscoring the importance of disaggregating these groups. A comparison of physical performance among native Japanese women, Japanese American women in Hawaii and Caucasian women in the United States found a distinct advantage in both groups of Japanese relative to the Caucasians (Aoyagi et al., 2001), whereas a study of physical function in elderly Chinese Americans living in Chicago found poorer performance compared to Caucasians (Dong et al., 2014).

SWAN, an observational, community-based, prospective study of the menopausal transition with a multiracial/ethnic cohort (African Americans, Caucasians, Chinese, Hispanics, and Japanese) of midlife women, provides a unique opportunity to address these gaps in understanding about racial/ethnic disparities in functional limitations at the onset of the aging process. This study aimed to use formal mediation analysis to provide insight into the degree to which racial/ethnic disparities in physical performance in midlife women are simultaneously accounted for by disparities in education, financial strain, health status, overweight/obesity, physical activity, and perceived stress.

Methods

Study Sample

The SWAN cohort, which consists of 3,302 initially pre- and early-perimenopausal women of diverse race/ethnicities, was formed in 1996–1997 to study the biological, psychological, and social and cultural antecedents and consequences of the menopausal transition. Details of the sampling and recruitment strategies have been previously described (Sowers et al., 2000). Briefly, women, aged 42–52 years, were recruited from defined sampling frames in seven geographic sites across the United States: Boston; Chicago; Detroit-area; Los Angeles; Newark, NJ; Oakland, CA; and Pittsburgh. To be eligible, women had to report having had a menstrual period and no use of exogenous hormones in the 3 months prior to recruitment, not be pregnant or lactating, and to identify their primary race/ethnicity as African American (Boston, Chicago, Detroit, and Pittsburgh), Caucasians (all sites), Chinese (Oakland), Hispanic (Newark), or Japanese (Los Angeles). The cohort participated in a baseline clinical examination and continues to be seen for annual or biannual exams. Retention at the Year 13 follow-up exam in 2012 (FU13) was 77%. All participants provided written informed consent, and all protocols were approved by the institutional review boards of the participating institutions.

The sample for the current analysis included 1,945 women in SWAN who attended FU13 and completed at least one measure of physical performance. Of these, 1,855 (26.6% African Americans, 48.0% Caucasian, 10.4% Chinese, 5% Hispanic, and 10.0% Japanese) completed all three performance measures included in the primary outcome measure (physical performance decile score—see later) and constitute the sample for the main analysis.

Physical Performance Decile Score

During the FU13 exam, SWAN participants completed a series of physical performance tasks conducted by trained, certified staff following a standardized protocol. The tests included a timed 4 m walk, a timed repeated chair stand, and grip strength. The timed 4 m walk and timed repeated chair stand are two elements of the Short Physical Performance Battery (SPPB), a well-established reliable and valid measure of lower extremity function necessary for daily activities (Guralnik et al., 1994); participants also performed the third element of the SPPB, a series of balance tests, but this was not used in the construction of the decile score because it does not yield a continuous measure. For the timed walk, participants were instructed to walk the specified distance at their usual pace. The walk was repeated a second time and the fastest time was used in the decile score. The repeated chair stand was defined by the time in seconds a participant took to rise five times from a chair with her arms folded across her chest. Grip strength in the dominant hand was measured with a Baseline hydraulic dynamometer adjusted for hand size with the participant seated with her arm bent at a 90° angle at the elbow. The maximum (kilogram) of three attempts was used in the decile score.

Using the procedure developed by Michael and colleagues (2011), a summary decile score was created by ranking participants into sample-specific deciles (1–10) for each of the components and then summing over all three components to produce scores with a range of 3–30. For example, a participant in the bottom decile of the cohort for all three components would have a decile score of 3 whereas a participant in the top decile for the three tasks would have a decile score of 30.

Definition of Race/Ethnicity

Race/ethnicity (African American, Caucasian, Chinese, Hispanic, and Japanese) is the primary exposure of interest in this analysis. Because SWAN was specifically designed to recruit and study a diverse cohort, race/ethnicity was defined during a screening interview prior to the baseline examination from participants’ response to the question, “How would you describe your primary racial or ethnic group?" The response categories included the following: black/African American; Puerto Rican/Dominican/Central American/Cuban or Cuban American/South American/Spanish or other Hispanic (all categorized as Hispanic), Chinese/Chinese American; Japanese/Japanese American, and Caucasian/white Non-Hispanic. Respondents who indicated they were Mexican/Mexican American, Mixed, or Other were not included in the SWAN cohort.

Assessment of Mediators

Factors broadly considered to be social determinants of health in the sense of being shaped by the social context in which individuals exist, and known to vary by race/ethnicity and influence later-life physical functioning were considered as potential mediators of racial/ethnic differences in physical performance; they included body mass index (BMI), bodily pain, comorbidities, perceived stress, education, financial strain, and physical activity. BMI during midlife was calculated from measured height and weight and defined as the area under the curve (AUC) cumulative estimate over all SWAN visits (baseline through FU13) with available data. Two other time varying, continuous mediators, perceived stress and physical activity, were also defined by the AUC cumulative estimate over all visits with available data. Perceived stress was assessed at all SWAN visits using the 4-item Cohen’s Perceived Stress Score (Cohen, Kamarck, & Mermelstein, 1983), and physical activity, measured at baseline, FU 1, 3, 5, 6, 9, 12, and 13, was defined as the total activity score from the Kaiser Physical Activity Survey (KPAS) (Sternfeld, Ainsworth, & Quesenberry, 1999). The KPAS total activity score assesses self-reported recreational activity, household and caregiving activity, and daily routine activity during the previous year, and has established reliability and validity comparable to other self-report instruments (Ainsworth, Sternfeld, Richardson, & Jackson, 2000). Bodily pain was assessed at FU13 only and was expressed as the sum of pain ratings from 0 to 10 for eight body locations (McCarthy, Bigal, Katz, Derby, & Lipton, 2009). Comorbidities were defined as the number of conditions (angina, anemia, high blood pressure, migraine, osteoarthritis, osteoporosis, thyroid problem, diabetes, cancer, heart attack, and stroke) ever reported since baseline. Depressive symptoms were also counted as a comorbidity based on having a score of 16 or above on the Center for Epidemiologic Studies Depression scale (Radloff, 1977) at any SWAN exam. Years of education were self-reported at baseline, and financial strain during midlife was defined as the proportion of SWAN visits at which respondents reported that it was very or somewhat hard to pay for basic necessities.

Covariates

Covariates that might confound any of the relations between race/ethnicity, mediators, and outcome included the following: clinical site, age, marital status, smoking status, and menopause status. Age was calculated from date of birth reported at baseline and the date of the FU13 examination. Marital status (married/living as married, single/separated/widowed/divorced), smoking status (never/former, current), and menopause status (natural post-menopause, surgical post-menopause, undetermined/hormone user/peri-menopause) were self-reported at FU13. The very small numbers of perimenopausal women (n = 14) precluded consideration of them as a separate category.

Data Analysis

Chi-square tests and analysis of variance (ANOVA) were used to compare women included in the analysis with those excluded as well as to describe racial/ethnic differences in FU13 categorical characteristics and physical performance decile score with Caucasian women as the reference group. A Bonferroni procedure was applied to correct for multiple comparisons. Except for age, which was described by means and standard deviations, all of the continuous variables (years of education, cumulative financial strain, BMI, number of comorbidities, pain index, physical activity, and perceived stress) were skewed and reported as medians and interquartile ranges. To test for significant differences by racial/ethnic group, ANOVA was used for age and nonparametric tests were used for years of education, cumulative financial strain, BMI, number of comorbidities, pain index, physical activity, and perceived stress.

For the mediation analysis, structural equation models (MPlus, version 8) provided estimates of the total, direct, and indirect effects of race/ethnicity through each mediator simultaneously (Hayes & Preacher, 2014) (Zahodne, Manly, Smith, Seeman, & Lachman, 2017); confidence intervals around these estimates were calculated from 1,000 bootstrap samples. As shown in Figure 1, race/ethnicity is the exposure variable with each group compared to Caucasians; using the language of Zahodne and colleagues (2017), the model estimates the direct effect of race/ethnicity on the physical performance decile score (dotted lines), as well as the indirect effects of race/ethnicity on each of the mediators (solid lines) and each of the mediators on the decile score (dashed lines), with the total effect estimated by the sum of the direct and indirect effects. The full mediation model includes adjustment for site in the race/ethnicity-mediator paths and adjustment for site, age, marital status, smoking, and menopause status in the race/ethnicity-decile score and mediator-decile score paths.

Figure 1.

Model of mediation analysis: Direct effects are shown as the dotted arrows going from the racial/ethnic groups (exposure variables) to the physical performance decile score (outcome); indirect effects are shown as the solid arrows going from the exposure variables to each of the mediators and the dashed arrows going from each mediator to the decile score; total effects are the sum of the indirect and direct paths.

Sensitivity analyses were conducted by examining each component of the decile score separately as an outcome, adjusted for outcome-specific covariates, such as floor surface and foot covering for the walk and chair stand, and dynamometer size setting and use of nondominant hand for grip strength. These analyses included, first, the entire SWAN sample with that specific physical performance outcome measure (n = 1,933 for grip strength, n = 1,905 for 4 m walk, and n = 1,864 for chair stand) and then were restricted to only those with the decile score (N = 1,855).

Model fit for all structural equation models was evaluated using the comparative fit index (> .95), the root mean square error of approximation (< .06), and the standardized root mean square residual (< .05).

Results

Compared to excluded SWAN participants, the women in the analytic sample were less likely to be African American (26.6% vs 30.5%) or Hispanic (5.0% vs 13.4%) and more likely to be Chinese (10% v 3.9%) or Japanese (10.0% vs 6.6%) (all ps < .0001). They were also less likely to experience financial strain (33.9% vs 47.8%), or to be obese (30.7% vs 39.4%) or current smokers (12.8% vs 23.3%). Conversely, they were more likely to have a college degree (48.2% vs 35.9%), to be in excellent or very good overall health (62.6% vs 51.3%), and to be physically active (51.5% vs 41.6%) (p < .0001 for all comparisons).

Among those in the analysis, there were significant differences between the Caucasian women and all or some of the other racial/ethnic groups in all of the mediators of interest (Table 1). Caucasian women had significantly more years of education than all of the other race/ethnic groups, and the African American and Hispanic women had a significantly greater proportion of visits with financial strain compared to the Caucasians. They also had a higher BMI than the Caucasians, reported more comorbidities, and had a higher pain index, whereas the Chinese and Japanese had lower BMI, fewer comorbidities, and a lower pain index. Except for the African Americans, the other race/ethnic groups had a significantly higher level of perceived stress than the Caucasians, and African American, Hispanic, and Chinese women had significantly lower physical activity scores than Caucasian women. Finally, African American and Hispanic women were less likely, and Japanese women more likely, than Caucasian women to be married or living as married; the Chinese women were substantially less likely to be current smokers whereas the African American women were more likely; and with the exception of a higher proportion of surgical menopause among the African American women, there were no differences in menopausal status.

Table 1.

Year 13 Characteristics of SWAN Participants With All Three Components of Physical Performance Decile Score^a (N = 1,855)

	Caucasian	African American	Hispanic	Chinese	Japanese
	(n = 891)	(n = 493)	(n = 92)	(n = 193)	(n = 186)
Age (yr), mean (SD)	61.9 (2.7)	61.6 (2.7)	62.1 (2.9)	62.2 (2.5)	62.3 (2.7)
Education (yr), median (IQR)	16 (15–18)	15 (14–16)^^^	12 (8–15)^^^	16 (12–16)^^	16 (15–16)**
Cumulative financial strainb, median (IQR)	0 (0–0.3)	0.3 (0–0.8)^^^	0.9 (0.5–1.0)^^^	0 (0–0.1)	0 (0–0.4)
BMIc (kg/m2), median (IQR)	27.2 (23.9–31.8)	32.0 (27.2–37.4)^^^	29.6 (26.2–32.8)**	23.0 (21.1–25.2)^^^	22.9 (21.2–26.1)^^^
Number of comorbiditiesd, median (IQR)	3 (2–4)	4 (3–5)^^^	4 (3–5)^^^	2 (1–3)^^^	2 (2–3)***
Pain indexc, median (IQR)	7.0 (3.8–12.4)	9.0 (2.9–15.9)^^^	16.5 (9.8–27)^^^	5.6 (2.2–10.4)	6.6 (3.2–12.4)
Perceived Stress Scalec, median (IQR)	6.9 (5.6–8.6)	7.3 (5.9–8.9)	8.5 (7.4–9.6)^^^	7.7 (6.4–9.1)^^	7.8 (6.4–9.4)^^^
Physical activityc, median (IQR)	8 (6.9–9.0)	7.2 (6.3–8.3) ^^^	6.7 (5.8–7.5) ^^^	7.3 (6.4–8.1) ^^^	7.7 (6.7–8.5)
Marital status, n (%)		^^^	^^^		**
Single/separated/widowed/divorced	299 (33.6)	297 (60.2)	57 (62.0)	48 (24.9)	39 (21.0)
Currently married or living as married	592 (66.4)	196 (39.8)	35 (38.0)	145 (75.1)	147 (79.0)
Current smoker, n (%)	45 (5.1)	73 (15.1)^^^	9 (9.8)	2 (1.0)*	8 (4.3)
Menopause status, n (%)		**
Natural post	813 (91.2)	424 (86)	84 (91.3)	183 (94.8)	177 (95.2)
Surgical post	65 (7.3)	66 (13.4)	8 (8.7)	9 (4.7)	8 (4.3)
Unknown, hormone user (or peri-menopause)	13 (1.5)	3 (0.6)	0 (0)	1 (0.5)	1 (0.5)

Note: Dunnet’s test used to correct for multiple comparisons for normally distributed continuous variables, Dwass-Steel-Critchlow-Fligner correction for skewed continuous variables, Bonferroni adjusted for four tests for categorical variables. BMI = body mass index; CES-D = Center for Epidemiology Studies Depression scale; IQR = interquartile range; SWAN = Study of Women’s Health Across the Nation.

^aSum of sample-specific percentile rankings on grip strength, 4 m walk and repeated chair stand.

^bProportion of SWAN visits women reported difficulty paying for very basics.

^cSummarized using area under the curve methods for all SWAN visits with data.

^dNumber of comorbidities (angina, anemia, high blood pressure, migraine, osteoarthritis, osteoporosis, thyroid, depression [score of 16 or more on CES-D], diabetes, cancer, heart attack, and stroke) reported since baseline.

p values for significant differences compared to whites: *≤.05. ≤.01. **≤.005. ^^≤.001. ***≤.0005. ^^^≤.0001.

As shown in Figure 2, the race/ethnic groups differed substantially in the unadjusted mean physical performance decile score. The Japanese women had the highest score (mean = 20.6, SD = 4.2) and performed significantly better than the Caucasian women whose mean score was 16.9 (SD = 5.6, p < .0001). Compared to Caucasians, the mean scores for the African American and Hispanic women (mean = 14.2, SD = 5.3 and mean = 8.9, SD = 4.3, respectively) were significantly lower (p < .0001 for both), whereas the mean score for the Chinese (mean = 16.4, SD = 5.7) was comparable (p = .29).

Figure 2.

Unadjusted physical performance decile score (mean, SD) by race/ethnicity. Mean scores of African Americans, Hispanics, and Japanese significantly different from Caucasians (p < .0001 for all).

Table 2 summarizes the results of the mediation analysis, showing, first, the estimated total difference in the physical performance decile score for each of the minority race/ethnic groups relative to Caucasians. Also shown are the estimates of the differences due to direct effects of race/ethnicity on the decile score, the differences due to the total indirect effects of race/ethnicity on the decile score through all of the mediators, and estimated differences in the decile score through each of the individual mediator variables.

Table 2.

Estimated Direct Effect of Race/Ethnicity and Indirect Effects of Mediators on Differences in Physical Performance Decile Score^a, With Whites as Referent^b

		Estimated difference	95% CI
African American	Total	−1.809	−2.398, −1.278
	Direct	−0.434	−1.031, 0.058
	Total indirect	−1.375	−1.621, −1.092
	Proportion difficulty paying for basics	−0.159	−0.296, −0.033
	BMI	−0.417	−0.585, −0.293
	Stress	−0.063	−0.140, −0.018
	Physical activity	−0.389	−0.530, −0.268
	Comorbidity	−0.101	−0.196, −0.024
	Pain index	−0.131	−0.220, −0.065
	Education	−0.113	−0.194, −0.054
Hispanics	Total	−2.599	−4.017, −1.207
	Direct	0.123	−1.144, 1.531
	Total indirect	−2.722	−3.472, −1.957
	Proportion difficulty paying for basics	−0.318	−0.617, −0.070
	BMI	−0.324	−0.620, −0.088
	Physical activity	−0.705	−1.040, −0.440
	Stress	0.056	−0.002, 0.197
	Comorbidity	−0.106	−0.278, −0.008
	Pain index	−0.771	−1.149, −0.485
	Education	−0.555	−0.939, −0.275
Chinese	Total	−2.142	−3.107, −1.115
	Direct	−2.172	−3.124, −1.247
	Total indirect	0.031	−0.390, 0.450
	Proportion difficulty paying for basics	0.014	−0.015, 0.072
	BMI	0.541	0.370, 0.757
	Physical activity	−0.343	−0.516, −0.164
	Stress	−0.069	−0.185, −0.013
	Comorbidity	0.033	−0.009, 0.129
	Pain index	0.103	0.008, 0.210
	Education	−0.248	−0.431, −0.130
Japanese	Total	1.340	0.396, 2.259
	Direct	1.309	0.442, 2.167
	Total indirect	0.031	−0.376, 0.376
	Proportion difficulty paying for basics	−0.078	−0.177, −0.020
	BMI	0.355	0.217, 0.539
	Physical activity	−0.122	−0.307, 0.049
	Stress	−0.106	−0.231, −0.024
	Comorbidity	0.062	0.008, 0.150
	Pain index	0.014	−0.080, 0.110
	Education	−0.093	−0.182, −0.035

Note: comparative fit index = 0.944; root mean square error of approximation = 0.06; standardized root mean square residual = 0.03. BMI = body mass index; CI = confidence interval.

^aSum of sample-specific percentile rankings on grip strength, 4 m walk and repeated chair stand.

^bEstimates from multiple mediation analysis; bold numbers are statistically significant differences compared to whites.

Compared with Caucasian women, African American women had a decile score that was 1.8 points lower. The indirect effect through all the mediators accounted for a reduced decile score of about 1.4 points, or approximately 76% of the total effect. The remaining direct effect of race/ethnicity on the difference in the decile score was not statistically significant. All the mediators, particularly BMI and physical activity (30% and 28.3%, respectively of the total indirect effect) significantly contributed to the indirect effect.

In Hispanic women, the total effect of race/ethnicity was a decrement of 2.6 points in physical performance compared with Caucasian women, but the direct effect of Hispanic race/ethnicity was only 4.7% of the total effect and was not statistically significant, meaning that essentially all of the disparity (95%) came indirectly through the mediators. More bodily pain and less physical activity together accounted for 55.4% of the total indirect effect; less education, higher BMI, more financial strain, and more comorbidity also significantly contributed to the total indirect effect.

For the Chinese women, whose unadjusted physical performance decile score was comparable to that of the Caucasian women, the estimate of the total difference in physical performance from the mediation analysis was 2.1 points below the Caucasians. This disparity was due entirely to a direct effect of race/ethnicity that was slightly larger than the total effect. Counteracting the negative direct effect of race/ethnicity was a positive indirect effect through lower BMI, which conferred a significant 0.5 point advantage, but which, when combined with the significant negative indirect effects of less physical activity and less education, resulted in an insignificant total indirect effect.

In contrast, the estimate of the total difference in physical performance decile score in the Japanese women was 1.3 points higher than the Caucasian women, but, as with the Chinese, essentially all of that total effect (98%) came from the direct effect of race/ethnicity. Significant positive indirect effects of race/ethnicity came through lower BMI and fewer comorbidities, and significant negative indirect effects came from greater stress, lower education, and more financial strain. Combined, the mediators had only a small and insignificant total indirect effect.

All sensitivity analyses yielded findings consistent with the main analyses described earlier (data not shown).

Discussion

This study of racial/ethnic differences in physical performance confirmed that disparities in physical function experienced by older African American and Hispanic women (Fuller-Thomson et al., 2009; Haas & Rohlfsen, 2010; Louie & Ward, 2011; Mendes de Leon et al., 2005) are apparent in midlife, and that midlife Chinese women also have a disparity in physical performance, whereas Japanese women do not. Specifically, this study found that African American women performed less well than Caucasian women, but this disparity was accounted for only indirectly by the effect of race/ethnicity on other factors. Similarly, the disparity in physical performance in the Hispanic women was almost entirely mediated by other factors, most notably greater bodily pain and less physical activity, as well as lower education, higher BMI and more financial strain. In contrast, the disparity in physical performance between the Chinese women and the Caucasian women was almost entirely due to the direct effect of race/ethnicity and would have been even greater if not for the positive indirect effect through lower BMI. Finally, most of the difference in physical performance between the Japanese women and the Caucasian women was due to race/ethnicity, but this difference favored the Japanese and, when combined with an indirect, positive effect through lower BMI, resulted in a higher physical performance score than the Caucasians.

A unique finding of this study is the advantage in physical performance observed in the Japanese women relative to the Caucasian women but the disparity in performance among the Chinese women which, in both cases, were attributable to the direct effect of race/ethnicity itself. The reasons for this difference between the two Asian subgroups are not entirely clear, but are consistent with two independent studies, one of which found better physical performance in Japanese American women living in Hawaii relative to U.S. Caucasian women (Aoyagi et al., 2001), and the other of which found poorer physical performance in older Chinese adults living in Chicago relative to similarly aged Caucasian adults also living in Chicago (Dong et al., 2014). Differences in the communities from which these two study samples were drawn may have contributed to the relative advantage of the Japanese and the relative disadvantage of the Chinese. Similarly, in this study, differences in the communities of Japanese women living in Los Angeles and Chinese women living in Oakland may account for some of the observed differences in physical performance. However, in an analysis that used data from the American Community Survey that found that differences in socioeconomic status, immigration history, and citizenship status accounted for a reduced risk of functional limitations and limitations in IADLs in the Japanese population, relative to the Chinese population (Fuller-Thomson, Brennenstuhl, & Hurd, 2011), adjustment for those factors showed the Japanese actually to have an increased risk of disability. In this study, for both the Chinese and the Japanese women, the mediating factors together accounted for only a small and insignificant amount of the differences in physical performance relative to Caucasian women. This is an area that deserves further study and emphasizes the importance of disaggregating Asian population subgroups in health disparities research.

The lack of significance of race/ethnicity itself as a direct effect on the disparity in physical performance between either the African American or the Hispanic and Caucasian women in SWAN is more generally consistent with previous health disparities research in Hispanics than in African Americans. For instance, in an analysis of health disparities in mortality in the Multi-Ethnic Study of Atherosclerosis, 30% of the higher mortality rate among African Americans was mediated by occupational complexity, leaving a substantial effect attributable to race (Fujishiro et al., 2017). In terms of disparities in physical function specifically, a comparison of walking speed in African American and Caucasian women with osteoarthritis found that, even with adjustment for demographic and disease factors, African Americans still had a significantly increased risk of a walking speed of less than 1.0 m/s (Kirkness & Ren, 2015). In contrast, disparities in walking speed between Mexican Americans and European Americans in the Sacramento Area Latino Study of Aging (SALSA) study were entirely explained by demographic, lifestyle, disease, and impairment factors (Quiben & Hazuda, 2015). On the other hand, an analysis from the Exploring Health Disparities in Integrated Communities Study found that, in this cohort where African Americans and Caucasians lived in the same environment and shared socioeconomic and behavioral characteristics, African Americans actually reported fewer IADL limitations than Caucasians (Thorpe et al., 2014). Overall, these findings underscore the relevance of social determinants of health, which estimates from the Centers for Disease Control and Prevention (CDC) suggest account for well over half of population health (https://www.cdc.gov/nchhstp/socialdeterminants/faq.html (last accessed August 22, 2019)).

The implication of the current study findings is that, although racial/ethnic disparities in physical performance may exist, they are unlikely to be genetic or biologically based; genetic variability within racial/ethnic groups is often greater than between them (Adler & Rehkopf, 2008; Stolley, 1999; Wu & Schimmele, 2005), and despite considerable expenditure of resources, genomic research has failed to identify significant genetic variations explaining racial/ethnic disparities in the most prevalent diseases, such as cardiovascular diseases (Kaufman, Dolman, Rushani, & Cooper, 2015; Merikangas & Risch, 2003), Rather, the disparities more likely arise from the particular socioeconomic and cultural/behavioral context of living as a person of color in the United States (Wu & Schimmele, 2005). Although the current analysis was designed specifically to assess the degree to which racial/ethnic disparities in physical performance were mediated by other factors, the fact that race/ethnicity itself was still a significant contributor to that disparity among the Chinese suggests that other, unmeasured, perhaps unknown, aspects of neighborhood context, such as safety concerns or toxic environmental exposures, may limit access to social capital and result in a position of social disadvantage due to race that, over time, can affect physical performance (Adler & Rehkopf, 2008; Chinn & Hummer, 2016; Do et al., 2008; Lo, Lara, & Cheng, 2017) (Shuey & Willson, 2014).

The current analysis has several limitations. First, because this analysis examines physical performance at one point in time, it does not take into account the very real possibility of improved performance over time, which is known to occur with self-reported functional limitations (Ylitalo et al., 2013) Also, physical performance itself as a measure of overall functional status does not account for adaptations individuals make in their behavior or their environment that allows for full functional independence despite a limitation or deficit. In addition, although our hypothesized model was intended to include all relevant mediators, as discussed earlier, it is possible that part of the racial/ethnic disparity in physical performance could be explained by unmeasured factors. Finally, the findings may be biased given that there were significant differences in all of the mediators between those included and those excluded from the analysis. However, this potential selection bias would tend to underestimate true associations, particularly the indirect and total effects, because those who were excluded tended to be considerably less advantaged in terms of all the mediators.

Nevertheless, this study has important strengths. Most notably, the multiracial/ethnic composition of this midlife cohort allowed for the simultaneous comparison of four different groups to Caucasian women, including, uniquely, Japanese and Chinese women, and revealed a striking difference in physical performance between the two Asian subgroups. It also allowed for the examination of disparities in physical function at a younger age, and earlier in the disablement process, than most other studies. In addition, the use of formal mediation analysis allowed for the simultaneous consideration of all specified mediating factors. Other strengths include the repeated examination of the cohort more than 13 years which allowed for the consideration of mediators that varied over time. Finally, the community-based sampling of the cohort ensured broad generalizability of the findings.

In conclusion, this study suggests that, although not all racial/ethnic disparities in physical performance at midlife in all race/ethnic groups can be explained by known socioeconomic, medical, behavioral, or psychosocial factors, much can be done to reduce those disparities. Specifically, addressing poverty and racial bias, providing access to high quality health care, and improving neighborhoods to facilitate healthy lifestyles are of high priority for ensuring optimal functional status and overall health for all and minimizing loss of independence in old age.

Funding

The Study of Women’s Health Across the Nation (SWAN) has grant support from the National Institutes of Health (NIH), Department of Health and Human Services (DHHS), through the National Institute on Aging (NIA), the National Institute of Nursing Research (NINR), and the NIH Office of Research on Women’s Health (ORWH) (U01NR004061, U01AG012505, U01AG012535, U01AG012531, U01AG012539, U01AG012546, U01AG012553, U01AG012554, U01AG012495).

Author Contributions

B. Sternfeld designed the study, planned data collection, supervised statistical analysis, and wrote the manuscript; A. Colvin directed statistical analyses and provided critical review of the manuscript; A. Stewart conducted statistical analyses and critical review of the manuscript; B. M. Appelhans, J. A. Cauley, S.E. Dugan, S. R. E. Khoudary, G. A. Greendale, E. S. Strotmeyer, and C. A. Karvonenen-Gutierrez all helped plan data collection and provided critical review of the manuscript.

Conflict of Interest

None reported.