| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
BRIEF REPORT |
1 College of Public Health, University of Iowa, Iowa City.
2 School of Medicine, Saint Louis University, Missouri.
3 Geriatric Research, Education, and Clinical Center, St. Louis, Missouri.
4 School of Public Health, Saint Louis University, Missouri.
5 Division of Biostatistics, Washington University School of Medicine, St. Louis, Missouri.
Address correspondence to Fredric D. Wolinsky, the John W. Colloton Chair, Department of Health Management and Policy, College of Public Health, University of Iowa, 200 Hawkins Drive, E205 General Hospital, Iowa City, IA 52242. E-mail: fredric-wolinsky{at}uiowa.edu
Abstract
Objectives. In this article we explore the measurement properties of the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) in 49- to 65-year-old African Americans, compare their health-related quality of life (HRQoL) with that of the nation, and evaluate the association of selected covariates with HRQoL.
Methods. A probability sample of 998 African Americans in St. Louis received comprehensive in-home assessments. We used an item analysis, exploratory and confirmatory factor analysis, and internal consistency reliability to evaluate the measurement properties of the eight SF-36 scales and their interrelationships. We used a multiple linear regression analysis to characterize the effects of the demographic, socioeconomic status, psychosocial attributes, and biomedical markers.
Results. Assessments averaged 2.5 hr. Each SF-36 scale was unidimensional, all items had robust factor loadings, and all but one scale achieved excellent (
.80) internal consistency reliability levels. The overall factor structure of the SF-36 scales was generally consistent with national norms. Substantial variance was explained by the covariates, mostly attributable to socioeconomic status and the biomedical markers.
Conclusions. The SF-36 is a reliable and valid measure of HRQoL for use with African Americans. In this sample, the HRQoL was below national averages. Future social epidemiologic studies should include grip strength, vision, and hearing assessments, which had substantial and consistent associations with the SF-36 scale scores.
RACIAL disparities in disease and illness behavior are well documented (Cain & Kington, 2003
; Department of Health and Human Services [DHHS], 1985; Lillie-Blanton, Parsons, Gayle, & Dievler, 1996; Williams, Lavizzo-Mourey, & Warren, 1994; Williams, Yu, Jackson, & Anderson, 1997). African Americans have shorter life expectancies and bear greater disease burdens than Whites. Despite considerable federal programmatic efforts, this gap has not narrowed (DHHS, 1991, 2000; National Center for Health Statistics [NCHS], 2001, 2003). Indeed, for several conditions and diseases, the chasm is increasing. Racial differences in health-related quality of life (HRQoL), however, have not been well documented (Hill, Njai, Neighbors, Williams, & Jackson, 2003; Williams, Neighbors, & Jackson, 2003). This is quite surprising for two reasons. First, HRQoL encompasses the World Health Organization's (WHO, 1947) classic and broader notion that health is not simply the absence of disease but also includes physical, social, and role functions, as well as mental health and general health perceptions. Second, HRQoL was selected by the Centers for Medicare and Medicaid Services (CMS) as the primary outcome measure for evaluating managed care delivery programs, such as Medicare+Choice (CMS, 2003). Thus, research on HRQoL among African Americans, especially with respect to the measurement properties of the instruments selected by the CMS and their correlates, is warranted (Hutchinson, 1996).
Data can be brought to bear on these issues from the African American Health (AAH) study. The AAH study consists of a large, multistage, probability sample of middle-aged African Americans living in St. Louis, Missouri. Its overall purpose is to examine the disablement process and to identify critical pathways at which intelligent interventional strategies can be targeted. At baseline, the AAH study included comprehensive demographic, socioeconomic, psychosocial, and biomedical assessments. Among these was Version 2.0 of the Medical Outcomes Study 36-Item Short-Form Health Survey, known as SF-36 (Ware, Kosinski, & Dewey, 2000), which serves as the core component of the CMS protocol (CMS, 2003). Moreover, to maximize socioeconomic diversity within the African American population, the AAH study sampled from two strata—a poor inner city area and a nearby middle-class suburban area.
In this brief report we use the AAH data to confirm the measurement properties of the SF-36 in this middle-aged African American population, compare their HRQoL with national norms, and evaluate the association of selected covariates with HRQoL. The latter is especially important for two reasons. First, it is well known that perceptions of health and illness are related to demographic, socioeconomic status, psychosocial, and biomedical factors (Cleary, 1996; Wilson & Cleary, 1995). Second, there is considerable controversy over whether observed racial differences in disease and illness behavior simply result from socioeconomic inequalities in American society (House, 2002; Mirowsky & Ross, 2003). If this is the case, then the association of HRQoL with socioeconomic status, which is measured at the individual, neighborhood, and strata levels in the AAH study, should be substantial.
METHODS
Sample
The sampling design of the AAH study has been described elsewhere (Miller et al., in press). Simply put, the AAH study includes 998 African Americans who were born between 1936 and 1950. All participants lived in either a poor, inner-city area that had previously been studied (Miller et al., 1996) or in suburbs just northwest of the city. Sampling proportions were set to recruit approximately equal numbers of participants from both strata, which resulted in higher probabilities of selection in the inner city because it had fewer eligible participants. Therefore, weighted data are used in these analyses. Besides birth dates, inclusion criteria involved self-reported Black or African American race, standardized Mini-Mental Status Examination (MMSE) scores 16 (Molloy, Silberfeld, & Darzins, 1996), and willingness to sign informed consent. All participants received in-home, baseline evaluations that averaged 2.5 hr and occurred between September of 2000 and July of 2001. The response rate was 76%.
The SF-36
The SF-36 comprehensively addresses the WHO (1947) definition of health, and it is the most widely used HRQoL instrument in the world (Brazier, Harper, & Jones, 1992; McHorney, Ware, & Raczek, 1993; Stewart, Hays, & Ware, 1988; Tarlov, Ware, & Greenfield, 1989; Ware & Kosinski, 1999; Ware, Kosinski, & Dewey, 2000; Ware & Sherbourne, 1992). It was developed through iterative data-reduction experiences with a variety of precursor measures used in the Medical Outcomes Study (MOS). Version 2.0 provides improved wording to minimize ambiguity, and it replaces the dichotomous response set for seven items with five levels to enhance internal consistency (Ware et al., 2000). A detailed description of the development procedures is available elsewhere (Ware, 1996), as is the exact wording of the items (Ware & Kosinski, 1999; Ware et al., 2000). Thirty-five of the 36 items make up eight scales: physical functioning (10 items), role limitations that are due to physical functioning (4 items), bodily pain (2 items), general health perceptions (5 items), vitality (4 items), social functioning (2 items), role limitations that are due to emotional problems (3 items), and mental health (5 items). The remaining item asks respondents about any health changes over the past year but is not used in any of the scales. Within scales, a proration imputation method is used for missing data. That is, as long as the respondent answers at least half of the items within a scale, the average of those items is imputed for any items not answered in that scale. Because of the different number of items and response options in each scale, raw scores are transformed to range from 0 (worst health) to 100 (best health). These can also be rescaled (normed) to national data having a mean of 50 and a standard deviation of 10 by applying the appropriate algorithm (Ware et al., 2000).
Selected Covariates
Various conceptual models posit that HRQoL is associated with a number of factors (Cleary, 1996; Wilson & Cleary, 1995). Of special interest here are socioeconomic status and biomedical markers. It has been suggested that the former may explain a substantial amount of racial disparities (House, 2002; Mirowsky & Ross, 2003). The potential confounding from biomedical markers, however, has not been adequately explored (Cleary, 1996), so it is critical to determine if the association of socioeconomic status with HRQoL is substantial after biomedical and other factors have been adjusted for. Thus, the multiple linear regression analysis includes demographics, socioeconomic status, psychosocial attributes, and biomedical markers.
The demographic characteristics included age, gender, marital status, and race consciousness. Age was measured in years (range = 49–65, M = 56.8, SD = 4.4). Gender was coded 1 for men (41.8%). Marital status was assessed with a set of dummy variables reflecting being single (11.7%), divorced or separated (28.5%), or widowed (12.8%), with being married as the reference category (47.0%). Race consciousness in this study of African Americans was measured by asking respondents how often they thought about their race, with those responding never or only once a year (42.2%) contrasted with all others.
Socioeconomic status was measured by education, subjective and objective income, perceived financial barriers to obtaining health care, the desirability of the neighborhood, and sampling strata. Education was measured in years (range = 0–25, M = 12.5, SD = 2.8) and subsequently recoded to contrast having a grade school education or less (6.4%) versus going to high school or beyond. Subjective income was assessed with a set of two dummy variables reflecting having a comfortable income (45.9%) or not enough to get by (15.1%), with having just enough to get by as the reference category (39.0%). Objective income was measured by response to an unfolding income question based on eight categories, which was collapsed into a set of two dummy variables reflecting $20,000 or less (27.8%) or refusing to report income (4.1%), with having more than $20,000 in annual income as the reference category (68.1%). Perceived financial barriers to health care were tapped by asking respondents if they had been unable to see a doctor when they needed to as a result of cost at any time during the prior year (7.7%). Neighborhood desirability was assessed by a four-item scale ( =.792), which was recoded to contrast living in the least desirable quartile (24.5%) versus all others. Sampling strata were coded as a binary contrast for living in the city stratum (21.3%).
Two psychological attributes were assessed: social support and religiosity. Social support was measured with five items (i.e., someone to confide in, get together with, help with daily chores, turn to for suggestions, and love and make you feel wanted; =.859) from the MOS instrument (Sherbourne & Stewart, 1991). The resulting scale score was recoded to contrast being in the lowest quintile (21.6%) versus all others. Religiosity was measured with a five-item scale ( =.703) derived from the Fetzer instrument (Fetzer Institute, 1999). The resulting scale score was also recoded to contrast being in the lowest quintile (20.0%) versus all others.
Biomedical factors included having a history of cancer, chronic obstructive pulmonary disease (COPD), or heart disease; high systolic or diastolic blood pressure; clinically relevant levels of depressive symptoms; and grip strength, vision, and hearing. The prevalence of the three diseases was 6.7% for cancer, 4.6% for COPD, and 9.5% for heart disease. When the average of two readings from an automated sphygmomanometer was used, the prevalence of high systolic blood pressure (140) was 49.3% and the prevalence of high diastolic blood pressure (90) was 31.6%. The presence of clinically relevant levels of depressive symptoms (21.1%) was evaluated by use of the 11-item version of the Center for Epidemiological Studies Depression (CES-D) scale ( =.836), for which a score of 9 or greater is equivalent to the traditional cutoff score of 16, using the 20-item CES-D (Kohout, Berkman, Evans, & Cornoni-Huntley, 1993). Grip strength was assessed as the average of three attempts by respondents to use a dynamometer with their stronger hand (range = 3–76 kg, M = 34.6 kg, SD = 12.1); the average was recoded into a set of three dummy variables reflecting the highest quintile (44.7 kg, 19.0%), the lowest quintile (
25.0 kg, 18.1%), or not performing the test (7.6%), versus the three middle quintiles (55.3%). Vision was measured by use of a three-item scale derived from the Health and Retirement Study (HRS, 2002; =.769), which was subsequently recoded to contrast being in the poorest quintile (18.2%) versus all others. Hearing was evaluated with the single-item HRS self-assessment (HRS, 2002), dichotomized to reflect fair or poor responses (13.1%) versus all others.
Analytic Approach
Psychometric evaluation was performed to determine whether the measurement properties of the SF-36 scales in the AAH were equivalent to national norms. Consistent with the national norming studies for the SF-36 (McHorney et al., 1993), this involved item and exploratory factor analysis (EFA) with oblique rotation as well as internal consistency reliability (Nunnally, 1967). Confirmatory factor analysis (CFA) was also used to estimate the model originally hypothesized for the SF-36 (McHorney et al., 1993; Ware, 1996), as well as three competing models using AMOS 5.0 (Arbuckle, 2003). Multiple linear regression analysis was used to model each SF-36 scale score (Allison, 1999). In those analyses, the demographics were entered first, followed by socioeconomic status, then the psychosocial attributes, and finally the biomedical markers. Although effect decomposition of the more distal covariates was evaluated as the more proximal covariates were introduced, only the results of the final models with all covariates entered are shown. All assumptions of the statistical model were evaluated by use of standard procedures, and no meaningful violations were found (Allison, 1999). Given the number of covariates, it is especially important to note that multicollinearity was not a problem (all tolerance statistics .610).
RESULTS
Item analyses (not shown) are consistent with national data. Typical of population-based samples, ceiling effects are relatively common, especially for the items in the physical function, social functioning, role emotional, and mental health scales. Table 1 summarizes the EFA conducted within SF-36 scales, as well as the internal consistency reliability evaluations. All eight scales are unidimensional, and all factor loadings exceed conservative standards (Nunnally, 1967). Internal consistency reliability generally reaches or exceeds levels considered excellent for basic research ( .80; Nunnally, 1967). The principal exception is the social functioning scale, for which alpha is marginally below standard levels of acceptability ( .70; Nunnally, 1967). Review of the means and standard deviations demonstrates that the individual item ceiling effects are overcome by reliance on the scale scores, where adequate variation exists. The normed means (population M = 50, SD = 10) indicate that the AAH population has somewhat poorer HRQoL than the nation as a whole.
|
). In subsequent discussions, Ware (1996) has argued that, conceptually, the physical function, role physical, bodily pain, and general health perception scales principally reflect the physical dimension, whereas the mental health, role emotional, social functioning, and vitality scales principally reflect the mental dimension. In the AAH study, the EFAs indicate that these expectations generally hold, if a two-factor solution is specified. Only the first factor, however, has an eigenvalue greater than unity and thus meets the traditional threshold for being a common factor.
For the structure of the SF-36 scale scores to be explored further, four competing CFA models were estimated (not shown). The first was an overly simplistic single common factor model consistent with the results obtained from the first EFA. Although all estimated factor loadings were robust (i.e., >.70), the model fit the data poorly (
2 = 647.4; Normed Fit Index [NFI] =.870; root mean square error of approximation, or RMSEA, =.178). The second model was consistent with Ware's (1996) two-factor conceptual specification but did not allow for factorial complexity (i.e., cross-loadings). Factor loadings obtained from this model were also robust, and although model fit improved, it remained relatively poor (2 = 370.2; NFI =.926; RMSEA =.137). The third model built on the second by allowing for the factorial complexity anticipated in developing the SF-36 (McHorney et al., 1993). Once again, although model fit improved, the need for further refinement was evident (2 = 288.8; NFI =.954; RMSEA =.116). The final model involved an empirical respecification of the second model based on its modification indices. In this two-factor model, the only factorial complexity allowed involved the mental health scale. This model fit the data best (2 = 178.1; NFI =.964; RMSEA =.095), and it is the only model acceptable by established standards (Browne & Cudeck, 1993).
Table 2 contains the partial, unstandardized (b) regression coefficients obtained in the final modeling stage of each SF-36 scale. In general, the models account for 32% to 44% of the variance in the SF-36 scale scores. The exception is the pain scale (R2 =.216). Although substantial portions of the SF-36 scales are explained, very little is attributable to demographics. Even when entered by themselves (not shown), the demographics account for only 1% to 5% of the variance, with most of this attributable to lower HRQoL for men. Similarly, when the psychosocial attributes are added to the demographic and socioeconomic status factors already in the model at the third stage, they generally add less than 2% to the explained variance, which is attributable to social support. Moreover, once the biomedical factors are introduced, neither social support nor religiosity retains statistical significance.
|
). When entered into the final models, the biomedical markers add 11.7% to 23.1% to the explained variance. Although all of the biomedical markers have significant effects in one or more of the models, the largest and most consistent associations involve having COPD, clinically relevant levels of depressive symptoms, low grip strength or the inability to perform that test, poor vision, and poor hearing. DISCUSSION
Four lessons can be learned from these results. First, each of the eight SF-36 scales was shown to be psychometrically sound in this population-based African American sample. Moreover, the relationships among the eight SF-36 scales observed in the EFA were quite consistent with those reported from the national norming study (McHorney et al., 1993) and postulated by Ware (1996). Results from the CFA were slightly different, in that the best-fitting model had two factors but only allowed for factorial complexity in the mental health scale. This difference is modest and is somewhat difficult to assess inasmuch as the national norming study did not use CFA. Thus, there should be no cause for concern about using the SF-36 to evaluate HRQoL in middle-aged African Americans.
Second, using the SF-36 permitted a direct comparison with national norms. This was accomplished when the AAH study SF-36 scale scores were rescaled by use of the formulas provided by Ware and colleagues (2000), based on a national mean of 50 with a standard deviation of 10. When this was done, the AAH SF-36 scale scores were below the national average (i.e., 50) for all but the vitality and mental health scales. Although the magnitude of the disparity may appear trivial (approximately 2 points on average), given that the standard deviation is set at 10, these differences actually reflect a 0.20 effect size, which Cohen considers to be small but clinically important in intervention studies (Cohen, 1988). These results also suggest that, despite experiencing poorer physical well-being, African Americans coped with their limitations better than Whites.
Third, the multiple regression analysis has shown that the two categories of covariates principally responsible for the substantial levels of explained variance were socioeconomic status and the biomedical factors. Moreover, the contributions of these two covariate categories were generally equivalent. The predictive ability of socioeconomic status reflects the diversity within the African American population and supports the conclusion that racial disparities in health and health behavior likely result from socioeconomic inequalities in American society (House, 2002; Mirowsky & Ross, 2003), whereas the predictive ability of the biomedical markers reflects the clinical validity of the SF-36 and its conceptualization.
Finally, the multiple regression analysis identified dramatic associations such that respondents with low incomes, COPD, clinically relevant levels of depressive symptoms, poor grip strength, poor vision, and poor hearing had lower HRQoL. Although income, COPD, and depressive symptoms are known to be strong correlates of HRQoL, to our knowledge, this is the first report of consistent and substantial deleterious associations involving grip strength, vision, and hearing. This suggests that these three assessments, which are relatively easy and inexpensive to obtain, should be given strong consideration for inclusion in future social and epidemiologic studies of African Americans. The reasons for this are straightforward. Grip strength is an important component of the frailty phenotype (Fried et al., 2001) and is predictive of subsequent mortality (Metter, Talbot, Schrager, & Conwit, 2002). Similarly, vision and hearing limitations are predictive of subsequent morbidity (Tinetti, Inouye, Gill, & Doucette, 1995). Thus, the onsets of decline in grip strength, vision, and hearing, which are not well described, may be the critical pathways at which intelligent interventional strategies can be targeted.
Acknowledgments
This research was supported by Grant R01 AG-10436 from the National Institutes of Health to Dr. D. K. Miller. The opinions expressed here are those of the authors only.
Footnotes
Decision Editor: Charles F. Longino, Jr., PhD
Received for publication May 7, 2003. Accepted for publication September 16, 2003.
References
This article has been cited by other articles:
 |
|
 |
|
  Yuri Jang, D. A. Chiriboga, A. R. Borenstein, B. J. Small, and J. A. Mortimer Health-Related Quality of Life in Community-Dwelling Older Whites and African Americans J Aging Health, April 1, 2009; 21(2): 336 - 349. [Abstract] [PDF]
|
|
|
|
 |
|
 M. Schootman, E. M. Andresen, F. D. Wolinsky, T. K. Malmstrom, J. P. Miller, Y. Yan, and D. K. Miller The Effect of Adverse Housing and Neighborhood Conditions on the Development of Diabetes Mellitus among Middle-aged African Americans Am. J. Epidemiol., August 15, 2007; 166(4): 379 - 387. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Schootman, E. M Andresen, F. D Wolinsky, T. K Malmstrom, J P. Miller, and D. K Miller Neighbourhood environment and the incidence of depressive symptoms among middle-aged African Americans J Epidemiol Community Health, June 1, 2007; 61(6): 527 - 532. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. M. Andresen, F. D. Wolinsky, J. P. Miller, M.-M. G. Wilson, T. K. Malmstrom, and D. K. Miller Cross-Sectional and Longitudinal Risk Factors for Falls, Fear of Falling, and Falls Efficacy in a Cohort of Middle-Aged African Americans Gerontologist, April 1, 2006; 46(2): 249 - 257. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Schootman, E. M. Andresen, F. D. Wolinsky, T. K. Malmstrom, J. P. Miller, and D. K. Miller Neighborhood Conditions and Risk of Incident Lower-Body Functional Limitations among Middle-aged African Americans Am. J. Epidemiol., March 1, 2006; 163(5): 450 - 458. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. M. Andresen and D. K. Miller The Future (History) of Socioeconomic Measurement and Implications for Improving Health Outcomes Among African Americans J. Gerontol. A Biol. Sci. Med. Sci., October 1, 2005; 60(10): 1345 - 1350. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
