Paper

Socioeconomic Status and Health: How Education, Income, and Occupation Contribute to Risk Factors for Cardiovascular Disease

Marilyn A. Wnkleby, PhD, Danus E. Jatulis, MS, Erica Frank, MD, MPH, and Stephen P. Fortmann, MD

Intodwton One of the strongest and most con-

sistent predictors of a person's morbidity and mortality experience is that person's socioeconomic status (SES).1-5 This find- ing persists across all diseases with few exceptions, continues throughout the en- tire life span,5 and extends across numer- ous risk factors for disease.6-9

The significant impact of SES on dis- ease makes its definition and measure- ment of critical importance. SES is a com- plex phenomenon predicted by a broad spectrum of variables that is often con- ceptualized as a combination of financial, occupational, and educational influ- ences.0-13 Although these dimensions of SES are interrelated, it has been proposed that each reflects somewhat different in- dividual and societal forces associated with health and disease. For example, in- come reflects spending power, housing, diet, and medical care; occupation mea- sures prestige, responsibility, physical ac- tivity, and work exposures; and education indicates skills requisite for acquiring pos- itive social, psychological, and economic resources.9,14

Much has been written about the un- derlying mechanisms through which SES may operate to affect disease. Many au- thors have suggested that certain dimen- sions ofSES are more predictive of health than others;1'14-'6 however, such propos- als tend to be theoretically based, without substantiating data. Over time, education has become the most commonly used measure of SES in epidemiological stud- ies,17 yet no investigators in the United States have conducted an empirical anal- ysis quantifying the relative contributions of different measures of SES to risk fac- tors or disease outcomes. (This paper does

not suggest which component ofSES may be the most reliable and valid measure be- cause this question has been examined, with a variety of conclusions, in previ- ously published reports.)'1-13, 17-18

The present study examines the as- sociation between income, education, oc- cupation, and a set of risk factors for car- diovascular disease-namely, cigarette smoking, systolic and diastolic blood pres- sure, and total and high-density lipopro- tein (HDL) cholesterol. Our study objec- tives are to (1) examine the impact of each separate dimension of SES on cardiovas- cular risk factors, (2) use a forward step- wise selection model to evaluate if one measure of SES is the strongest predictor of risk factors, and (3) offer guidance to researchers about selection of SES mea- sures. This guidance is critical because nearly all epidemiological studies use SES as an explanatory or a control variable, or for the selection of subjects or matching criteria.

Methds Subjects aged 25 to 64 were drawn

from the two control cities of the Stanford Five-City Project,'9 a communitywide cardiovascular disease intervention study that contains data from four separate

The authors are with the Stanford Center for Research in Disease Prevention, Stanford Uni- versity School of Medicine, in Palo Alto, CA.

Requests for reprints should be sent to Marilyn A. Winkleby, Stanford Center for Re- search in Disease Prevention, Stanford Univer- sity School of Medicine, 1000 Welch Road, Palo Alto, CA 94304-1885.

This paper was submitted to the Journal April 19,1991, and acceptedwith revisions Sep- tember 13, 1991.

Editor's Note. See related editorial on p 785 of this issue.

June 1992, Vol. 82, No. 6

Socoeconomic Status and Cardiovascular Disease

cross-sectional surveys, conducted from 1979 to 1986. Participants who were un- employed (n = 98), students (n = 130), or retirees (n = 146) were excluded because they had no occupation that could be ranked. Data from the four cross-sectional surveys were pooled because demo- graphic characteristics, including mea- sures of SES, showed no significant inter- actions over time.7 As previously reported,20 the educational attainment of respondents from the control cities were well matched to those from the treatment cities. A nonrespondent questionnaire in- dicated that participants were generally more educated than those in both treat- ment and control communities who re- fused to participate.20

Information on education, income, and occupation was ascertained through questionnaires. Education was recorded as the highest number of years of schooling completed. It was used as a continuous variable in regression analyses and was di- vided into the following four categories in stratified analyses: less than 12 years, 12 years, 13 to 15 years, and 16years or more.

Income information, defined as gross annual household income, was collected in intervals of $5000. For multivariate analyses, the midpoint of the income cat- egories was used; for the remaining anal- yses, income was divided into five cate- gories ranging from less than $10 000 to $40 000 or more per year.

Current occupation was collected as an open-ended variable and coded using the 1980 US Bureau of the Census occu- pational scaling system, which creates a hierarchy of occupations ordered on the basis of education and inCome.21 To ex- amine associations between separate oc- cupations and risk factors, we coded each occupation, including homemaker, as an indicator variable, with executives as the reference category. In multivariate mod- els, we excluded homemakers and used occupation as a ranked variable ranging from executives to unskilled workers.

Participantswere considered cigarette smokers if they reported ever smoldng on a daily basis and had smoked one or more cigarettes in the last week. Participants whose plasma thiocyanate exceeded 100 pLmol/L and whose expired-air carbon monoxide level exceeded 8 ppmwere clas- sified as smokers regardless of their self- reported responses.22 Total and HDL plasma cholesterol were derived from non- fasting venous samples, analyzed fresh by

right arm using a semiautomatic recorder, and the average of the second and third readings were used for analyses.24

Resuls An overall response rate of 69o was

achieved. Approximately 600 individuals participated in each survey, resulting in a total of 2380 participants for all surveys combined. Because the study population was predominantly White, non-Hispanic (85%), findings are not generalizable to populations representing a broad spec- trum of racial groups.

All pairwise correlations between ed- ucation, income, and occupation were positive and were stronger for men than for women (Table 1). The lowest correla- tion was between education and income, indicating that education is not a primary determinant of wage. Higher correlations were shown for education and occupa- tion, suggesting that skills acquired during education may help determine occupa- tion. Although correlations ranged from .23 to .67, their relatively low magnitude (highest adjusted R2 = 45%) indicates that the three dimensions are not redun- dant measures of SES.

Although the study population had relatively high educational and financial levels, individuals from all education, in- come, and occupation categories were well-represented (Table 2). However, while men were represented fairly evenly across occupations, approximately 75% of women employed outside the home

held nonprofessional white-collar jobs. Men tended to have more years of educa- tion than women and to be from higher- income households.

In general, those with the lowest ed- ucational attainment exhibited the highest prevalence of risk factors (Table 3). Clear gradients were seen between educational level and smoking for both sexes, and be- tween education and total and HDL cho- lesterol forwomen. Across all riskfactors, men consistently exhibited higher risk than did women.

Income and occupation were less consistent risk predictors. Higher riskwas associated with lower incomes for smok- ing and HDLcholesterol in both sexes but with higher incomes for total cholesterol in men. Within occupations, men and women white-collar executives and man- agers exhibited the lowest levels of smok- ing. Among men, executives and manag- ers showed the lowest mean levels of blood pressure.

standard methods established by the Lipid Research Clinics Program32 Three blood pressure measurements were taken on the

June 1992, Vol. 82, No. 6 American Joumal of Pubfic Health 817

Wvieby et aL

To compare the strength of the in- terrelationships between the three di- mensions of SES, we conducted regres- sion analyses (Table 4), adjusted for age and time of survey. Partial correlations are presented with two-tailed P values, with significance defined asP < .05. The univariate relationship between SES and risk factors was strongest and most con- sistent for education, showing higher risk associated with lower levels of educa- tion. Using a forward selection model that allowed for inclusion of all three measures ofSES after adjustment for age and time of survey, education was the

only measure of SES that was signifi- cantly associated with the risk factors. This finding is consistent with a previ- ously conducted multivariate analysis, which showed that the strength and sig- nificance of the associations between ed- ucation and a set of disease risk factors remained virtually unchanged after ad- justment for income and occupation.7

When stratified by White, non-His- panic and Hispanic ethnicities (not shown), education remained the strongest SES predictor but became nonsignificant forsmokng, possiblybecause ofthe small sample of Hispanics.

Diswcusion

Strengths and Limitations of Using Education as the Markerfor SES

We caution that, in some studies, us- ing only one indicator of SES may yield misleading results or provide less informa- tion than using multiple measures. How- ever, using multiple or composite measures10-14 requires the cost and time of collecting data on several SES parameters and may not significantly explain more about a population than would a single, weil-chosen parameter. As noted by a re-

818 American Journal of Public Health June 1992, Vol. 82, No. 6

cent working group of the National Heart, Lung, and Blood Institute, use ofcompos- ite measures may obscure important dif- ferences in associations.25

Based onourfindings, education may be the mostjudicious SES measure for use in epidemiological studies (unless the study hypothesis dictates which dimen- sion of SES is to be chosen). In studies that have a cost or time restraint but need a measure ofSES as a potential confound- ing variable, education is an expeditious choice. In addition, education is available for all individuals regardless of employ- ment status, has high reliability and valid- ity,17 is generally stable after early adult- hood, is easily reported, and can be collected as a continuous variable. Fur- thermore, because education is often available in epidemiological studies,17 it also permits opportunities for meta-anal- yses and interstudy comparisons.

However, there are potential limita- tions to using education as a sole indicator of SES. Its stability can mask important changes in individuals' circumstances. There may also be a cohort effect distort- ing differences between populations of various ages. For example, the percent of the population obtaining at least a high school education has increased nearly threefold since 1940.17 This has led to in- creasing homogeneity in the amount ofed- ucation obtained, making differentiation between educational strata more difficult. Other potential problems include regional differences in education, the question of whether degrees or certification are better measurement parameters than years of schooling,1726,27 and the possibility that other dimensions of SES are more sensi- tive markers for health in some population subgroups. (For example, for foreign-born female Hispanics, acculturation may be a stronger measure of SES.)13 Companson with Past Studies

To our knowledge, this is the only study in the United States to examine as- sociations between separate SES dimen- sions and risk factors or disease outcomes. Other studies, however, have exaniined associations between one measure of SES and one disease risk factor,6828-32 morbid- ity,33 or mortality.1,33-35 In general, these studies have found that education is more strongly associated with disease than is in- come or occupation. One ofthe most com- plete studies ofmortality differentials found that lower SES groups exhibited higher rates of all-cause mortality than did higher SES groups, irrespetve of whether edu- cation, income, or occupation was used as

Junle 1992, Vol. 82, No. 6

the measure of SES.1 Other studies have documented strong inverse associations between education and all-cause mortali- ty3l,34,36 and life expectancy.37 Framing- ham study data shows that, of 23 potential contnrbutors to morbidity, only education and age at study enrollment were related in both sexes to "survival with good func- tion."38 Cardiovascular disease studies have shown that lower levels of education are associated with hypertension,6-m829-31 39 cgarette smoking,63l132,4041 and hi cho- lesterol,6'32 as well as with cardiovascula morbidity33 and mortaity.3l,33,35 Why Education May Be the Strongest Predictor ofGood Health

Several different mechanisms through which education may positively influence health have been proposed. Fuchs has suggested that both education and health are markers for willingness to delay gratification in order to "invest in human capital."'15 Others have argued that education may simply serve as a marker for intelligence; however, studies showing that environmental factors are the strong- est predictors of school dropout lend little support to this hypothesis.42 Some have suggested that higher education may im- prove health by conferring economic ad- vantages, but our low correlation coeffi- cients between education and income, as well as the lack of significant income ef- fects on risk factors, argue against this hy- pothesis. Neither does the health knowl- edge acquisition that accompanies higher education appear to explain the relation- ship between education and health, given that provision of information alone ap- pears to be a weak stimulus to human be- havior change.16,37

One hypothesis we find most plausi- ble is that education may protect against disease by influencing life-style behaviors, problem-solving abilities, and values.17 Moreover, education mayfacilitate the ac- quisition of positive social, psychological, and economic skills and assets, and may provide insulation from adverse influenc- es.7 Such skills and assets that may ac- company higher educational attainment include positive attitudes about health, ac- cess to preventive health services,15 mem- bership in peer groups that promote the adoption or continuation ofpositive health behaviors, and higher self-esteem and self-efficacy.43,44 Suinrnai

There can be no SES measure that is universally valid and suitable for all pop- ulations. However, if economics and time

Socioeconomic Status and Cardiovascular Diseas

dictate that a single parameter be chosen and if the research hypothesis does not dictate otherwise, this study suggests that higher education, rather than income or oc- cupation, may be the strongest and most consistent predictor of good health. C1

Acknowledgments This research was supported by Public Health Service Grant 1RO1-HL-21906 from the Na- tional Heart, Lung, and Blood Institute to Dr. John W. Farquhar.

The authors thank Drs. Lawrence Green and David Ragland for comments on an earlier draft, Dr. Helena Kraemer and Ms. Beverly Rockhill for statistical advice, and Ms. Cindy German-Kung for the preparation of tables.

References 1. Kitagawa EM, Hauser PM. Differential

Mortality in the United States:A Study in SocioeconomicEpidemiology. Cambridge, Mass: Harvard University Press; 1973.

2. Blaxter M. Evidence on inequality in health from a national survey. Lancet 1987;ii:30- 33.

3. Black D. Inequaliies in Healtk Report of a Research WoHding Group. London, En- gland: Department ofHealth and Social Se- curity; 1980.

4. Haan M, Kaplan G, Camacho T. Poverty and health: prospective evidence from the Alameda County Study. Am JEpiemioL 1987;125:989-998.

5. Marmot MG, Kogevinas M, Elston MA. Social/economic status and disease. Ann Rev Public Health. 1987;8:111-135.

6. Matthews KA, Kelsey SF, Meilahn EN, Kuller LH, Wing RR. Educational attain- ment and behavioral and biological risk fac- tors for coronary heart disease in middle- aged women. Am J EpidemioL 1989;129: 1132-1144.

7. Winkleby MA, Fortmann SP, Barrett DC. Social class disparities in risk factors for dis- ease: eight-yearprevalence patterns bylevel of education. Prev Med 1990;19:1-12.

8. Helmert U, Herman B, Joeckel KH, Gre- iser E, Madans J. Social class and risk fac- tors for coronaxy heart disease in the Fed- eral Republic of Germany: results of the baseline survey of the German Cardiovas- cular Prevention Study. JEpidemiol Com- munity Health. 1989;43:37-42.

9. Antonovsky A. Social class, life expect- ancy and overall mortality. Mbank Mem Fund Q. 1967;45:31-73.

10. Mueller CW, Parcel TL. Measures of so- cioeconomic status: alternatives and rec- ,ommendations. Chil Dev. 1981;52:13-30.

11. Hollingshead AB. The indiscriminate state of social class measurement. Soc Forces. 1971;49:563-567.

12. Duncan OD. Asocioeconomic indexfor all occupations. In: Reiss AJ Jr, Duncan OD, Hatt PK, North CC, eds. Occupation and Social Status. New York: The Free Press of Glencoe; 1%61:109-138.

13. Green L. Manual for scoring socioeco- nomic status for research on health behav- ior. Public Health ReCp. 1970;85:815-827.

14. Susser MW, Watson W, Hopper K. Soci-

American Journal of Public Health 819

Wikieby et aL

ologyin Medicine. NewYork, NY: Oxford University Press; 1985.

15. Fuchs VR. Economics, health, and post- industrial society. Mibank Mem Fund QlHealth and Society. 1979;57:153-182.

16. Williams DR. Socioeconomic differentials in health: a review and redirection. Soc Psych Q. 1990;53:81-99.

17. Liberatos P, Link BG, Kelsey JL. The measurement of social class in epidemiol- ogy. Epidenio Rev. 1988;10:87-121.

18. Morgenstern H. Socioeconomic factors: concepts, measurement and health effects. NHLBI Workshop on "Measuring Psy- chosocial Variables in Epidemiological Studies of Cardiovascular Disease." De- cember 1983; Galveston, TX.

19. Farquhar JW, Fortmann SP, Maccoby N, et al. The Stanford Five-City Project: de- sign and methods. Am J EpidehnioL 1985; 122:323-334.

20. Farquhar JW, Fortmann SP, Flora JA, et al. Effects ofcommunitywide education on cardiovascular disease risk factors.JAM4. 1990;264:359-365.

21. US Bureau of the Census. 1980 Census of Population: A4labeical Inder of Indus- ries and Occupations. Washington, DC: US Government Printing Office; 1982.

22. Fortmann SP, Rogers T, Haskeli WL, So- lomon DS, Vranizan K, Farquhar JW. In- direct measures ofcigarette use: expired air carbon monoxde vs plasma thiocyanate. Prev Med. 1984;13:127-135.

23. US Department of Health, Education, and Welfare. LipidResearch ClidcsManual of Laboratory Opemations. Vol 1: Lipid and Lpoprotein Analysis. Washington, DC: US Government Printing Office; 1974.

24. Fortmann SP, Marcuson R, Bitter PH, Haskell WL. A comparison of the Sphyg- metrics SR-2 automatic blood pressure re-

corder to the mercury sphygmomanometer inpopulationstudiesA4mJEpidemioL 1981; 114:836-84.

25. Socioeconomic factors: working group summary. In: Ostfeld AM, Eaker ED, eds. Measwiurg Psychosocial Variables in Epi- demioogic Studes ofCardiovascular Dis- ease: Proeedigs ofa Workshop. Wash- ington, DC: US Government Printing Office; March 1985. NIH publication 85- 2270.

26. Morgan M. Measuring social inequality: occupational classifications and their alter- natives. Community Med 1983;5:116-124.

27. Faia MA. Selection by certification: a ne- glected variable in stratification research. AmJSocioL 1981;86:1093-1111.

28. Hypertension Detection and Follow-Up Program Cooperative Group. Race, educa- tion and prevalence of hypertension.AmJ EpidemioL 1977;106:351-361.

29. Dyer AR, Stamler J, Shekelle RB, Schoe- nberger J. The relationship of education to blood pressure: findings on 40 000 em- ployed Chicagoans. Ciwulation. 1976;54: 987-992.

30. Abramson JH, Gofin R, Habib J, Pridan H, Gofin J. A comparative appraisal of mea- sures for use in epidemiological studies. Soc Sci Med 1982;16:1739-1746.

31. Liu K, Cedres LB, Stamler J, et al. Rela- tionship of education to major risk factors and death fromcoronaxyheart disease, car- diovascular diseases, and all causes. Cir- culation 1982;66:1308-1314.

32. Jacobsen BK, Thelle DS. Risk factors for coronary heart disease and level of educa- tion. AmJEpiemnoL 1988;127:923-932.

33. Hinkle LE, Whitney LH, Lehman EW, et al. Occupation, education, and coronary heart disease. Science. 1968;161:238-246.

34. Snowdon DA, Ostwald SK, Kane RL. Ed- ucation, survival and independence in el- derly Catholic sisters, 1936-1988.AmJEp- idenioL 1989;130:999-1012.

35. Hypertension Detection and Follow-Up Progam Cooperative Group. Educational level and 5-year all-cause mortality in the Hypertension Detection and Follow-Up Program. HypeIension. 1987;9:641-646.

36. Feldman JJ, Makuc DM, Kleinman JC, Cornoni-Huntley J. National trends in ed- ucational differentials in mortality. Am J EpidemioL 1989;129:919-933.

37. Sagan LA. 7he Healh of Nations. New York, NY: Basic Books, Inc; 1987.

38. Pinsky JL, Leaverton PE, Stokes J. Pre- dictors of good function: the Framingham study. J Chmnic Dis. 1987;40:159S-167S.

39. Berger MC, Leigh JP. Schooling, self-se- lection, and health. J Hum Res. 1989;24: 433-455.

40. Millar WJ, Wigle DT. Socioeconomic dis- parities in risk factors for cardiovascular disease. Can MedAssoc J. 1986;134:127- 132.

41. Wagenknecht LE, Perkins LL, Cutler GR, et al. Cigarette smoking is strongly related to educational status: the CARDIA study. Prev Med. 1990;19:158-169.

42. Howard MA, Anderson RJ. Early identifi- cation of potential school dropouts: a liter- ature review. Child Welfare. 1978;57:221- 231.

43. Bunker JP, Gomby DS. Preface. In: Bun- ker JP, Gomby DS, Kehrer BH, ed. Path- ways to Healtk The Role of Social Fac- tors. Menlo Park, Calif: Hemy J. Kaiser Foundation; 1989:15.

44. Cohen S, Syme SL. Social Support and Healt New York, NY: Academic Press; 1985.

820 American Journal of Public Health June 1992, Vol. 82, No.6