- Email: [email protected]
Socioeconomic Variation in the Magnitude of the Association between Self-Rated Health and Mortality
Socioeconomic Variation in the Magnitude of the Association between Self-Rated Health and Mortality ´ N, JUAN L. GUTIE´RREZ-FISAC, ENRIQUE REGIDOR, PILAR GUALLAR-CASTILLO ´ ´ JOSE R. BANEGAS, AND FERNANDO RODRIGUEZ-ARTALEJO
PURPOSE: To assess socioeconomic variation in the association between self-rated health (SRH) and mortality and to determine whether socioeconomic inequalities in SRH and socioeconomic inequalities in mortality differ in magnitude. METHODS: We used data from a cohort of Spanish people 60 years of age and older with an 8-year follow-up of mortality. The association between SRH at baseline and mortality was estimated by the age-adjusted relative risk of mortality in people with low, medium, and high education. The measures of health inequalities were the prevalence ratio of poor SRH and the age-adjusted relative risk of mortality according to educational level. The validity of SRH to reflect life-threatening and non–life-threatening health conditions was summarized with the likelihood ratio for poor SRH in each educational category. RESULTS: The relative risk of mortality according to SRH in subjects with high and low education was 3.24 and 1.62 in men and 2.25 and 1.50 in women, respectively. Inequalities in poor self-rated health were larger than inequalities in mortality: –1.63 versus 1.07 in men and 1.45 versus 1.30 in women. The highest likelihood ratio for SRH was seen in persons with high education in the case of life-threatening conditions, and for those with low education, in the case of non–life-threatening conditions. CONCLUSIONS: Socioeconomic variation in the validity of SRH to reflect life-threatening and non–life-threatening conditions could explain the greater ability of SRH to predict mortality in persons with high education and why inequalities in poor SRH are larger than inequalities in mortality. Ann Epidemiol 2010;20:395–400. Ó 2010 Elsevier Inc. All rights reserved. KEY WORDS:
Self-rated Health, Mortality, Education, Health Inequalities, Health Problems.
INTRODUCTION Many investigations have shown a strong association between self-rated health (SRH) and mortality (1, 2). In contrast, only a few studies have evaluated whether this relation varies by socioeconomic position. Four of five studies on this issue found a socioeconomic variation in the relation between SRH and mortality (3–7). Three studies in the general population of Sweden, The Netherlands, and the United States, respectively, observed that the relative risk of mortality according to SRH was highest in subjects in the highest socioeconomic position (3–5). However, another study, conducted in an occupational cohort in France, found the highest relative risk for mortality in those with the lowest socioeconomic position (6). Since the results in the French cohort could have been affected by From the Department of Preventive Medicine and Public Health, Universidad Complutense de Madrid (E.R.); CIBER Epidemiologı´a y Salud Pu´blica (CIBERESP) (E.R., P.G-C., J.L.G-F., J.R.B., F.R-A); and the Department of Preventive Medicine and Public Health, Universidad Auto´noma de Madrid (P.G-C., J.L.G-F., J.R.B., F.R-A), Spain. Address correspondence to: Enrique Regidor, Department of Preventive Medicine and Public Health, Universidad Complutense de Madrid, Ciudad Universitaria, 28040 Madrid, Spain. Tel: þ34 913941521. E-mail: [email protected]. Received September 15, 2009; accepted January 30, 2010. Ó 2010 Elsevier Inc. All rights reserved. 360 Park Avenue South, New York, NY 10010
different selection biases (8), the consistency of the findings of the other three studies suggests that the validity of SRH to reflect life-threatening conditions may be higher in subjects with high socioeconomic position. These findings are also relevant to the debate about whether the use of SRH, rather than mortality, leads to an underestimate or overestimate of socioeconomic inequalities in health (8–12). In theory, if there is socioeconomic variation in the validity of SRH to reflect life-threatening conditions, the magnitude of socioeconomic inequalities in health should be different depending on whether SRH or mortality is used. However, the information available is not conclusive. In one study, socioeconomic inequalities in health showed the same magnitude for SRH and for mortality (9). In addition, in other studies, inequalities using one or another measure of health gave different results depending on the indicator of socioeconomic position used or the SRH cut-off point (10, 11). In any case, results for SRH and for mortality are not comparable because SRH reports were obtained at the end of a follow-up period. SRH was reported only by those who were alive at the end of this periodda highly health-selected group; thus, SRH and mortality were measured in different samples. The present work uses information from a cohort of older residents in Spain: first, to assess the socioeconomic 1047-2797/10/$–see front matter doi:10.1016/j.annepidem.2010.01.007
396
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
variation in the association between SRH and mortality after an 8-year follow-up period and second, to determine whether the magnitude of socioeconomic inequalities in SRH is different from the magnitude of socioeconomic inequalities in mortality.
METHODS Study Population A cohort of 4,008 persons, representative of the noninstitutionalized Spanish population 60 years of age or older, was selected between 1 October 2000 and 31 March 2001 and was followed prospectively up to 31 October 2008. At baseline subjects were selected using probabilistic sampling with multistage clusters. The clusters were obtained according to region of residence and size of municipality. Census sections were then chosen randomly within each cluster, and the households in which information was finally obtained from the subjects were chosen within each section. Study participants were selected in age and sex strata. The study response rate was 71%. Baseline information was collected in the home through personal interviews and physical examination by trained and certified personnel. Informed consent was obtained in writing from each study participant and from an accompanying family member. The study was approved by the Clinical Research Ethics Committee of ‘‘La Paz’’ University Hospital in Madrid, Spain.
Measures At baseline, people were asked to rate their health by answering the following question: ‘‘How is your health in general?’’ The response categories were ‘‘excellent,’’ ‘‘very good,’’ ‘‘good,’’ ‘‘fair,’’ and ‘‘poor’’. This variable was grouped into two categories for the analysis: ‘‘good (excellent/very good/good)’’ and ‘‘poor (fair/poor)’’ SRH. Information was also collected at baseline on several non–life-threatening conditionsdosteoarthritis, cataract, and depressiondand life-threatening conditionsddiabetes, myocardial infarction and other serious heart conditions, obesity, and hypertension. The presence of osteoarthritis, cataract, depression, diabetes, and heart conditions was self-reported; subjects were shown a list of various diseases and were asked whether a physician had ever told them they suffered any of them; respondents replied ‘‘yes’’ or ‘‘no’’ for each disease. Information on obesity and hypertension was obtained by physical examination. The measurement methods have been reported elsewhere (13, 14). A study participant was considered to be obese when waist circumference was greater than 102 cm in men and 88 cm in women. Subjects were deemed to be hypertensive when their systolic blood pressure was >140 mm Hg or
AEP Vol. 20, No. 5 May 2010: 395–400
their diastolic blood pressure was >90 mm Hg, or if they were currently receiving antihypertensive drug treatment. Mortality data were obtained from the National Death Index, a computerized database with information on the vital status of all residents in Spain (15). Up to October 31, 2008, the vital status of 3,991 individuals (99.6% of the cohort) had been identified. A total of 972 deaths occurred during this period. Education was chosen as the indicator of socioeconomic position because information on this variable was available for most subjects. This is especially important in our study since the majority of subjects are outside the labor market. Moreover, education is an important discriminator of individual health status (16). Educational level was grouped into three categories based on the subject’s highest educational level attained: lowdilliteracy and incomplete primary education; mediumdcompleted primary education; and highdcompleted secondary or higher level education.
Statistical Analysis Of the initial cohort, 5% of subjects were excluded from the analysis because of lack of data on education or SRH. Men and women were analyzed separately. First, we calculated the age-adjusted mortality in each educational group for people with good SRH and those with poor SRH. The age-adjusted rate (per 100 person-years) was calculated by the direct method using the age distribution of the European standard population. We assessed the association between SRH and mortality using the Cox proportional hazard method of regression. The age-adjusted relative risk of mortality was estimated separately in each of the three educational groups. To evaluate whether the difference in the age-adjusted relative risk of mortality between educational groups was significant, we then estimated the possible interaction between SRH and education in a single regression model, by introducing an interaction term between SRH and education. The difference in the association between SRH and mortality may be due to variation in the validity of SRH to reflect life-threatening conditions by educational level. It may be that in some socioeconomic groups poor SRH mainly reflects the presence of life-threatening conditions, while in others it mainly reflects the presence of non–life-threatening conditions. To evaluate the validity of SRH to reflect each health condition, we estimated the likelihood ratio for poor SRH in each educational group. The likelihood ratio represents the probability of reporting poor SRH among individuals who have a particular health condition (sensitivity) divided by the probability of declaring poor SRH among individuals who do not have this health condition (1–specificity) (17). That is, the likelihood ratio reflects how many times
AEP Vol. 20, No. 5 May 2010: 395–400
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
more probable it is that poor SRH has been reported in the presence of a particular health condition than in its absence. Finally, socioeconomic inequalities were evaluated using SRH and mortality. Inequalities in SRH were estimated by the age-adjusted prevalence ratio for poor SRH by educational level. The direct method and the age distribution of the European standard population were used. The confidence intervals for the prevalence ratios were calculated by the Mantel-Haenszel method (18). Inequalities in mortality were estimated by the age-adjusted relative risk of mortality by educational level using Cox regression analysis.
397
is seen in subjects with high education, except for heart disease in men where it is seen in those with medium education; the highest likelihood ratio in non–life-threatening conditions is seen in subjects with low education. Socioeconomic inequalities were larger for SRH than for mortality (Table 5). The prevalence ratio for poor SRH in subjects with low versus high education was 1.63 in men and 1.45 in women. In contrast, the mortality rate ratio was 1.07 in men and 1.30 in women.
DISCUSSION RESULTS Table 1 shows the distribution of baseline SRH by education and of mortality by SRH and education. In persons with good SRH the highest mortality is found in those with low education, and the lowest mortality is found in those with high education. In subjects with poor SRH, however, no clear pattern is observed in the percentage of deaths by educational level. The association between SRH and mortality across educational categories is presented in Table 2. In both sexes, the highest and lowest relative risk of mortality was found in persons with the highest and lowest educational level, respectively. The magnitude of the relative risk of mortality was 3.24, 1.77, and 1.62 in men and 2.25, 1.65, and 1.50 in women with high, medium, and low education, respectively. In the association between SRH and mortality there was evidence of interaction between SRH and education in men (p Z 0.07), but not in women. Table 3 shows that a large number of subjects free of life-threatening and non–life-threatening conditions report poor SRH. The sensitivity, specificity, and likelihood ratio of SRH to reflect these conditions are presented in Table 4. The highest likelihood ratio for life-threatening conditions
With regard to the first objective, our study showed that the association between SRH and mortality was strongest in persons with high education. Although the confidence intervals overlap for the relative risk of mortality in the different educational groups, and no interaction between SRH and education was found in predicting mortality in women, the consistency of our findings in men and women provides additional evidence of socioeconomic differences in the ability of SRH to predict mortality. Three of the five studies that evaluated this relation also found the strongest association in subjects of high socioeconomic position, whether the indicator used was social class, education, or income (3–5). These findings can probably be attributed to the fact that among subjects with poor SRH those in higher socioeconomic position experience relatively higher mortality. This has been found in the only two studies of this type that reported mortality rates: one study in Sweden (3) and our own study in Spain. In both studies the mortality rate showed an inverse socioeconomic gradient in subjects with good SRH, but no gradient was seen in those with poor SRH. In the other two of the five studies on the relation between SRH and mortality, the results were different. One of them, also conducted in Sweden, found no
TABLE 1. Number of subjects with ‘‘good’’ and ‘‘poor’’ self-rated health and mortality over an 8-year follow-up, by sex and educational level* Follow-upy
At baseline Educational level Men High Medium Low Women High Medium Low
No.
No. (%) with good SRH
No. (%) with poor SRH
No. (%) with good SRH who died during follow-up
No. (%) with poor SRH who died during follow-up
308 600 743
232 (75.3) 355 (59.2) 386 (52.0)
76 (24.7) 245 (40.8) 357 (48.0)
41 (17.7) 71 (20.0) 89 (23.1)
35 (43.1) 76 (31.0) 129 (36.1)
177 736 1245
111 (62.7) 367 (49.9) 542 (43.5)
66 (37.3) 369 (50.1) 703 (56.5)
10 (9.0) 49 (13.4) 108 (20.0)
18 (27.3) 81 (22.0) 191 (27.2)
SRH Z self-rated health. *Subjects were persons 60 years of age and older in Spain. y Losses to follow-up: Five men and one woman with high education; four men with medium education; one man and five women with low education.
398
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
TABLE 2. Age-adjusted mortality rate per 100 person-years and relative risk of mortality for poor versus good SRH by educational level Good SRH Educational level Men High Medium Low Women High Medium Low
Poor SRH
Rate
Personyears
Rate
Relative risk of mortality (95% CI)
1599.1 2458.8 2659.6
2.8 3.2 3.4
434.3 1528.2 2233.9
8.0 5.4 5.3
3.24 (2.04–5.14) 1.77 (1.28–2.46) 1.62 (1.24–2.13)
820.4 2655.2 3754.0
1.8 1.7 2.1
439.4 2522.7 4610.8
3.5 3.0 3.4
2.25 (0.99–5.11) 1.65 (1.15–2.35) 1.50 (1.18–1.90)
Personyears
SRH Z self-rated health; CI Z confidence interval.
socioeconomic variation in the relation between SRH and mortality (7), even though the data source was the same as that used in the study mentioned above (3). In addition, another investigation, in a cohort of workers in a company in France, found that the association between SRH and mortality was higher in subjects with low socioeconomic position (6). However, these findings have been attributed to selection bias resulting from downward internal mobility TABLE 3. Association between SRH and seven health problems,* by sex Health problems Men SRH Life-threatening conditions Abdominal obesity Poor Good Hypertension Poor Good Diabetes Poor Good Heart disease Poor Good Non–life-threatening conditions Osteoarthritis Poor Good Cataract Poor Good Depression Poor Good
Women
Yes
No
Yes
No
302 440
309 485
822 727
204 232
430 614
224 344
780 657
328 341
128 108
546 453
225 121
898 883
93 42
584 923
66 22
1053 992
360 316
310 643
855 558
263 438
141 174
526 790
300 220
810 779
60 24
616 942
222 93
901 917
*The total number of subjects varies for each health problem because of different missing values.
AEP Vol. 20, No. 5 May 2010: 395–400
within the company in response to a limiting illness occurring during employment (8). Our findings may be due to higher validity of SRH to reflect life-threatening health conditions in subjects with higher education. As has been noted, the likelihood ratio for life-threatening health conditions is highest in subjects with high education. This could be due to a greater awareness of such disorders by these subjects. In contrast, the validity of SRH to reflect non–life-threatening health conditions is higher in subjects with low education. It is possible that, when reporting poor SRH, individuals with low education emphasize health dimensions other than life-threatening problems. This would explain why the association between SRH and mortality is weakest in subjects with low education. And the opposite would occur in subjects with high education. With regard to the second objective, our findings show that health inequalities are larger for poor SRH than for mortality. These findings were to be expected because in developed countries the prevalence of both poor SRH and mortality is higher in persons with lower socioeconomic position. However, mortality inequalities are smaller than inequalities in poor SRH (19–21). Thus, it is surprising that this matter is still controversial (8–12). In several studies SRH is evaluated at the end of the follow-up period, with the result that different population samples are used to estimate inequalities in poor SRH and in mortality (9–11). Inequalities in poor SRH should have been assessed at baseline, since the controversy was whether socioeconomic variations in the prediction of mortality by SRH indicated that SRH misestimates health inequalities. The findings of the likelihood ratio of SRH to reflect the different health conditions suggest that the greater frequency of poor SRH in the low educational group is attributable to non–life-threatening conditions. Likewise, the lower validity of SRH to reflect life-threatening conditions in this group may explain why inequality in poor SRH is greater than inequality in mortality. On the other hand, the findings of this study about the validity of SRH raise doubts about the appropriateness of SRH to assess health inequalities, since SRH does not reflect the same phenomena in subjects with different socioeconomic position. This does not mean that inequalities in mortality are the ‘‘only’’ and ‘‘true’’ inequalities in health, but that, given the variety of health problems captured by SRH, the estimate of socioeconomic inequalities in SRH may not be meaningful without an explicit previous description of the type of health conditions that SRH is meant to measure. Some methodological aspects of this study should also be noted. An important strength is that both the large cohort size and the long follow-up duration allowed for a sufficient number of deaths to obtain precise mortality estimates in each educational stratum in men and women. In addition,
AEP Vol. 20, No. 5 May 2010: 395–400
399
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
TABLE 4. Sensitivity, specificity, and likelihood ratio of poor SRH to reflect life-threatening and non–life-threatening conditions, by education and sex Specificity (%)y
Sensitivity* (%) Educational level Life-threatening conditions Abdominal obesity High Medium Low Hypertension High Medium Low Diabetes High Medium Low Heart disease High Medium Low Non–life-threatening conditions Osteoarthritis High Medium Low Cataract High Medium Low Depression High Medium Low
Likelihood ratioz
Men
Women
Men
Women
Men
Women
23.3 39.0 48.7
41.9 50.9 55.6
78.6 60.6 53.9
69.4 57.5 45.8
1.09 0.99 1.06
1.37 1.20 1.03
30.2 53.0 46.0
46.8 50.9 57.2
81.3 45.7 50.0
80.8 51.4 45.3
1.61 0.98 0.92
2.44 1.05 1.05
59.8 51.8 54.3
65.0 60.5 66.9
80.7 51.4 52.5
65.1 51.3 45.9
3.10 1.07 1.14
1.86 1.24 1.24
43.8 81.9 67.1
100 73.2 75.8
76.2 62.5 53.9
63.0 51.0 44.8
1.84 2.18 1.46
2.70 1.49 1.37
38.2 54.4 56.7
50.9 59.2 62.5
60.1 67.1 81.7
56.0 67.8 79.2
0.96 1.65 3.10
1.16 1.84 3.00
24.6 50.0 50.0
44.1 59.7 58.0
52.8 61.3 74.9
44.1 53.3 64.3
0.52 1.29 1.99
0.79 1.28 1.62
65.6 60.1 82.3
63.5 70.7 71.2
53.4 60.6 76.7
46.2 53.4 67.3
1.41 1.53 3.53
1.18 1.52 2.18
*Sensitivity is the probability of correctly identifying a diseased person by self-rated health; it is calculated as the percentage of subjects with the health condition who reported poor self-rated health. y Specificity is the probability of correctly identifying a non-diseased person by self-rated health; it is calculated as the percentage of subjects without the health condition who did not report poor self-rated health. z Likelihood ratio: sensitivity/(100–specificity).
the losses to follow-up were so few that their influence on the results must have been insignificant. One of the surprising findings is the lack of socioeconomic differences in mortality. Several investigations in different European populations have found that the magnitude of the differences in mortality by educational level is smaller in Spain and Italy than in other countries in Europe (19– 21). Moreover, in the early 1990s no socioeconomic differences in mortality were observed in men 45 to 64 years of age residing in Barcelona (22) or in women 45 to 64 years of age residing in Madrid (23). In any case, a possible selection bias cannot be ruled out in this study. Since this was a cohort of people aged 60 years and older, mortality before that age in subjects with high education is likely to have been lower than in persons with medium and low education. That is, the frailest persons with low and medium education are likely to die before age 60, leaving more robust members to survive
to old age. This is one of the explanations for the blackwhite mortality crossover in the United States (24, 25) (throughout most of their lives, black persons have higher mortality than do white persons, but in later life the opposite occurs), and this may also be the reason why mortality among those with poor SRH in our study was lower in persons with low or medium education than in those with high education. Better socioeconomic conditions throughout life or better access to health services in persons with high education may have delayed the occurrence of life-threatening problems or may have reduced the likelihood of a fatal outcome once a life-threatening problem occurred. Information on obesity and hypertension was obtained by physical examination, and information on the other health problems studied was obtained by self-reports. Given that self-reports of diabetes, heart disease, osteoarthritis, and mental problems are moderately valid (26, 27), it can be
400
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
AEP Vol. 20, No. 5 May 2010: 395–400
TABLE 5. Age-adjusted prevalence and prevalence ratios for poor self-rated health and age-adjusted mortality raters per 100 personyears and relative risk of mortality, by educational level Poor SRH Educational level Men High Medium Low Women High Medium Low
Mortality
Prevalence (%)
Prevalence ratio (95% CI)
Rate
Relative risk of mortality (95% CI)
22.2 32.6 36.1
1.00 1.47 (1.11–1.95) 1.63 (1.24–2.13)
4.0 4.0 4.3
1.00 1.02 (0.77–1.34) 1.07 (0.82–1.39)
44.3 53.2 64.4
1.00 1.20 (0.94–1.53) 1.45 (1.15–1.83)
2.4 2.4 2.9
1.00 1.10 (0.74–1.65) 1.30 (0.89–1.91)
CI Z confidence interval.
inferred that the sensitivity and specificity of SRH to reflect the health problems estimated in this study might be a good proxy for the true sensitivity and specificity. In conclusion, the ability of SRH to predict mortality is highest in subjects with high education, and inequalities in poor SRH are greater than inequalities in mortality. Socioeconomic variation in the validity of SRH to reflect life-threatening and non–life-threatening conditions may underlie these findings. REFERENCES 1. Idler EL, Benyamini Y. Self-rated health and mortality: a review of twentyseven community studies. J Health Soc Behav. 1997;38:21–37. 2. DeSalvo KB, Bloser N, Reynolds K, He J, Muntner P. Mortality prediction with a single general self-rated health question: a meta-analysis. J Gen Intern Med. 2006;21:267–275. 3. Burstrom B, Fredlund P. Self rated health: is it as good a predictor of subsequent mortality among adults in lower as well as in higher social classes? J Epidemiol Community Health. 2001;55:836. 4. Huisman M, van Lenthe F, Mackenbach J. The predictive ability of selfassessed health for mortality in different educational groups. Int J Epidemiol. 2007;36:1207–1213. 5. Beam Dowd J, Zajacova A. Does the predictive power of self-rated health for subsequent mortality risk vary by socioeconomic status in the US? Int J Epidemiol. 2007;36:1214–1221. 6. Singh-Manoux A, Dugravot A, Shipley MJ, Ferrie JE, Martikainen P, Goldberg M, et al. The association between self-related health and mortality in different socioeconomic groups in the Gazel Cohort study. Int J Epidemiol. 2007;36:1222–1228. 7. Van Doorslaer E, Gerdtham U-G. Does inequality in self-assessed health predict inequality in survival by income? Evidence from Swedish data. Soc Sci Med. 2003;57:1621–1629. 8. Quesnel-Valle´e A. Self-rated health: caught in the crossfire of the quest for ‘true’ health? Int J Epidemiol. 2007;36:1161–1164. 9. Subramanian SV, Ertel K. Is the use of self-rated measures to assess health inequalities misleading? Int J Epidemiol. 2008;37:1436–1437. 10. Huisman M, Van Lenthe F, Mackenbach JP. Is the use of self-rated measures to assess health inequalities misleading? Int J Epidemiol. 2008;37:1437–1438. 11. Singh-Manoux A, Shipley MJ, Zins M, Ferrie JE. Is the use of self-rated measures to assess health inequalities misleading? Int J Epidemiol. 2008;37:1439–1440. 12. Dowd JB, Zajacova A. Is the use of self-rated measures to assess health inequalities misleading? Int J Epidemiol. 2008;37:1438–1439.
13. Gutie´rrez-Fisac JL, Lo´pez E, Banegas JR, Graciani A, Rodrı´guez-Artalejo F. Prevalence of overweight and obesity in elderly people in Spain. Obes Res. 2004;12:710–715. 14. Regidor E, Banegas JR, Gutie´rrez-Fisac JL, Domı´nguez V, Rodrı´guez-Artalejo F. Influence of childhood socioeconomic circumstances, height, and obesity on pulse pressure and systolic and diastolic blood pressure in older people. J Hum Hypertens. 2006;20:73–82. 15. Navarro C. [The National Death Index: a largely expected advance in the access to mortality data.]. Gac Sanit. 2006;20:421–423 In Spanish. 16. Galobardes B, Shaw ME, Lawlor DE, Lynch JW, Davey Smith G. Indicators of socioeconomic position (I). J Epidemiol Community Health. 2006;60:7–12. 17. Porta P, ed. A dictionary of epidemiology. 5th ed. New York: Oxford University Press; 2008:145. 18. Greenland S, Rothman KJ. Introduction to stratified analysis. In: Rothman KJ, Greenland S, eds. Modern epidemiology. Philadelphia: Lippincott Williams & Wilkins; 1998:253–280. 19. Mackenbach JP, Kunst AE, Cavelaars AEJM, Groenhof F, Geurts JJM, EU Working Group on Socioeconomic Inequalities in Health. Socioeconomic inequalities in morbidity and mortality in Western Europe. Lancet. 1997;349:1655–1659. 20. Dalstra AA, Kunst AE, Mackenbach JP, EU Working Group on Socioeconomic Inequalities in Health. A comparative appraisal of the relationship of education, income and housing tenure with less than good health among the elderly in Europe. Soc Sci Med. 2006;62:2046–2060. 21. Huisman M, Kunst AE, Andersen O, Bopp M, Borgan J-K, Borrell C, et al. Socioeconomic inequalities in mortality among the elderly in eleven European populations. J Epidemiol Community Health. 2004;58:468–475. 22. Puigpino´s R, Borrell C, Pasarı´n MI, Montella` N, Pe´rez G, Plase`ncia A, et al. Inequalities in mortality by social class in men in Barcelona, Spain. Eur J Epidemiol. 2000;16:751–756. 23. Regidor E, De la Fuente L, Calle ME, Navarro P, Domı´nguez V. Unusually strong association between education and mortality in young adults in a community with a high rate of injection-drug users. Eur J Public Health. 2003;13:334–339. 24. Liu X, Witten M. A biologically based explanation for mortality crossover in human populations. Gerontologist. 1995;35:609–615. 25. Lynch SM, Brown JS, Harmsen KG. Black-white differences in mortality compression and deceleration and the mortality crossover reconsidered. Res Aging. 2003;25:456–483. 26. Helio¨vaara M, Aromaa A, Klaukka T, Knekt P, Joukamaa M, Impivaara O. Reliability and validity of interview data on chronic diseases. The MiniFinland Health Survey. J Clin Epidemiol. 1993;46:181–191. 27. Van der Velden J, Abrahamse H, Donker G, Van der Steen J, Van Sonsbeek J, Van den Bos G. What do health interview surveys tell us about the prevalence of somatic chronic diseases? Eur J Public Health. 1998;8:52–58.
PURPOSE: To assess socioeconomic variation in the association between self-rated health (SRH) and mortality and to determine whether socioeconomic inequalities in SRH and socioeconomic inequalities in mortality differ in magnitude. METHODS: We used data from a cohort of Spanish people 60 years of age and older with an 8-year follow-up of mortality. The association between SRH at baseline and mortality was estimated by the age-adjusted relative risk of mortality in people with low, medium, and high education. The measures of health inequalities were the prevalence ratio of poor SRH and the age-adjusted relative risk of mortality according to educational level. The validity of SRH to reflect life-threatening and non–life-threatening health conditions was summarized with the likelihood ratio for poor SRH in each educational category. RESULTS: The relative risk of mortality according to SRH in subjects with high and low education was 3.24 and 1.62 in men and 2.25 and 1.50 in women, respectively. Inequalities in poor self-rated health were larger than inequalities in mortality: –1.63 versus 1.07 in men and 1.45 versus 1.30 in women. The highest likelihood ratio for SRH was seen in persons with high education in the case of life-threatening conditions, and for those with low education, in the case of non–life-threatening conditions. CONCLUSIONS: Socioeconomic variation in the validity of SRH to reflect life-threatening and non–life-threatening conditions could explain the greater ability of SRH to predict mortality in persons with high education and why inequalities in poor SRH are larger than inequalities in mortality. Ann Epidemiol 2010;20:395–400. Ó 2010 Elsevier Inc. All rights reserved. KEY WORDS:
Self-rated Health, Mortality, Education, Health Inequalities, Health Problems.
INTRODUCTION Many investigations have shown a strong association between self-rated health (SRH) and mortality (1, 2). In contrast, only a few studies have evaluated whether this relation varies by socioeconomic position. Four of five studies on this issue found a socioeconomic variation in the relation between SRH and mortality (3–7). Three studies in the general population of Sweden, The Netherlands, and the United States, respectively, observed that the relative risk of mortality according to SRH was highest in subjects in the highest socioeconomic position (3–5). However, another study, conducted in an occupational cohort in France, found the highest relative risk for mortality in those with the lowest socioeconomic position (6). Since the results in the French cohort could have been affected by From the Department of Preventive Medicine and Public Health, Universidad Complutense de Madrid (E.R.); CIBER Epidemiologı´a y Salud Pu´blica (CIBERESP) (E.R., P.G-C., J.L.G-F., J.R.B., F.R-A); and the Department of Preventive Medicine and Public Health, Universidad Auto´noma de Madrid (P.G-C., J.L.G-F., J.R.B., F.R-A), Spain. Address correspondence to: Enrique Regidor, Department of Preventive Medicine and Public Health, Universidad Complutense de Madrid, Ciudad Universitaria, 28040 Madrid, Spain. Tel: þ34 913941521. E-mail: [email protected]. Received September 15, 2009; accepted January 30, 2010. Ó 2010 Elsevier Inc. All rights reserved. 360 Park Avenue South, New York, NY 10010
different selection biases (8), the consistency of the findings of the other three studies suggests that the validity of SRH to reflect life-threatening conditions may be higher in subjects with high socioeconomic position. These findings are also relevant to the debate about whether the use of SRH, rather than mortality, leads to an underestimate or overestimate of socioeconomic inequalities in health (8–12). In theory, if there is socioeconomic variation in the validity of SRH to reflect life-threatening conditions, the magnitude of socioeconomic inequalities in health should be different depending on whether SRH or mortality is used. However, the information available is not conclusive. In one study, socioeconomic inequalities in health showed the same magnitude for SRH and for mortality (9). In addition, in other studies, inequalities using one or another measure of health gave different results depending on the indicator of socioeconomic position used or the SRH cut-off point (10, 11). In any case, results for SRH and for mortality are not comparable because SRH reports were obtained at the end of a follow-up period. SRH was reported only by those who were alive at the end of this periodda highly health-selected group; thus, SRH and mortality were measured in different samples. The present work uses information from a cohort of older residents in Spain: first, to assess the socioeconomic 1047-2797/10/$–see front matter doi:10.1016/j.annepidem.2010.01.007
396
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
variation in the association between SRH and mortality after an 8-year follow-up period and second, to determine whether the magnitude of socioeconomic inequalities in SRH is different from the magnitude of socioeconomic inequalities in mortality.
METHODS Study Population A cohort of 4,008 persons, representative of the noninstitutionalized Spanish population 60 years of age or older, was selected between 1 October 2000 and 31 March 2001 and was followed prospectively up to 31 October 2008. At baseline subjects were selected using probabilistic sampling with multistage clusters. The clusters were obtained according to region of residence and size of municipality. Census sections were then chosen randomly within each cluster, and the households in which information was finally obtained from the subjects were chosen within each section. Study participants were selected in age and sex strata. The study response rate was 71%. Baseline information was collected in the home through personal interviews and physical examination by trained and certified personnel. Informed consent was obtained in writing from each study participant and from an accompanying family member. The study was approved by the Clinical Research Ethics Committee of ‘‘La Paz’’ University Hospital in Madrid, Spain.
Measures At baseline, people were asked to rate their health by answering the following question: ‘‘How is your health in general?’’ The response categories were ‘‘excellent,’’ ‘‘very good,’’ ‘‘good,’’ ‘‘fair,’’ and ‘‘poor’’. This variable was grouped into two categories for the analysis: ‘‘good (excellent/very good/good)’’ and ‘‘poor (fair/poor)’’ SRH. Information was also collected at baseline on several non–life-threatening conditionsdosteoarthritis, cataract, and depressiondand life-threatening conditionsddiabetes, myocardial infarction and other serious heart conditions, obesity, and hypertension. The presence of osteoarthritis, cataract, depression, diabetes, and heart conditions was self-reported; subjects were shown a list of various diseases and were asked whether a physician had ever told them they suffered any of them; respondents replied ‘‘yes’’ or ‘‘no’’ for each disease. Information on obesity and hypertension was obtained by physical examination. The measurement methods have been reported elsewhere (13, 14). A study participant was considered to be obese when waist circumference was greater than 102 cm in men and 88 cm in women. Subjects were deemed to be hypertensive when their systolic blood pressure was >140 mm Hg or
AEP Vol. 20, No. 5 May 2010: 395–400
their diastolic blood pressure was >90 mm Hg, or if they were currently receiving antihypertensive drug treatment. Mortality data were obtained from the National Death Index, a computerized database with information on the vital status of all residents in Spain (15). Up to October 31, 2008, the vital status of 3,991 individuals (99.6% of the cohort) had been identified. A total of 972 deaths occurred during this period. Education was chosen as the indicator of socioeconomic position because information on this variable was available for most subjects. This is especially important in our study since the majority of subjects are outside the labor market. Moreover, education is an important discriminator of individual health status (16). Educational level was grouped into three categories based on the subject’s highest educational level attained: lowdilliteracy and incomplete primary education; mediumdcompleted primary education; and highdcompleted secondary or higher level education.
Statistical Analysis Of the initial cohort, 5% of subjects were excluded from the analysis because of lack of data on education or SRH. Men and women were analyzed separately. First, we calculated the age-adjusted mortality in each educational group for people with good SRH and those with poor SRH. The age-adjusted rate (per 100 person-years) was calculated by the direct method using the age distribution of the European standard population. We assessed the association between SRH and mortality using the Cox proportional hazard method of regression. The age-adjusted relative risk of mortality was estimated separately in each of the three educational groups. To evaluate whether the difference in the age-adjusted relative risk of mortality between educational groups was significant, we then estimated the possible interaction between SRH and education in a single regression model, by introducing an interaction term between SRH and education. The difference in the association between SRH and mortality may be due to variation in the validity of SRH to reflect life-threatening conditions by educational level. It may be that in some socioeconomic groups poor SRH mainly reflects the presence of life-threatening conditions, while in others it mainly reflects the presence of non–life-threatening conditions. To evaluate the validity of SRH to reflect each health condition, we estimated the likelihood ratio for poor SRH in each educational group. The likelihood ratio represents the probability of reporting poor SRH among individuals who have a particular health condition (sensitivity) divided by the probability of declaring poor SRH among individuals who do not have this health condition (1–specificity) (17). That is, the likelihood ratio reflects how many times
AEP Vol. 20, No. 5 May 2010: 395–400
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
more probable it is that poor SRH has been reported in the presence of a particular health condition than in its absence. Finally, socioeconomic inequalities were evaluated using SRH and mortality. Inequalities in SRH were estimated by the age-adjusted prevalence ratio for poor SRH by educational level. The direct method and the age distribution of the European standard population were used. The confidence intervals for the prevalence ratios were calculated by the Mantel-Haenszel method (18). Inequalities in mortality were estimated by the age-adjusted relative risk of mortality by educational level using Cox regression analysis.
397
is seen in subjects with high education, except for heart disease in men where it is seen in those with medium education; the highest likelihood ratio in non–life-threatening conditions is seen in subjects with low education. Socioeconomic inequalities were larger for SRH than for mortality (Table 5). The prevalence ratio for poor SRH in subjects with low versus high education was 1.63 in men and 1.45 in women. In contrast, the mortality rate ratio was 1.07 in men and 1.30 in women.
DISCUSSION RESULTS Table 1 shows the distribution of baseline SRH by education and of mortality by SRH and education. In persons with good SRH the highest mortality is found in those with low education, and the lowest mortality is found in those with high education. In subjects with poor SRH, however, no clear pattern is observed in the percentage of deaths by educational level. The association between SRH and mortality across educational categories is presented in Table 2. In both sexes, the highest and lowest relative risk of mortality was found in persons with the highest and lowest educational level, respectively. The magnitude of the relative risk of mortality was 3.24, 1.77, and 1.62 in men and 2.25, 1.65, and 1.50 in women with high, medium, and low education, respectively. In the association between SRH and mortality there was evidence of interaction between SRH and education in men (p Z 0.07), but not in women. Table 3 shows that a large number of subjects free of life-threatening and non–life-threatening conditions report poor SRH. The sensitivity, specificity, and likelihood ratio of SRH to reflect these conditions are presented in Table 4. The highest likelihood ratio for life-threatening conditions
With regard to the first objective, our study showed that the association between SRH and mortality was strongest in persons with high education. Although the confidence intervals overlap for the relative risk of mortality in the different educational groups, and no interaction between SRH and education was found in predicting mortality in women, the consistency of our findings in men and women provides additional evidence of socioeconomic differences in the ability of SRH to predict mortality. Three of the five studies that evaluated this relation also found the strongest association in subjects of high socioeconomic position, whether the indicator used was social class, education, or income (3–5). These findings can probably be attributed to the fact that among subjects with poor SRH those in higher socioeconomic position experience relatively higher mortality. This has been found in the only two studies of this type that reported mortality rates: one study in Sweden (3) and our own study in Spain. In both studies the mortality rate showed an inverse socioeconomic gradient in subjects with good SRH, but no gradient was seen in those with poor SRH. In the other two of the five studies on the relation between SRH and mortality, the results were different. One of them, also conducted in Sweden, found no
TABLE 1. Number of subjects with ‘‘good’’ and ‘‘poor’’ self-rated health and mortality over an 8-year follow-up, by sex and educational level* Follow-upy
At baseline Educational level Men High Medium Low Women High Medium Low
No.
No. (%) with good SRH
No. (%) with poor SRH
No. (%) with good SRH who died during follow-up
No. (%) with poor SRH who died during follow-up
308 600 743
232 (75.3) 355 (59.2) 386 (52.0)
76 (24.7) 245 (40.8) 357 (48.0)
41 (17.7) 71 (20.0) 89 (23.1)
35 (43.1) 76 (31.0) 129 (36.1)
177 736 1245
111 (62.7) 367 (49.9) 542 (43.5)
66 (37.3) 369 (50.1) 703 (56.5)
10 (9.0) 49 (13.4) 108 (20.0)
18 (27.3) 81 (22.0) 191 (27.2)
SRH Z self-rated health. *Subjects were persons 60 years of age and older in Spain. y Losses to follow-up: Five men and one woman with high education; four men with medium education; one man and five women with low education.
398
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
TABLE 2. Age-adjusted mortality rate per 100 person-years and relative risk of mortality for poor versus good SRH by educational level Good SRH Educational level Men High Medium Low Women High Medium Low
Poor SRH
Rate
Personyears
Rate
Relative risk of mortality (95% CI)
1599.1 2458.8 2659.6
2.8 3.2 3.4
434.3 1528.2 2233.9
8.0 5.4 5.3
3.24 (2.04–5.14) 1.77 (1.28–2.46) 1.62 (1.24–2.13)
820.4 2655.2 3754.0
1.8 1.7 2.1
439.4 2522.7 4610.8
3.5 3.0 3.4
2.25 (0.99–5.11) 1.65 (1.15–2.35) 1.50 (1.18–1.90)
Personyears
SRH Z self-rated health; CI Z confidence interval.
socioeconomic variation in the relation between SRH and mortality (7), even though the data source was the same as that used in the study mentioned above (3). In addition, another investigation, in a cohort of workers in a company in France, found that the association between SRH and mortality was higher in subjects with low socioeconomic position (6). However, these findings have been attributed to selection bias resulting from downward internal mobility TABLE 3. Association between SRH and seven health problems,* by sex Health problems Men SRH Life-threatening conditions Abdominal obesity Poor Good Hypertension Poor Good Diabetes Poor Good Heart disease Poor Good Non–life-threatening conditions Osteoarthritis Poor Good Cataract Poor Good Depression Poor Good
Women
Yes
No
Yes
No
302 440
309 485
822 727
204 232
430 614
224 344
780 657
328 341
128 108
546 453
225 121
898 883
93 42
584 923
66 22
1053 992
360 316
310 643
855 558
263 438
141 174
526 790
300 220
810 779
60 24
616 942
222 93
901 917
*The total number of subjects varies for each health problem because of different missing values.
AEP Vol. 20, No. 5 May 2010: 395–400
within the company in response to a limiting illness occurring during employment (8). Our findings may be due to higher validity of SRH to reflect life-threatening health conditions in subjects with higher education. As has been noted, the likelihood ratio for life-threatening health conditions is highest in subjects with high education. This could be due to a greater awareness of such disorders by these subjects. In contrast, the validity of SRH to reflect non–life-threatening health conditions is higher in subjects with low education. It is possible that, when reporting poor SRH, individuals with low education emphasize health dimensions other than life-threatening problems. This would explain why the association between SRH and mortality is weakest in subjects with low education. And the opposite would occur in subjects with high education. With regard to the second objective, our findings show that health inequalities are larger for poor SRH than for mortality. These findings were to be expected because in developed countries the prevalence of both poor SRH and mortality is higher in persons with lower socioeconomic position. However, mortality inequalities are smaller than inequalities in poor SRH (19–21). Thus, it is surprising that this matter is still controversial (8–12). In several studies SRH is evaluated at the end of the follow-up period, with the result that different population samples are used to estimate inequalities in poor SRH and in mortality (9–11). Inequalities in poor SRH should have been assessed at baseline, since the controversy was whether socioeconomic variations in the prediction of mortality by SRH indicated that SRH misestimates health inequalities. The findings of the likelihood ratio of SRH to reflect the different health conditions suggest that the greater frequency of poor SRH in the low educational group is attributable to non–life-threatening conditions. Likewise, the lower validity of SRH to reflect life-threatening conditions in this group may explain why inequality in poor SRH is greater than inequality in mortality. On the other hand, the findings of this study about the validity of SRH raise doubts about the appropriateness of SRH to assess health inequalities, since SRH does not reflect the same phenomena in subjects with different socioeconomic position. This does not mean that inequalities in mortality are the ‘‘only’’ and ‘‘true’’ inequalities in health, but that, given the variety of health problems captured by SRH, the estimate of socioeconomic inequalities in SRH may not be meaningful without an explicit previous description of the type of health conditions that SRH is meant to measure. Some methodological aspects of this study should also be noted. An important strength is that both the large cohort size and the long follow-up duration allowed for a sufficient number of deaths to obtain precise mortality estimates in each educational stratum in men and women. In addition,
AEP Vol. 20, No. 5 May 2010: 395–400
399
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
TABLE 4. Sensitivity, specificity, and likelihood ratio of poor SRH to reflect life-threatening and non–life-threatening conditions, by education and sex Specificity (%)y
Sensitivity* (%) Educational level Life-threatening conditions Abdominal obesity High Medium Low Hypertension High Medium Low Diabetes High Medium Low Heart disease High Medium Low Non–life-threatening conditions Osteoarthritis High Medium Low Cataract High Medium Low Depression High Medium Low
Likelihood ratioz
Men
Women
Men
Women
Men
Women
23.3 39.0 48.7
41.9 50.9 55.6
78.6 60.6 53.9
69.4 57.5 45.8
1.09 0.99 1.06
1.37 1.20 1.03
30.2 53.0 46.0
46.8 50.9 57.2
81.3 45.7 50.0
80.8 51.4 45.3
1.61 0.98 0.92
2.44 1.05 1.05
59.8 51.8 54.3
65.0 60.5 66.9
80.7 51.4 52.5
65.1 51.3 45.9
3.10 1.07 1.14
1.86 1.24 1.24
43.8 81.9 67.1
100 73.2 75.8
76.2 62.5 53.9
63.0 51.0 44.8
1.84 2.18 1.46
2.70 1.49 1.37
38.2 54.4 56.7
50.9 59.2 62.5
60.1 67.1 81.7
56.0 67.8 79.2
0.96 1.65 3.10
1.16 1.84 3.00
24.6 50.0 50.0
44.1 59.7 58.0
52.8 61.3 74.9
44.1 53.3 64.3
0.52 1.29 1.99
0.79 1.28 1.62
65.6 60.1 82.3
63.5 70.7 71.2
53.4 60.6 76.7
46.2 53.4 67.3
1.41 1.53 3.53
1.18 1.52 2.18
*Sensitivity is the probability of correctly identifying a diseased person by self-rated health; it is calculated as the percentage of subjects with the health condition who reported poor self-rated health. y Specificity is the probability of correctly identifying a non-diseased person by self-rated health; it is calculated as the percentage of subjects without the health condition who did not report poor self-rated health. z Likelihood ratio: sensitivity/(100–specificity).
the losses to follow-up were so few that their influence on the results must have been insignificant. One of the surprising findings is the lack of socioeconomic differences in mortality. Several investigations in different European populations have found that the magnitude of the differences in mortality by educational level is smaller in Spain and Italy than in other countries in Europe (19– 21). Moreover, in the early 1990s no socioeconomic differences in mortality were observed in men 45 to 64 years of age residing in Barcelona (22) or in women 45 to 64 years of age residing in Madrid (23). In any case, a possible selection bias cannot be ruled out in this study. Since this was a cohort of people aged 60 years and older, mortality before that age in subjects with high education is likely to have been lower than in persons with medium and low education. That is, the frailest persons with low and medium education are likely to die before age 60, leaving more robust members to survive
to old age. This is one of the explanations for the blackwhite mortality crossover in the United States (24, 25) (throughout most of their lives, black persons have higher mortality than do white persons, but in later life the opposite occurs), and this may also be the reason why mortality among those with poor SRH in our study was lower in persons with low or medium education than in those with high education. Better socioeconomic conditions throughout life or better access to health services in persons with high education may have delayed the occurrence of life-threatening problems or may have reduced the likelihood of a fatal outcome once a life-threatening problem occurred. Information on obesity and hypertension was obtained by physical examination, and information on the other health problems studied was obtained by self-reports. Given that self-reports of diabetes, heart disease, osteoarthritis, and mental problems are moderately valid (26, 27), it can be
400
Regidor et al. SOCIOECONOMIC VARIATION IN MORTALITY BY SELF-RATED HEALTH
AEP Vol. 20, No. 5 May 2010: 395–400
TABLE 5. Age-adjusted prevalence and prevalence ratios for poor self-rated health and age-adjusted mortality raters per 100 personyears and relative risk of mortality, by educational level Poor SRH Educational level Men High Medium Low Women High Medium Low
Mortality
Prevalence (%)
Prevalence ratio (95% CI)
Rate
Relative risk of mortality (95% CI)
22.2 32.6 36.1
1.00 1.47 (1.11–1.95) 1.63 (1.24–2.13)
4.0 4.0 4.3
1.00 1.02 (0.77–1.34) 1.07 (0.82–1.39)
44.3 53.2 64.4
1.00 1.20 (0.94–1.53) 1.45 (1.15–1.83)
2.4 2.4 2.9
1.00 1.10 (0.74–1.65) 1.30 (0.89–1.91)
CI Z confidence interval.
inferred that the sensitivity and specificity of SRH to reflect the health problems estimated in this study might be a good proxy for the true sensitivity and specificity. In conclusion, the ability of SRH to predict mortality is highest in subjects with high education, and inequalities in poor SRH are greater than inequalities in mortality. Socioeconomic variation in the validity of SRH to reflect life-threatening and non–life-threatening conditions may underlie these findings. REFERENCES 1. Idler EL, Benyamini Y. Self-rated health and mortality: a review of twentyseven community studies. J Health Soc Behav. 1997;38:21–37. 2. DeSalvo KB, Bloser N, Reynolds K, He J, Muntner P. Mortality prediction with a single general self-rated health question: a meta-analysis. J Gen Intern Med. 2006;21:267–275. 3. Burstrom B, Fredlund P. Self rated health: is it as good a predictor of subsequent mortality among adults in lower as well as in higher social classes? J Epidemiol Community Health. 2001;55:836. 4. Huisman M, van Lenthe F, Mackenbach J. The predictive ability of selfassessed health for mortality in different educational groups. Int J Epidemiol. 2007;36:1207–1213. 5. Beam Dowd J, Zajacova A. Does the predictive power of self-rated health for subsequent mortality risk vary by socioeconomic status in the US? Int J Epidemiol. 2007;36:1214–1221. 6. Singh-Manoux A, Dugravot A, Shipley MJ, Ferrie JE, Martikainen P, Goldberg M, et al. The association between self-related health and mortality in different socioeconomic groups in the Gazel Cohort study. Int J Epidemiol. 2007;36:1222–1228. 7. Van Doorslaer E, Gerdtham U-G. Does inequality in self-assessed health predict inequality in survival by income? Evidence from Swedish data. Soc Sci Med. 2003;57:1621–1629. 8. Quesnel-Valle´e A. Self-rated health: caught in the crossfire of the quest for ‘true’ health? Int J Epidemiol. 2007;36:1161–1164. 9. Subramanian SV, Ertel K. Is the use of self-rated measures to assess health inequalities misleading? Int J Epidemiol. 2008;37:1436–1437. 10. Huisman M, Van Lenthe F, Mackenbach JP. Is the use of self-rated measures to assess health inequalities misleading? Int J Epidemiol. 2008;37:1437–1438. 11. Singh-Manoux A, Shipley MJ, Zins M, Ferrie JE. Is the use of self-rated measures to assess health inequalities misleading? Int J Epidemiol. 2008;37:1439–1440. 12. Dowd JB, Zajacova A. Is the use of self-rated measures to assess health inequalities misleading? Int J Epidemiol. 2008;37:1438–1439.
13. Gutie´rrez-Fisac JL, Lo´pez E, Banegas JR, Graciani A, Rodrı´guez-Artalejo F. Prevalence of overweight and obesity in elderly people in Spain. Obes Res. 2004;12:710–715. 14. Regidor E, Banegas JR, Gutie´rrez-Fisac JL, Domı´nguez V, Rodrı´guez-Artalejo F. Influence of childhood socioeconomic circumstances, height, and obesity on pulse pressure and systolic and diastolic blood pressure in older people. J Hum Hypertens. 2006;20:73–82. 15. Navarro C. [The National Death Index: a largely expected advance in the access to mortality data.]. Gac Sanit. 2006;20:421–423 In Spanish. 16. Galobardes B, Shaw ME, Lawlor DE, Lynch JW, Davey Smith G. Indicators of socioeconomic position (I). J Epidemiol Community Health. 2006;60:7–12. 17. Porta P, ed. A dictionary of epidemiology. 5th ed. New York: Oxford University Press; 2008:145. 18. Greenland S, Rothman KJ. Introduction to stratified analysis. In: Rothman KJ, Greenland S, eds. Modern epidemiology. Philadelphia: Lippincott Williams & Wilkins; 1998:253–280. 19. Mackenbach JP, Kunst AE, Cavelaars AEJM, Groenhof F, Geurts JJM, EU Working Group on Socioeconomic Inequalities in Health. Socioeconomic inequalities in morbidity and mortality in Western Europe. Lancet. 1997;349:1655–1659. 20. Dalstra AA, Kunst AE, Mackenbach JP, EU Working Group on Socioeconomic Inequalities in Health. A comparative appraisal of the relationship of education, income and housing tenure with less than good health among the elderly in Europe. Soc Sci Med. 2006;62:2046–2060. 21. Huisman M, Kunst AE, Andersen O, Bopp M, Borgan J-K, Borrell C, et al. Socioeconomic inequalities in mortality among the elderly in eleven European populations. J Epidemiol Community Health. 2004;58:468–475. 22. Puigpino´s R, Borrell C, Pasarı´n MI, Montella` N, Pe´rez G, Plase`ncia A, et al. Inequalities in mortality by social class in men in Barcelona, Spain. Eur J Epidemiol. 2000;16:751–756. 23. Regidor E, De la Fuente L, Calle ME, Navarro P, Domı´nguez V. Unusually strong association between education and mortality in young adults in a community with a high rate of injection-drug users. Eur J Public Health. 2003;13:334–339. 24. Liu X, Witten M. A biologically based explanation for mortality crossover in human populations. Gerontologist. 1995;35:609–615. 25. Lynch SM, Brown JS, Harmsen KG. Black-white differences in mortality compression and deceleration and the mortality crossover reconsidered. Res Aging. 2003;25:456–483. 26. Helio¨vaara M, Aromaa A, Klaukka T, Knekt P, Joukamaa M, Impivaara O. Reliability and validity of interview data on chronic diseases. The MiniFinland Health Survey. J Clin Epidemiol. 1993;46:181–191. 27. Van der Velden J, Abrahamse H, Donker G, Van der Steen J, Van Sonsbeek J, Van den Bos G. What do health interview surveys tell us about the prevalence of somatic chronic diseases? Eur J Public Health. 1998;8:52–58.