Discussion: Gender Differences in Self-Rated Health, in Mortality, and in the Relationship Between the Two

The set of articles making up this special section were all driven by the authors' interest in self-ratings of health and these ratings' potential for indicating risk for mortality. Studies of this association began to appear in the mid 1980s, and in a relatively short time, dozens had been published, primarily in North America, Europe, and Asia. There was a consistent finding, with few exceptions, that respondents to health surveys who rated their health as “poor,” “fair,” or sometimes just “good” had a significantly higher risk of mortality than those who considered their health “excellent” or the equivalent, even when other measures of health were considered. In some of these studies, the effects were similar for men and women; in others, there were differences. Contributors to this special section have chosen to look at their data with an eye to the differences and with the hope that an analysis of the differences may help reseachers better understand the similarities.

Should one expect a gender difference? For multiple reasons, beginning with the dependent variable (mortality), the answer would be yes. Life expectancy is greater for women in virtually every country on earth, at birth and at every other age. It is certainly true for all of the countries for which there are published reports of self-rated health–mortality analyses (Benyamini & Idler, 1999; Idler & Benyamini, 1997). As a result, the age-adjusted risk of mortality in these samples will be lower for the women than for the men, and the number of deaths among women will be proportionately fewer given the number of women in the sample. Another consequence of gender differentials in mortality risk is that (unless sampling weights are used) the number of women in a representative sample will be larger, providing more statistical power and reducing the possibility of Type II errors in subsamples of women compared with subsamples of men.

A second reason for expecting to see gender differences is due to characteristics of the independent variable, self-rated health. The U.S. National Health Interview Surveys throughout the 1990s show slightly higher percentages of women rating their health as fair or poor, compared with men (National Center for Health Statistics, 2001a). Among published self-rated health–mortality studies that report these distributions, there are occasionally more men in the poorest health category, but more often it is women reporting poorer health (Benyamini & Idler, 1999; Idler & Benyamini, 1997). Among the studies in this special section, the three that report frequencies for the self-rated health variable (the Israeli Cross Sectional and Longitudinal Study [CALAS; Benyamini, Blumstein, Lusky, & Modan, 2003], the Longitudinal Aging Study Amsterdam [LASA; Deeg & Kriegsman, 2003), and the Nottingham Longitudinal Study of Activity and Ageing [NLSAA; Bath, 2003]) report higher proportions of women with poor or fair self-ratings. In addition, rates of functional disability, disabling chronic conditions, and comorbidities are higher for women in each of the studies that report them (NLSAA, CALAS, LASA). So in general, and in these studies specifically, women's rates of disease, disability, and poor self-rated health might lead one to argue that women's health—during their lifetime—is poorer than men's health.

Third, there are gender differences in the self-rated health–mortality relationship to consider. Do men's generally better self-ratings of health have a different relationship with their higher mortality rates when compared with the poorer self-rated health and lower mortality rates of women? Figure 1 presents hazard or risk ratios for mortality for the studies reviewed in Idler and Benyamini (1997) and Benyamini and Idler (1999) that analyzed data separately by gender. Of 16 studies reporting data in this way, 10 show higher risks for men, 2 show no differences, and just 4 show higher hazard or risk ratios for women.

Among the four studies presented in this special section, there is a similar, mixed pattern that tends toward stronger results for men. In the final model adjusted for all covariates in the Melton Mowbray Ageing Project data (Spiers, Jagger, Clarke, & Arthur, 2003), the hazard ratio for men is a statistically significant 1.8, whereas for women it is a smaller and nonsignificant 1.4. Moreover, the test for the gender interaction was significant. In the full follow-up LASA data (Deeg & Kriegsman, 2003), several forms of the self-rated health variable are tested; in models adjusted for age only, the risk ratios are higher for men in four cases out of five. In fully adjusted models the pattern is the same, but in no case is the risk ratio significant for women. In the NLSAA data (Bath, 2003) for the full follow-up period, although the hazard ratio for poor self-rated health was initially larger for women than for men in the unadjusted model, it became nonsignificant for women at an earlier step than for men (i.e., as soon as health status and physical activity were adjusted). In the fully adjusted models in this study, poor self-rated health was not a significant risk for mortality for either men or women. And finally, in the CALAS data (Benyamini et al., 2003), unadjusted risk ratios are higher for women aged 75 to 84 compared with same age men and quite similar for men and women aged 85–94; however, final adjusted models show no significant risk ratios for either males or females. Thus, in two of four cases, when all other relevant factors are considered, poor self-rated health poses a greater risk for men than for women. In the remaining two cases, there is no difference between men and women.

Two alternative interpretations are possible. On the one hand, the studies showing higher hazard or risk ratios for men may demonstrate that men know more about their health than women know about theirs. Deeg and Kriegsman (2003) conduct supplementary analyses showing that women's self-ratings of health are more closely linked with health conditions that are disabling, rather than fatal, whereas men's self-ratings tended to take the mortality risks and lifestyle factors into account. Moreover, they found that the biggest difference in the predictive ability of men's and women's self-ratings of health came when respondents were asked to rate their health compared with others their age. This suggests that women's self-ratings may be unduly influenced by the prevalent chronic and disabling but nonfatal conditions among their female age peers. With this interpretation, men's self-ratings of health are better predictors of their mortality because men base their self-ratings of health on the conditions that are mortality relevant and do not take other, irrelevant health information into account.

A second interpretation is based on the more popular view that women know more about health than men. Women are the primary consumers of health services and health information. Advocates for women's health have noted that women's higher educational levels and increased entry into the labor force have increased access to and the use of health care (Strobino, Grason, & Minkovitz, 2002). U.S. data on ambulatory medical care shows that adult women are much more likely than men to have had any visit to a physician in the previous year and are also more likely to have had four or more visits (National Center for Health Statistics, 2001b). Popular health information in the media clearly favors women; the market for magazines devoted to women's and children's health overwhelmingly favors women (Simmons Market Research Bureau, 1994). A hallmark of women's health care in the last three decades has been the teaching of techniques for self-examination and early detection of conditions that can be treated successfully if caught early. Moreover, recent efforts call attention to the undertreatment of major causes of morbidity such as heart disease in women (Matthews et al. 1997), suggesting that women's consciousness of health will continue to be stimulated by public health education efforts.

Thus, the alternative explanation for the gender differences in these studies is that men's risk ratios for poor self-ratings of health are better predictors of their mortality precisely because women know more about their health. The reasoning is this: Because women know more about health in general, and their own health specifically, they respond to inquiries about their health in surveys with greater accuracy. As a consequence, when multivariate analyses are conducted, and health status information from the respondent and other sources is included, the association of global self-ratings of health with mortality will be more fully accounted for in samples of women than in samples of men. Support for this hypothesis can be seen only when initial, unadjusted hazard or risk ratios are seen to decrease as covariates for sociodemographic and health status are added.

This analysis strategy was used in some of the studies for this special section, and the results largely support the second explanation. In the NLSAA (Bath, 2003), for example, the unadjusted hazard ratios for poor health are 2.2 for men and a higher 2.9 for women, both strongly significant. When age, health index score, and physical activity are adjusted, the relative size of the hazard ratios reverses. After adjustment for health status information—that comes directly from the respondent—the hazard ratio for women declines to a nonsignificant 1.6, whereas the men's hazard ratio remains significant at 1.8. In the LASA (Deeg & Kriegsman, 2003), the initial age-adjusted risk ratios are greater for the men than for the women in almost every case, although the risk of poor health for women was significant in three of five cases. In the final adjusted models, none of the risk ratios remained significant for women, although two of five were still predictive for men. In the Melton Mowbray data (Spiers et al., 2003) the initial, unadjusted models showed significant hazard ratios for poor health for both men and women, with the men's hazard ratio being larger (2.7) than the women's (1.9). In the fully adjusted model for the whole sample, the hazard ratio is reduced to still-significant 2.3 for men and just 1.6 for women. Similarly, in the models stratified by health problem, none of the hazard ratios are significant for women, but there is a significant hazard ratio in the largest group (disabling problem only) for men. The CALAS data (Benyamini et al., 2003), on the other hand, shows the opposite pattern: The risk ratios for women are consistently higher for women in both age groups and at all steps in the analysis. However, hazard ratios for neither men nor women are significant in the final step, with health status fully adjusted. Thus, in general, the sequential steps of these analyses show that the effect of self-rated health on mortality is partially accounted for by health status information reported by respondents, and that, in three of the four cases, the inclusion of health status information reduces the effect of self-rated health more for women than it does for men.

In gender and health there are paradoxes. Women's higher disability and chronic condition prevalence rates are, in a sense, the price that they pay for their longer life expectancy, through their longer period of exposure to the processes that produce chronic and comorbid conditions. Another possible paradox is that many self-rated health studies appear to show that women's reported global health is a poorer indicator of their future health and/or survival than is men's—in other words, it appears that men have more knowledge about their health, when in fact they may have less. Between these two paradoxes may be a link. Women's greater disease burden and longer life expectancy causes a longer period of decline. This longer period of coping with chronic illness may provide elderly women with a longer opportunity for appraisal, information-seeking, and continually refined feedback loops, resulting in a high level of health status knowledge relative to men's. This in turn provides women with very accurate information in responding to health interviews that reduce the predictiveness of their global self-ratings of health. And at the same time, men's self-ratings of health (more so than their reports of chronic and comorbid conditions) may indeed summarize their known mortality risks very effectively, leading to the very significant mortality risks for those who rate their overall health as poor. In other words, either explanation could account for the data, and there may even be some truth in both.

Recognizing gender differences as an important issue in this field of research suggests some specific recommendations for future research. First, analyses should be both gender-specific and should also pool data and conduct tests for a self-rated health by gender interaction, as the Melton Mowbray study did. Such strategies would (a) establish whether gender moderates the effect of self-rated health in a particular sample, and (b) permit the investigation of gender-specific mediating effects. A second recommendation is that future studies should include the unadjusted or simple age-adjusted association of self-rated health with mortality. Then, when covariates for health status and disability are added in a hierarchical approach, the differential impact of these sets of variables on the associations among men and women can be assessed.

The accumulation of evidence of gender differences in self-rated health has raised critical questions about the ways in which self-ratings of health are derived and produced, through physical perceptions, personal knowledge, and social comparisons. These differences may be exactly what are needed to move the field forward, by finding out which components of health status information possessed by individuals explain the largest portion of the differences in mortality. Along the way, we may further our knowledge of men's and women's health.

Figure 1.

Open in new tabDownload slide

Relative risk of mortality for males and females reporting poorest health. ▪ Males, Females. Source: “Community Studies Reporting Association Between Self-Rated Health and Mortality,” by Y. Benyamini and E. Idler, 1999, Research on Aging, 21, and “Self-Rated Health and Mortality: A Review of Twenty-Seven Community Studies,” by E. Idler and Y. Benyamini, 1997, Journal of Health and Social Behavior, 38

References