METHODS

Data sources and study subjects

Census data, available electronically from the Korean Statistical Information System,¹³ and death certificate data available to the public were used for this study. To explore changes over time, 1990, 1995, and 2000 census data were used as denominators and 1990–91, 1995–96, and 2000–01 (all November—October) death certificate data were used as numerators. For mortality data decedent’s age was not determined by age at death, but by calculation of the numbers of years between the date of birth on death certificate data and the time of the last census. The results of using date of birth information will be similar to those from mortality follow up in cohort studies, if (1) census and death certificate data cover all Koreans residing in Korea, (2) emigration and immigration numbers are comparatively small compared with the total population, and (3) misclassification bias for educational level is slight. As individual death certificate information was not directly linked to information from the census, numerator/denominator bias may occur in calculating education specific mortality rates, producing biased educational mortality differentials. However, according to a previous study that used these methods,¹¹ these possibilities have been reviewed and were expected to be small.

For trends in morbidity inequality Social Statistics Survey (SSS) data from Korea’s National Statistical Office were used. These face to face interviews are conducted nationally from randomly selected households each year on the 10th day after the week that includes the 15th day of September. Sections regarding health were included in 1989, 1992, 1995, and 1999. Non-response rates for these surveys are very low (1.7% in 1999). SSS data from 1989 onward are available to the public.

This study included men and women aged 30–59. Adults aged less than 30 were excluded as they may not yet have completed their education. Adults aged 60+ were not included because college graduates in this group were very rare, especially among women. In the 1989 SSS data, only 10 of 5692 women aged 60+ (0.2%) reported completion of a college education. In addition, the Korean Statistical Information System¹³ did not provide education and one year specific numbers for adults aged 60+ in the 1990 and 1995 census, presenting age collapsed population numbers for the over 60 education groups.

Study subjects for mortality consisted of 146 151 deaths (from death certificate data) of 26 852 658 men (from three census surveys) and 53 789 deaths (from death certificate data) of 26 314 593 women (from three census surveys). For morbidity, 82 012 men and 84 403 women from four SSS rounds were included as study subjects.

Socioeconomic position measures

Education was used as the indicator of socioeconomic position. Educational level was measured as the highest level of education completed. Because middle and high school categories were combined in 1990–1991 death certificate data, these two education groups were categorised together into secondary education for comparison over time. Education was classified into three categories (elementary or less, middle or high school, and college or higher) for mortality and four categories (elementary or less, middle school, high school, and college or higher) for morbidity.

After examination of the data, it was determined that income and occupation could not be included as socioeconomic position indicators. Income information is not included on death certificates in Korea, and questionnaires about income were inconsistent across waves of the SSS. In addition, occupational classifications in survey data from the National Statistical Office (census and SSS data) and registration data (death certificate data) were changed in 1993. These changes and inconsistencies thus made using income or occupation as socioeconomic indicators difficult.

Health outcome measures

Health outcome measures used in this study were death (from death certificate data) and two indicators of morbidity (self rated health and self reported illness in the past two weeks in SSS data). Self rated health was measured by the question “How would you rate your health compared to others your age?” with five answer categories ranging from “very good” to “very poor”. The categories “fair”, “poor”, and “very poor” were combined to yield a measure of self rated health less than good. The question “Did you have any illness in the past two weeks?” with answer categories of “yes” and “no” was used to establish self reported illness. Questions regarding long term illness were not included in SSS.

Analysis

Education specific mortality rates were calculated using mortality data. These rates were directly age adjusted to one year age groups, with distribution of 2000 census data as the standard, producing age standardised mortality rates. Prevalence rates of self rated health less than good and self reported acute illness were calculated for morbidity data. These rates were also directly standardised to five year age groups, using the age distribution of 1999 SSS data. Confidence intervals of these age adjusted prevalence rates were estimated.

Men and women were analysed separately for both mortality and morbidity rates. Education specific relative risks for mortality were computed using Poisson regression analyses with the data of sex, one year age, and education specific numbers of population and deaths. Odds ratios for self rated health less than good and self reported acute illness were estimated by means of logistic regression.

The relative index of inequality (RII), a measure of effect that permits meaningful comparison of socioeconomic health inequalities over time,^{1,4–7,9,14,15} was computed. To calculate the RII, a relative educational position indicator was computed. This indicator is a value between 0 and 1, assigned by calculating the relative position in the cumulative population distribution of the central subject in each group of the educational hierarchy, and was entered as an independent variable in our analyses. The regression coefficient of the relative educational position indicator and the standard error were used to calculate the rate ratio with 95% confidence intervals. This rate ratio is the RII for mortality data.^1,2,9 The RII in morbidity data is an odds ratio that is computed from logistic regression analysis.^4–7 The RII is the relative risk of dying, or the odds of reporting morbidity at the lowest end of the educational hierarchy as compared with the risk of dying, or the odds of reporting morbidity at the very top of the educational hierarchy. The trend in the RII was estimated by including an interaction term of the relative educational position indicator and the variable that identified the year of the data in the model. All analyses were performed with SAS statistical software.¹⁶

RESULTS

As table 1 shows, age standardised mortality rates in both sexes decreased over time. Male mortality rates were greater than those of women. Graded educational mortality differences were found in both sexes at different times. Based on RIIs, the relative level of socioeconomic mortality inequality remained unchanged in men for the past 10 years (p = 0.66). The trend in RII among women showed a decreasing pattern but was not significant (p = 0.11). However, differences in age standardised mortality rates between the lowest and highest educational group, that is absolute level of inequalities, provide a different picture. In women, mortality rates of the elementary or less group remained virtually unchanged (from 368 in 1990–91 to 355 in 2000–01) while mortality rates of college graduates decreased from 157 in 1990–91 to 98 in 2000–01. As the result, the absolute mortality inequalities widened over time.

Graded patterns of educational inequalities in the prevalence rates of self rated health being less than good were detected in men at different time points (table 2). In women, the gradient was not clear in 1989 (RII = 0.96 (0.86 to 1.07)). Self rated health inequalities measured by RII trends increased over time in both men and women (p = 0.001 and p<0.0001). Moreover, the increase in RII was most notable between 1995 and 1999 in both sexes. RII increased in men (RII increase = 0.55, from 1.71 to 2.26) and women (RII increase = 0.59, from 1.35 to 1.94) between 1995 and 1999 accounting for 92% of male (RII increase = 0.61, from 1.67 to 2.26) and 60% of female total increases (RII increase = 0.98, from 0.96 to1.94) between 1989 and 1999. Furthermore, the RII increase between 1995 and 1999 was statistically significant in both sexes while all other comparisons between successive surveys (except for the comparison between 1989 and 1992 in women) did not yield statistically significant RII differences.

As table 3 shows, women with more education were more likely to report illness in the past two weeks in 1989 and 1992. However, this pattern was reversed in 1995 and 1999 with less educated women reporting acute illness more easily. According to RII trends, levels of inequalities for self reported acute illness have increased in the past 10 years in both sexes (men p = 0.0012, women p<0.0001). The RII in 1999 was statistically greater in women as compared with the RII in 1995. In addition, age adjusted prevalence rates of self reported acute illness increased over time in both sexes. This pattern was more notable in women who recorded a 4.7% rise (20.4% in 1989 to 25.1% in 1999) in the prevalence rate compared with men (2.5% increase, 14.4%–16.9%). This increase was attributable to the rise in self reported illness from women with less education (that is, elementary or less education).

According to the comparison of RIIs by year, levels of relative educational inequalities were greater in self rated health than those of self reported acute illness. However, the greatest level of inequality was found in mortality. Based on RIIs, levels of relative educational inequalities in mortality and morbidity were greater in men than women and persisted over time.

DISCUSSION

Results of this study showed that, based on RII trends from 1990 to 2001, relative educational mortality inequalities in Korean men and women remained virtually unchanged. The level of mortality inequality in women appeared to have narrowed, but the RII trend was not statistically significant. In contrast, absolute mortality inequality—a mortality rate difference—was found to have widened between women with elementary or less education and those with college or higher education. These contrasting relative and absolute socioeconomic differentials highlight how attention needs to be paid to how inequality is measured, in regard to assessing the magnitude and trends in socioeconomic health inequalities.^14,15

In contrast with mortality, the level of relative inequality in self rated health being less than good increased during the past 10 years among both sexes. Based on RII, relative inequalities in self reported acute illness increased in the past 10 years. An interesting finding was that the relation between education and self reported acute illness reversed in women, but not in men. Korean women with less education reported less acute illness than those with more education in 1989 and 1992 but disclosed more acute illness in 1999. This reversal in socioeconomic health gradient appeared to happen to women in a comparatively short period. One explanation may be attributable to the fact that self rated illness is a function of both the burden of disease pathology and individual perceptions. The individual’s perception is further influenced by social and cultural context. A systematic under-reporting of illness by people with lower socioeconomic position has been highlighted in other studies,^17,18 suggesting that the more disadvantaged may report less poor health despite their objectively worse health status. This may be related to people matching their expectations for health with their perceived health. Despite this underreporting, several studies in industrialised countries have revealed that poor people reported greater illness burden than rich people.^19,20 However, the rich report more ill health than the poor in many underdeveloped countries even though the rich have a better objective health status and lower mortality rates.^21,22 It has been suggested that this may reflect a richer (or more educated) person’s greater likelihood to have access to health services, receive diagnoses, and thus report illness.^23,24

Our result may be related to the expansion of Korea’s national health insurance, whereby universal coverage started in late 1989 along with the final inclusion of the self employed as beneficiary.²⁵ Given that the coverage of Korea’s national health insurance was expanded from workers in large companies in 1977 to the self employed,²⁵ the difference in labour market participation rates between men and women might contribute to the gender difference in educational inequalities of self reported acute illness. Furthermore, this expansion of national health insurance may explain both the upsurge in the prevalence rates of poor self rated health between 1989 and 1992 and greater increase in self reported acute illness in women than men in the past 10 years in this study. The lack of socioeconomic self rated health gradient among women in 1989 can be also understood in this context.

This study includes periods of Korean economic crisis, which started in November 1997 and ended in August 2001, as study period. The population health footprint of that economic crisis needs investigating. Despite a potential negative health impact of the economic crisis, mortality rates among all educational groups continued to decline in both sexes. There is no evidence that the Korean economic crisis widened socioeconomic inequalities in all cause mortality. On the other hand, the latest Korean economic crisis may have widened educational inequalities in self rated health. Most of the total increase in relative inequalities in self rated health was made between 1995 and 1999. However, this increase in relative socioeconomic health inequality was not accompanied by a general deterioration of self rated health for all educational groups. While in 1999, college graduates had similar levels of self rated health compared with 1995, those with middle or less education expressed worse levels of self rated health. A definitive conclusion on the relation of the Korean economic crisis with socioeconomic mortality inequalities will not be possible until further analyses of cause specific mortality in the midst of the economic crisis are completed. Future studies should include more objective morbidity indicators and income and/or occupation as indicators of socioeconomic position.

Health inequalities in self rated health were greater than those in self reported acute illness among both sexes over time. When more objective criteria of suffering or disability were used, steeper gradients were evident, with the poorer groups displaying worse health status.^4,10,26 A previous study suggested that perceived general health was related to chronic diseases but not to acute conditions.²⁷

This study explored educational inequality trends in mortality and two subjective morbidity indicators. Educational health inequalities were found to be greater in regard to the objective measure (all cause mortality) than in the subjective ones (two morbidity indicators in this study) among both sexes. This was also true in regard to trends over time. This highlights the importance of considering the outcomes used to monitor health inequalities in populations.

Socioeconomic health inequalities in women have appeared to be smaller than those of men when researchers used mortality data^28,29 rather than morbidity data as the outcome.³⁰ The idea that gender differences in socioeconomic health inequalities depend on the measure of inequality used has been investigated.^28,31 Given that socioeconomic health gradients for mortality and morbidity measures remained steeper over time among men than women in this study, inequality differences between women and men may not be artefacts but interesting observations that can trigger thinking about their aetiological processes.³⁰

Although this study did not find any rise in relative mortality inequalities, the importance of increasing socioeconomic inequalities in subjective morbidity measures should not be overlooked in Korea, where life expectancies approach the level of western industrialised countries and the burden of people’s subjective health status is perhaps growing in importance.³² The rise in inequalities in subjective health between socioeconomic groups, regardless of concurrent increased inequalities in objective health measures, requires increased social discourse and policy discussions about health inequalities in Korean society where socioeconomic health inequalities have not taken priority in the policy agenda.