Social Science & Medicine 104 (2014) 201e209
Contents lists available at ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Health inequalities in Japan: The role of material, psychosocial, social relational and behavioural factors Ayako Hiyoshi a, b, c, *, Yoshiharu Fukuda d, Martin J. Shipley a, Eric J. Brunner a a
Research Department of Epidemiology and Public Health, University College London, London, United Kingdom Clinical Epidemiology and Biostatistics, Örebro University Hospital, Örebro, Sweden c School of Health and Medical Sciences, Örebro University, Örebro, Sweden d Department of Community Health and Medicine, Yamaguchi University School of Medicine, Ube, Japan b
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 8 January 2014
The extent that risk factors, identified in Western countries, account for health inequalities in Japan remains unclear. We analysed a nationally representative sample (Comprehensive Survey of Living Conditions surveyed in 2001 (n ¼ 40,243)). The cross-sectional association between self-rated fair or poor health and household income and a theory-based occupational social class was summarised using the relative index of inequality [RII]. The percentage attenuation in RII accounted for by candidate contributory factors e material, psychosocial, social relational and behavioural e was computed. The results showed that the RII for household income based on self-rated fair or poor health was reduced after including the four candidate contributory factors in the model by 20% (95% CI 2.1, 43.6) and 44% (95% CI 18.2, 92.5) in men and women, respectively. The RII for the Japanese Socioeconomic Classification [J-SEC] was reduced, not significantly, by 22% (95% CI 6.3, 100.0) in men in the corresponding model, while J-SEC was not associated with self-rated health in women. Material factors produced the most consistent and strong attenuation in RII for both socioeconomic indicators, while the contributions attributable to behaviour alone were modest. Social relational factors consistently attenuated the RII for both socioeconomic indicators in men whereas they did not make an independent contribution in women. The influence of perceived stress was inconsistent and depended on the socioeconomic indicator used. In summary, social inequalities in self-rated fair or poor health were reduced to a degree by the factors included. The results indicate that the levelling of health across the socioeconomic hierarchy needs to consider a wide range of factors, including material and psychosocial factors, in addition to the behavioural factors upon which the current public health policies in Japan focus. The analyses in this study need to be replicated using a longitudinal study design to confirm the roles of different factors in health inequalities. Ó 2014 Elsevier Ltd. All rights reserved.
Keywords: Japan Health inequalities Socioeconomic Social class Income Self-rated health
Introduction Although Japan is one of the countries with the longest life expectancy in the world, with 83 years at birth (World Health Organization, 2013), socioeconomic inequalities in health are still evident across a number of health outcomes. Individual-level analyses have shown inequalities in all-cause and chronic disease mortality and incidence (Kagamimori, Gaina, & Nasermoaddeli, 2009), and such disparities appear to have persisted for over two decades (Hiyoshi, Fukuda, Shipley, Bartley, & Brunner, 2013a). * Corresponding author. Clinical Epidemiology and Biostatistics, S-huset, Örebro University Hospital, 701 85 Örebro, Sweden. E-mail addresses: [email protected], [email protected] (A. Hiyoshi). 0277-9536/$ e see front matter Ó 2014 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.socscimed.2013.12.028
Despite these persisting health inequalities, perspectives on health inequalities had been absent in national strategies on health policies, with strategies to improve health relying heavily on an individualistic approach and focused on behavioural aspects. The second stage of the ‘Healthy Japan 21’ was implemented in 2012 with one of its objectives being to reduce area (prefectural) disparities in healthy life expectancy, defined by the absence of limitations in daily living or self-rated fair or poor health (Ministry of Health, Labour and Welfare [MHLW], 2012). Although it is not yet clear whether aspects other than area disparities in health are considered or how area disparities in healthy life expectancy are redressed, the inclusion of the concept of social inequalities in health represents a considerable shift in the policy discussion in Japan.
202
A. Hiyoshi et al. / Social Science & Medicine 104 (2014) 201e209
Health inequalities are not ‘reducible’ to behaviours alone (Dunn, 2010), and public health interventions targeting individual’s health behaviour may even widen social patterning in behaviours. After controlling for behaviours rigorously, some have shown that health inequalities still remain (Stringhini et al., 2010). In addition, material, psychosocial and social relational factors have been identified to explain systematic differences in health according to socioeconomic position [SEP] (Aldabe et al., 2011; Brunner, 2007; Laaksonen, Roos, Rahkonen, Martikainen, & Lahelma, 2005). Housing conditions have been reported to relate to mental and physical health through house temperature, noise, cleanliness and hygiene (Thomson, Thomas, Sellstrom, & Petticrew, 2013), and those who were not homeowners have been exposed to a greater number of health damaging factors (Macintyre et al., 2003). The psychological approach recognises that the availability of resources to cope with stressful situations is closely associated with socially patterned emotions, the distribution of power and control, various forms of discrimination and the fairness of society (Brunner, 2007), and psychosocial factors seem to attenuate health inequalities in Western countries (Marmot, Bosma, Hemingway, Brunner, & Stansfeld, 1997; Power, Matthews, & Manor, 1998; Wen, Hawkley, & Cacioppo, 2006). Social relational factors such as marital status and living alone are important determinants of health (Holt-Lunstad, Smith, & Layton, 2010), and socioeconomic variation in marital status has been reported (Fieder, Huber, & Bookstein, 2011). In Japan, although there are many studies of factors linking SEP and health, to the best of our knowledge, none of these studies have explicitly tested all four dimensions of mechanisms simultaneously or calculated the extent of the attenuation in health inequalities by factors included. Health inequalities have most extensively examined in relation to education (Aida et al., 2011; Fujino, Iso, et al., 2005; Fujino, Tamakoshi, et al., 2005; Fujisawa, Hamano, & Takegawa, 2009; Hamano et al., 2010; Hirokawa, Tsutusmi, & Kayaba, 2006; Honjo, Tsutsumi, & Kayaba, 2010; Ichida et al., 2009; Ito et al., 2008; Iwasaki et al., 2002; Liang et al., 2002; Liang, Bennett, Sugisawa, Kobayashi, & Fukaya, 2003; Nishi et al., 2012; Wang et al., 2005) and, to a lesser extent, income (Liang et al., 2002, 2003; Oshio & Kobayash, 2009; Wang et al., 2005). Educational and income inequalities in health have been most consistently found in all-cause mortality and subjective health status, and these associations are attenuated somewhat by health behaviours, biomarkers, occupational factors and stress. The findings for occupation vary due to the differing occupational classifications employed (Hirokawa et al., 2006; Honjo et al., 2010; Ishizaki et al., 2006, 2001; Iwasaki et al., 2002; Sekine, Chandola, Martikainen, Marmot, & Kagamimori, 2006, Sekine, Chandola, Martikainen, Marmot, & Kagamimori, 2009; Sekine et al., 2011), but no study has used a theory-based occupational classification, which has the advantage of clarity when describing the dimension of inequality that was actually measured (Hiyoshi et al., 2013a). We consider that a study examining the four explanatory dimensions explicitly will contribute to advancing policy discussion for reducing health inequalities, which has just begun in Japan. The aim of the present paper is to assess the contribution of material, psychosocial, social relational and behavioural factors on health inequalities for household social class and income in a working age population in Japan. We utilise social class and income as socioeconomic indicators as they may describe important aspects of health inequalities in Japan after substantial social changes occurred in the 1990s during which there appeared to be increases in job insecurity and income inequality. We calculate the attenuation that the four domains of candidate contributory factors have on social inequality in self-rated health.
Methods We analysed data from the Comprehensive Survey of Living Conditions [CSLC], a triennial survey that has been conducted since 1986. In particular, we used data from the 2001 CSLC as this was the only time that data on perceived stress and behaviours, including a detailed question on smoking, were collected. The CSLC employs multi-stage stratified random cluster sampling with the primary sampling unit being the census Enumeration Districts [EDs] which divide Japan into approximately one million areas. After stratifying by prefecture and large cities, 5000 EDs were randomly selected and all households and household members living in these areas were approached to complete a Demography & Health questionnaire. In addition, 2000 EDs were randomly selected from the 5000 EDs to complete an Income & Savings questionnaire, and we used this subset sample for our analyses. Response rates were 87.4% for the Demography & Health questionnaire in 2001, and 79.5% for the Income & Savings questionnaire, respectively (MHLW, 2009). Having excluded individuals with missing data in relevant variables, the sample size was n ¼ 40,243 (51.6% women). We tested mediating models / / SELF-RATED HEALTH and calculated the extent to which income and social class inequalities in health were accounted for by 1) material, 2) psychosocial, 3) social relational, and 4) behavioural factors. Outcome Self-rated health is used as the outcome. In various countries including Japan, the determinants of self-rated health appear to be similar (French et al., 2012), suggesting that the perception of selfrated health as a concept is not different for the Japanese from some Western countries. Self-rated health has been shown to be a strong predictor of all cause and cause specific mortality in many countries including Japan (Idler & Benyamini, 1997; Murata, Kondo, Tamakoshi, Yatsuya, & Toyoshima, 2006). In the present study, self-rated health was assessed from the single question: ‘what is your current health (condition)?’. The five categories of response were: excellent, very good, good, fair, and poor. The variable was dichotomised, setting ‘poor’ and ‘fair’ responses as the outcome and expressed as ‘suboptimal health’ hereafter (Perlman & Bobak, 2008). Socioeconomic measures We used the Japanese Socioeconomic Classification [J-SEC]. It adopted the conceptual basis of the UK’s National Statistics Socioeconomic Classification [NS-SEC] (National Statistics, 2005), which differentiates the social position of individuals in terms of employment conditions and relations. In particular, J-SEC was based on the NS-SEC three category version since it is a hierarchical construct (National Statistics, 2005) which is suitable to be summarised using the relative index of inequality [RII] explained below. It was constructed using the Japanese Standard Classification of Occupation together with employment status (such as executives of companies, self-employed, employee, and limited term contract) and predicted economic and health differences for the Japanese population (Hiyoshi et al., 2013a). We used household social class, assigned by taking the highest social class value for any household member aged 15 or greater. Individuals who had missing data in variables used to derive J-SEC or lacked a household member having classifiable jobs were not assigned to a class (n ¼ 7,192, 17.9%). Annual household income, including benefits and inheritance before tax, was equivalised by dividing by the square root of household size. The study population was grouped into income
A. Hiyoshi et al. / Social Science & Medicine 104 (2014) 201e209
203
deciles, separately for the young (20e39 years) and old (40e59 years), to account for age differences in income.
candidate contributory factors and health to differ by sex in the present study.
Candidate contributory factors
Statistical analysis
Material factors Material factors included homeownership and living density. Homeownership consisted of five levels: 1) owned house, 2) renting, 3) work-related accommodation (let by an employer for employee), 4) social housing, and 5) lodging (renting a room in a house or flat). Social housing includes public housing provision to accommodate those with financial constraints and to achieve swift housing provision to the middle class. Living density was calculated by dividing the number of household members by the number of rooms.
In preliminary analyses, chi-squared tests for heterogeneity and trend were used to examine the association of suboptimal selfrated health according to levels of exposure and candidate contributory factors. Since the prevalence of suboptimal self-rated health increased monotonically with poorer SEP, the effect of the SEP measures (household J-SEC and household income) were summarised using the relative index of inequality [RII]. To estimate the RII, a rank variable was created for each socioeconomic indicator where each category was assigned the value, between 0 and 1, of the cumulative midpoint of their ranges; for example, the lowest income decile was assigned a score of 0.95, and the highest decile a score of 0.05. This score was then used in logistic regression models and the RII estimated using the regression coefficient associated with this score. A strength of using the RII is the ability to provide a single summary measure of health disparity, including direction and magnitude, which uses all data (Mackenbach & Kunst, 1997). Initially, associations between candidate contributory factors and self-rated health, as well as those between SEP indicators and candidate contributory factors, were examined. Logistic regression models were fitted to assess the former. To assess the latter, the SEP rank variable was used in logistic, multinomial logistic and linear regression models, depending upon the variable type of the candidate contributory factors which were used as the outcome. The RII in these models summarises the relative difference in candidate contributory factors between the extreme bottom versus the extreme top of the socioeconomic hierarchy. A series of logistic regression models with suboptimal self-rated health as the outcome were then fitted for the SEP measures, starting with an initial base model which adjusted for age, prefecture and clustering within households. Adjustment for household clustering was included to allow for the intra-household correlation between individuals, which was much stronger (intraclass correlation ¼ 0.336) than that for prefecture (intraclass correlation ¼ 0.003). Subsequent models then additionally adjusted for each of the four domains of candidate contributory factors separately and, finally, all together. The RII in these models summarises the relative difference in the risk of suboptimal self-rated health between the extreme bottom versus the extreme top of the socioeconomic hierarchy. The degree to which the RII was attenuated when adjusted for a candidate contributory factor was calculated by 100*(b0b1)/b0, where (b1) is the coefficient for the SEP score in a model with the contributory variable and (b0) is the coefficient for the SEP score in an initial base model without the candidate contributory factor, but still adjusted for age, prefecture and household clustering. The confidence interval for the percentage attenuation in the RII was obtained using a bias-corrected bootstrap method, with 2000 times re-samplings. Attributable percentage attenuations of the candidate contributory factors were calculated by subtracting the percentage attenuation in the RII of a model including all variables except for the given factor, from a model including all variables (van Oort, van Lenthe, & Mackenbach, 2005; Skalicka, van Lenthe, Bambra, Krokstad, & Mackenbach, 2009). This indicates the percentage of contribution that is attributable to a given factor alone, within the observed factors in our models. The population attributable risk fraction (‘population proportional attributable risk’ in Kirkwood and Sterne (2003)) due to having a social class not at the highest level or an income below the median income was calculated using the standard formula. Although the comparison of estimates derived from different
Behavioural factors Behavioural variables included insufficient sleep, unbalanced diet, irregular intake of meals, physical inactivity, excessive alcohol intake, smoking, and attendance at health check-ups. The first five of these behaviour variables were sub-questions of a lifestyle question: ‘Do you carry out daily the things specified below in order to maintain your health?’. The response was ‘yes’ when a respondent considered he/she agreed with the statement, otherwise a ‘no’ response is given to each question: I sleep enough; I eat a balanced diet; I regularly eat a morning, lunch and evening meal; I exercise an appropriate amount (or moderate amount) or am actively mobile; and, I tend not to drink too much. Smoking was measured by a four-level question: ‘Do you smoke?’. Respondents selected an answer from ‘I don’t smoke’, ‘I smoke every day’, ‘I smoke sometimes’, and ‘I have smoked in the past but have not smoked for more than one month’. The variable is used as a categorical variable having four levels. Health check-up was assessed by the question: ‘Have you taken a general health check-up or a comprehensive medical check-up in the last year’. The response was either ‘yes’ or ‘no’. These variables were coded so that the reference category was the healthier behaviour. Psychosocial factor Perceived stress was the only psychosocial factor ascertained and was measured by a single question: ‘Currently, do you have anxiety or stress in your daily life?’. The response was either ‘yes’ or ‘no’, coded with ‘no’ as the reference category. The measure has shown to be significantly associated with self-reported symptoms such as fatigue, irritated, headache and difficulty to sleep (Hamanishi, 2013). Social relational factors The two relational factors used were marital status (married, single, widowed, and separated) and living alone (1 vs 2 people in the household). Covariates Age, categorised into five-year intervals, and prefecture, the largest administrative division separating Japan into 47 areas, were treated as a series of binary dummy variables and considered as confounding factors. In previous analyses using other CSLC survey waves between 1986 and 2007 (Hiyoshi, Fukuda, Shipley, & Brunner, 2013b) we observed sex interactions between SEP and self-rated health. Therefore, to examine health inequalities, as well as the influence of candidate contributory factors, we stratified our analyses by sex. This then allowed the associations between SEP and candidate contributory factors as well as between the
204
A. Hiyoshi et al. / Social Science & Medicine 104 (2014) 201e209
Table 1 Characteristics of the sample and prevalence of suboptimal self-rated health according to all measures.
Age (mean, years) Prefecture (1e47) (mean observations (range) Self-rated health Optimal (excellent, very good, or good) Suboptimal (fair or poor) Variables Household J-SEC I II III P-value for x2 trend Household income (decile) 1 (highest) 2 3 4 5 6 7 8 9 10 (lowest) P-value trend Age (categorical) 20e24 25e29 30e34 35e39 40e44 45e49 50e54 55e59 P-value trend Material factors Homeownership Owning Renting Work-related Social housing Lodging P-value Living density Living density (mean, (SD), persons per room) P-value (t-test) Psychosocial factors Perceived stress Not stressed Stressed P-value Social relational factors Marital status Married Single Widowed Separated P-value Living alone Not alone Alone P-value Behavioural factors Sleep Enough sleep Not enough sleep P-value Diet Balanced diet Unbalanced diet P-value Meals Regular meals Irregular meals P-value
Men (n ¼ 19,486)
Women (n ¼ 20,757)
40.9 415 (128, 816)
40.7 442 (157, 868)
17,696 1790 (9.2%) %a or mean
b
Suboptimal health (%)
18,434 2323 (11.2%) %a or mean
Suboptimal health (%)b
52.3 27.2 20.5
8.6 8.5 9.9 0.062
52.4 28.8 18.8
10.9 10.0 11.6 0.66
10.2 10.4 10.3 10.2 10.0 10.0 10.2 9.9 9.6 9.3
7.7 8.5 9.7 9.3 7.7 8.0 8.6 10.8 9.2 12.8
Contents lists available at ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Health inequalities in Japan: The role of material, psychosocial, social relational and behavioural factors Ayako Hiyoshi a, b, c, *, Yoshiharu Fukuda d, Martin J. Shipley a, Eric J. Brunner a a
Research Department of Epidemiology and Public Health, University College London, London, United Kingdom Clinical Epidemiology and Biostatistics, Örebro University Hospital, Örebro, Sweden c School of Health and Medical Sciences, Örebro University, Örebro, Sweden d Department of Community Health and Medicine, Yamaguchi University School of Medicine, Ube, Japan b
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 8 January 2014
The extent that risk factors, identified in Western countries, account for health inequalities in Japan remains unclear. We analysed a nationally representative sample (Comprehensive Survey of Living Conditions surveyed in 2001 (n ¼ 40,243)). The cross-sectional association between self-rated fair or poor health and household income and a theory-based occupational social class was summarised using the relative index of inequality [RII]. The percentage attenuation in RII accounted for by candidate contributory factors e material, psychosocial, social relational and behavioural e was computed. The results showed that the RII for household income based on self-rated fair or poor health was reduced after including the four candidate contributory factors in the model by 20% (95% CI 2.1, 43.6) and 44% (95% CI 18.2, 92.5) in men and women, respectively. The RII for the Japanese Socioeconomic Classification [J-SEC] was reduced, not significantly, by 22% (95% CI 6.3, 100.0) in men in the corresponding model, while J-SEC was not associated with self-rated health in women. Material factors produced the most consistent and strong attenuation in RII for both socioeconomic indicators, while the contributions attributable to behaviour alone were modest. Social relational factors consistently attenuated the RII for both socioeconomic indicators in men whereas they did not make an independent contribution in women. The influence of perceived stress was inconsistent and depended on the socioeconomic indicator used. In summary, social inequalities in self-rated fair or poor health were reduced to a degree by the factors included. The results indicate that the levelling of health across the socioeconomic hierarchy needs to consider a wide range of factors, including material and psychosocial factors, in addition to the behavioural factors upon which the current public health policies in Japan focus. The analyses in this study need to be replicated using a longitudinal study design to confirm the roles of different factors in health inequalities. Ó 2014 Elsevier Ltd. All rights reserved.
Keywords: Japan Health inequalities Socioeconomic Social class Income Self-rated health
Introduction Although Japan is one of the countries with the longest life expectancy in the world, with 83 years at birth (World Health Organization, 2013), socioeconomic inequalities in health are still evident across a number of health outcomes. Individual-level analyses have shown inequalities in all-cause and chronic disease mortality and incidence (Kagamimori, Gaina, & Nasermoaddeli, 2009), and such disparities appear to have persisted for over two decades (Hiyoshi, Fukuda, Shipley, Bartley, & Brunner, 2013a). * Corresponding author. Clinical Epidemiology and Biostatistics, S-huset, Örebro University Hospital, 701 85 Örebro, Sweden. E-mail addresses: [email protected], [email protected] (A. Hiyoshi). 0277-9536/$ e see front matter Ó 2014 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.socscimed.2013.12.028
Despite these persisting health inequalities, perspectives on health inequalities had been absent in national strategies on health policies, with strategies to improve health relying heavily on an individualistic approach and focused on behavioural aspects. The second stage of the ‘Healthy Japan 21’ was implemented in 2012 with one of its objectives being to reduce area (prefectural) disparities in healthy life expectancy, defined by the absence of limitations in daily living or self-rated fair or poor health (Ministry of Health, Labour and Welfare [MHLW], 2012). Although it is not yet clear whether aspects other than area disparities in health are considered or how area disparities in healthy life expectancy are redressed, the inclusion of the concept of social inequalities in health represents a considerable shift in the policy discussion in Japan.
202
A. Hiyoshi et al. / Social Science & Medicine 104 (2014) 201e209
Health inequalities are not ‘reducible’ to behaviours alone (Dunn, 2010), and public health interventions targeting individual’s health behaviour may even widen social patterning in behaviours. After controlling for behaviours rigorously, some have shown that health inequalities still remain (Stringhini et al., 2010). In addition, material, psychosocial and social relational factors have been identified to explain systematic differences in health according to socioeconomic position [SEP] (Aldabe et al., 2011; Brunner, 2007; Laaksonen, Roos, Rahkonen, Martikainen, & Lahelma, 2005). Housing conditions have been reported to relate to mental and physical health through house temperature, noise, cleanliness and hygiene (Thomson, Thomas, Sellstrom, & Petticrew, 2013), and those who were not homeowners have been exposed to a greater number of health damaging factors (Macintyre et al., 2003). The psychological approach recognises that the availability of resources to cope with stressful situations is closely associated with socially patterned emotions, the distribution of power and control, various forms of discrimination and the fairness of society (Brunner, 2007), and psychosocial factors seem to attenuate health inequalities in Western countries (Marmot, Bosma, Hemingway, Brunner, & Stansfeld, 1997; Power, Matthews, & Manor, 1998; Wen, Hawkley, & Cacioppo, 2006). Social relational factors such as marital status and living alone are important determinants of health (Holt-Lunstad, Smith, & Layton, 2010), and socioeconomic variation in marital status has been reported (Fieder, Huber, & Bookstein, 2011). In Japan, although there are many studies of factors linking SEP and health, to the best of our knowledge, none of these studies have explicitly tested all four dimensions of mechanisms simultaneously or calculated the extent of the attenuation in health inequalities by factors included. Health inequalities have most extensively examined in relation to education (Aida et al., 2011; Fujino, Iso, et al., 2005; Fujino, Tamakoshi, et al., 2005; Fujisawa, Hamano, & Takegawa, 2009; Hamano et al., 2010; Hirokawa, Tsutusmi, & Kayaba, 2006; Honjo, Tsutsumi, & Kayaba, 2010; Ichida et al., 2009; Ito et al., 2008; Iwasaki et al., 2002; Liang et al., 2002; Liang, Bennett, Sugisawa, Kobayashi, & Fukaya, 2003; Nishi et al., 2012; Wang et al., 2005) and, to a lesser extent, income (Liang et al., 2002, 2003; Oshio & Kobayash, 2009; Wang et al., 2005). Educational and income inequalities in health have been most consistently found in all-cause mortality and subjective health status, and these associations are attenuated somewhat by health behaviours, biomarkers, occupational factors and stress. The findings for occupation vary due to the differing occupational classifications employed (Hirokawa et al., 2006; Honjo et al., 2010; Ishizaki et al., 2006, 2001; Iwasaki et al., 2002; Sekine, Chandola, Martikainen, Marmot, & Kagamimori, 2006, Sekine, Chandola, Martikainen, Marmot, & Kagamimori, 2009; Sekine et al., 2011), but no study has used a theory-based occupational classification, which has the advantage of clarity when describing the dimension of inequality that was actually measured (Hiyoshi et al., 2013a). We consider that a study examining the four explanatory dimensions explicitly will contribute to advancing policy discussion for reducing health inequalities, which has just begun in Japan. The aim of the present paper is to assess the contribution of material, psychosocial, social relational and behavioural factors on health inequalities for household social class and income in a working age population in Japan. We utilise social class and income as socioeconomic indicators as they may describe important aspects of health inequalities in Japan after substantial social changes occurred in the 1990s during which there appeared to be increases in job insecurity and income inequality. We calculate the attenuation that the four domains of candidate contributory factors have on social inequality in self-rated health.
Methods We analysed data from the Comprehensive Survey of Living Conditions [CSLC], a triennial survey that has been conducted since 1986. In particular, we used data from the 2001 CSLC as this was the only time that data on perceived stress and behaviours, including a detailed question on smoking, were collected. The CSLC employs multi-stage stratified random cluster sampling with the primary sampling unit being the census Enumeration Districts [EDs] which divide Japan into approximately one million areas. After stratifying by prefecture and large cities, 5000 EDs were randomly selected and all households and household members living in these areas were approached to complete a Demography & Health questionnaire. In addition, 2000 EDs were randomly selected from the 5000 EDs to complete an Income & Savings questionnaire, and we used this subset sample for our analyses. Response rates were 87.4% for the Demography & Health questionnaire in 2001, and 79.5% for the Income & Savings questionnaire, respectively (MHLW, 2009). Having excluded individuals with missing data in relevant variables, the sample size was n ¼ 40,243 (51.6% women). We tested mediating models / / SELF-RATED HEALTH and calculated the extent to which income and social class inequalities in health were accounted for by 1) material, 2) psychosocial, 3) social relational, and 4) behavioural factors. Outcome Self-rated health is used as the outcome. In various countries including Japan, the determinants of self-rated health appear to be similar (French et al., 2012), suggesting that the perception of selfrated health as a concept is not different for the Japanese from some Western countries. Self-rated health has been shown to be a strong predictor of all cause and cause specific mortality in many countries including Japan (Idler & Benyamini, 1997; Murata, Kondo, Tamakoshi, Yatsuya, & Toyoshima, 2006). In the present study, self-rated health was assessed from the single question: ‘what is your current health (condition)?’. The five categories of response were: excellent, very good, good, fair, and poor. The variable was dichotomised, setting ‘poor’ and ‘fair’ responses as the outcome and expressed as ‘suboptimal health’ hereafter (Perlman & Bobak, 2008). Socioeconomic measures We used the Japanese Socioeconomic Classification [J-SEC]. It adopted the conceptual basis of the UK’s National Statistics Socioeconomic Classification [NS-SEC] (National Statistics, 2005), which differentiates the social position of individuals in terms of employment conditions and relations. In particular, J-SEC was based on the NS-SEC three category version since it is a hierarchical construct (National Statistics, 2005) which is suitable to be summarised using the relative index of inequality [RII] explained below. It was constructed using the Japanese Standard Classification of Occupation together with employment status (such as executives of companies, self-employed, employee, and limited term contract) and predicted economic and health differences for the Japanese population (Hiyoshi et al., 2013a). We used household social class, assigned by taking the highest social class value for any household member aged 15 or greater. Individuals who had missing data in variables used to derive J-SEC or lacked a household member having classifiable jobs were not assigned to a class (n ¼ 7,192, 17.9%). Annual household income, including benefits and inheritance before tax, was equivalised by dividing by the square root of household size. The study population was grouped into income
A. Hiyoshi et al. / Social Science & Medicine 104 (2014) 201e209
203
deciles, separately for the young (20e39 years) and old (40e59 years), to account for age differences in income.
candidate contributory factors and health to differ by sex in the present study.
Candidate contributory factors
Statistical analysis
Material factors Material factors included homeownership and living density. Homeownership consisted of five levels: 1) owned house, 2) renting, 3) work-related accommodation (let by an employer for employee), 4) social housing, and 5) lodging (renting a room in a house or flat). Social housing includes public housing provision to accommodate those with financial constraints and to achieve swift housing provision to the middle class. Living density was calculated by dividing the number of household members by the number of rooms.
In preliminary analyses, chi-squared tests for heterogeneity and trend were used to examine the association of suboptimal selfrated health according to levels of exposure and candidate contributory factors. Since the prevalence of suboptimal self-rated health increased monotonically with poorer SEP, the effect of the SEP measures (household J-SEC and household income) were summarised using the relative index of inequality [RII]. To estimate the RII, a rank variable was created for each socioeconomic indicator where each category was assigned the value, between 0 and 1, of the cumulative midpoint of their ranges; for example, the lowest income decile was assigned a score of 0.95, and the highest decile a score of 0.05. This score was then used in logistic regression models and the RII estimated using the regression coefficient associated with this score. A strength of using the RII is the ability to provide a single summary measure of health disparity, including direction and magnitude, which uses all data (Mackenbach & Kunst, 1997). Initially, associations between candidate contributory factors and self-rated health, as well as those between SEP indicators and candidate contributory factors, were examined. Logistic regression models were fitted to assess the former. To assess the latter, the SEP rank variable was used in logistic, multinomial logistic and linear regression models, depending upon the variable type of the candidate contributory factors which were used as the outcome. The RII in these models summarises the relative difference in candidate contributory factors between the extreme bottom versus the extreme top of the socioeconomic hierarchy. A series of logistic regression models with suboptimal self-rated health as the outcome were then fitted for the SEP measures, starting with an initial base model which adjusted for age, prefecture and clustering within households. Adjustment for household clustering was included to allow for the intra-household correlation between individuals, which was much stronger (intraclass correlation ¼ 0.336) than that for prefecture (intraclass correlation ¼ 0.003). Subsequent models then additionally adjusted for each of the four domains of candidate contributory factors separately and, finally, all together. The RII in these models summarises the relative difference in the risk of suboptimal self-rated health between the extreme bottom versus the extreme top of the socioeconomic hierarchy. The degree to which the RII was attenuated when adjusted for a candidate contributory factor was calculated by 100*(b0b1)/b0, where (b1) is the coefficient for the SEP score in a model with the contributory variable and (b0) is the coefficient for the SEP score in an initial base model without the candidate contributory factor, but still adjusted for age, prefecture and household clustering. The confidence interval for the percentage attenuation in the RII was obtained using a bias-corrected bootstrap method, with 2000 times re-samplings. Attributable percentage attenuations of the candidate contributory factors were calculated by subtracting the percentage attenuation in the RII of a model including all variables except for the given factor, from a model including all variables (van Oort, van Lenthe, & Mackenbach, 2005; Skalicka, van Lenthe, Bambra, Krokstad, & Mackenbach, 2009). This indicates the percentage of contribution that is attributable to a given factor alone, within the observed factors in our models. The population attributable risk fraction (‘population proportional attributable risk’ in Kirkwood and Sterne (2003)) due to having a social class not at the highest level or an income below the median income was calculated using the standard formula. Although the comparison of estimates derived from different
Behavioural factors Behavioural variables included insufficient sleep, unbalanced diet, irregular intake of meals, physical inactivity, excessive alcohol intake, smoking, and attendance at health check-ups. The first five of these behaviour variables were sub-questions of a lifestyle question: ‘Do you carry out daily the things specified below in order to maintain your health?’. The response was ‘yes’ when a respondent considered he/she agreed with the statement, otherwise a ‘no’ response is given to each question: I sleep enough; I eat a balanced diet; I regularly eat a morning, lunch and evening meal; I exercise an appropriate amount (or moderate amount) or am actively mobile; and, I tend not to drink too much. Smoking was measured by a four-level question: ‘Do you smoke?’. Respondents selected an answer from ‘I don’t smoke’, ‘I smoke every day’, ‘I smoke sometimes’, and ‘I have smoked in the past but have not smoked for more than one month’. The variable is used as a categorical variable having four levels. Health check-up was assessed by the question: ‘Have you taken a general health check-up or a comprehensive medical check-up in the last year’. The response was either ‘yes’ or ‘no’. These variables were coded so that the reference category was the healthier behaviour. Psychosocial factor Perceived stress was the only psychosocial factor ascertained and was measured by a single question: ‘Currently, do you have anxiety or stress in your daily life?’. The response was either ‘yes’ or ‘no’, coded with ‘no’ as the reference category. The measure has shown to be significantly associated with self-reported symptoms such as fatigue, irritated, headache and difficulty to sleep (Hamanishi, 2013). Social relational factors The two relational factors used were marital status (married, single, widowed, and separated) and living alone (1 vs 2 people in the household). Covariates Age, categorised into five-year intervals, and prefecture, the largest administrative division separating Japan into 47 areas, were treated as a series of binary dummy variables and considered as confounding factors. In previous analyses using other CSLC survey waves between 1986 and 2007 (Hiyoshi, Fukuda, Shipley, & Brunner, 2013b) we observed sex interactions between SEP and self-rated health. Therefore, to examine health inequalities, as well as the influence of candidate contributory factors, we stratified our analyses by sex. This then allowed the associations between SEP and candidate contributory factors as well as between the
204
A. Hiyoshi et al. / Social Science & Medicine 104 (2014) 201e209
Table 1 Characteristics of the sample and prevalence of suboptimal self-rated health according to all measures.
Age (mean, years) Prefecture (1e47) (mean observations (range) Self-rated health Optimal (excellent, very good, or good) Suboptimal (fair or poor) Variables Household J-SEC I II III P-value for x2 trend Household income (decile) 1 (highest) 2 3 4 5 6 7 8 9 10 (lowest) P-value trend Age (categorical) 20e24 25e29 30e34 35e39 40e44 45e49 50e54 55e59 P-value trend Material factors Homeownership Owning Renting Work-related Social housing Lodging P-value Living density Living density (mean, (SD), persons per room) P-value (t-test) Psychosocial factors Perceived stress Not stressed Stressed P-value Social relational factors Marital status Married Single Widowed Separated P-value Living alone Not alone Alone P-value Behavioural factors Sleep Enough sleep Not enough sleep P-value Diet Balanced diet Unbalanced diet P-value Meals Regular meals Irregular meals P-value
Men (n ¼ 19,486)
Women (n ¼ 20,757)
40.9 415 (128, 816)
40.7 442 (157, 868)
17,696 1790 (9.2%) %a or mean
b
Suboptimal health (%)
18,434 2323 (11.2%) %a or mean
Suboptimal health (%)b
52.3 27.2 20.5
8.6 8.5 9.9 0.062
52.4 28.8 18.8
10.9 10.0 11.6 0.66
10.2 10.4 10.3 10.2 10.0 10.0 10.2 9.9 9.6 9.3
7.7 8.5 9.7 9.3 7.7 8.0 8.6 10.8 9.2 12.8
