Changes in Socioeconomic Differences in Hospital Days With Age: Cumulative Disadvantage, Age-as-Leveler, or Both?

Changes in Socioeconomic Differences in Hospital Days With Age: Cumulative Disadvantage,... Abstract Objectives Length of hospital stay is inversely associated with socioeconomic status (SES). It is less clear whether socioeconomic disparities in numbers of hospital days diverge or converge with age. Method Longitudinal linked Finnish registry data (1988–2007) from 137,653 men and women aged 50–79 years at the end of 1987 were used. Trajectories of annual total hospital days by education, household income, and occupational class were estimated using negative binomial models. Results Men and women with higher education, household income, and occupational class had fewer hospital days in 1988 than those with lower SES. Hospital days increased between 1988 and 2007. For some age groups, higher SES was associated with a faster annual rate of increase, resulting in narrowing rate ratios of hospital days between SES groups (relative differences); the rate ratios remained stable for other groups. Absolute SES differences in numbers of hospital days appeared to diverge with age among those aged 50–69 years at baseline, but converge among those aged 70–79 years at baseline. Discussion The hypotheses that socioeconomic disparities in health diverge or converge with age may not be mutually exclusive; we demonstrated convergence/maintenance in relative differences for all age groups, but divergence or convergence in absolute differences depending on age. Socioeconomic disparities, Hospital days, Longitudinal registry-based data Population aging is a global phenomenon, and a major challenge for the sustainability of health care and social security systems (Rechel et al., 2013; World Health Organization, 2015). Ensuring that older adults remain in good health as long as possible not only improves the quality of life for the aging population but also buffer against the expected cost pressures on the care system (Rechel et al., 2013; World Health Organization, 2015). Research on the determinants of healthy aging can greatly contribute to this goal. Socioeconomic status (SES) is one of the most established and persistent determinants of health for older adults. SES is multidimensional and can be reflected by education, occupation, and income. These SES indicators are correlated but they represent different types of resources and are associated with health through both shared and independent pathways (Braveman, Egerter, & Williams, 2011). Education is clearly associated with employment opportunities, access to information and ability to process information, ability to develop and change behaviors, and control of life. Occupation is linked with physical working environment, employment-related earnings and benefits, and work-related stress and social support. Whereas income is related to direct access to health care, nutrition, housing, transport, and other resources. As a result, to better understand how SES is linked to health, it is beneficial to investigate various SES indicators. For high-income countries, 55% of the total loss of disability-adjusted life years in 2015 was attributable to health conditions among the older population aged more than 60 years (World Health Organization, 2016). Reducing health differentials at older ages and improving health of the more disadvantaged older adults in particular could reduce the total disease burden and ameliorate health of the whole population (Feinstein, 1993; Grundy & Sloggett, 2003; Huisman, Kunst, & Mackenbach, 2003; Marmot, 2005; World Health Organization, 2015). A number of studies have examined how socioeconomic disparities in health change with age among older adults, but the evidence remains mixed (Corna, 2013; Pavalko & Wilson, 2011). Some studies have found that socioeconomic disparities in health increase with age, that is, a further divergence of health differentials between SES groups or the effect of SES on health increases with age (Dupre, 2007; Lynch, 2003; Mirowsky & Ross, 2008; Willson, Shuey, & Elder, 2007). This finding of a further divergence supports the cumulative advantages/disadvantages (CAD) hypothesis: advantages/disadvantages in social, economic, behavioral, and psychosocial resources gradually accumulate over the life course and thus produce enlarging differences in health with age. However, narrowing disparities of health between SES groups with age also have been reported in previous studies, that is, a convergence of the health differentials between SES groups or the effect of SES on health diminishes with age (Herd, 2006; House et al., 1994; Ross & Wu, 1996). This observation of a convergence provides evidence for the age-as-leveler (AAL) hypothesis: there is a biological ceiling in late life that older adults from different SES groups become universally fragile with age, placing less importance of social determinants to aging. The convergence, however, can also be related to mortality selection that less healthy people from lower SES groups are more likely to die prematurely than their counterparts from higher SES groups (Beckett, 2000; Lauderdale, 2001). Still other studies have shown that there are constant socioeconomic gaps in health at older ages (status maintenance; Leopold & Engelhardt, 2013; Pavalko & Wilson, 2011; Stolz, Mayerl, Waxenegger, Rasky, & Freidl, 2017). The inconsistency of these findings is largely attributable to differences in study design; that is, to differences in the health outcomes studied and in whether the data used were cross-sectional or longitudinal (Leopold & Engelhardt, 2013; Pavalko & Wilson, 2011). Cross-sectional data provide only a snapshot of age-specific health patterns, whereas the comparability of longitudinal studies is hampered by the varying characteristics of the study samples (e.g., countries and age groups) and the statistical methods used to analyze the data. Long-term panel data are preferable when investigating changes in socioeconomic differences in health at older ages, as they allow for the separation of age and birth cohort effects (Dupre, 2007; House, Lantz, & Herd, 2005; Leopold & Engelhardt, 2013; Pavalko & Wilson, 2011). However, problems related to nonresponse and attrition can threaten the validity of findings from panel studies; an issue that only a few studies have addressed explicitly (Howe, Tilling, Galobardes, & Lawlor, 2013; Lynch & Brown, 2011; Pavalko & Wilson, 2011; Willson et al., 2007). The association between SES and health are often measured by relative and absolute differences. The relative difference focuses on the equality, independent of the actual level of health in each SES group, whereas the absolute difference emphasizes the difference in the actual level of health between lower and higher SES groups (Mackenbach, 2015). It is important, if possible, to estimate both absolute and relative SES differences in health for better monitoring of health inequalities, evaluating policy interventions, and improving our understanding of the causes of health differentials (Regidor, 2004). However, we are not aware of research on changes in relative differences in health with age using long-term longitudinal data with repeated measures of health. It remains unclear whether the changes of absolute and relative SES differences in health with age are similar to those observed for mortality; that is, exhibiting opposing trends for the relative and absolute difference with age (Bor, Cohen, & Galea, 2017; Mackenbach, Kulhánová, et al., 2016). Older adults are disproportionately heavy users of health care and long-term care services (Wolinsky, Culler, Callahan, & Johnson, 1994; World Health Organization, 2015). Length of hospital stay, an important index of morbidity and resource consumption, has been found to be inversely associated with education, income, and occupational class (Epstein, Stern, & Weissman, 1990; Liao, McGee, Kaufman, Cao, & Cooper, 1999; Roos & Mustard, 1997). The recent World Report on Ageing and Health (World Health Organization, 2015) revealed pronounced inequalities that, compared to more advantaged older adults, disadvantaged older adults have more difficulties accessing health services, over and above their higher health risks and more severe health problems. To our knowledge, no previous study has investigated how absolute and relative SES differences in hospital use change as individuals grow older. In this study, we examined these issues using 20 years of Finnish registry data. Methods Study Population We used data from a linked register-based 11% random sample of the population residing in Finland at the end of each year between 1987 and 2007. We restricted our study population to individuals aged 50–79 years at the end of 1987 (i.e., born between January 1908 and December 1937). We also excluded individuals who (a) were not part of the dwelling population of Finland (e.g., being institutionalized or imprisoned) at the end of 1985 or 1987; (b) were not residing in Finland at the end of any given year in the 1987–2007 period; or (c) had died in 1988 with no hospitalizations as they contributed no information toward hospital use in the 1988–2007 period. The final size of the study population was 137,653 (59,586 men and 78,067 women). The cohort was followed up annually between 1988 and 2007. Hospital Days Hospitalization episodes between January 1, 1988, and December 31, 2007, were extracted from hospital discharge records. We calculated the total number of hospital days for each year of the 1988–2007 period. The annual hospital days could include multiple hospitalization episodes in a given year, and the days from all episodes were added up. If no hospitalization occurred in a given year, the number of hospital days was coded as zero. Socioeconomic Status To reflect the multidimensional nature of SES, we selected three indicators from the labor market file: highest educational attainment, household income, and occupational class. Highest educational attainment and household income were measured at the end of 1987. Individuals with basic education (i.e., less than upper secondary school) were the dominant group. We thus dichotomized educational attainment into basic education and beyond basic education (i.e., upper secondary school or higher). Household income was derived by dividing the taxable household income by the number of consumption units in the household using the Organisation for Economic Co-operation and Development (OECD)-modified scale (Hagenaars, de Vos, & Zaidi, 1994). Household income was further categorized into tertiles by sex and 5-year age groups. Information on occupational class was available every 5 years, and we used the most recent information measured at the end of 1985 to our baseline (i.e., between the end of 1987 and the end of 1988). The occupational classes were as follows: nonspecialized manual class or specialization unknown, specialized manual class, white-collar class, and other (including farmers, self-employed, students, other occupational classes, or occupational class unknown). For individuals who were retired or unemployed in 1985, the most recent information available on their previous occupational class was used. Covariates Native language and region of residence were controlled as covariates. Native language was time-invariant, and was dichotomized into Finnish and Swedish or other languages. Region of residence was time-varying, and was updated annually at the end of each year in the 1987–2007 period. In line with Statistics Finland, we used the following region of residence categories: eastern, western, southern, and northern Finland. Statistical Analysis In each year of the 1988–2007 period, the distribution of hospital days was highly skewed, as 66%–80% of our study population was not hospitalized (i.e., had zero hospital days). We modeled the hospital days as a count response. The overdispersion and excess of zeroes of the hospital days can be handled by using the negative binomial (NB) model and two zero-altered models: the zero-inflated Poisson (ZIP) model and zero-inflated negative binomial (ZINB) model (Hilbe, 2007). We compared the model fit of the three models using information criteria and found that the ZIP and ZINB models did not outperform the NB model. We therefore used the most parsimonious NB model throughout this study. The key variables in the NB model are (a) the intercept, which captures the hospital days at baseline (i.e., the year 1988 coded as time zero); and (b) the slope, which represents the annual rate of change in hospital days over the follow-up years (i.e., the years 1989–2007 coded as time 1–19, respectively). All models were fitted using Mplus 7 (Muthén & Muthén, 1998–2015). For more details on the model comparison, see Section 1.1 of the Supplementary Materials. The “empty” model containing only the intercept and slope with no SES indicators or covariates (unconditional models) was fitted first. Because of the log link function used in the NB models, the exponentiated estimates of the intercept and the slope respectively represent the expected hospital days at baseline and the expected annual rate of increase in hospital days over the follow-up period. The hospital-day trajectories over the years 1988–2007 were thus exponential on the scale of actual days (Liu & Powers, 2007). Both the intercept and slope were then regressed on education, household income, and occupational class; adjusting for native language and region of residence (conditional models). All models were fitted separately for each SES indicator, for men and women, and for three age groups at baseline (50–59, 60–69, and 70–79 years). As a measure of relative difference, the exponentiated coefficients of the intercept and the slope for the SES indicators (conditional models) represent the ratio of expected hospital days at baseline and the ratio of the expected annual rate of increase in hospital days between SES groups respectively. The absolute SES differences in hospital days at baseline and changes in these differences over the follow-up period were plotted in figures. Results Table 1 summarizes the sociodemographic characteristics of our study population at baseline, as well as the mean hospital days in each year of the 1988–2007 period. The men in our study population were younger than the women. The mean hospital days at baseline were higher in older than in younger age groups. More than 70% of the study population had basic education. More women than men were in the white-collar class, and more men than women were in the specialized manual class. For both sexes, the mean hospital days at baseline were smaller in the higher than in the lower socioeconomic groups. The mean hospital days increased gradually with age. The mean hospital days were 5.06 days higher for men and 6.62 days higher for women in 2007 than in 1988. Table 1. Baseline Sociodemographic Characteristics and Mean Hospital Days in 1988–2007   Men  Women  N (%)  Hospital days  N (%)  Hospital days  Total  59,586    78,067    Baseline (1988)   Age (years)    50–59  27,506 (46.2)  2.39  29,367 (37.6)  2.18    60–69  20,600 (34.6)  5.06  27,915 (35.8)  3.85    70–79  11,480 (19.3)  11.08  20,785 (26.6)  10.69   Native language    Finnish  54,852 (92.1)  5.04  72,323 (92.6)  5.07    Swedish/other  4,734 (7.9)  4.40  5,744 (7.4)  4.77   Region of residence    Southern  26,676 (44.8)  4.78  36,335 (46.5)  4.98    Northern  16,750 (28.1)  4.58  21,653 (27.7)  5.03    Eastern  9,186 (15.4)  5.72  11,828 (15.2)  5.27    Western  6,974 (11.7)  5.79  8,251 (10.6)  5.03   Education    Basic education  43,898 (73.7)  5.43  60,417 (77.4)  5.51    Beyond basic education  15,688 (26.3)  3.75  17,650 (22.6)  3.46   Household income    Low  19,899 (33.4)  6.08  26,306 (33.7)  5.92    Middle  19,822 (33.3)  4.80  25,753 (33.0)  4.58    High  19,865 (33.3)  4.08  26,008 (33.3)  4.62   Occupational class    Manual, nonspecialized/unknown  10,351 (17.4)  4.90  19,301 (24.7)  4.56    Manual, specialized  18,155 (30.5)  5.53  14,139 (18.1)  5.68    White collar  15,052 (25.3)  3.93  26,852 (34.4)  4.05    Other  16,028 (26.9)  5.41  17,775 (22.8)  6.57  Year   1988  59,586 (100.0)  4.99  78,067 (100.0)  5.04   1989  58,475 (98.1)  5.26  77,088 (98.8)  5.50   1990  56,773 (95.3)  5.67  75,858 (97.2)  5.98   1991  54,967 (92.3)  5.75  74,537 (95.5)  6.38   1992  53,307 (89.5)  5.96  73,120 (93.7)  6.80   1993  51,510 (86.5)  6.16  71,604 (91.7)  7.36   1994  49,761 (83.5)  6.69  69,905 (89.5)  7.89   1995  48,009 (80.6)  7.01  68,238 (87.4)  8.56   1996  46,210 (77.6)  7.65  66,378 (85.0)  9.36   1997  44,421 (74.6)  7.84  64,586 (82.7)  9.83   1998  42,501 (71.3)  7.93  62,615 (80.2)  10.34   1999  40,643 (68.2)  8.35  60,654 (77.7)  10.89   2000  38,773 (65.1)  8.54  58,578 (75.0)  11.42   2001  37,040 (62.2)  9.02  56,353 (72.2)  11.87   2002  35,267 (59.2)  9.44  54,230 (69.5)  12.12   2003  33,469 (56.2)  9.59  52,033 (66.7)  12.32   2004  31,777 (53.3)  9.58  49,770 (63.8)  11.98   2005  30,174 (50.6)  9.93  47,656 (61.0)  12.12   2006  28,573 (48.0)  10.35  45,580 (58.4)  12.03   2007  27,022 (45.4)  10.05  43,503 (55.7)  11.66    Men  Women  N (%)  Hospital days  N (%)  Hospital days  Total  59,586    78,067    Baseline (1988)   Age (years)    50–59  27,506 (46.2)  2.39  29,367 (37.6)  2.18    60–69  20,600 (34.6)  5.06  27,915 (35.8)  3.85    70–79  11,480 (19.3)  11.08  20,785 (26.6)  10.69   Native language    Finnish  54,852 (92.1)  5.04  72,323 (92.6)  5.07    Swedish/other  4,734 (7.9)  4.40  5,744 (7.4)  4.77   Region of residence    Southern  26,676 (44.8)  4.78  36,335 (46.5)  4.98    Northern  16,750 (28.1)  4.58  21,653 (27.7)  5.03    Eastern  9,186 (15.4)  5.72  11,828 (15.2)  5.27    Western  6,974 (11.7)  5.79  8,251 (10.6)  5.03   Education    Basic education  43,898 (73.7)  5.43  60,417 (77.4)  5.51    Beyond basic education  15,688 (26.3)  3.75  17,650 (22.6)  3.46   Household income    Low  19,899 (33.4)  6.08  26,306 (33.7)  5.92    Middle  19,822 (33.3)  4.80  25,753 (33.0)  4.58    High  19,865 (33.3)  4.08  26,008 (33.3)  4.62   Occupational class    Manual, nonspecialized/unknown  10,351 (17.4)  4.90  19,301 (24.7)  4.56    Manual, specialized  18,155 (30.5)  5.53  14,139 (18.1)  5.68    White collar  15,052 (25.3)  3.93  26,852 (34.4)  4.05    Other  16,028 (26.9)  5.41  17,775 (22.8)  6.57  Year   1988  59,586 (100.0)  4.99  78,067 (100.0)  5.04   1989  58,475 (98.1)  5.26  77,088 (98.8)  5.50   1990  56,773 (95.3)  5.67  75,858 (97.2)  5.98   1991  54,967 (92.3)  5.75  74,537 (95.5)  6.38   1992  53,307 (89.5)  5.96  73,120 (93.7)  6.80   1993  51,510 (86.5)  6.16  71,604 (91.7)  7.36   1994  49,761 (83.5)  6.69  69,905 (89.5)  7.89   1995  48,009 (80.6)  7.01  68,238 (87.4)  8.56   1996  46,210 (77.6)  7.65  66,378 (85.0)  9.36   1997  44,421 (74.6)  7.84  64,586 (82.7)  9.83   1998  42,501 (71.3)  7.93  62,615 (80.2)  10.34   1999  40,643 (68.2)  8.35  60,654 (77.7)  10.89   2000  38,773 (65.1)  8.54  58,578 (75.0)  11.42   2001  37,040 (62.2)  9.02  56,353 (72.2)  11.87   2002  35,267 (59.2)  9.44  54,230 (69.5)  12.12   2003  33,469 (56.2)  9.59  52,033 (66.7)  12.32   2004  31,777 (53.3)  9.58  49,770 (63.8)  11.98   2005  30,174 (50.6)  9.93  47,656 (61.0)  12.12   2006  28,573 (48.0)  10.35  45,580 (58.4)  12.03   2007  27,022 (45.4)  10.05  43,503 (55.7)  11.66  View Large Table 1. Baseline Sociodemographic Characteristics and Mean Hospital Days in 1988–2007   Men  Women  N (%)  Hospital days  N (%)  Hospital days  Total  59,586    78,067    Baseline (1988)   Age (years)    50–59  27,506 (46.2)  2.39  29,367 (37.6)  2.18    60–69  20,600 (34.6)  5.06  27,915 (35.8)  3.85    70–79  11,480 (19.3)  11.08  20,785 (26.6)  10.69   Native language    Finnish  54,852 (92.1)  5.04  72,323 (92.6)  5.07    Swedish/other  4,734 (7.9)  4.40  5,744 (7.4)  4.77   Region of residence    Southern  26,676 (44.8)  4.78  36,335 (46.5)  4.98    Northern  16,750 (28.1)  4.58  21,653 (27.7)  5.03    Eastern  9,186 (15.4)  5.72  11,828 (15.2)  5.27    Western  6,974 (11.7)  5.79  8,251 (10.6)  5.03   Education    Basic education  43,898 (73.7)  5.43  60,417 (77.4)  5.51    Beyond basic education  15,688 (26.3)  3.75  17,650 (22.6)  3.46   Household income    Low  19,899 (33.4)  6.08  26,306 (33.7)  5.92    Middle  19,822 (33.3)  4.80  25,753 (33.0)  4.58    High  19,865 (33.3)  4.08  26,008 (33.3)  4.62   Occupational class    Manual, nonspecialized/unknown  10,351 (17.4)  4.90  19,301 (24.7)  4.56    Manual, specialized  18,155 (30.5)  5.53  14,139 (18.1)  5.68    White collar  15,052 (25.3)  3.93  26,852 (34.4)  4.05    Other  16,028 (26.9)  5.41  17,775 (22.8)  6.57  Year   1988  59,586 (100.0)  4.99  78,067 (100.0)  5.04   1989  58,475 (98.1)  5.26  77,088 (98.8)  5.50   1990  56,773 (95.3)  5.67  75,858 (97.2)  5.98   1991  54,967 (92.3)  5.75  74,537 (95.5)  6.38   1992  53,307 (89.5)  5.96  73,120 (93.7)  6.80   1993  51,510 (86.5)  6.16  71,604 (91.7)  7.36   1994  49,761 (83.5)  6.69  69,905 (89.5)  7.89   1995  48,009 (80.6)  7.01  68,238 (87.4)  8.56   1996  46,210 (77.6)  7.65  66,378 (85.0)  9.36   1997  44,421 (74.6)  7.84  64,586 (82.7)  9.83   1998  42,501 (71.3)  7.93  62,615 (80.2)  10.34   1999  40,643 (68.2)  8.35  60,654 (77.7)  10.89   2000  38,773 (65.1)  8.54  58,578 (75.0)  11.42   2001  37,040 (62.2)  9.02  56,353 (72.2)  11.87   2002  35,267 (59.2)  9.44  54,230 (69.5)  12.12   2003  33,469 (56.2)  9.59  52,033 (66.7)  12.32   2004  31,777 (53.3)  9.58  49,770 (63.8)  11.98   2005  30,174 (50.6)  9.93  47,656 (61.0)  12.12   2006  28,573 (48.0)  10.35  45,580 (58.4)  12.03   2007  27,022 (45.4)  10.05  43,503 (55.7)  11.66    Men  Women  N (%)  Hospital days  N (%)  Hospital days  Total  59,586    78,067    Baseline (1988)   Age (years)    50–59  27,506 (46.2)  2.39  29,367 (37.6)  2.18    60–69  20,600 (34.6)  5.06  27,915 (35.8)  3.85    70–79  11,480 (19.3)  11.08  20,785 (26.6)  10.69   Native language    Finnish  54,852 (92.1)  5.04  72,323 (92.6)  5.07    Swedish/other  4,734 (7.9)  4.40  5,744 (7.4)  4.77   Region of residence    Southern  26,676 (44.8)  4.78  36,335 (46.5)  4.98    Northern  16,750 (28.1)  4.58  21,653 (27.7)  5.03    Eastern  9,186 (15.4)  5.72  11,828 (15.2)  5.27    Western  6,974 (11.7)  5.79  8,251 (10.6)  5.03   Education    Basic education  43,898 (73.7)  5.43  60,417 (77.4)  5.51    Beyond basic education  15,688 (26.3)  3.75  17,650 (22.6)  3.46   Household income    Low  19,899 (33.4)  6.08  26,306 (33.7)  5.92    Middle  19,822 (33.3)  4.80  25,753 (33.0)  4.58    High  19,865 (33.3)  4.08  26,008 (33.3)  4.62   Occupational class    Manual, nonspecialized/unknown  10,351 (17.4)  4.90  19,301 (24.7)  4.56    Manual, specialized  18,155 (30.5)  5.53  14,139 (18.1)  5.68    White collar  15,052 (25.3)  3.93  26,852 (34.4)  4.05    Other  16,028 (26.9)  5.41  17,775 (22.8)  6.57  Year   1988  59,586 (100.0)  4.99  78,067 (100.0)  5.04   1989  58,475 (98.1)  5.26  77,088 (98.8)  5.50   1990  56,773 (95.3)  5.67  75,858 (97.2)  5.98   1991  54,967 (92.3)  5.75  74,537 (95.5)  6.38   1992  53,307 (89.5)  5.96  73,120 (93.7)  6.80   1993  51,510 (86.5)  6.16  71,604 (91.7)  7.36   1994  49,761 (83.5)  6.69  69,905 (89.5)  7.89   1995  48,009 (80.6)  7.01  68,238 (87.4)  8.56   1996  46,210 (77.6)  7.65  66,378 (85.0)  9.36   1997  44,421 (74.6)  7.84  64,586 (82.7)  9.83   1998  42,501 (71.3)  7.93  62,615 (80.2)  10.34   1999  40,643 (68.2)  8.35  60,654 (77.7)  10.89   2000  38,773 (65.1)  8.54  58,578 (75.0)  11.42   2001  37,040 (62.2)  9.02  56,353 (72.2)  11.87   2002  35,267 (59.2)  9.44  54,230 (69.5)  12.12   2003  33,469 (56.2)  9.59  52,033 (66.7)  12.32   2004  31,777 (53.3)  9.58  49,770 (63.8)  11.98   2005  30,174 (50.6)  9.93  47,656 (61.0)  12.12   2006  28,573 (48.0)  10.35  45,580 (58.4)  12.03   2007  27,022 (45.4)  10.05  43,503 (55.7)  11.66  View Large Table 2 shows the association between SES and the trajectories of hospital days (relative difference) for men and women. The unconditional estimates of both the hospital days at baseline (i.e., intercept) and the annual rate of increase in hospital days (i.e., slope) from the “empty” models were similar for men and women aged 50–59 years at baseline. On average, women aged 60–69 years at baseline had fewer hospital days at baseline, but a faster rate of increase than their male counterparts (difference in the slope p <.001). Women aged 70–79 years at baseline had both more hospital days at baseline and a faster rate of increase (difference in the slope p = .01) than men in the same age band. Table 2. SES and Trajectory of Hospital Days (Relative Difference), Stratified by Age Groups at Baseline   Men  Women  Intercept  Slope  Intercept  Slope  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  Age at baseline  50–59 Years   Unconditional estimates  2.24 (2.11, 2.37)***  1.06 (1.06, 1.06)***  2.02 (1.90, 2.16)***  1.06 (1.05, 1.06)***   Conditional estimates    Educationa     Beyond basic  0.75 (0.66, 0.85)***  1.01 (1.00, 1.02)  0.72 (0.63, 0.82)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)  0.55 (0.48, 0.63)***  1.01 (1.00, 1.02)*     High  0.43 (0.38, 0.49)***  1.02 (1.01, 1.03)**  0.44 (0.38, 0.52)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.01 (0.85, 1.19)  0.99 (0.98, 1.00)  1.22 (1.01, 1.47)*  1.00 (0.98, 1.01)     White collar  0.64 (0.54, 0.75)***  1.00 (0.99, 1.01)  0.83 (0.71, 0.97)*  0.99 (0.98, 1.01)     Other  0.87 (0.74, 1.03)  1.00 (0.99, 1.01)  1.35 (1.09, 1.67)**  0.99 (0.97, 1.00)  60–69 Years   Unconditional estimates  5.08 (4.83, 5.35)***  1.07 (1.06, 1.07)***  3.95 (3.76, 4.15)***  1.09 (1.09, 1.09)***   Conditional estimates    Educationa     Beyond basic  0.88 (0.77, 1.00)  1.00 (0.99, 1.01)  0.74 (0.65, 0.84)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.77 (0.69, 0.87)***  1.01 (1.00, 1.02)*  0.80 (0.72, 0.90)***  1.01 (1.00, 1.02)     High  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)*  0.73 (0.65, 0.83)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.04 (0.96, 1.20)  0.99 (0.98, 1.01)  1.25 (1.08, 1.45)**  0.99 (0.98, 1.00)     White collar  0.88 (0.75, 1.04)  0.99 (0.98, 1.01)  0.89 (0.78, 1.02)  1.00 (0.99, 1.01)     Other  0.96 (0.83, 1.11)  1.00 (0.98, 1.01)  1.23 (1.08, 1.41)**  0.99 (0.98, 1.00)*  70–79 Years   Unconditional estimates  11.73 (11.11, 12.39)***  1.07 (1.06, 1.07)***  12.33 (11.81, 12.88)***  1.08 (1.07, 1.08)***   Conditional estimates    Educationa     Beyond basic  0.79 (0.69, 0.91)**  1.01 (0.99, 1.03)  0.78 (0.69, 0.89)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.89 (0.78, 1.02)  1.00 (0.99, 1.02)  0.86 (0.78, 0.96)**  1.01 (1.00, 1.02)     High  0.88 (0.77, 1.00)*  1.01 (0.99, 1.02)  0.86 (0.77, 0.95)**  1.00 (0.99, 1.01)    Occupational classc     Manual, specialized  1.02 (0.85, 1.22)  0.99 (0.97, 1.01)  0.83 (0.72, 0.95)**  1.02 (1.00, 1.03)*     White collar  0.83 (0.68, 1.02)  1.00 (0.97, 1.02)  0.83 (0.72, 0.95)**  1.01 (1.00, 1.03)*     Other  0.91 (0.76, 1.09)  1.01 (0.98, 1.03)  0.83 (0.73, 0.95)**  1.02 (1.00, 1.03)*    Men  Women  Intercept  Slope  Intercept  Slope  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  Age at baseline  50–59 Years   Unconditional estimates  2.24 (2.11, 2.37)***  1.06 (1.06, 1.06)***  2.02 (1.90, 2.16)***  1.06 (1.05, 1.06)***   Conditional estimates    Educationa     Beyond basic  0.75 (0.66, 0.85)***  1.01 (1.00, 1.02)  0.72 (0.63, 0.82)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)  0.55 (0.48, 0.63)***  1.01 (1.00, 1.02)*     High  0.43 (0.38, 0.49)***  1.02 (1.01, 1.03)**  0.44 (0.38, 0.52)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.01 (0.85, 1.19)  0.99 (0.98, 1.00)  1.22 (1.01, 1.47)*  1.00 (0.98, 1.01)     White collar  0.64 (0.54, 0.75)***  1.00 (0.99, 1.01)  0.83 (0.71, 0.97)*  0.99 (0.98, 1.01)     Other  0.87 (0.74, 1.03)  1.00 (0.99, 1.01)  1.35 (1.09, 1.67)**  0.99 (0.97, 1.00)  60–69 Years   Unconditional estimates  5.08 (4.83, 5.35)***  1.07 (1.06, 1.07)***  3.95 (3.76, 4.15)***  1.09 (1.09, 1.09)***   Conditional estimates    Educationa     Beyond basic  0.88 (0.77, 1.00)  1.00 (0.99, 1.01)  0.74 (0.65, 0.84)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.77 (0.69, 0.87)***  1.01 (1.00, 1.02)*  0.80 (0.72, 0.90)***  1.01 (1.00, 1.02)     High  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)*  0.73 (0.65, 0.83)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.04 (0.96, 1.20)  0.99 (0.98, 1.01)  1.25 (1.08, 1.45)**  0.99 (0.98, 1.00)     White collar  0.88 (0.75, 1.04)  0.99 (0.98, 1.01)  0.89 (0.78, 1.02)  1.00 (0.99, 1.01)     Other  0.96 (0.83, 1.11)  1.00 (0.98, 1.01)  1.23 (1.08, 1.41)**  0.99 (0.98, 1.00)*  70–79 Years   Unconditional estimates  11.73 (11.11, 12.39)***  1.07 (1.06, 1.07)***  12.33 (11.81, 12.88)***  1.08 (1.07, 1.08)***   Conditional estimates    Educationa     Beyond basic  0.79 (0.69, 0.91)**  1.01 (0.99, 1.03)  0.78 (0.69, 0.89)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.89 (0.78, 1.02)  1.00 (0.99, 1.02)  0.86 (0.78, 0.96)**  1.01 (1.00, 1.02)     High  0.88 (0.77, 1.00)*  1.01 (0.99, 1.02)  0.86 (0.77, 0.95)**  1.00 (0.99, 1.01)    Occupational classc     Manual, specialized  1.02 (0.85, 1.22)  0.99 (0.97, 1.01)  0.83 (0.72, 0.95)**  1.02 (1.00, 1.03)*     White collar  0.83 (0.68, 1.02)  1.00 (0.97, 1.02)  0.83 (0.72, 0.95)**  1.01 (1.00, 1.03)*     Other  0.91 (0.76, 1.09)  1.01 (0.98, 1.03)  0.83 (0.73, 0.95)**  1.02 (1.00, 1.03)*  Conditional models adjusted for native language and region of residence; all models were fitted separately for education, household income, and occupational class. IRR: Incidence rate ratio. *** p < .001 ** p < .01 * p < .05. Reference category a Basic education b Low household income tertile c Nonspecialized manual class or specialization unknown. View Large Table 2. SES and Trajectory of Hospital Days (Relative Difference), Stratified by Age Groups at Baseline   Men  Women  Intercept  Slope  Intercept  Slope  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  Age at baseline  50–59 Years   Unconditional estimates  2.24 (2.11, 2.37)***  1.06 (1.06, 1.06)***  2.02 (1.90, 2.16)***  1.06 (1.05, 1.06)***   Conditional estimates    Educationa     Beyond basic  0.75 (0.66, 0.85)***  1.01 (1.00, 1.02)  0.72 (0.63, 0.82)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)  0.55 (0.48, 0.63)***  1.01 (1.00, 1.02)*     High  0.43 (0.38, 0.49)***  1.02 (1.01, 1.03)**  0.44 (0.38, 0.52)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.01 (0.85, 1.19)  0.99 (0.98, 1.00)  1.22 (1.01, 1.47)*  1.00 (0.98, 1.01)     White collar  0.64 (0.54, 0.75)***  1.00 (0.99, 1.01)  0.83 (0.71, 0.97)*  0.99 (0.98, 1.01)     Other  0.87 (0.74, 1.03)  1.00 (0.99, 1.01)  1.35 (1.09, 1.67)**  0.99 (0.97, 1.00)  60–69 Years   Unconditional estimates  5.08 (4.83, 5.35)***  1.07 (1.06, 1.07)***  3.95 (3.76, 4.15)***  1.09 (1.09, 1.09)***   Conditional estimates    Educationa     Beyond basic  0.88 (0.77, 1.00)  1.00 (0.99, 1.01)  0.74 (0.65, 0.84)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.77 (0.69, 0.87)***  1.01 (1.00, 1.02)*  0.80 (0.72, 0.90)***  1.01 (1.00, 1.02)     High  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)*  0.73 (0.65, 0.83)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.04 (0.96, 1.20)  0.99 (0.98, 1.01)  1.25 (1.08, 1.45)**  0.99 (0.98, 1.00)     White collar  0.88 (0.75, 1.04)  0.99 (0.98, 1.01)  0.89 (0.78, 1.02)  1.00 (0.99, 1.01)     Other  0.96 (0.83, 1.11)  1.00 (0.98, 1.01)  1.23 (1.08, 1.41)**  0.99 (0.98, 1.00)*  70–79 Years   Unconditional estimates  11.73 (11.11, 12.39)***  1.07 (1.06, 1.07)***  12.33 (11.81, 12.88)***  1.08 (1.07, 1.08)***   Conditional estimates    Educationa     Beyond basic  0.79 (0.69, 0.91)**  1.01 (0.99, 1.03)  0.78 (0.69, 0.89)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.89 (0.78, 1.02)  1.00 (0.99, 1.02)  0.86 (0.78, 0.96)**  1.01 (1.00, 1.02)     High  0.88 (0.77, 1.00)*  1.01 (0.99, 1.02)  0.86 (0.77, 0.95)**  1.00 (0.99, 1.01)    Occupational classc     Manual, specialized  1.02 (0.85, 1.22)  0.99 (0.97, 1.01)  0.83 (0.72, 0.95)**  1.02 (1.00, 1.03)*     White collar  0.83 (0.68, 1.02)  1.00 (0.97, 1.02)  0.83 (0.72, 0.95)**  1.01 (1.00, 1.03)*     Other  0.91 (0.76, 1.09)  1.01 (0.98, 1.03)  0.83 (0.73, 0.95)**  1.02 (1.00, 1.03)*    Men  Women  Intercept  Slope  Intercept  Slope  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  Age at baseline  50–59 Years   Unconditional estimates  2.24 (2.11, 2.37)***  1.06 (1.06, 1.06)***  2.02 (1.90, 2.16)***  1.06 (1.05, 1.06)***   Conditional estimates    Educationa     Beyond basic  0.75 (0.66, 0.85)***  1.01 (1.00, 1.02)  0.72 (0.63, 0.82)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)  0.55 (0.48, 0.63)***  1.01 (1.00, 1.02)*     High  0.43 (0.38, 0.49)***  1.02 (1.01, 1.03)**  0.44 (0.38, 0.52)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.01 (0.85, 1.19)  0.99 (0.98, 1.00)  1.22 (1.01, 1.47)*  1.00 (0.98, 1.01)     White collar  0.64 (0.54, 0.75)***  1.00 (0.99, 1.01)  0.83 (0.71, 0.97)*  0.99 (0.98, 1.01)     Other  0.87 (0.74, 1.03)  1.00 (0.99, 1.01)  1.35 (1.09, 1.67)**  0.99 (0.97, 1.00)  60–69 Years   Unconditional estimates  5.08 (4.83, 5.35)***  1.07 (1.06, 1.07)***  3.95 (3.76, 4.15)***  1.09 (1.09, 1.09)***   Conditional estimates    Educationa     Beyond basic  0.88 (0.77, 1.00)  1.00 (0.99, 1.01)  0.74 (0.65, 0.84)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.77 (0.69, 0.87)***  1.01 (1.00, 1.02)*  0.80 (0.72, 0.90)***  1.01 (1.00, 1.02)     High  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)*  0.73 (0.65, 0.83)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.04 (0.96, 1.20)  0.99 (0.98, 1.01)  1.25 (1.08, 1.45)**  0.99 (0.98, 1.00)     White collar  0.88 (0.75, 1.04)  0.99 (0.98, 1.01)  0.89 (0.78, 1.02)  1.00 (0.99, 1.01)     Other  0.96 (0.83, 1.11)  1.00 (0.98, 1.01)  1.23 (1.08, 1.41)**  0.99 (0.98, 1.00)*  70–79 Years   Unconditional estimates  11.73 (11.11, 12.39)***  1.07 (1.06, 1.07)***  12.33 (11.81, 12.88)***  1.08 (1.07, 1.08)***   Conditional estimates    Educationa     Beyond basic  0.79 (0.69, 0.91)**  1.01 (0.99, 1.03)  0.78 (0.69, 0.89)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.89 (0.78, 1.02)  1.00 (0.99, 1.02)  0.86 (0.78, 0.96)**  1.01 (1.00, 1.02)     High  0.88 (0.77, 1.00)*  1.01 (0.99, 1.02)  0.86 (0.77, 0.95)**  1.00 (0.99, 1.01)    Occupational classc     Manual, specialized  1.02 (0.85, 1.22)  0.99 (0.97, 1.01)  0.83 (0.72, 0.95)**  1.02 (1.00, 1.03)*     White collar  0.83 (0.68, 1.02)  1.00 (0.97, 1.02)  0.83 (0.72, 0.95)**  1.01 (1.00, 1.03)*     Other  0.91 (0.76, 1.09)  1.01 (0.98, 1.03)  0.83 (0.73, 0.95)**  1.02 (1.00, 1.03)*  Conditional models adjusted for native language and region of residence; all models were fitted separately for education, household income, and occupational class. IRR: Incidence rate ratio. *** p < .001 ** p < .01 * p < .05. Reference category a Basic education b Low household income tertile c Nonspecialized manual class or specialization unknown. View Large As we can see from the conditional estimates, there was a clear gradient in the intercept indicating that among both sexes, individuals with higher education, household income, and occupational class had around 10%–60% fewer hospital days at baseline than their counterparts with lower SES. However, among men, most of the differences in the intercept between occupational classes were not statistically significant. For the slope, among some subgroups, higher SES was associated with a faster annual rate of increase in hospital days. For example, the rate of increase was 2% faster for men aged 50–59 years at baseline in the high household income tertile than for those in the low household income tertile (ratio of annual rate of increase: 1.02, 95% confidence interval [CI]: 1.01, 1.03). For men aged 60–69 years at baseline, being in the middle or high household income tertiles was associated with a 1% faster rate of increase than being in the low household income tertile (middle household income: 1.01, 95% CI: 1.00, 1.02; high household income: 1.01, 95% CI: 1.00, 1.03). For women aged 50–59 years at baseline, being in the middle household income tertile was associated with a 1% faster rate of increase than being in the low household income tertile (1.01; 95% CI: 1.00, 1.02). For women aged 70–79 years at baseline, compared to being in the nonspecialized manual class/specialization unknown category, the rate of increase was 1% faster for being in the specialized manual class (1.01, 95% CI: 1.00, 1.03) and 2% faster for being in the white-collar class (1.02, 95% CI: 1.00, 1.03). Accordingly, relative SES differences in hospital days diminished over the follow-up period for these subgroups. For other subgroups, the relative SES differences appeared to be unchanged, given the statistically nonsignificant ratios of the rate of increase. The interactions of each SES indicator with sex and with age were tested on both the intercept and slope. The effect of SES on the trajectories of hospital days did not differ by age groups among either men or women, except for household income on the intercept (both sex p < .001), for occupational class on the intercept (men p = .04; women p < .001), and for occupational class on the slope among women (p = .02). Similarly, we did not find sex differences in the effect of SES in any age group, except for occupational class on the intercept among individuals aged 50–59 years (p = .01) and 60–69 years (p = .03) and for education on the intercept among those aged 60–69 years (p = .04). The results from the pooled study population of men and women are shown in Supplementary Table 2. Figures 1 and 2 show how absolute differences in hospital days by education and household income changed in the years 1988–2007 (see Supplementary Figure 2 for occupational class). Although relative SES differences in hospital days largely maintained and only decreased in some subgroups, absolute differences appeared to enlarge over the 1988–2007 period among men and women aged 50–69 years at baseline, but narrow among those aged 70–79 years baseline. Figure 1. View largeDownload slide Educational attainment and trajectories of hospital days in 1988–2007 (absolute difference), by sex and age groups at baseline (in 1988). Figure 1. View largeDownload slide Educational attainment and trajectories of hospital days in 1988–2007 (absolute difference), by sex and age groups at baseline (in 1988). Figure 2. View largeDownload slide Household income and trajectories of hospital days in 1988–2007 (absolute difference), by sex and age groups at baseline (in 1988). Figure 2. View largeDownload slide Household income and trajectories of hospital days in 1988–2007 (absolute difference), by sex and age groups at baseline (in 1988). Number of hospitalization episodes is another important aspect of hospital care. We repeated our analysis using the annual total number of hospitalizations as the outcome, and the findings were similar to those for hospital days (Supplementary Table 3). However, a faster increase in the number of hospitalizations associated with higher education was found among men and women aged 70–79 years at baseline but not among those aged 50–59 years at baseline. For women aged 70–79 years at baseline, a faster increase in the number of hospitalizations was associated with higher household income but not with occupational class. For the absolute differences in the number of hospitalization between education groups, the tendency of a convergence over the 1988–2007 period became more evident in the group aged 70–79 years at baseline (Supplementary Figure 3). Over the follow-up period, 34,103 men and 36,631 women died. Hospital days were missing for these individuals after their deaths. Mplus handles missing data using full information maximum likelihood; a statistical estimation technique that is valid under the assumption of missing at random (MAR, i.e., missingness depends on covariates and the outcome variable observed at previous time points; Muthén & Muthén, 1998–2015). If MAR does not hold, missing not at random (MNAR) models should be used to perform sensitivity analysis (for more details, see Section 1.2 of the Supplementary Materials; Muthén, Asparouhov, Hunter, & Leuchter, 2011). We therefore fitted pattern-mixture models within the framework of latent growth curve modeling (MNAR models), in which both the intercept and slope of the hospital days were modeled as a function of the indicators of the year of death (Little, 2009; Muthén et al., 2011). More details are provided in Section 1.2 of the Supplementary Materials. The results of the pattern-mixture models deviated little from those of the MAR models (see results in Supplementary Tables 4). For both sexes and all age groups at baseline, relative SES differences for both the intercept and slope were smaller in the pattern-mixture models than in the MAR models. The age-specific patterns of changes in absolute differences in the pattern-mixture models were generally consistent with those in the MAR models (Supplementary Figures 4–6). However, a continuous divergence was seen between household income tertiles and occupational classes for men aged 70–79 years at baseline. Discussion In this study using large-scale longitudinal registry data, we found that older adults with higher education, household income, and occupational class had fewer hospital days at baseline than their counterparts with lower SES. Over the years 1989–2007, relative differences in hospital days declined between the lowest and higher household income tertiles for men aged 50–69 years at baseline; between the lowest and the middle household income tertiles for women aged 50–59 years at baseline; and between the nonspecialized manual class and the specialized manual/white-collar class for women aged 70–79 years at baseline. For the other subgroups, relative SES differences in hospital days were unchanged. There was, however, a further divergence in absolute differences in hospital days by education, household income, and occupational class groups for men and women aged 50–69 years at baseline, but a convergence for those aged 70–79 years at baseline. Our findings suggest that the divergence and convergence of socioeconomic disparities in health at older ages may not be mutually exclusive; rather, which trend is found may depend on the measurement scale of socioeconomic disparities in health (relative or absolute difference) and the ages used. Similar to previous observations on mortality (Bor et al., 2017; Mackenbach, Kulhánová, et al., 2016), for some SES subgroups at ages 50–69 years, we found that the relative difference narrowed while the absolute difference increased. For other subgroups, the relative difference remained stable at all ages, whereas the absolute difference increased at ages 50–69 years but diminished at ages older than 70 years. It is not straightforward to link these opposing trends to AAL, CAD, or status maintenance hypotheses because inference based on relative and absolute difference may lead to different conclusions. Mackenbach, Martikainen, Menvielle, and de Gelder (2016) showed that different combinations of starting levels and changes of mortality by SES groups over time lead to different patterns of changes in relative and absolute differences in mortality. Our findings fall into one of these patterns. Whereas relative SES differences in hospital days maintained/diminished in the 1988–2007 period, changes in absolute differences were less monotonic: absolute differences started to decline only after the ratio of hospital days in the higher versus the lower SES groups became larger than the ratio of the increase in hospital days in a year in the lower versus the higher SES groups. Since the opposing trends of change in relative and absolute differences are still not fully understood (Mackenbach, Martikainen, et al., 2016), it remains for further research to answer the question of which measurement scale, relative or absolute, is more appropriate to use for testing the AAL and CAD hypotheses. From the perspective of policy makers, actual absolute hospital days are a more relevant metric for policies aiming to reduce the SES differentials in hospital care and are more meaningful and easy to implement in monitoring success in policy interventions. We compared our findings on changes in absolute differences with the results of selected previous studies that used longitudinal data with repeated measures and similar statistical analytic approaches (Benzeval, Green, & Leyland, 2011; Chandola, Ferrie, Sacker, & Marmot, 2007; Herd, 2006; Kim & Durden, 2007; Stolz et al., 2017; Willson et al., 2007). One study reported a continuous divergence in the self-rated health trajectories between employment grades among British civil servants aged 39–74 years (Chandola et al., 2007), whereas another reported a continuous divergence in the physical impairment trajectories between education and income groups across ages 25–89 years (Kim & Durden, 2007). But Stolz and colleagues (2017) found stable differences in frailty with age across educational, wealth, and occupational class groups, and decreasing differences across income groups among Europeans older than 50 years. In line with our findings, other studies have shown that after diverging at earlier ages, disparities in physical functioning and self-rated health converge across education groups and manual and nonmanual classes when people reach their 60s or 70s (Benzeval et al., 2011; Herd, 2006; Willson et al., 2007). The studies conducted by Kim and Durden (2007) and Willson and colleagues (2007) are the only studies that covered most of the adult ages (ages 25–89 years and 26–92 years, respectively), whereas the other studies tracked changes in socioeconomic disparities in health only until people reached their 60s or early 70s. Our data enabled us to follow men and women aged 50–59, 60–69, and 70–79 years over a 20-year period starting at the end of 1987; and thus, until ages 70–79, 80–89, and 90–99 years, respectively. Moreover, we found important differences between the three 10-year age groups. Kim and Durden (2007) assumed common physical impairment trajectories for all ages, even though 43% of their sample were aged 25–49 years at baseline. Our results suggest that the divergence Kim and Durden (2007) found may be driven by these relatively healthy young adults, and that the convergence at older ages is hidden. Although Willson and colleagues (2007) did not make such assumption of common trajectories for all ages, the long-term panel data they used covered a relatively small sample of older adults aged 46–75 years at baseline (1,695, 31% of the whole sample), with nonresponse and attrition over the follow-up period. Uniquely, we were able to evaluate the competing hypotheses of convergence and divergence separately for different age groups with large sample sizes and no attrition. A divergence in absolute differences in hospital days over the 1988–2007 period for men and women aged 50–69 years at baseline was observed across education, household income, and occupational class groups. Of these SES indicators, income had the greatest effects on the rate of increase in hospital days. Our results support the idea that education, occupation, and income represent different types of resources and affect health through different pathways: income is more directly related to material resources available for accessing formal care services, treating health problems, or changing life circumstances to slow down the progression of health problems (House et al., 2005). The trend toward convergence in absolute SES disparities in hospital days in the group aged 70–79 years at baseline supports the AAL hypothesis, which asserts that physiological factors play a larger role in aging than social determinants (Dupre, 2007; Hoffmann, 2008). Benzeval and colleagues (2011) and Willson and colleagues (2007) argued that the convergence they observed was artificially caused by mortality selection and/or attrition. In contrast, Rohwer (2016) contended that survival should be viewed as a necessary precondition instead of as a possible source of bias, and that the growth curve models do not estimate health trajectories conditioned on survival. The pattern-mixture models we used for our sensitivity analysis are fully conditional on survival (Kurland, Johnson, Egleston, & Diehr, 2009). When mortality was taken into account in a series of pattern-mixture models, the changes in relative and absolute SES differences in hospital days found in these models were generally consistent with the results of the MAR models. This suggests that the MAR assumption (i.e., missing hospital days due to death were random after taking into account covariates and hospital days at previous time points) is reasonable. These observations partially confirm the assumption that mortality selection is not the main reason for the convergence in socioeconomic disparities in health at advanced ages (e.g., at older than 75 years; Beckett, 2000; Herd, 2006; House et al., 2005). There are several possible reasons for the divergence at younger ages and the convergence at older ages in absolute SES differences. Differences in levels of exposure to health risk factors (e.g., health behaviors and psychosocial factors) between socioeconomic groups peak at early and middle ages, and impacts of these health risk factors may fade over the life course (Chandola et al., 2007; House et al., 2005). In addition, the compression of morbidity (i.e., the time between the onset of chronic diseases or disability and the time an individual dies is compressed) at advanced ages may be greater in higher than in lower SES groups (Fries, 1980, 1996). In other words, compared to their counterparts with lower SES, older adults with higher SES may live longer without chronic diseases, disability, or hospitalization; but may experience a steeper terminal decline in health (Fries, 1996). This pattern would lead to a convergence in socioeconomic disparities in health at advanced ages. This trend toward convergence may also be related to the leveling effects of social security and health benefits provided by welfare (Hoffmann, 2008). Another well-known issue is the need to disentangle birth cohort effects from age effects (House et al., 2005; Lauderdale, 2001; Lynch, 2003; Rohwer, 2016; Willson et al., 2007). Ross and Wu (1996) acknowledged that their findings, using cross-sectional age, of a further divergence in educational differences in self-reported health with age might reflect both age and birth cohort effects. Lauderdale (2001) reported increasing educational differences in survival with age, and that the effect of education was stronger in later than in earlier birth cohorts. Lynch (2003) also showed that ignoring birth cohorts suppressed the widening of socioeconomic gaps in health with age. We took an approach similar to that of Lynch, stratifying our study population by 10-year age groups at baseline (i.e., corresponding to the cohorts born in 1908–1917, 1918–1927, 1928–1937) and estimating the trajectories of hospital days over the follow-up years. It is possible that our approach could not differentiate the true age effect from the period effect. The decline of somatic hospital beds and hospital admissions observed in Finland over the years 2001–2011 could reflect the improvement of population health, technological advances, and the shift from inpatient care to ambulatory care (Keskimäki, Forssas, Rautiainen, Rasilainen, & Gissler, 2014; Manderbacka, Arffman, & Keskimaki, 2014). Also in the U.K., advances in technology, medical treatment, and medical care have been found to reduce the length of hospital stay (Lewis & Edwards, 2015). It is unlikely that the increase in hospital days we observed over the 1988–2007 period was driven by period effects. However, the advances in medical care made in Finland over our study period may complicate the interpretation of our analyses. Since Finland has a universal health care system, it seems less likely that individuals with higher SES had significantly better access to these advances than those with lower SES. If there was a significant faster increase in access to care among older adults with higher SES, we would observe a convergence across all older ages. This cannot explain the further divergence among those aged 50–69 years at baseline. Remarkable regional variations in health care usage have been observed in Finland due to the differences in population morbidity pattern, medical practices, health care resources, and efficiency (Keskimäki et al., 2014). Living in rural areas far away from hospitals (e.g., more than 40 km) is found to be associated with fewer days spent in hospital (Zielinski, Borgquist, & Halling, 2013). Finnish individuals who moved from urban to rural areas tend to be older (e.g., after retirement) and with lower education and income (Nivalainen, 2003a, 2003b). However, it is unclear to what extent these processes affect our results because of other urban–rural differences. For instance, the reduction of somatic hospital beds and shift to ambulatory care may be greater in urban than in rural areas, and the medical practices in rural areas may prefer to refer patient to inpatient hospital care more readily than in urban areas. Unfortunately, we do not have information on urban/rural areas to formally test these possibilities; a topic that requires further research. Our study has some important strengths. First, following recommendations made in previous studies (Dupre, 2007; House et al., 2005; Leopold, Engelhardt, & Engelhartdt, 2013), our findings were based on 20-year longitudinal data from a large-scale, national representative random sample of older adults. More importantly, we advanced our understanding of changes in relative SES differences in hospital days with age. Since our data came from administrative registers, our findings are not affected by the bias that can arise in panel studies because individuals with low SES or poor health are less likely to participate in studies. We used hospital days to reflect older adults’ health status and morbidity. Hospital days is an objective measure that is less prone to measurement error than the subjective measures commonly used in previous studies, such as self-rated health and self-reported disability. It could be argued that hospital days reflect serious health problems only. However, this issue may be less problematic in our study population, as older adults tend to have more serious health conditions than younger adults. We examined educational attainment, household income, and occupational class in order to capture the multidimensional character of SES. Particularly for women, household income may be a better indicator of material circumstances than personal income. We acknowledge that our study also has some limitations. First, the models we used could not describe trajectories over time-varying characteristics. The SES indicators in our study were measured at baseline. Given the ages of our study population, we can assume that their educational attainment was fixed; however, their occupational class and household income may have changed (e.g., after retirement). The occupational class of retirees was based on their previous job status, and remains unchanged. Working-aged older adults may change jobs or become unemployed, but their occupational class is unlikely to change over time. Our data showed that the occupational class remained unchanged for 97% of our study population. Change in household income can be influenced, for example, by retirement and widowhood. We used the age-specific household income tertiles at baseline rather than the annually updated household income; over 80% of our study population stayed in the same tertile over the follow-up period. Overall, SES thus appears to be relative stable at these older ages. Second, in Finland, older adults from lower SES groups are more likely to transit into long-term institutional care (Martikainen, Nihtilä, & Moustgaard, 2008). We did not exclude institutionalized older adults from this study; instead we coded their hospital days as zero if they did not receive hospital care after they were institutionalized. Therefore, assuming that long-term care substitutes for the care given in hospitals, the hospital days and the rate of increase in hospital days may be downward biased among older adults; in particular among those with lower SES compared to those with higher SES. However, of our study population, only 1,200 men (2.0%) and 3,470 women (4.4%) had ever been institutionalized, and in 39% of the years that they were institutionalized they also received hospital care. Hence, the possibly downward bias in hospital days and rate of increase in hospital days in the lower SES groups due to institutionalization is unlikely to substantially affect our results. Third, because we were using registry data, we were unable to provide insights into the behavioral mechanisms and the interplay between behavioral risk factors and SES over the life course that could affect health in later life. Thus, future research using panel data with multiple repeated measures on health and related risk factors, and with long-term follow-up to track the health of the older population to advanced ages, is needed to fully explain changes in socioeconomic disparities in late life with age. Moreover, to better explain the convergence in absolute SES differences in health at advanced ages, additional research that investigates the roles of welfare state policies and compression of morbidity is required. Overall, our findings suggest that the divergence and the convergence in socioeconomic disparities in health at older ages may not be mutually exclusive; as they may depend on the measure of health disparities used and the ages of the populations studied. This implies that more efforts are needed to tackle socioeconomic differences in health in later life, including among the oldest old. A particular focus should be on early old ages; a stage of life when absolute SES differentials are still diverging. Supplementary Material Supplementary data is available at Journals of Gerontology, Series B: Psychological and Social Sciences online. Funding This work is supported by the European Research Council Starting Grant to M. Myrskylä (COSTPOST) [#336475]. The research leading to this publication has been supported by the Max Planck Society within the framework of the project “On the edge of societies: New vulnerable populations, emerging challenges for social policies and future demands for social innovation. The experience of the Baltic Sea States” (2016–2021). P. Martikainen is funded by the Academy of Finland, Strategic Research Council PROMEQ project (#303615) and the Signe and Ane Gyllenberg Foundation. Conflict of Interest None reported. Acknowledgment The authors would like to thank Statistics Finland and the National Institute for Health and Welfare for providing the data. Y. Hu and P. Martikainen conceptualized the research project. Y. Hu performed the data analysis and drafted the manuscript. Y. Hu, T. Leinonen, M. Myrskylä, and P. Martikainen interpreted the results. T. Leinonen, M. Myrskylä, and P. Martikainen critically revised and commented on all versions of the manuscript. All authors approved the final version of the manuscript. References Beckett, M. ( 2000). Converging health inequalities in later life–an artifact of mortality selection. Journal of Health and Social Behavior , 41, 106– 119. doi:10.2307/2676363 Google Scholar CrossRef Search ADS PubMed  Benzeval, M., Green, M. J., & Leyland, A. H. ( 2011). Do social inequalities in health widen or converge with age? Longitudinal evidence from three cohorts in the West of Scotland. BMC Public Health , 11, 947. doi:10.1186/1471-2458-11-947 Google Scholar CrossRef Search ADS PubMed  Bor, J., Cohen, G. H., & Galea, S. ( 2017). Population health in an era of rising income inequality: USA, 1980-2015. Lancet (London, England) , 389, 1475– 1490. doi:10.1016/S0140-6736(17)30571-8 Google Scholar CrossRef Search ADS PubMed  Braveman, P., Egerter, S., & Williams, D. R. ( 2011). The social determinants of health: Coming of age. Annual Review of Public Health , 32, 381– 398. doi:10.1146/annurev-publhealth-031210-101218 Google Scholar CrossRef Search ADS PubMed  Chandola, T., Ferrie, J., Sacker, A., & Marmot, M. ( 2007). Social inequalities in self reported health in early old age: Follow-up of prospective cohort study. BMJ (Clinical research ed.) , 334, 990. doi:10.1136/bmj.39167.439792.55 Google Scholar CrossRef Search ADS PubMed  Corna, L. M. ( 2013). A life course perspective on socioeconomic inequalities in health: A critical review of conceptual frameworks. Advances in Life Course Research , 18, 150– 159. doi:10.1016/j.alcr.2013.01.002 Google Scholar CrossRef Search ADS PubMed  Dupre, M. E. ( 2007). Educational differences in age-related patterns of disease: Reconsidering the cumulative disadvantage and age-as-leveler hypotheses. Journal of Health and Social Behavior , 48, 1– 15. doi:10.1177/002214650704800101 Google Scholar CrossRef Search ADS PubMed  Epstein, A. M., Stern, R. S., & Weissman, J. S. ( 1990). Do the poor cost more? A multihospital study of patients’ socioeconomic status and use of hospital resources. The New England Journal of Medicine , 322, 1122– 1128. doi:10.1056/NEJM199004193221606 Google Scholar CrossRef Search ADS PubMed  Feinstein, J. S. ( 1993). The relationship between socioeconomic status and health: A review of the literature. The Milbank quarterly , 71, 279– 322. doi: 10.2307/3350401 Google Scholar CrossRef Search ADS PubMed  Fries, J. F. ( 1980). Aging, natural death, and the compression of morbidity. The New England Journal of Medicine , 303, 130– 135. doi:10.1056/NEJM198007173030304 Google Scholar CrossRef Search ADS PubMed  Fries, J. F. ( 1996). Physical activity, the compression of morbidity, and the health of the elderly. Journal of the Royal Society of Medicine , 89, 64– 68. PMCID: PMC1295658. Google Scholar PubMed  Grundy, E., & Sloggett, A. ( 2003). Health inequalities in the older population: The role of personal capital, social resources and socio-economic circumstances. Social Science and Medicine (1982) , 56, 935– 947. doi:10.1016/S0277-9536(02) 00093-X Google Scholar CrossRef Search ADS   Hagenaars, A., de Vos, K., & Zaidi, M. A. ( 1994). Poverty statistics in the late 1980s: Research based on micro-data . Luxembourg: Office for Official Publications of the European Communities. Herd, P. ( 2006). Do functional health inequalities decrease in old age? Educational status and functional decline among the 1931–1941 birth cohort. Research on Aging , 28, 375– 392. doi:10.1177/0164027505285845 Google Scholar CrossRef Search ADS   Hilbe, J. M. ( 2007). Negative binomial regression . Cambridge, U.K.: Cambridge University Press. Google Scholar CrossRef Search ADS   Hoffmann, R. ( 2008). Socioeconomic differences in old age mortality . Dordrecht: Springer. Google Scholar CrossRef Search ADS   House, J. S., Lantz, P. M., & Herd, P. ( 2005). Continuity and change in the social stratification of aging and health over the life course: Evidence from a nationally representative longitudinal study from 1986 to 2001/2002 (Americans’ Changing Lives Study). The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences , 60 Spec No 2, 15– 26. doi:10.1093/geronb/60.Special_Issue_2.S15 Google Scholar CrossRef Search ADS   House, J. S., Lepkowski, J. M., Kinney, A. M., Mero, R. P., Kessler, R. C., & Herzog, A. R. ( 1994). The social stratification of aging and health. Journal of Health and Social Behavior , 35, 213– 234. doi:10.2307/2137277 Google Scholar CrossRef Search ADS PubMed  Howe, L. D., Tilling, K., Galobardes, B., & Lawlor, D. A. ( 2013). Loss to follow-up in cohort studies: Bias in estimates of socioeconomic inequalities. Epidemiology (Cambridge, Mass.) , 24, 1– 9. doi:10.1097/EDE.0b013e31827623b1 Google Scholar CrossRef Search ADS PubMed  Huisman, M., Kunst, A. E., & Mackenbach, J. P. ( 2003). Socioeconomic inequalities in morbidity among the elderly; a European overview. Social Science and Medicine (1982) , 57, 861– 873. doi:10.1016/S0277-9536(02)00454-9 Google Scholar CrossRef Search ADS   Keskimäki, I., Forssas, E., Rautiainen, H., Rasilainen, J., & Gissler, M. ( 2014). Finland: Geographic variations in health care. In Geographic variations in health care: What do we know and what can be done to improve health system performance?  Paris: OECD Publishing. Google Scholar CrossRef Search ADS   Kim, J., & Durden, E. ( 2007). Socioeconomic status and age trajectories of health. Social Science and Medicine (1982) , 65, 2489– 2502. doi:10.1016/j.socscimed.2007.07.022 Google Scholar CrossRef Search ADS   Kurland, B. F., Johnson, L. L., Egleston, B. L., & Diehr, P. H. ( 2009). Longitudinal data with follow-up truncated by death: Match the analysis method to research aims. Statistical Science: A Review Journal of the Institute of Mathematical Statistics , 24, 211. doi:10.1214/09-STS293 Google Scholar PubMed  Lauderdale, D. S. ( 2001). Education and survival: Birth cohort, period, and age effects. Demography , 38, 551– 561. doi:10.1353/dem.2001.0035 Google Scholar CrossRef Search ADS PubMed  Leopold, L., & Engelhardt. ( 2013). Education and physical health trajectories in old age. Evidence from the Survey of Health, Ageing and Retirement in Europe (SHARE). International Journal of Public Health , 58, 23– 31. doi:10.1007/s00038-012-0399-0 Google Scholar CrossRef Search ADS PubMed  Lewis, R., & Edwards, N. ( 2015). Improving length of stay: What can hospitals do? Retrieved May 24, 2017 from https://www.nuffieldtrust.org.uk/files/2017-01/improving-length-of-stay-hospitals-web-final.pdf Liao, Y., McGee, D. L., Kaufman, J. S., Cao, G., & Cooper, R. S. ( 1999). Socioeconomic status and morbidity in the last years of life. American Journal of Public Health , 89, 569– 572. doi:10.2105/Ajph.89.4.569 Google Scholar CrossRef Search ADS PubMed  Little, R. ( 2009). Selection and pattern mixture models. In G. Fitzmaurice, M. Davidian, G. Verbeke & G. Molenberghs (Eds.), Longitudinal data analysis  (pp. 409– 431). Boca Raton, FL: Chapman & Hall. Google Scholar CrossRef Search ADS   Liu, H., & Powers, D. A. ( 2007). Growth curve models for zero-inflated count data: An application to smoking behavior. Structural Equation Modeling: A Multidisciplinary Journal , 14, 247– 279. doi:10.1080/10705510709336746 Google Scholar CrossRef Search ADS   Lynch, S. M. ( 2003). Cohort and life-course patterns in the relationship between education and health: A hierarchical approach. Demography , 40, 309– 331. doi:10.2307/3180803 Google Scholar CrossRef Search ADS PubMed  Lynch, S. M., & Brown, J. S. ( 2011). Stratification and inequality over the life course. In R. H. Binstock & L. K. George (Eds.), Handbook of aging and the social sciences  ( 7th ed., pp. 105– 120). San Diego, CA: Elsevier. Google Scholar CrossRef Search ADS   Mackenbach, J. P. ( 2015). Should we aim to reduce relative or absolute inequalities in mortality? European Journal of Public Health , 25, 185. doi:10.1093/eurpub/cku217 Google Scholar CrossRef Search ADS PubMed  Mackenbach, J. P., Martikainen, P., Menvielle, G., & de Gelder, R. ( 2016). The arithmetic of reducing relative and absolute inequalities in health: A theoretical analysis illustrated with European mortality data. Journal of Epidemiology and Community Health , 70, 730– 736. doi:10.1136/jech-2015-207018 Google Scholar CrossRef Search ADS PubMed  Mackenbach, J. P., Kulhánová, I., Artnik, B., Bopp, M., Borrell, C., Clemens, T., … de Gelder, R. ( 2016). Changes in mortality inequalities over two decades: register based study of European countries. British Medical Journal , 353, i1732. doi:10.1136/bmj.i1732 Google Scholar CrossRef Search ADS PubMed  Manderbacka, K., Arffman, M., & Keskimäki, I. ( 2014). Has socioeconomic equity increased in somatic specialist care: A register-based cohort study from Finland in 1995-2010. BMC Health Services Research , 14, 430. doi:10.1186/1472-6963-14-430 Google Scholar CrossRef Search ADS PubMed  Marmot, M. ( 2005). Social determinants of health inequalities. Lancet , 365, 1099– 1104. doi:10.1016/S0140-6736(05)71146-6 Google Scholar CrossRef Search ADS PubMed  Martikainen, P., Nihtilä, E., & Moustgaard, H. ( 2008). The effects of socioeconomic status and health on transitions in living arrangements and mortality: A longitudinal analysis of elderly Finnish men and women from 1997 to 2002. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences , 63, S99– S109. doi:10.1093/geronb/63.2.S99 Google Scholar CrossRef Search ADS   Mirowsky, J., & Ross, C. E. ( 2008). Education and self-rated health: Cumulative advantage and its rising importance. Research on Aging , 30, 93– 122. doi:10.1177/0164027507309649 Google Scholar CrossRef Search ADS   Muthén, B., Asparouhov, T., Hunter, A. M., & Leuchter, A. F. ( 2011). Growth modeling with nonignorable dropout: Alternative analyses of the STAR*D antidepressant trial. Psychological Methods , 16, 17– 33. doi:10.1037/a0022634 Google Scholar CrossRef Search ADS PubMed  Muthén, L. K., & Muthén, B. O. ( 1998–2015). Mplus user’s guide  ( 7th ed.). Los Angeles, CA: Muthén & Muthén. Nivalainen, S. ( 2003a). Where do migrants go? An Analysis of Rural and Urban Destined/Originated Migration in Finland 1996–1999 (Working Papers No. 66) . Helsinki, Finland: Pellervo Economic Research Institute. Nivalainen, S. ( 2003b, August). Who move to rural areas? Micro Evidence from Finland . Paper presented at the 43rd Congress of the European Regional Science Association: “Peripheries, Centres, and Spatial Development in the New Europe”. Jyväskylä, Finland. Retrieved August 15, 2017 from http://www-sre.wu.ac.at/ersa/ersaconfs/ersa03/cdrom/papers/214.pdf Pavalko, E. K., & Wilson, A. E. ( 2011). Life course approaches to health, illness and healing. In B. A. Pescosolido, J. K. Martin, J. D. McLeod & A. Rogers (Eds.), Handbook of the sociology of health, illness, and healing . New York: Springer. Google Scholar CrossRef Search ADS   Rechel, B., Grundy, E., Robine, J. M., Cylus, J., Mackenbach, J. P., Knai, C., & McKee, M. ( 2013). Ageing in the European Union. Lancet (London, England) , 381, 1312– 1322. doi:10.1016/S0140-6736(12)62087-X Google Scholar CrossRef Search ADS PubMed  Regidor, E. ( 2004). Measures of health inequalities: Part 2. Journal of Epidemiology and Community Health , 58, 900– 903. doi:10.1136/jech.2004.023036 Google Scholar CrossRef Search ADS PubMed  Rohwer, G. ( 2016). A note on the dependence of health on age and education. Demography , 53, 325– 335. doi:10.1007/s13524-016-0457-y Google Scholar CrossRef Search ADS PubMed  Roos, N. P., & Mustard, C. A. ( 1997). Variation in health and health care use by socioeconomic status in Winnipeg, Canada: Does the system work well? Yes and no. The Milbank Quarterly , 75, 89– 111. doi:10.1111/1468-0009.00045 Google Scholar CrossRef Search ADS PubMed  Ross, C. E., & Wu, C. L. ( 1996). Education, age, and the cumulative advantage in health. Journal of Health and Social Behavior , 37, 104– 120. doi:10.2307/2137234 Google Scholar CrossRef Search ADS PubMed  Stolz, E., Mayerl, H., Waxenegger, A., Rásky, É., & Freidl, W. ( 2017). Impact of socioeconomic position on frailty trajectories in 10 European countries: Evidence from the Survey of Health, Ageing and Retirement in Europe (2004-2013). Journal of Epidemiology and Community Health , 71, 73– 80. doi:10.1136/jech-2016-207712 Google Scholar CrossRef Search ADS PubMed  Willson, A. E., Shuey, K. M., & Elder, G. H. ( 2007). Cumulative advantage processes as mechanisms of inequality in life course health. American Journal of Sociology , 112, 1886– 1924. doi:10.1086/512712 Google Scholar CrossRef Search ADS   Wolinsky, F. D., Culler, S. D., Callahan, C. M., & Johnson, R. J. ( 1994). Hospital resource consumption among older adults: A prospective analysis of episodes, length of stay, and charges over a seven-year period. Journal of Gerontology , 49, S240– S252. doi:10.1093/geronj/49.5.S240 Google Scholar CrossRef Search ADS PubMed  World Health Organization.( 2015). World Report on Ageing and Health. Luxembourg: World Health Organization. Retrieved February 15, 2017 from http://apps.who.int/iris/bitstream/10665/186463/1/9789240694811_eng.pdf. World Health Organization.( 2016). Global Health Estimates 2015: Disease burden by Cause, Age, Sex, by Country and by Region, 2000–2015. Retrieved Auguest 14, 2017 from http://www.who.int/healthinfo/global_burden_disease/estimates/en/index2.html Zielinski, A., Borgquist, L., & Halling, A. ( 2013). Distance to hospital and socioeconomic status influence secondary health care use. Scandinavian Journal of Primary Health Care , 31, 83– 88. doi:10.3109/02813432.2012.759712 Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The Journals of Gerontology Series B: Psychological Sciences and Social Sciences Oxford University Press

Changes in Socioeconomic Differences in Hospital Days With Age: Cumulative Disadvantage, Age-as-Leveler, or Both?

Loading next page...
 
/lp/ou_press/changes-in-socioeconomic-differences-in-hospital-days-with-age-6kivmZJr4Y
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
1079-5014
eISSN
1758-5368
D.O.I.
10.1093/geronb/gbx161
Publisher site
See Article on Publisher Site

Abstract

Abstract Objectives Length of hospital stay is inversely associated with socioeconomic status (SES). It is less clear whether socioeconomic disparities in numbers of hospital days diverge or converge with age. Method Longitudinal linked Finnish registry data (1988–2007) from 137,653 men and women aged 50–79 years at the end of 1987 were used. Trajectories of annual total hospital days by education, household income, and occupational class were estimated using negative binomial models. Results Men and women with higher education, household income, and occupational class had fewer hospital days in 1988 than those with lower SES. Hospital days increased between 1988 and 2007. For some age groups, higher SES was associated with a faster annual rate of increase, resulting in narrowing rate ratios of hospital days between SES groups (relative differences); the rate ratios remained stable for other groups. Absolute SES differences in numbers of hospital days appeared to diverge with age among those aged 50–69 years at baseline, but converge among those aged 70–79 years at baseline. Discussion The hypotheses that socioeconomic disparities in health diverge or converge with age may not be mutually exclusive; we demonstrated convergence/maintenance in relative differences for all age groups, but divergence or convergence in absolute differences depending on age. Socioeconomic disparities, Hospital days, Longitudinal registry-based data Population aging is a global phenomenon, and a major challenge for the sustainability of health care and social security systems (Rechel et al., 2013; World Health Organization, 2015). Ensuring that older adults remain in good health as long as possible not only improves the quality of life for the aging population but also buffer against the expected cost pressures on the care system (Rechel et al., 2013; World Health Organization, 2015). Research on the determinants of healthy aging can greatly contribute to this goal. Socioeconomic status (SES) is one of the most established and persistent determinants of health for older adults. SES is multidimensional and can be reflected by education, occupation, and income. These SES indicators are correlated but they represent different types of resources and are associated with health through both shared and independent pathways (Braveman, Egerter, & Williams, 2011). Education is clearly associated with employment opportunities, access to information and ability to process information, ability to develop and change behaviors, and control of life. Occupation is linked with physical working environment, employment-related earnings and benefits, and work-related stress and social support. Whereas income is related to direct access to health care, nutrition, housing, transport, and other resources. As a result, to better understand how SES is linked to health, it is beneficial to investigate various SES indicators. For high-income countries, 55% of the total loss of disability-adjusted life years in 2015 was attributable to health conditions among the older population aged more than 60 years (World Health Organization, 2016). Reducing health differentials at older ages and improving health of the more disadvantaged older adults in particular could reduce the total disease burden and ameliorate health of the whole population (Feinstein, 1993; Grundy & Sloggett, 2003; Huisman, Kunst, & Mackenbach, 2003; Marmot, 2005; World Health Organization, 2015). A number of studies have examined how socioeconomic disparities in health change with age among older adults, but the evidence remains mixed (Corna, 2013; Pavalko & Wilson, 2011). Some studies have found that socioeconomic disparities in health increase with age, that is, a further divergence of health differentials between SES groups or the effect of SES on health increases with age (Dupre, 2007; Lynch, 2003; Mirowsky & Ross, 2008; Willson, Shuey, & Elder, 2007). This finding of a further divergence supports the cumulative advantages/disadvantages (CAD) hypothesis: advantages/disadvantages in social, economic, behavioral, and psychosocial resources gradually accumulate over the life course and thus produce enlarging differences in health with age. However, narrowing disparities of health between SES groups with age also have been reported in previous studies, that is, a convergence of the health differentials between SES groups or the effect of SES on health diminishes with age (Herd, 2006; House et al., 1994; Ross & Wu, 1996). This observation of a convergence provides evidence for the age-as-leveler (AAL) hypothesis: there is a biological ceiling in late life that older adults from different SES groups become universally fragile with age, placing less importance of social determinants to aging. The convergence, however, can also be related to mortality selection that less healthy people from lower SES groups are more likely to die prematurely than their counterparts from higher SES groups (Beckett, 2000; Lauderdale, 2001). Still other studies have shown that there are constant socioeconomic gaps in health at older ages (status maintenance; Leopold & Engelhardt, 2013; Pavalko & Wilson, 2011; Stolz, Mayerl, Waxenegger, Rasky, & Freidl, 2017). The inconsistency of these findings is largely attributable to differences in study design; that is, to differences in the health outcomes studied and in whether the data used were cross-sectional or longitudinal (Leopold & Engelhardt, 2013; Pavalko & Wilson, 2011). Cross-sectional data provide only a snapshot of age-specific health patterns, whereas the comparability of longitudinal studies is hampered by the varying characteristics of the study samples (e.g., countries and age groups) and the statistical methods used to analyze the data. Long-term panel data are preferable when investigating changes in socioeconomic differences in health at older ages, as they allow for the separation of age and birth cohort effects (Dupre, 2007; House, Lantz, & Herd, 2005; Leopold & Engelhardt, 2013; Pavalko & Wilson, 2011). However, problems related to nonresponse and attrition can threaten the validity of findings from panel studies; an issue that only a few studies have addressed explicitly (Howe, Tilling, Galobardes, & Lawlor, 2013; Lynch & Brown, 2011; Pavalko & Wilson, 2011; Willson et al., 2007). The association between SES and health are often measured by relative and absolute differences. The relative difference focuses on the equality, independent of the actual level of health in each SES group, whereas the absolute difference emphasizes the difference in the actual level of health between lower and higher SES groups (Mackenbach, 2015). It is important, if possible, to estimate both absolute and relative SES differences in health for better monitoring of health inequalities, evaluating policy interventions, and improving our understanding of the causes of health differentials (Regidor, 2004). However, we are not aware of research on changes in relative differences in health with age using long-term longitudinal data with repeated measures of health. It remains unclear whether the changes of absolute and relative SES differences in health with age are similar to those observed for mortality; that is, exhibiting opposing trends for the relative and absolute difference with age (Bor, Cohen, & Galea, 2017; Mackenbach, Kulhánová, et al., 2016). Older adults are disproportionately heavy users of health care and long-term care services (Wolinsky, Culler, Callahan, & Johnson, 1994; World Health Organization, 2015). Length of hospital stay, an important index of morbidity and resource consumption, has been found to be inversely associated with education, income, and occupational class (Epstein, Stern, & Weissman, 1990; Liao, McGee, Kaufman, Cao, & Cooper, 1999; Roos & Mustard, 1997). The recent World Report on Ageing and Health (World Health Organization, 2015) revealed pronounced inequalities that, compared to more advantaged older adults, disadvantaged older adults have more difficulties accessing health services, over and above their higher health risks and more severe health problems. To our knowledge, no previous study has investigated how absolute and relative SES differences in hospital use change as individuals grow older. In this study, we examined these issues using 20 years of Finnish registry data. Methods Study Population We used data from a linked register-based 11% random sample of the population residing in Finland at the end of each year between 1987 and 2007. We restricted our study population to individuals aged 50–79 years at the end of 1987 (i.e., born between January 1908 and December 1937). We also excluded individuals who (a) were not part of the dwelling population of Finland (e.g., being institutionalized or imprisoned) at the end of 1985 or 1987; (b) were not residing in Finland at the end of any given year in the 1987–2007 period; or (c) had died in 1988 with no hospitalizations as they contributed no information toward hospital use in the 1988–2007 period. The final size of the study population was 137,653 (59,586 men and 78,067 women). The cohort was followed up annually between 1988 and 2007. Hospital Days Hospitalization episodes between January 1, 1988, and December 31, 2007, were extracted from hospital discharge records. We calculated the total number of hospital days for each year of the 1988–2007 period. The annual hospital days could include multiple hospitalization episodes in a given year, and the days from all episodes were added up. If no hospitalization occurred in a given year, the number of hospital days was coded as zero. Socioeconomic Status To reflect the multidimensional nature of SES, we selected three indicators from the labor market file: highest educational attainment, household income, and occupational class. Highest educational attainment and household income were measured at the end of 1987. Individuals with basic education (i.e., less than upper secondary school) were the dominant group. We thus dichotomized educational attainment into basic education and beyond basic education (i.e., upper secondary school or higher). Household income was derived by dividing the taxable household income by the number of consumption units in the household using the Organisation for Economic Co-operation and Development (OECD)-modified scale (Hagenaars, de Vos, & Zaidi, 1994). Household income was further categorized into tertiles by sex and 5-year age groups. Information on occupational class was available every 5 years, and we used the most recent information measured at the end of 1985 to our baseline (i.e., between the end of 1987 and the end of 1988). The occupational classes were as follows: nonspecialized manual class or specialization unknown, specialized manual class, white-collar class, and other (including farmers, self-employed, students, other occupational classes, or occupational class unknown). For individuals who were retired or unemployed in 1985, the most recent information available on their previous occupational class was used. Covariates Native language and region of residence were controlled as covariates. Native language was time-invariant, and was dichotomized into Finnish and Swedish or other languages. Region of residence was time-varying, and was updated annually at the end of each year in the 1987–2007 period. In line with Statistics Finland, we used the following region of residence categories: eastern, western, southern, and northern Finland. Statistical Analysis In each year of the 1988–2007 period, the distribution of hospital days was highly skewed, as 66%–80% of our study population was not hospitalized (i.e., had zero hospital days). We modeled the hospital days as a count response. The overdispersion and excess of zeroes of the hospital days can be handled by using the negative binomial (NB) model and two zero-altered models: the zero-inflated Poisson (ZIP) model and zero-inflated negative binomial (ZINB) model (Hilbe, 2007). We compared the model fit of the three models using information criteria and found that the ZIP and ZINB models did not outperform the NB model. We therefore used the most parsimonious NB model throughout this study. The key variables in the NB model are (a) the intercept, which captures the hospital days at baseline (i.e., the year 1988 coded as time zero); and (b) the slope, which represents the annual rate of change in hospital days over the follow-up years (i.e., the years 1989–2007 coded as time 1–19, respectively). All models were fitted using Mplus 7 (Muthén & Muthén, 1998–2015). For more details on the model comparison, see Section 1.1 of the Supplementary Materials. The “empty” model containing only the intercept and slope with no SES indicators or covariates (unconditional models) was fitted first. Because of the log link function used in the NB models, the exponentiated estimates of the intercept and the slope respectively represent the expected hospital days at baseline and the expected annual rate of increase in hospital days over the follow-up period. The hospital-day trajectories over the years 1988–2007 were thus exponential on the scale of actual days (Liu & Powers, 2007). Both the intercept and slope were then regressed on education, household income, and occupational class; adjusting for native language and region of residence (conditional models). All models were fitted separately for each SES indicator, for men and women, and for three age groups at baseline (50–59, 60–69, and 70–79 years). As a measure of relative difference, the exponentiated coefficients of the intercept and the slope for the SES indicators (conditional models) represent the ratio of expected hospital days at baseline and the ratio of the expected annual rate of increase in hospital days between SES groups respectively. The absolute SES differences in hospital days at baseline and changes in these differences over the follow-up period were plotted in figures. Results Table 1 summarizes the sociodemographic characteristics of our study population at baseline, as well as the mean hospital days in each year of the 1988–2007 period. The men in our study population were younger than the women. The mean hospital days at baseline were higher in older than in younger age groups. More than 70% of the study population had basic education. More women than men were in the white-collar class, and more men than women were in the specialized manual class. For both sexes, the mean hospital days at baseline were smaller in the higher than in the lower socioeconomic groups. The mean hospital days increased gradually with age. The mean hospital days were 5.06 days higher for men and 6.62 days higher for women in 2007 than in 1988. Table 1. Baseline Sociodemographic Characteristics and Mean Hospital Days in 1988–2007   Men  Women  N (%)  Hospital days  N (%)  Hospital days  Total  59,586    78,067    Baseline (1988)   Age (years)    50–59  27,506 (46.2)  2.39  29,367 (37.6)  2.18    60–69  20,600 (34.6)  5.06  27,915 (35.8)  3.85    70–79  11,480 (19.3)  11.08  20,785 (26.6)  10.69   Native language    Finnish  54,852 (92.1)  5.04  72,323 (92.6)  5.07    Swedish/other  4,734 (7.9)  4.40  5,744 (7.4)  4.77   Region of residence    Southern  26,676 (44.8)  4.78  36,335 (46.5)  4.98    Northern  16,750 (28.1)  4.58  21,653 (27.7)  5.03    Eastern  9,186 (15.4)  5.72  11,828 (15.2)  5.27    Western  6,974 (11.7)  5.79  8,251 (10.6)  5.03   Education    Basic education  43,898 (73.7)  5.43  60,417 (77.4)  5.51    Beyond basic education  15,688 (26.3)  3.75  17,650 (22.6)  3.46   Household income    Low  19,899 (33.4)  6.08  26,306 (33.7)  5.92    Middle  19,822 (33.3)  4.80  25,753 (33.0)  4.58    High  19,865 (33.3)  4.08  26,008 (33.3)  4.62   Occupational class    Manual, nonspecialized/unknown  10,351 (17.4)  4.90  19,301 (24.7)  4.56    Manual, specialized  18,155 (30.5)  5.53  14,139 (18.1)  5.68    White collar  15,052 (25.3)  3.93  26,852 (34.4)  4.05    Other  16,028 (26.9)  5.41  17,775 (22.8)  6.57  Year   1988  59,586 (100.0)  4.99  78,067 (100.0)  5.04   1989  58,475 (98.1)  5.26  77,088 (98.8)  5.50   1990  56,773 (95.3)  5.67  75,858 (97.2)  5.98   1991  54,967 (92.3)  5.75  74,537 (95.5)  6.38   1992  53,307 (89.5)  5.96  73,120 (93.7)  6.80   1993  51,510 (86.5)  6.16  71,604 (91.7)  7.36   1994  49,761 (83.5)  6.69  69,905 (89.5)  7.89   1995  48,009 (80.6)  7.01  68,238 (87.4)  8.56   1996  46,210 (77.6)  7.65  66,378 (85.0)  9.36   1997  44,421 (74.6)  7.84  64,586 (82.7)  9.83   1998  42,501 (71.3)  7.93  62,615 (80.2)  10.34   1999  40,643 (68.2)  8.35  60,654 (77.7)  10.89   2000  38,773 (65.1)  8.54  58,578 (75.0)  11.42   2001  37,040 (62.2)  9.02  56,353 (72.2)  11.87   2002  35,267 (59.2)  9.44  54,230 (69.5)  12.12   2003  33,469 (56.2)  9.59  52,033 (66.7)  12.32   2004  31,777 (53.3)  9.58  49,770 (63.8)  11.98   2005  30,174 (50.6)  9.93  47,656 (61.0)  12.12   2006  28,573 (48.0)  10.35  45,580 (58.4)  12.03   2007  27,022 (45.4)  10.05  43,503 (55.7)  11.66    Men  Women  N (%)  Hospital days  N (%)  Hospital days  Total  59,586    78,067    Baseline (1988)   Age (years)    50–59  27,506 (46.2)  2.39  29,367 (37.6)  2.18    60–69  20,600 (34.6)  5.06  27,915 (35.8)  3.85    70–79  11,480 (19.3)  11.08  20,785 (26.6)  10.69   Native language    Finnish  54,852 (92.1)  5.04  72,323 (92.6)  5.07    Swedish/other  4,734 (7.9)  4.40  5,744 (7.4)  4.77   Region of residence    Southern  26,676 (44.8)  4.78  36,335 (46.5)  4.98    Northern  16,750 (28.1)  4.58  21,653 (27.7)  5.03    Eastern  9,186 (15.4)  5.72  11,828 (15.2)  5.27    Western  6,974 (11.7)  5.79  8,251 (10.6)  5.03   Education    Basic education  43,898 (73.7)  5.43  60,417 (77.4)  5.51    Beyond basic education  15,688 (26.3)  3.75  17,650 (22.6)  3.46   Household income    Low  19,899 (33.4)  6.08  26,306 (33.7)  5.92    Middle  19,822 (33.3)  4.80  25,753 (33.0)  4.58    High  19,865 (33.3)  4.08  26,008 (33.3)  4.62   Occupational class    Manual, nonspecialized/unknown  10,351 (17.4)  4.90  19,301 (24.7)  4.56    Manual, specialized  18,155 (30.5)  5.53  14,139 (18.1)  5.68    White collar  15,052 (25.3)  3.93  26,852 (34.4)  4.05    Other  16,028 (26.9)  5.41  17,775 (22.8)  6.57  Year   1988  59,586 (100.0)  4.99  78,067 (100.0)  5.04   1989  58,475 (98.1)  5.26  77,088 (98.8)  5.50   1990  56,773 (95.3)  5.67  75,858 (97.2)  5.98   1991  54,967 (92.3)  5.75  74,537 (95.5)  6.38   1992  53,307 (89.5)  5.96  73,120 (93.7)  6.80   1993  51,510 (86.5)  6.16  71,604 (91.7)  7.36   1994  49,761 (83.5)  6.69  69,905 (89.5)  7.89   1995  48,009 (80.6)  7.01  68,238 (87.4)  8.56   1996  46,210 (77.6)  7.65  66,378 (85.0)  9.36   1997  44,421 (74.6)  7.84  64,586 (82.7)  9.83   1998  42,501 (71.3)  7.93  62,615 (80.2)  10.34   1999  40,643 (68.2)  8.35  60,654 (77.7)  10.89   2000  38,773 (65.1)  8.54  58,578 (75.0)  11.42   2001  37,040 (62.2)  9.02  56,353 (72.2)  11.87   2002  35,267 (59.2)  9.44  54,230 (69.5)  12.12   2003  33,469 (56.2)  9.59  52,033 (66.7)  12.32   2004  31,777 (53.3)  9.58  49,770 (63.8)  11.98   2005  30,174 (50.6)  9.93  47,656 (61.0)  12.12   2006  28,573 (48.0)  10.35  45,580 (58.4)  12.03   2007  27,022 (45.4)  10.05  43,503 (55.7)  11.66  View Large Table 1. Baseline Sociodemographic Characteristics and Mean Hospital Days in 1988–2007   Men  Women  N (%)  Hospital days  N (%)  Hospital days  Total  59,586    78,067    Baseline (1988)   Age (years)    50–59  27,506 (46.2)  2.39  29,367 (37.6)  2.18    60–69  20,600 (34.6)  5.06  27,915 (35.8)  3.85    70–79  11,480 (19.3)  11.08  20,785 (26.6)  10.69   Native language    Finnish  54,852 (92.1)  5.04  72,323 (92.6)  5.07    Swedish/other  4,734 (7.9)  4.40  5,744 (7.4)  4.77   Region of residence    Southern  26,676 (44.8)  4.78  36,335 (46.5)  4.98    Northern  16,750 (28.1)  4.58  21,653 (27.7)  5.03    Eastern  9,186 (15.4)  5.72  11,828 (15.2)  5.27    Western  6,974 (11.7)  5.79  8,251 (10.6)  5.03   Education    Basic education  43,898 (73.7)  5.43  60,417 (77.4)  5.51    Beyond basic education  15,688 (26.3)  3.75  17,650 (22.6)  3.46   Household income    Low  19,899 (33.4)  6.08  26,306 (33.7)  5.92    Middle  19,822 (33.3)  4.80  25,753 (33.0)  4.58    High  19,865 (33.3)  4.08  26,008 (33.3)  4.62   Occupational class    Manual, nonspecialized/unknown  10,351 (17.4)  4.90  19,301 (24.7)  4.56    Manual, specialized  18,155 (30.5)  5.53  14,139 (18.1)  5.68    White collar  15,052 (25.3)  3.93  26,852 (34.4)  4.05    Other  16,028 (26.9)  5.41  17,775 (22.8)  6.57  Year   1988  59,586 (100.0)  4.99  78,067 (100.0)  5.04   1989  58,475 (98.1)  5.26  77,088 (98.8)  5.50   1990  56,773 (95.3)  5.67  75,858 (97.2)  5.98   1991  54,967 (92.3)  5.75  74,537 (95.5)  6.38   1992  53,307 (89.5)  5.96  73,120 (93.7)  6.80   1993  51,510 (86.5)  6.16  71,604 (91.7)  7.36   1994  49,761 (83.5)  6.69  69,905 (89.5)  7.89   1995  48,009 (80.6)  7.01  68,238 (87.4)  8.56   1996  46,210 (77.6)  7.65  66,378 (85.0)  9.36   1997  44,421 (74.6)  7.84  64,586 (82.7)  9.83   1998  42,501 (71.3)  7.93  62,615 (80.2)  10.34   1999  40,643 (68.2)  8.35  60,654 (77.7)  10.89   2000  38,773 (65.1)  8.54  58,578 (75.0)  11.42   2001  37,040 (62.2)  9.02  56,353 (72.2)  11.87   2002  35,267 (59.2)  9.44  54,230 (69.5)  12.12   2003  33,469 (56.2)  9.59  52,033 (66.7)  12.32   2004  31,777 (53.3)  9.58  49,770 (63.8)  11.98   2005  30,174 (50.6)  9.93  47,656 (61.0)  12.12   2006  28,573 (48.0)  10.35  45,580 (58.4)  12.03   2007  27,022 (45.4)  10.05  43,503 (55.7)  11.66    Men  Women  N (%)  Hospital days  N (%)  Hospital days  Total  59,586    78,067    Baseline (1988)   Age (years)    50–59  27,506 (46.2)  2.39  29,367 (37.6)  2.18    60–69  20,600 (34.6)  5.06  27,915 (35.8)  3.85    70–79  11,480 (19.3)  11.08  20,785 (26.6)  10.69   Native language    Finnish  54,852 (92.1)  5.04  72,323 (92.6)  5.07    Swedish/other  4,734 (7.9)  4.40  5,744 (7.4)  4.77   Region of residence    Southern  26,676 (44.8)  4.78  36,335 (46.5)  4.98    Northern  16,750 (28.1)  4.58  21,653 (27.7)  5.03    Eastern  9,186 (15.4)  5.72  11,828 (15.2)  5.27    Western  6,974 (11.7)  5.79  8,251 (10.6)  5.03   Education    Basic education  43,898 (73.7)  5.43  60,417 (77.4)  5.51    Beyond basic education  15,688 (26.3)  3.75  17,650 (22.6)  3.46   Household income    Low  19,899 (33.4)  6.08  26,306 (33.7)  5.92    Middle  19,822 (33.3)  4.80  25,753 (33.0)  4.58    High  19,865 (33.3)  4.08  26,008 (33.3)  4.62   Occupational class    Manual, nonspecialized/unknown  10,351 (17.4)  4.90  19,301 (24.7)  4.56    Manual, specialized  18,155 (30.5)  5.53  14,139 (18.1)  5.68    White collar  15,052 (25.3)  3.93  26,852 (34.4)  4.05    Other  16,028 (26.9)  5.41  17,775 (22.8)  6.57  Year   1988  59,586 (100.0)  4.99  78,067 (100.0)  5.04   1989  58,475 (98.1)  5.26  77,088 (98.8)  5.50   1990  56,773 (95.3)  5.67  75,858 (97.2)  5.98   1991  54,967 (92.3)  5.75  74,537 (95.5)  6.38   1992  53,307 (89.5)  5.96  73,120 (93.7)  6.80   1993  51,510 (86.5)  6.16  71,604 (91.7)  7.36   1994  49,761 (83.5)  6.69  69,905 (89.5)  7.89   1995  48,009 (80.6)  7.01  68,238 (87.4)  8.56   1996  46,210 (77.6)  7.65  66,378 (85.0)  9.36   1997  44,421 (74.6)  7.84  64,586 (82.7)  9.83   1998  42,501 (71.3)  7.93  62,615 (80.2)  10.34   1999  40,643 (68.2)  8.35  60,654 (77.7)  10.89   2000  38,773 (65.1)  8.54  58,578 (75.0)  11.42   2001  37,040 (62.2)  9.02  56,353 (72.2)  11.87   2002  35,267 (59.2)  9.44  54,230 (69.5)  12.12   2003  33,469 (56.2)  9.59  52,033 (66.7)  12.32   2004  31,777 (53.3)  9.58  49,770 (63.8)  11.98   2005  30,174 (50.6)  9.93  47,656 (61.0)  12.12   2006  28,573 (48.0)  10.35  45,580 (58.4)  12.03   2007  27,022 (45.4)  10.05  43,503 (55.7)  11.66  View Large Table 2 shows the association between SES and the trajectories of hospital days (relative difference) for men and women. The unconditional estimates of both the hospital days at baseline (i.e., intercept) and the annual rate of increase in hospital days (i.e., slope) from the “empty” models were similar for men and women aged 50–59 years at baseline. On average, women aged 60–69 years at baseline had fewer hospital days at baseline, but a faster rate of increase than their male counterparts (difference in the slope p <.001). Women aged 70–79 years at baseline had both more hospital days at baseline and a faster rate of increase (difference in the slope p = .01) than men in the same age band. Table 2. SES and Trajectory of Hospital Days (Relative Difference), Stratified by Age Groups at Baseline   Men  Women  Intercept  Slope  Intercept  Slope  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  Age at baseline  50–59 Years   Unconditional estimates  2.24 (2.11, 2.37)***  1.06 (1.06, 1.06)***  2.02 (1.90, 2.16)***  1.06 (1.05, 1.06)***   Conditional estimates    Educationa     Beyond basic  0.75 (0.66, 0.85)***  1.01 (1.00, 1.02)  0.72 (0.63, 0.82)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)  0.55 (0.48, 0.63)***  1.01 (1.00, 1.02)*     High  0.43 (0.38, 0.49)***  1.02 (1.01, 1.03)**  0.44 (0.38, 0.52)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.01 (0.85, 1.19)  0.99 (0.98, 1.00)  1.22 (1.01, 1.47)*  1.00 (0.98, 1.01)     White collar  0.64 (0.54, 0.75)***  1.00 (0.99, 1.01)  0.83 (0.71, 0.97)*  0.99 (0.98, 1.01)     Other  0.87 (0.74, 1.03)  1.00 (0.99, 1.01)  1.35 (1.09, 1.67)**  0.99 (0.97, 1.00)  60–69 Years   Unconditional estimates  5.08 (4.83, 5.35)***  1.07 (1.06, 1.07)***  3.95 (3.76, 4.15)***  1.09 (1.09, 1.09)***   Conditional estimates    Educationa     Beyond basic  0.88 (0.77, 1.00)  1.00 (0.99, 1.01)  0.74 (0.65, 0.84)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.77 (0.69, 0.87)***  1.01 (1.00, 1.02)*  0.80 (0.72, 0.90)***  1.01 (1.00, 1.02)     High  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)*  0.73 (0.65, 0.83)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.04 (0.96, 1.20)  0.99 (0.98, 1.01)  1.25 (1.08, 1.45)**  0.99 (0.98, 1.00)     White collar  0.88 (0.75, 1.04)  0.99 (0.98, 1.01)  0.89 (0.78, 1.02)  1.00 (0.99, 1.01)     Other  0.96 (0.83, 1.11)  1.00 (0.98, 1.01)  1.23 (1.08, 1.41)**  0.99 (0.98, 1.00)*  70–79 Years   Unconditional estimates  11.73 (11.11, 12.39)***  1.07 (1.06, 1.07)***  12.33 (11.81, 12.88)***  1.08 (1.07, 1.08)***   Conditional estimates    Educationa     Beyond basic  0.79 (0.69, 0.91)**  1.01 (0.99, 1.03)  0.78 (0.69, 0.89)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.89 (0.78, 1.02)  1.00 (0.99, 1.02)  0.86 (0.78, 0.96)**  1.01 (1.00, 1.02)     High  0.88 (0.77, 1.00)*  1.01 (0.99, 1.02)  0.86 (0.77, 0.95)**  1.00 (0.99, 1.01)    Occupational classc     Manual, specialized  1.02 (0.85, 1.22)  0.99 (0.97, 1.01)  0.83 (0.72, 0.95)**  1.02 (1.00, 1.03)*     White collar  0.83 (0.68, 1.02)  1.00 (0.97, 1.02)  0.83 (0.72, 0.95)**  1.01 (1.00, 1.03)*     Other  0.91 (0.76, 1.09)  1.01 (0.98, 1.03)  0.83 (0.73, 0.95)**  1.02 (1.00, 1.03)*    Men  Women  Intercept  Slope  Intercept  Slope  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  Age at baseline  50–59 Years   Unconditional estimates  2.24 (2.11, 2.37)***  1.06 (1.06, 1.06)***  2.02 (1.90, 2.16)***  1.06 (1.05, 1.06)***   Conditional estimates    Educationa     Beyond basic  0.75 (0.66, 0.85)***  1.01 (1.00, 1.02)  0.72 (0.63, 0.82)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)  0.55 (0.48, 0.63)***  1.01 (1.00, 1.02)*     High  0.43 (0.38, 0.49)***  1.02 (1.01, 1.03)**  0.44 (0.38, 0.52)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.01 (0.85, 1.19)  0.99 (0.98, 1.00)  1.22 (1.01, 1.47)*  1.00 (0.98, 1.01)     White collar  0.64 (0.54, 0.75)***  1.00 (0.99, 1.01)  0.83 (0.71, 0.97)*  0.99 (0.98, 1.01)     Other  0.87 (0.74, 1.03)  1.00 (0.99, 1.01)  1.35 (1.09, 1.67)**  0.99 (0.97, 1.00)  60–69 Years   Unconditional estimates  5.08 (4.83, 5.35)***  1.07 (1.06, 1.07)***  3.95 (3.76, 4.15)***  1.09 (1.09, 1.09)***   Conditional estimates    Educationa     Beyond basic  0.88 (0.77, 1.00)  1.00 (0.99, 1.01)  0.74 (0.65, 0.84)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.77 (0.69, 0.87)***  1.01 (1.00, 1.02)*  0.80 (0.72, 0.90)***  1.01 (1.00, 1.02)     High  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)*  0.73 (0.65, 0.83)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.04 (0.96, 1.20)  0.99 (0.98, 1.01)  1.25 (1.08, 1.45)**  0.99 (0.98, 1.00)     White collar  0.88 (0.75, 1.04)  0.99 (0.98, 1.01)  0.89 (0.78, 1.02)  1.00 (0.99, 1.01)     Other  0.96 (0.83, 1.11)  1.00 (0.98, 1.01)  1.23 (1.08, 1.41)**  0.99 (0.98, 1.00)*  70–79 Years   Unconditional estimates  11.73 (11.11, 12.39)***  1.07 (1.06, 1.07)***  12.33 (11.81, 12.88)***  1.08 (1.07, 1.08)***   Conditional estimates    Educationa     Beyond basic  0.79 (0.69, 0.91)**  1.01 (0.99, 1.03)  0.78 (0.69, 0.89)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.89 (0.78, 1.02)  1.00 (0.99, 1.02)  0.86 (0.78, 0.96)**  1.01 (1.00, 1.02)     High  0.88 (0.77, 1.00)*  1.01 (0.99, 1.02)  0.86 (0.77, 0.95)**  1.00 (0.99, 1.01)    Occupational classc     Manual, specialized  1.02 (0.85, 1.22)  0.99 (0.97, 1.01)  0.83 (0.72, 0.95)**  1.02 (1.00, 1.03)*     White collar  0.83 (0.68, 1.02)  1.00 (0.97, 1.02)  0.83 (0.72, 0.95)**  1.01 (1.00, 1.03)*     Other  0.91 (0.76, 1.09)  1.01 (0.98, 1.03)  0.83 (0.73, 0.95)**  1.02 (1.00, 1.03)*  Conditional models adjusted for native language and region of residence; all models were fitted separately for education, household income, and occupational class. IRR: Incidence rate ratio. *** p < .001 ** p < .01 * p < .05. Reference category a Basic education b Low household income tertile c Nonspecialized manual class or specialization unknown. View Large Table 2. SES and Trajectory of Hospital Days (Relative Difference), Stratified by Age Groups at Baseline   Men  Women  Intercept  Slope  Intercept  Slope  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  Age at baseline  50–59 Years   Unconditional estimates  2.24 (2.11, 2.37)***  1.06 (1.06, 1.06)***  2.02 (1.90, 2.16)***  1.06 (1.05, 1.06)***   Conditional estimates    Educationa     Beyond basic  0.75 (0.66, 0.85)***  1.01 (1.00, 1.02)  0.72 (0.63, 0.82)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)  0.55 (0.48, 0.63)***  1.01 (1.00, 1.02)*     High  0.43 (0.38, 0.49)***  1.02 (1.01, 1.03)**  0.44 (0.38, 0.52)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.01 (0.85, 1.19)  0.99 (0.98, 1.00)  1.22 (1.01, 1.47)*  1.00 (0.98, 1.01)     White collar  0.64 (0.54, 0.75)***  1.00 (0.99, 1.01)  0.83 (0.71, 0.97)*  0.99 (0.98, 1.01)     Other  0.87 (0.74, 1.03)  1.00 (0.99, 1.01)  1.35 (1.09, 1.67)**  0.99 (0.97, 1.00)  60–69 Years   Unconditional estimates  5.08 (4.83, 5.35)***  1.07 (1.06, 1.07)***  3.95 (3.76, 4.15)***  1.09 (1.09, 1.09)***   Conditional estimates    Educationa     Beyond basic  0.88 (0.77, 1.00)  1.00 (0.99, 1.01)  0.74 (0.65, 0.84)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.77 (0.69, 0.87)***  1.01 (1.00, 1.02)*  0.80 (0.72, 0.90)***  1.01 (1.00, 1.02)     High  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)*  0.73 (0.65, 0.83)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.04 (0.96, 1.20)  0.99 (0.98, 1.01)  1.25 (1.08, 1.45)**  0.99 (0.98, 1.00)     White collar  0.88 (0.75, 1.04)  0.99 (0.98, 1.01)  0.89 (0.78, 1.02)  1.00 (0.99, 1.01)     Other  0.96 (0.83, 1.11)  1.00 (0.98, 1.01)  1.23 (1.08, 1.41)**  0.99 (0.98, 1.00)*  70–79 Years   Unconditional estimates  11.73 (11.11, 12.39)***  1.07 (1.06, 1.07)***  12.33 (11.81, 12.88)***  1.08 (1.07, 1.08)***   Conditional estimates    Educationa     Beyond basic  0.79 (0.69, 0.91)**  1.01 (0.99, 1.03)  0.78 (0.69, 0.89)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.89 (0.78, 1.02)  1.00 (0.99, 1.02)  0.86 (0.78, 0.96)**  1.01 (1.00, 1.02)     High  0.88 (0.77, 1.00)*  1.01 (0.99, 1.02)  0.86 (0.77, 0.95)**  1.00 (0.99, 1.01)    Occupational classc     Manual, specialized  1.02 (0.85, 1.22)  0.99 (0.97, 1.01)  0.83 (0.72, 0.95)**  1.02 (1.00, 1.03)*     White collar  0.83 (0.68, 1.02)  1.00 (0.97, 1.02)  0.83 (0.72, 0.95)**  1.01 (1.00, 1.03)*     Other  0.91 (0.76, 1.09)  1.01 (0.98, 1.03)  0.83 (0.73, 0.95)**  1.02 (1.00, 1.03)*    Men  Women  Intercept  Slope  Intercept  Slope  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  IRR (95% CI)  Age at baseline  50–59 Years   Unconditional estimates  2.24 (2.11, 2.37)***  1.06 (1.06, 1.06)***  2.02 (1.90, 2.16)***  1.06 (1.05, 1.06)***   Conditional estimates    Educationa     Beyond basic  0.75 (0.66, 0.85)***  1.01 (1.00, 1.02)  0.72 (0.63, 0.82)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)  0.55 (0.48, 0.63)***  1.01 (1.00, 1.02)*     High  0.43 (0.38, 0.49)***  1.02 (1.01, 1.03)**  0.44 (0.38, 0.52)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.01 (0.85, 1.19)  0.99 (0.98, 1.00)  1.22 (1.01, 1.47)*  1.00 (0.98, 1.01)     White collar  0.64 (0.54, 0.75)***  1.00 (0.99, 1.01)  0.83 (0.71, 0.97)*  0.99 (0.98, 1.01)     Other  0.87 (0.74, 1.03)  1.00 (0.99, 1.01)  1.35 (1.09, 1.67)**  0.99 (0.97, 1.00)  60–69 Years   Unconditional estimates  5.08 (4.83, 5.35)***  1.07 (1.06, 1.07)***  3.95 (3.76, 4.15)***  1.09 (1.09, 1.09)***   Conditional estimates    Educationa     Beyond basic  0.88 (0.77, 1.00)  1.00 (0.99, 1.01)  0.74 (0.65, 0.84)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.77 (0.69, 0.87)***  1.01 (1.00, 1.02)*  0.80 (0.72, 0.90)***  1.01 (1.00, 1.02)     High  0.62 (0.55, 0.71)***  1.01 (1.00, 1.02)*  0.73 (0.65, 0.83)***  1.01 (1.00, 1.02)    Occupational classc     Manual, specialized  1.04 (0.96, 1.20)  0.99 (0.98, 1.01)  1.25 (1.08, 1.45)**  0.99 (0.98, 1.00)     White collar  0.88 (0.75, 1.04)  0.99 (0.98, 1.01)  0.89 (0.78, 1.02)  1.00 (0.99, 1.01)     Other  0.96 (0.83, 1.11)  1.00 (0.98, 1.01)  1.23 (1.08, 1.41)**  0.99 (0.98, 1.00)*  70–79 Years   Unconditional estimates  11.73 (11.11, 12.39)***  1.07 (1.06, 1.07)***  12.33 (11.81, 12.88)***  1.08 (1.07, 1.08)***   Conditional estimates    Educationa     Beyond basic  0.79 (0.69, 0.91)**  1.01 (0.99, 1.03)  0.78 (0.69, 0.89)***  1.01 (1.00, 1.02)    Household incomeb     Middle  0.89 (0.78, 1.02)  1.00 (0.99, 1.02)  0.86 (0.78, 0.96)**  1.01 (1.00, 1.02)     High  0.88 (0.77, 1.00)*  1.01 (0.99, 1.02)  0.86 (0.77, 0.95)**  1.00 (0.99, 1.01)    Occupational classc     Manual, specialized  1.02 (0.85, 1.22)  0.99 (0.97, 1.01)  0.83 (0.72, 0.95)**  1.02 (1.00, 1.03)*     White collar  0.83 (0.68, 1.02)  1.00 (0.97, 1.02)  0.83 (0.72, 0.95)**  1.01 (1.00, 1.03)*     Other  0.91 (0.76, 1.09)  1.01 (0.98, 1.03)  0.83 (0.73, 0.95)**  1.02 (1.00, 1.03)*  Conditional models adjusted for native language and region of residence; all models were fitted separately for education, household income, and occupational class. IRR: Incidence rate ratio. *** p < .001 ** p < .01 * p < .05. Reference category a Basic education b Low household income tertile c Nonspecialized manual class or specialization unknown. View Large As we can see from the conditional estimates, there was a clear gradient in the intercept indicating that among both sexes, individuals with higher education, household income, and occupational class had around 10%–60% fewer hospital days at baseline than their counterparts with lower SES. However, among men, most of the differences in the intercept between occupational classes were not statistically significant. For the slope, among some subgroups, higher SES was associated with a faster annual rate of increase in hospital days. For example, the rate of increase was 2% faster for men aged 50–59 years at baseline in the high household income tertile than for those in the low household income tertile (ratio of annual rate of increase: 1.02, 95% confidence interval [CI]: 1.01, 1.03). For men aged 60–69 years at baseline, being in the middle or high household income tertiles was associated with a 1% faster rate of increase than being in the low household income tertile (middle household income: 1.01, 95% CI: 1.00, 1.02; high household income: 1.01, 95% CI: 1.00, 1.03). For women aged 50–59 years at baseline, being in the middle household income tertile was associated with a 1% faster rate of increase than being in the low household income tertile (1.01; 95% CI: 1.00, 1.02). For women aged 70–79 years at baseline, compared to being in the nonspecialized manual class/specialization unknown category, the rate of increase was 1% faster for being in the specialized manual class (1.01, 95% CI: 1.00, 1.03) and 2% faster for being in the white-collar class (1.02, 95% CI: 1.00, 1.03). Accordingly, relative SES differences in hospital days diminished over the follow-up period for these subgroups. For other subgroups, the relative SES differences appeared to be unchanged, given the statistically nonsignificant ratios of the rate of increase. The interactions of each SES indicator with sex and with age were tested on both the intercept and slope. The effect of SES on the trajectories of hospital days did not differ by age groups among either men or women, except for household income on the intercept (both sex p < .001), for occupational class on the intercept (men p = .04; women p < .001), and for occupational class on the slope among women (p = .02). Similarly, we did not find sex differences in the effect of SES in any age group, except for occupational class on the intercept among individuals aged 50–59 years (p = .01) and 60–69 years (p = .03) and for education on the intercept among those aged 60–69 years (p = .04). The results from the pooled study population of men and women are shown in Supplementary Table 2. Figures 1 and 2 show how absolute differences in hospital days by education and household income changed in the years 1988–2007 (see Supplementary Figure 2 for occupational class). Although relative SES differences in hospital days largely maintained and only decreased in some subgroups, absolute differences appeared to enlarge over the 1988–2007 period among men and women aged 50–69 years at baseline, but narrow among those aged 70–79 years baseline. Figure 1. View largeDownload slide Educational attainment and trajectories of hospital days in 1988–2007 (absolute difference), by sex and age groups at baseline (in 1988). Figure 1. View largeDownload slide Educational attainment and trajectories of hospital days in 1988–2007 (absolute difference), by sex and age groups at baseline (in 1988). Figure 2. View largeDownload slide Household income and trajectories of hospital days in 1988–2007 (absolute difference), by sex and age groups at baseline (in 1988). Figure 2. View largeDownload slide Household income and trajectories of hospital days in 1988–2007 (absolute difference), by sex and age groups at baseline (in 1988). Number of hospitalization episodes is another important aspect of hospital care. We repeated our analysis using the annual total number of hospitalizations as the outcome, and the findings were similar to those for hospital days (Supplementary Table 3). However, a faster increase in the number of hospitalizations associated with higher education was found among men and women aged 70–79 years at baseline but not among those aged 50–59 years at baseline. For women aged 70–79 years at baseline, a faster increase in the number of hospitalizations was associated with higher household income but not with occupational class. For the absolute differences in the number of hospitalization between education groups, the tendency of a convergence over the 1988–2007 period became more evident in the group aged 70–79 years at baseline (Supplementary Figure 3). Over the follow-up period, 34,103 men and 36,631 women died. Hospital days were missing for these individuals after their deaths. Mplus handles missing data using full information maximum likelihood; a statistical estimation technique that is valid under the assumption of missing at random (MAR, i.e., missingness depends on covariates and the outcome variable observed at previous time points; Muthén & Muthén, 1998–2015). If MAR does not hold, missing not at random (MNAR) models should be used to perform sensitivity analysis (for more details, see Section 1.2 of the Supplementary Materials; Muthén, Asparouhov, Hunter, & Leuchter, 2011). We therefore fitted pattern-mixture models within the framework of latent growth curve modeling (MNAR models), in which both the intercept and slope of the hospital days were modeled as a function of the indicators of the year of death (Little, 2009; Muthén et al., 2011). More details are provided in Section 1.2 of the Supplementary Materials. The results of the pattern-mixture models deviated little from those of the MAR models (see results in Supplementary Tables 4). For both sexes and all age groups at baseline, relative SES differences for both the intercept and slope were smaller in the pattern-mixture models than in the MAR models. The age-specific patterns of changes in absolute differences in the pattern-mixture models were generally consistent with those in the MAR models (Supplementary Figures 4–6). However, a continuous divergence was seen between household income tertiles and occupational classes for men aged 70–79 years at baseline. Discussion In this study using large-scale longitudinal registry data, we found that older adults with higher education, household income, and occupational class had fewer hospital days at baseline than their counterparts with lower SES. Over the years 1989–2007, relative differences in hospital days declined between the lowest and higher household income tertiles for men aged 50–69 years at baseline; between the lowest and the middle household income tertiles for women aged 50–59 years at baseline; and between the nonspecialized manual class and the specialized manual/white-collar class for women aged 70–79 years at baseline. For the other subgroups, relative SES differences in hospital days were unchanged. There was, however, a further divergence in absolute differences in hospital days by education, household income, and occupational class groups for men and women aged 50–69 years at baseline, but a convergence for those aged 70–79 years at baseline. Our findings suggest that the divergence and convergence of socioeconomic disparities in health at older ages may not be mutually exclusive; rather, which trend is found may depend on the measurement scale of socioeconomic disparities in health (relative or absolute difference) and the ages used. Similar to previous observations on mortality (Bor et al., 2017; Mackenbach, Kulhánová, et al., 2016), for some SES subgroups at ages 50–69 years, we found that the relative difference narrowed while the absolute difference increased. For other subgroups, the relative difference remained stable at all ages, whereas the absolute difference increased at ages 50–69 years but diminished at ages older than 70 years. It is not straightforward to link these opposing trends to AAL, CAD, or status maintenance hypotheses because inference based on relative and absolute difference may lead to different conclusions. Mackenbach, Martikainen, Menvielle, and de Gelder (2016) showed that different combinations of starting levels and changes of mortality by SES groups over time lead to different patterns of changes in relative and absolute differences in mortality. Our findings fall into one of these patterns. Whereas relative SES differences in hospital days maintained/diminished in the 1988–2007 period, changes in absolute differences were less monotonic: absolute differences started to decline only after the ratio of hospital days in the higher versus the lower SES groups became larger than the ratio of the increase in hospital days in a year in the lower versus the higher SES groups. Since the opposing trends of change in relative and absolute differences are still not fully understood (Mackenbach, Martikainen, et al., 2016), it remains for further research to answer the question of which measurement scale, relative or absolute, is more appropriate to use for testing the AAL and CAD hypotheses. From the perspective of policy makers, actual absolute hospital days are a more relevant metric for policies aiming to reduce the SES differentials in hospital care and are more meaningful and easy to implement in monitoring success in policy interventions. We compared our findings on changes in absolute differences with the results of selected previous studies that used longitudinal data with repeated measures and similar statistical analytic approaches (Benzeval, Green, & Leyland, 2011; Chandola, Ferrie, Sacker, & Marmot, 2007; Herd, 2006; Kim & Durden, 2007; Stolz et al., 2017; Willson et al., 2007). One study reported a continuous divergence in the self-rated health trajectories between employment grades among British civil servants aged 39–74 years (Chandola et al., 2007), whereas another reported a continuous divergence in the physical impairment trajectories between education and income groups across ages 25–89 years (Kim & Durden, 2007). But Stolz and colleagues (2017) found stable differences in frailty with age across educational, wealth, and occupational class groups, and decreasing differences across income groups among Europeans older than 50 years. In line with our findings, other studies have shown that after diverging at earlier ages, disparities in physical functioning and self-rated health converge across education groups and manual and nonmanual classes when people reach their 60s or 70s (Benzeval et al., 2011; Herd, 2006; Willson et al., 2007). The studies conducted by Kim and Durden (2007) and Willson and colleagues (2007) are the only studies that covered most of the adult ages (ages 25–89 years and 26–92 years, respectively), whereas the other studies tracked changes in socioeconomic disparities in health only until people reached their 60s or early 70s. Our data enabled us to follow men and women aged 50–59, 60–69, and 70–79 years over a 20-year period starting at the end of 1987; and thus, until ages 70–79, 80–89, and 90–99 years, respectively. Moreover, we found important differences between the three 10-year age groups. Kim and Durden (2007) assumed common physical impairment trajectories for all ages, even though 43% of their sample were aged 25–49 years at baseline. Our results suggest that the divergence Kim and Durden (2007) found may be driven by these relatively healthy young adults, and that the convergence at older ages is hidden. Although Willson and colleagues (2007) did not make such assumption of common trajectories for all ages, the long-term panel data they used covered a relatively small sample of older adults aged 46–75 years at baseline (1,695, 31% of the whole sample), with nonresponse and attrition over the follow-up period. Uniquely, we were able to evaluate the competing hypotheses of convergence and divergence separately for different age groups with large sample sizes and no attrition. A divergence in absolute differences in hospital days over the 1988–2007 period for men and women aged 50–69 years at baseline was observed across education, household income, and occupational class groups. Of these SES indicators, income had the greatest effects on the rate of increase in hospital days. Our results support the idea that education, occupation, and income represent different types of resources and affect health through different pathways: income is more directly related to material resources available for accessing formal care services, treating health problems, or changing life circumstances to slow down the progression of health problems (House et al., 2005). The trend toward convergence in absolute SES disparities in hospital days in the group aged 70–79 years at baseline supports the AAL hypothesis, which asserts that physiological factors play a larger role in aging than social determinants (Dupre, 2007; Hoffmann, 2008). Benzeval and colleagues (2011) and Willson and colleagues (2007) argued that the convergence they observed was artificially caused by mortality selection and/or attrition. In contrast, Rohwer (2016) contended that survival should be viewed as a necessary precondition instead of as a possible source of bias, and that the growth curve models do not estimate health trajectories conditioned on survival. The pattern-mixture models we used for our sensitivity analysis are fully conditional on survival (Kurland, Johnson, Egleston, & Diehr, 2009). When mortality was taken into account in a series of pattern-mixture models, the changes in relative and absolute SES differences in hospital days found in these models were generally consistent with the results of the MAR models. This suggests that the MAR assumption (i.e., missing hospital days due to death were random after taking into account covariates and hospital days at previous time points) is reasonable. These observations partially confirm the assumption that mortality selection is not the main reason for the convergence in socioeconomic disparities in health at advanced ages (e.g., at older than 75 years; Beckett, 2000; Herd, 2006; House et al., 2005). There are several possible reasons for the divergence at younger ages and the convergence at older ages in absolute SES differences. Differences in levels of exposure to health risk factors (e.g., health behaviors and psychosocial factors) between socioeconomic groups peak at early and middle ages, and impacts of these health risk factors may fade over the life course (Chandola et al., 2007; House et al., 2005). In addition, the compression of morbidity (i.e., the time between the onset of chronic diseases or disability and the time an individual dies is compressed) at advanced ages may be greater in higher than in lower SES groups (Fries, 1980, 1996). In other words, compared to their counterparts with lower SES, older adults with higher SES may live longer without chronic diseases, disability, or hospitalization; but may experience a steeper terminal decline in health (Fries, 1996). This pattern would lead to a convergence in socioeconomic disparities in health at advanced ages. This trend toward convergence may also be related to the leveling effects of social security and health benefits provided by welfare (Hoffmann, 2008). Another well-known issue is the need to disentangle birth cohort effects from age effects (House et al., 2005; Lauderdale, 2001; Lynch, 2003; Rohwer, 2016; Willson et al., 2007). Ross and Wu (1996) acknowledged that their findings, using cross-sectional age, of a further divergence in educational differences in self-reported health with age might reflect both age and birth cohort effects. Lauderdale (2001) reported increasing educational differences in survival with age, and that the effect of education was stronger in later than in earlier birth cohorts. Lynch (2003) also showed that ignoring birth cohorts suppressed the widening of socioeconomic gaps in health with age. We took an approach similar to that of Lynch, stratifying our study population by 10-year age groups at baseline (i.e., corresponding to the cohorts born in 1908–1917, 1918–1927, 1928–1937) and estimating the trajectories of hospital days over the follow-up years. It is possible that our approach could not differentiate the true age effect from the period effect. The decline of somatic hospital beds and hospital admissions observed in Finland over the years 2001–2011 could reflect the improvement of population health, technological advances, and the shift from inpatient care to ambulatory care (Keskimäki, Forssas, Rautiainen, Rasilainen, & Gissler, 2014; Manderbacka, Arffman, & Keskimaki, 2014). Also in the U.K., advances in technology, medical treatment, and medical care have been found to reduce the length of hospital stay (Lewis & Edwards, 2015). It is unlikely that the increase in hospital days we observed over the 1988–2007 period was driven by period effects. However, the advances in medical care made in Finland over our study period may complicate the interpretation of our analyses. Since Finland has a universal health care system, it seems less likely that individuals with higher SES had significantly better access to these advances than those with lower SES. If there was a significant faster increase in access to care among older adults with higher SES, we would observe a convergence across all older ages. This cannot explain the further divergence among those aged 50–69 years at baseline. Remarkable regional variations in health care usage have been observed in Finland due to the differences in population morbidity pattern, medical practices, health care resources, and efficiency (Keskimäki et al., 2014). Living in rural areas far away from hospitals (e.g., more than 40 km) is found to be associated with fewer days spent in hospital (Zielinski, Borgquist, & Halling, 2013). Finnish individuals who moved from urban to rural areas tend to be older (e.g., after retirement) and with lower education and income (Nivalainen, 2003a, 2003b). However, it is unclear to what extent these processes affect our results because of other urban–rural differences. For instance, the reduction of somatic hospital beds and shift to ambulatory care may be greater in urban than in rural areas, and the medical practices in rural areas may prefer to refer patient to inpatient hospital care more readily than in urban areas. Unfortunately, we do not have information on urban/rural areas to formally test these possibilities; a topic that requires further research. Our study has some important strengths. First, following recommendations made in previous studies (Dupre, 2007; House et al., 2005; Leopold, Engelhardt, & Engelhartdt, 2013), our findings were based on 20-year longitudinal data from a large-scale, national representative random sample of older adults. More importantly, we advanced our understanding of changes in relative SES differences in hospital days with age. Since our data came from administrative registers, our findings are not affected by the bias that can arise in panel studies because individuals with low SES or poor health are less likely to participate in studies. We used hospital days to reflect older adults’ health status and morbidity. Hospital days is an objective measure that is less prone to measurement error than the subjective measures commonly used in previous studies, such as self-rated health and self-reported disability. It could be argued that hospital days reflect serious health problems only. However, this issue may be less problematic in our study population, as older adults tend to have more serious health conditions than younger adults. We examined educational attainment, household income, and occupational class in order to capture the multidimensional character of SES. Particularly for women, household income may be a better indicator of material circumstances than personal income. We acknowledge that our study also has some limitations. First, the models we used could not describe trajectories over time-varying characteristics. The SES indicators in our study were measured at baseline. Given the ages of our study population, we can assume that their educational attainment was fixed; however, their occupational class and household income may have changed (e.g., after retirement). The occupational class of retirees was based on their previous job status, and remains unchanged. Working-aged older adults may change jobs or become unemployed, but their occupational class is unlikely to change over time. Our data showed that the occupational class remained unchanged for 97% of our study population. Change in household income can be influenced, for example, by retirement and widowhood. We used the age-specific household income tertiles at baseline rather than the annually updated household income; over 80% of our study population stayed in the same tertile over the follow-up period. Overall, SES thus appears to be relative stable at these older ages. Second, in Finland, older adults from lower SES groups are more likely to transit into long-term institutional care (Martikainen, Nihtilä, & Moustgaard, 2008). We did not exclude institutionalized older adults from this study; instead we coded their hospital days as zero if they did not receive hospital care after they were institutionalized. Therefore, assuming that long-term care substitutes for the care given in hospitals, the hospital days and the rate of increase in hospital days may be downward biased among older adults; in particular among those with lower SES compared to those with higher SES. However, of our study population, only 1,200 men (2.0%) and 3,470 women (4.4%) had ever been institutionalized, and in 39% of the years that they were institutionalized they also received hospital care. Hence, the possibly downward bias in hospital days and rate of increase in hospital days in the lower SES groups due to institutionalization is unlikely to substantially affect our results. Third, because we were using registry data, we were unable to provide insights into the behavioral mechanisms and the interplay between behavioral risk factors and SES over the life course that could affect health in later life. Thus, future research using panel data with multiple repeated measures on health and related risk factors, and with long-term follow-up to track the health of the older population to advanced ages, is needed to fully explain changes in socioeconomic disparities in late life with age. Moreover, to better explain the convergence in absolute SES differences in health at advanced ages, additional research that investigates the roles of welfare state policies and compression of morbidity is required. Overall, our findings suggest that the divergence and the convergence in socioeconomic disparities in health at older ages may not be mutually exclusive; as they may depend on the measure of health disparities used and the ages of the populations studied. This implies that more efforts are needed to tackle socioeconomic differences in health in later life, including among the oldest old. A particular focus should be on early old ages; a stage of life when absolute SES differentials are still diverging. Supplementary Material Supplementary data is available at Journals of Gerontology, Series B: Psychological and Social Sciences online. Funding This work is supported by the European Research Council Starting Grant to M. Myrskylä (COSTPOST) [#336475]. The research leading to this publication has been supported by the Max Planck Society within the framework of the project “On the edge of societies: New vulnerable populations, emerging challenges for social policies and future demands for social innovation. The experience of the Baltic Sea States” (2016–2021). P. Martikainen is funded by the Academy of Finland, Strategic Research Council PROMEQ project (#303615) and the Signe and Ane Gyllenberg Foundation. Conflict of Interest None reported. Acknowledgment The authors would like to thank Statistics Finland and the National Institute for Health and Welfare for providing the data. Y. Hu and P. Martikainen conceptualized the research project. Y. Hu performed the data analysis and drafted the manuscript. Y. Hu, T. Leinonen, M. Myrskylä, and P. Martikainen interpreted the results. T. Leinonen, M. Myrskylä, and P. Martikainen critically revised and commented on all versions of the manuscript. All authors approved the final version of the manuscript. References Beckett, M. ( 2000). Converging health inequalities in later life–an artifact of mortality selection. Journal of Health and Social Behavior , 41, 106– 119. doi:10.2307/2676363 Google Scholar CrossRef Search ADS PubMed  Benzeval, M., Green, M. J., & Leyland, A. H. ( 2011). Do social inequalities in health widen or converge with age? Longitudinal evidence from three cohorts in the West of Scotland. BMC Public Health , 11, 947. doi:10.1186/1471-2458-11-947 Google Scholar CrossRef Search ADS PubMed  Bor, J., Cohen, G. H., & Galea, S. ( 2017). Population health in an era of rising income inequality: USA, 1980-2015. Lancet (London, England) , 389, 1475– 1490. doi:10.1016/S0140-6736(17)30571-8 Google Scholar CrossRef Search ADS PubMed  Braveman, P., Egerter, S., & Williams, D. R. ( 2011). The social determinants of health: Coming of age. Annual Review of Public Health , 32, 381– 398. doi:10.1146/annurev-publhealth-031210-101218 Google Scholar CrossRef Search ADS PubMed  Chandola, T., Ferrie, J., Sacker, A., & Marmot, M. ( 2007). Social inequalities in self reported health in early old age: Follow-up of prospective cohort study. BMJ (Clinical research ed.) , 334, 990. doi:10.1136/bmj.39167.439792.55 Google Scholar CrossRef Search ADS PubMed  Corna, L. M. ( 2013). A life course perspective on socioeconomic inequalities in health: A critical review of conceptual frameworks. Advances in Life Course Research , 18, 150– 159. doi:10.1016/j.alcr.2013.01.002 Google Scholar CrossRef Search ADS PubMed  Dupre, M. E. ( 2007). Educational differences in age-related patterns of disease: Reconsidering the cumulative disadvantage and age-as-leveler hypotheses. Journal of Health and Social Behavior , 48, 1– 15. doi:10.1177/002214650704800101 Google Scholar CrossRef Search ADS PubMed  Epstein, A. M., Stern, R. S., & Weissman, J. S. ( 1990). Do the poor cost more? A multihospital study of patients’ socioeconomic status and use of hospital resources. The New England Journal of Medicine , 322, 1122– 1128. doi:10.1056/NEJM199004193221606 Google Scholar CrossRef Search ADS PubMed  Feinstein, J. S. ( 1993). The relationship between socioeconomic status and health: A review of the literature. The Milbank quarterly , 71, 279– 322. doi: 10.2307/3350401 Google Scholar CrossRef Search ADS PubMed  Fries, J. F. ( 1980). Aging, natural death, and the compression of morbidity. The New England Journal of Medicine , 303, 130– 135. doi:10.1056/NEJM198007173030304 Google Scholar CrossRef Search ADS PubMed  Fries, J. F. ( 1996). Physical activity, the compression of morbidity, and the health of the elderly. Journal of the Royal Society of Medicine , 89, 64– 68. PMCID: PMC1295658. Google Scholar PubMed  Grundy, E., & Sloggett, A. ( 2003). Health inequalities in the older population: The role of personal capital, social resources and socio-economic circumstances. Social Science and Medicine (1982) , 56, 935– 947. doi:10.1016/S0277-9536(02) 00093-X Google Scholar CrossRef Search ADS   Hagenaars, A., de Vos, K., & Zaidi, M. A. ( 1994). Poverty statistics in the late 1980s: Research based on micro-data . Luxembourg: Office for Official Publications of the European Communities. Herd, P. ( 2006). Do functional health inequalities decrease in old age? Educational status and functional decline among the 1931–1941 birth cohort. Research on Aging , 28, 375– 392. doi:10.1177/0164027505285845 Google Scholar CrossRef Search ADS   Hilbe, J. M. ( 2007). Negative binomial regression . Cambridge, U.K.: Cambridge University Press. Google Scholar CrossRef Search ADS   Hoffmann, R. ( 2008). Socioeconomic differences in old age mortality . Dordrecht: Springer. Google Scholar CrossRef Search ADS   House, J. S., Lantz, P. M., & Herd, P. ( 2005). Continuity and change in the social stratification of aging and health over the life course: Evidence from a nationally representative longitudinal study from 1986 to 2001/2002 (Americans’ Changing Lives Study). The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences , 60 Spec No 2, 15– 26. doi:10.1093/geronb/60.Special_Issue_2.S15 Google Scholar CrossRef Search ADS   House, J. S., Lepkowski, J. M., Kinney, A. M., Mero, R. P., Kessler, R. C., & Herzog, A. R. ( 1994). The social stratification of aging and health. Journal of Health and Social Behavior , 35, 213– 234. doi:10.2307/2137277 Google Scholar CrossRef Search ADS PubMed  Howe, L. D., Tilling, K., Galobardes, B., & Lawlor, D. A. ( 2013). Loss to follow-up in cohort studies: Bias in estimates of socioeconomic inequalities. Epidemiology (Cambridge, Mass.) , 24, 1– 9. doi:10.1097/EDE.0b013e31827623b1 Google Scholar CrossRef Search ADS PubMed  Huisman, M., Kunst, A. E., & Mackenbach, J. P. ( 2003). Socioeconomic inequalities in morbidity among the elderly; a European overview. Social Science and Medicine (1982) , 57, 861– 873. doi:10.1016/S0277-9536(02)00454-9 Google Scholar CrossRef Search ADS   Keskimäki, I., Forssas, E., Rautiainen, H., Rasilainen, J., & Gissler, M. ( 2014). Finland: Geographic variations in health care. In Geographic variations in health care: What do we know and what can be done to improve health system performance?  Paris: OECD Publishing. Google Scholar CrossRef Search ADS   Kim, J., & Durden, E. ( 2007). Socioeconomic status and age trajectories of health. Social Science and Medicine (1982) , 65, 2489– 2502. doi:10.1016/j.socscimed.2007.07.022 Google Scholar CrossRef Search ADS   Kurland, B. F., Johnson, L. L., Egleston, B. L., & Diehr, P. H. ( 2009). Longitudinal data with follow-up truncated by death: Match the analysis method to research aims. Statistical Science: A Review Journal of the Institute of Mathematical Statistics , 24, 211. doi:10.1214/09-STS293 Google Scholar PubMed  Lauderdale, D. S. ( 2001). Education and survival: Birth cohort, period, and age effects. Demography , 38, 551– 561. doi:10.1353/dem.2001.0035 Google Scholar CrossRef Search ADS PubMed  Leopold, L., & Engelhardt. ( 2013). Education and physical health trajectories in old age. Evidence from the Survey of Health, Ageing and Retirement in Europe (SHARE). International Journal of Public Health , 58, 23– 31. doi:10.1007/s00038-012-0399-0 Google Scholar CrossRef Search ADS PubMed  Lewis, R., & Edwards, N. ( 2015). Improving length of stay: What can hospitals do? Retrieved May 24, 2017 from https://www.nuffieldtrust.org.uk/files/2017-01/improving-length-of-stay-hospitals-web-final.pdf Liao, Y., McGee, D. L., Kaufman, J. S., Cao, G., & Cooper, R. S. ( 1999). Socioeconomic status and morbidity in the last years of life. American Journal of Public Health , 89, 569– 572. doi:10.2105/Ajph.89.4.569 Google Scholar CrossRef Search ADS PubMed  Little, R. ( 2009). Selection and pattern mixture models. In G. Fitzmaurice, M. Davidian, G. Verbeke & G. Molenberghs (Eds.), Longitudinal data analysis  (pp. 409– 431). Boca Raton, FL: Chapman & Hall. Google Scholar CrossRef Search ADS   Liu, H., & Powers, D. A. ( 2007). Growth curve models for zero-inflated count data: An application to smoking behavior. Structural Equation Modeling: A Multidisciplinary Journal , 14, 247– 279. doi:10.1080/10705510709336746 Google Scholar CrossRef Search ADS   Lynch, S. M. ( 2003). Cohort and life-course patterns in the relationship between education and health: A hierarchical approach. Demography , 40, 309– 331. doi:10.2307/3180803 Google Scholar CrossRef Search ADS PubMed  Lynch, S. M., & Brown, J. S. ( 2011). Stratification and inequality over the life course. In R. H. Binstock & L. K. George (Eds.), Handbook of aging and the social sciences  ( 7th ed., pp. 105– 120). San Diego, CA: Elsevier. Google Scholar CrossRef Search ADS   Mackenbach, J. P. ( 2015). Should we aim to reduce relative or absolute inequalities in mortality? European Journal of Public Health , 25, 185. doi:10.1093/eurpub/cku217 Google Scholar CrossRef Search ADS PubMed  Mackenbach, J. P., Martikainen, P., Menvielle, G., & de Gelder, R. ( 2016). The arithmetic of reducing relative and absolute inequalities in health: A theoretical analysis illustrated with European mortality data. Journal of Epidemiology and Community Health , 70, 730– 736. doi:10.1136/jech-2015-207018 Google Scholar CrossRef Search ADS PubMed  Mackenbach, J. P., Kulhánová, I., Artnik, B., Bopp, M., Borrell, C., Clemens, T., … de Gelder, R. ( 2016). Changes in mortality inequalities over two decades: register based study of European countries. British Medical Journal , 353, i1732. doi:10.1136/bmj.i1732 Google Scholar CrossRef Search ADS PubMed  Manderbacka, K., Arffman, M., & Keskimäki, I. ( 2014). Has socioeconomic equity increased in somatic specialist care: A register-based cohort study from Finland in 1995-2010. BMC Health Services Research , 14, 430. doi:10.1186/1472-6963-14-430 Google Scholar CrossRef Search ADS PubMed  Marmot, M. ( 2005). Social determinants of health inequalities. Lancet , 365, 1099– 1104. doi:10.1016/S0140-6736(05)71146-6 Google Scholar CrossRef Search ADS PubMed  Martikainen, P., Nihtilä, E., & Moustgaard, H. ( 2008). The effects of socioeconomic status and health on transitions in living arrangements and mortality: A longitudinal analysis of elderly Finnish men and women from 1997 to 2002. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences , 63, S99– S109. doi:10.1093/geronb/63.2.S99 Google Scholar CrossRef Search ADS   Mirowsky, J., & Ross, C. E. ( 2008). Education and self-rated health: Cumulative advantage and its rising importance. Research on Aging , 30, 93– 122. doi:10.1177/0164027507309649 Google Scholar CrossRef Search ADS   Muthén, B., Asparouhov, T., Hunter, A. M., & Leuchter, A. F. ( 2011). Growth modeling with nonignorable dropout: Alternative analyses of the STAR*D antidepressant trial. Psychological Methods , 16, 17– 33. doi:10.1037/a0022634 Google Scholar CrossRef Search ADS PubMed  Muthén, L. K., & Muthén, B. O. ( 1998–2015). Mplus user’s guide  ( 7th ed.). Los Angeles, CA: Muthén & Muthén. Nivalainen, S. ( 2003a). Where do migrants go? An Analysis of Rural and Urban Destined/Originated Migration in Finland 1996–1999 (Working Papers No. 66) . Helsinki, Finland: Pellervo Economic Research Institute. Nivalainen, S. ( 2003b, August). Who move to rural areas? Micro Evidence from Finland . Paper presented at the 43rd Congress of the European Regional Science Association: “Peripheries, Centres, and Spatial Development in the New Europe”. Jyväskylä, Finland. Retrieved August 15, 2017 from http://www-sre.wu.ac.at/ersa/ersaconfs/ersa03/cdrom/papers/214.pdf Pavalko, E. K., & Wilson, A. E. ( 2011). Life course approaches to health, illness and healing. In B. A. Pescosolido, J. K. Martin, J. D. McLeod & A. Rogers (Eds.), Handbook of the sociology of health, illness, and healing . New York: Springer. Google Scholar CrossRef Search ADS   Rechel, B., Grundy, E., Robine, J. M., Cylus, J., Mackenbach, J. P., Knai, C., & McKee, M. ( 2013). Ageing in the European Union. Lancet (London, England) , 381, 1312– 1322. doi:10.1016/S0140-6736(12)62087-X Google Scholar CrossRef Search ADS PubMed  Regidor, E. ( 2004). Measures of health inequalities: Part 2. Journal of Epidemiology and Community Health , 58, 900– 903. doi:10.1136/jech.2004.023036 Google Scholar CrossRef Search ADS PubMed  Rohwer, G. ( 2016). A note on the dependence of health on age and education. Demography , 53, 325– 335. doi:10.1007/s13524-016-0457-y Google Scholar CrossRef Search ADS PubMed  Roos, N. P., & Mustard, C. A. ( 1997). Variation in health and health care use by socioeconomic status in Winnipeg, Canada: Does the system work well? Yes and no. The Milbank Quarterly , 75, 89– 111. doi:10.1111/1468-0009.00045 Google Scholar CrossRef Search ADS PubMed  Ross, C. E., & Wu, C. L. ( 1996). Education, age, and the cumulative advantage in health. Journal of Health and Social Behavior , 37, 104– 120. doi:10.2307/2137234 Google Scholar CrossRef Search ADS PubMed  Stolz, E., Mayerl, H., Waxenegger, A., Rásky, É., & Freidl, W. ( 2017). Impact of socioeconomic position on frailty trajectories in 10 European countries: Evidence from the Survey of Health, Ageing and Retirement in Europe (2004-2013). Journal of Epidemiology and Community Health , 71, 73– 80. doi:10.1136/jech-2016-207712 Google Scholar CrossRef Search ADS PubMed  Willson, A. E., Shuey, K. M., & Elder, G. H. ( 2007). Cumulative advantage processes as mechanisms of inequality in life course health. American Journal of Sociology , 112, 1886– 1924. doi:10.1086/512712 Google Scholar CrossRef Search ADS   Wolinsky, F. D., Culler, S. D., Callahan, C. M., & Johnson, R. J. ( 1994). Hospital resource consumption among older adults: A prospective analysis of episodes, length of stay, and charges over a seven-year period. Journal of Gerontology , 49, S240– S252. doi:10.1093/geronj/49.5.S240 Google Scholar CrossRef Search ADS PubMed  World Health Organization.( 2015). World Report on Ageing and Health. Luxembourg: World Health Organization. Retrieved February 15, 2017 from http://apps.who.int/iris/bitstream/10665/186463/1/9789240694811_eng.pdf. World Health Organization.( 2016). Global Health Estimates 2015: Disease burden by Cause, Age, Sex, by Country and by Region, 2000–2015. Retrieved Auguest 14, 2017 from http://www.who.int/healthinfo/global_burden_disease/estimates/en/index2.html Zielinski, A., Borgquist, L., & Halling, A. ( 2013). Distance to hospital and socioeconomic status influence secondary health care use. Scandinavian Journal of Primary Health Care , 31, 83– 88. doi:10.3109/02813432.2012.759712 Google Scholar CrossRef Search ADS PubMed  © The Author(s) 2018. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.

Journal

The Journals of Gerontology Series B: Psychological Sciences and Social SciencesOxford University Press

Published: Jan 11, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off