Health inequalities in old age: the relative contribution of material, behavioral and psychosocial factors in a German sample

Abstract

Background

Whereas the association between education and health in later life is well described, investigations about the underlying mechanisms of these health inequalities are scarce. This study examines the relative contribution of material, behavioral and psychosocial factors to health inequalities in older Germans.

Methods

Data were drawn from the fifth wave of the Survey of Health, Ageing and Retirement in Europe (SHARE). The analytic sample included 3246 participants aged 60–85 years. We examined the independent and indirect contribution of material, behavioral and psychosocial factors to the association between education and self-rated health based on logistic regression models.

Results

Material factors were most important as they were additionally working through behavioral and psychosocial factors whereas the independent contribution of behavioral and psychosocial factors was much lower than suggested in the separate analyses of the three explanatory pathways.

Conclusions

Policy interventions that focus on the improvement of material living conditions might reduce health inequalities in old age. In studies on the underlying mechanisms of health inequalities, material, behavioral and psychosocial factors should be modeled as inter-related predictors as the separate analysis does not reveal their actual contribution so that the relevance of single explanatory pathways might be overestimated.

older people, social determinants, socioeconomic factors

Introduction

Against the background of rapid population aging in Europe and as older people constitute the vast majority of those with health problems in developed countries, understanding the determinants of health in old age has become one major concern of policy-makers. One such determinant is a person’s educational attainment and numerous studies have documented an educational gradient in health: the higher the educational level, the lower the risk of long-standing illness, functional limitations, low self-rated health (SRH) and mortality.^1–3 These health inequalities are even present in the oldest old although it is less clear if and how these relationships change during the aging process as compared to middle adulthood.^2,4

Several studies have identified material, behavioral and psychosocial factors as key pathways for explaining health inequalities.⁵ The materialist explanation underlines the importance of financial resources, working and housing conditions or access to goods, services and healthcare.^6,7 The behavioral explanation claims that health inequalities result from the higher prevalence of smoking, excessive alcohol consumption, physical inactivity and inadequate nutrition in lower educational groups.¹ Psychosocial explanations stress the unequal distribution of risk factors such as a lack of social support and social participation or insufficient control beliefs which affect health through various pathways.^7–9

Current explanatory approaches postulate that material, behavioral and psychosocial factors exert an independent influence on health (direct effect), while also being interrelated and working through one another (indirect effect) (Fig. 1, modified after Moor et al.⁵). Material factors affect health indirectly through psychosocial factors, e.g. when the lack of financial resources leads to social exclusion which in turn has a negative impact on health. Furthermore, material factors work indirectly through behavioral factors, e.g. in the case of an unhealthy nutrition as paying for high quality foods cannot be afforded.⁸ Psychosocial factors exert an indirect effect through health behaviors, e.g. when smoking or overeating are applied to cope with psychosocial stressors.⁸ Therefore, the relevance of single explanatory approaches might be overestimated when only one pathway is taken into account.⁵

Fig. 1

Simplified causal model for educational inequalities in health with independent (direct) and indirect effects of material, behavioral and psychosocial factors.

Most previous studies on the underlying mechanisms of educational inequalities in health are based on study samples including the population of working age only, whereas studies on the older population are scarce. The existing studies argue for the importance of material living conditions as they are widely working through behavioral and psychosocial factors.^10–14

According to a recent literature review,⁵ no study has analyzed the relative contribution of material, behavioral and psychosocial factors to educational inequalities in SRH in the older population. Thus, the aim of this study was to examine contribution of these explanatory pathways to the association between education and SRH in old age.

Method

Study population

The data were drawn from the fifth wave of the Survey of Health, Ageing and Retirement in Europe (SHARE), a multidisciplinary panel study on the living conditions of people aged 50+. The data are collected in computer-assisted personal interviews supplemented by self-completion questionnaires. A detailed description of the survey methodology can be found elsewhere.¹⁵ For this study, the German sample aged 60–85 years was selected. In contrast to the ‘oldest old’ (age 85+), people of this age group are often characterized by a relatively good health status and cognitive abilities.¹⁶

We limited our analysis to only one country as previous research has shown that, besides individual characteristics, there are also macro-level influences on health inequalities,^17,18 which speaks against examining the pooled sample of all participating countries. Germany with its conservative welfare state in which the living conditions in old age are strongly influenced by resources from prior life stages (such as education) offers an interesting opportunity for studying health inequalities in old age.

For some variables relevant for our analysis (material deprivation, access to healthcare and social capital), the number of missing values was much higher than for others. We ran the analysis with missing data excluded which eliminated nearly 300 cases (in addition to the n = 152 with missing values on the other variables). Additionally, we ran the analysis with a dataset in which missing values were replaced by the mean score of the sample. As there were only small differences in the size of coefficient estimates and their statistical significance, we decided to rely on the dataset which maintains cases with missing values and therefore provides a less biased sample. Participants with missing values on the other variables (n = 152) and with a BMI lower than 18 (n = 22) were excluded, so that an analytic sample of 3246 participants resulted.

Variables

Education and health

Education was measured by the highest educational level using the International Standard Classification of Educational Degrees (ISCED-97).¹⁹ Participants were categorized into three groups: ‘low education’ (ISCED-97-levels 0–2: pre-primary, primary and lower secondary education), ‘medium education’ (ISCED-97-levels 3–4: upper and post-secondary education) and ‘high education’ (ISCED-97-levels 5–6: tertiary education).^20,21

SRH was chosen as indicator of health as it is a strong predictor for morbidity and mortality that is widely used in public health research.^22,23 SRH was assessed by asking ‘Would you say your health is…excellent/very good/good/fair/poor’. A binary variable was coded into ‘good health’ (good or better) versus ‘poor health’ (fair/poor).^10,20

Material, behavioral and psychosocial factors

We did a literature-based selection of explanatory variables and covered a large number of variables that have been examined previously. The material factors were house ownership,^14,24 financial situation, material deprivation, access to healthcare and type of health insurance.^14,25

Behavioral factors were smoking habits,¹ alcohol consumption,²⁶ moderate or vigorous physical activity^27,28 and body mass index (BMI) as a proxy of the quantity and quality of food intake.^24,29

Psychosocial factors were civil status,^25,30 parenthood,¹⁰ social participation, control beliefs and social capital. See Table 1 for the operationalization and coding of the variables.

Table 1

Operationalization and coding of explanatory variables

Variable	Operationalization in SHARE and coding for analysis
Material factors
Housing tenure	0 = Owners, 1 = non-owners
Financial situation	‘Thinking of your household’s total monthly income, would you say that your household is able to make ends meet…with great difficulty/with some difficulty/fairly easily/easily?’ 0 = no financial problems (easily), 1 = some financial problems (fairly easy/with some difficulty), 3 = great financial problems (with great difficulty)
Material deprivation	Several questions about problems in affording goods and amenities of daily living (replacing worn out clothes or shoes; replacing glasses; buying necessary groceries and household supplies; heating; eating meat, fish, poultry, fruits or vegetables more than once a week; doing a week long holiday once a year; paying unexpected expenses without borrowing money), 0 = not deprived (less than three items mentioned), 1 = deprived (three or more items mentioned)
Access to healthcare	Several questions that inquire if the participants could not see a doctor or dentist because of costs or long waiting times in the past 12 months, or if they had difficulties to get to their general practitioner, the nearest health center or pharmacy 0 = no problems, 1 = at least one problem
Type of health insurance	0 = Private health insurance, 1 = statutory health insurance
Behavioral factors
Smoking habits	0 = Non-smoker, 1 = smoker
Alcohol consumption	1 = Never / one to two times per month, 2 = one to four times per week, 3 = five times a week or more
Moderate or vigorous physical activity	0 = More than once a week, 1 = less than once a week
BMI	0 = Normal weight (18–25), 1 = overweight (25 < 30), 2 = obesity (≥ 30)
Psychosocial factors
Marital status	0 = Married/in a relationship, 1 = divorced/separated/widowed/single
Parenthood	0 = At least one child alive, 1 = no child alive
Social participation	‘How often in the past 12 months…did you do voluntary or charity work/have you attended an educational or training course/did you go to a sport, social or other kind of club/haven you taken part in a political or community-related organization/have you played cards or games such as chess?’ 0 = any of these activities at least once a week, 1 = none of these activities at least once a week
Control beliefs	‘How often do you feel that what happens to you is out of your control?’; ‘How often do you feel left out of things?’ (both coded 1 = often, 2 = sometimes, 3 = rarely, 4 = never) Index summing up the two scores (range 2–8) with 0 = sufficient control beliefs (index values 5–8), 1 = insufficient control beliefs (index values 2–4)
Social capital	‘I really feel part of this area.’; ‘If I were in trouble, there are people in this area who would help me.’ (both coded 1 = agree, 2 = agree, 3 = disagree, 4 = strongly disagree) Index summing up the two scores (range 2–8) with 0 = sufficient social capital (index values 2–4), 1 = insufficient social capital (index values 5–8)

Statistical analysis

We estimated logistic regression models using the highest educational group as the reference category. In accordance with previous studies, a mediator analysis was performed to investigate the relative contribution of material, behavioral and psychosocial factors to the association between education and SRH.^13,14,31 Only those variables were selected for mediator analysis which (i) had a negative and significant association with education (bivariate analysis, χ²-test) and (ii) were significantly associated with SRH (logistic regression controlled for sex, age, employment status and chronic diseases).^10,14,31 Several models were estimated to calculate the contribution of the different explanatory pathways. The baseline model (Model 1) included educational status and control variables (sex, age, employment status and chronic diseases). Afterwards, we separately added material factors, behavioral factors and psychosocial factors (Models 2–4). The reduction of the effect size of the coefficient estimates for educational status was interpreted as the effect of education that is mediated by the variable group included. Furthermore, we adjusted the baseline model for several combinations of two groups of variables (Models 5–7). By comparing the models with two groups of variables with the corresponding models with each one group of explanatory factors we distinguished between the independent and indirect effects. For a detailed description of the calculation method, see Table 3. Finally, we included all variables simultaneously to estimate their total contribution (Model 8).^10,13,14 For each model, we calculated the relative contribution to the association between education and SRH by: ((β_{(Model 1)} – β_{(Model 2}–₈)/β_{(Model 1)}) × 100). Although interrelations between variables were found, there was no serious problem of multicollinearity (mean variance inflation factor < 3.1).

Analyses were performed with Stata V.13 (StatCorp, Texas, USA) using the decomposition method of Karlson, Holm and Breen (KHB) which allows a comparison of regression coefficients between same-sample nested non-linear models.³² In contrast to linear models, changes in regression coefficients in logit models cannot readily be attributed to the effect of including mediator variables as parameter rescaling by itself tends to increase the regression coefficients.^32,33

The KHB-method corrects for the issue of parameter rescaling and provides an unbiased decomposition of the total effect into direct and indirect effects. Additionally, we use beta-coefficients as odds ratios (OR) are not symmetric on a linear scale which results in a biased interpretation of the relative contribution of mediator variables.

As we use cross-sectional data, it is not possible to determine the direction of causation between education, mediator variables and SRH. When the terms ‘effect’ or ‘explain’ are used in the upcoming sections, they should be understood in a statistical sense rather than in a causal relationship sense.

Results

Inequalities in SRH were found with a higher risk of reporting poor health with an OR of 2.06 (95% CI = 1.71–2.48) in the middle educational group and an OR of 3.44 (95% CI = 2.63–4.50) in the low educational group as compared to the highly educated when controlling for sex, age, employment status and chronic diseases (descriptive statistics of the study sample in Supplementary online file).

After the first step of variable selection (χ² test), all material factors and most of the behavioral factors remained for further analysis. Only alcohol consumption was excluded as there was an inverse social gradient (excessive alcohol consumption was less common in the middle and low educational group). Of the psychosocial factors, parenthood and social capital were excluded as they were not significantly associated with education. After the second step of variable selection (logistic regression), all of the remaining factors except of a BMI 25 <30 were included in the mediator analysis as they were significantly associated with SRH (results of variable selection in Supplementary online file).

Table 2 shows the results of the mediator analysis. In the separate analyses (Models 2–4), the contribution of material factors was the highest (middle educational group: 18%, low educational group: 23%), followed by behavioral factors (middle educational group: 13%, low educational group: 19%) and psychosocial factors (middle educational group: 12%, low educational group: 17%). All pathways together (Model 8) contributed by 31% to the association between education and SRH in the middle educational group, with material factors (12%) being slightly more important than behavioral factors (9%) and psychosocial factors (9%). In the low educational group the three explanatory pathways together contributed by 42% to the association between education and SRH with material factors being most important (16%), followed by behavioral factors (14%) and psychosocial factors (13%). The most relevant variables were financial problems, type of health insurance (especially in the low educational group), low social participation, insufficient control beliefs, lacking physical activity and a BMI ≥ 30.

Table 2

Logistic regression models of poor SRH by education, adjusted for material, behavioral and psychosocial factors (n = 3246)

Model	Middle educational group			Low educational group
	β	95% CI	Reduction (%)	β	95% CI	Reduction (%)
Model 1a	0.793	0.595–0.992		1.426	1.132–1.721
Model 2 (Model 1 + material factors)	0.651	0.451–0.850	17.98	1.097	0.801–1.393	23.11
Model 3 (Model 1 + psychosocial factors)	0.695	0.497–0.893	12.39	1.180	0.886–1.474	17.29
Model 4 (Model 1 + behavioral factors)	0.691	0.492–0.890	12.86	1.161	0.866–1.457	18.57
Model 5 (Model 1 + material factors + psychosocial factors)	0.602	0.403–0.801	24.08	0.969	0.672–1.265	32.09
Model 6 (Model 1 + psychosocial factors + behavioral factors)	0.623	0.424–0.822	21.46	0.988	0.692–1.284	30.73
Model 7 (Model 1 + material factors + behavioral factors)	0.583	0.383–0.783	26.54	0.913	0.616–1.210	35.98
Model 8 (Model 1 + material + behavioral + psychosocial factors)	0.549	0.350–0.748	30.84	0.822	0.524–1.119	42.41
Contribution of single explanatory variables in Model 8
Housing tenure	0.791	0.593–0.990	0.28	1.420	1.125–1.716	0.43
Some financial problems	0.751	0.552–0.951	5.31	1.358	1.062–1.654	4.79
Great financial problems	0.780	0.582–0.979	1.67	1.398	1.103–1.693	2.00
Material deprivation	0.774	0.575–0.973	2.43	1.387	1.091–1.683	2.76
Access to healthcare	0.790	0.592–0.989	0.39	1.402	1.107–1.696	1.74
Type of health insurance	0.775	0.577–0.974	2.27	1.368	1.072–1.663	4.10
Total material factors	0.695	0.496–0.895	12.34	1.201	0.903–1.499	15.83
Marital status	0.793	0.595–0.992	0.03	1.425	1.130–1.720	0.10
Social participation	0.764	0.565–0.962	3.76	1.320	1.025–1.616	7.44
Control beliefs	0.749	0.551–0.948	5.56	1.352	1.058–1.646	5.20
Total psychosocial factors	0.719	0.521–0.918	9.35	1.245	0.949–1.540	12.75
Smoking	0.786	0.587–0.985	0.90	1.416	1.121–1.712	0.72
Physical activity	0.764	0.566–0.963	3.68	1.336	1.041–1.631	6.36
BMI ≥ 30	0.757	0.559–0.956	4.56	1.330	1.035–1.625	6.76
Total behavioral factors	0.7208	0.522–0.920	9.15	1.2291	0.933–1.526	13.83

^aβ in Model 1 refers to the difference in SRH (controlled for sex, age, employment status and chronic diseases) between the middle resp. low educational group as compared to the high educational group.

By comparing the models with two groups of variables with the models with each one group of explanatory factors it became obvious that the independent contribution of material factors was higher than that of behavioral and psychosocial factors.

When adjusting for material and psychosocial factors simultaneously (Model 5), the independent effect of psychosocial factors net of material factors contributed by 6% in the middle educational group and by 9% in the low educational group whereas the independent effect of material factors was higher (middle educational group: 12%, low educational group: 15%). Furthermore, material factors were working through psychosocial factors with an indirect contribution of 6% in the middle educational group and 8% in the low educational group.

When comparing the relative contribution of behavioral and psychosocial factors (Model 6), the independent effect of psychosocial factors net of behavioral amounted to 9% in the middle educational group and to 12% in the low educational group. Additionally, psychosocial factors had an indirect effect through behavioral factors (middle educational group: 4%, low educational group: 5%). The independent effect of behavioral factors was comparable to the independent effect of psychosocial factors (middle educational group: 9%, low educational group: 13%).

When material and behavioral factors were included simultaneously (Model 7), the independent effect of material factors amounted to 14% in the middle educational group and to 17% in the low educational group. Additionally, material factors were indirectly working through health behaviors with an additional effect of 4% in the middle educational group and 6% in the low educational group. The independent effect of behavioral factors net of material factors amounted to 9% in the middle educational group and 13% in the low educational group (Table 3).

Table 3

Independent and indirect effects of material, behavioral and psychosocial factors (n = 3246)

Model comparison	Middle educational group (%)	Low educational group (%)
Independent effect of psychosocial factors net of material factors: Model 5–Model 2	6.10	8.98
Independent effect of material factors net of psychosocial factors: Model 5–Model 3	11.70	14.80
Indirect effect of material factors via psychosocial factors: Model 5–ind. effect of material factors—ind. effect of psychosocial factors	6.29	8.31
Independent effect of psychosocial factors net of behavioral factors: Model 6–Model 4	8.60	12.16
Independent effect of behavioral factors net of psychosocial factors: Model 6–Model 3	9.07	13.44
Indirect effect of psychosocial factors via behavioral factors: Model 6–ind. effect of psychosocial factors—ind. effect of behavioral factors	3.79	5.13
Independent effect of material factors net of behavioral factors: Model 7–Model 4	13.67	17.40
Independent effect of behavioral factors net of material factors: Model 7–Model 2	8.55	12.87
Indirect effect of material factors via behavioral factors: Model 7–ind. effect of material factors—ind. effect of behavioral factors	4.31	5.70

Discussion

Main findings

The three explanatory pathways together contributed by 31% to the association between education and SRH in the middle educational group and by 42% in the low educational group. The higher contribution in the low educational group is due to the higher prevalence of health risks which hints at the fact that there are health risks especially relevant in the middle educational group that we did not consider.

Our analysis revealed that material factors were the most important as they exerted the largest independent effect and as they were additionally working indirectly through behavioral and psychosocial factors. The independent contribution of behavioral and psychosocial factors was much lower than suggested by the separate analyses.

We conclude that material factors are of major importance for explaining health inequalities as they most adequately reflect the life situation of the socially disadvantaged. In old age, the available material resources reflect to a certain extend the accumulation of (dis-)advantage over the life course so that their contribution to health inequalities is of particular relevance.³⁴

Interventions for the reduction of health inequalities in old age should thus focus on improving material living conditions. Such interventions should focus on all age groups as living conditions from prior life stages are important determinants of health in later life.^35–37

What is already known on this topic??

Studies with the population of working age have identified material, behavioral and psychosocial factors as key pathways for explaining health inequalities. According to a recent literature review,⁵ there are few studies that examine the contribution of all of the three explanatory pathways to the association between education and SRH. All of the existing studies focus on the population of working age and they highlight the importance of material factors as they are also working through behavioral and psychosocial factors.^11,12 The results are similar with regard to educational inequalities in other health indicators such as mortality¹⁴ or chronic diseases.¹¹

What does this study add?

This study is the first that provides insights in the underlying mechanisms of educational inequalities in SRH in old age by including material, behavioral and psychosocial factors. In general, our results are consistent with studies on the population of working age which show that material factors are the most important in the association between education and SRH due to their relatively large independent effect and their indirect effect through behavioral and psychosocial factors.^10,11 However, the relative contribution of the explanatory pathways differs between the studies which is probably due to the amount of variables included to represent one explanatory pathway.⁵

Furthermore, study results might depend on the age group under study as the relevance of single variables could differ between people in middle adulthood as compared to the older population. Unfortunately, most of the existing studies with the population of working age do not provide results on the contribution of single variables. In our analysis, the most relevant mediators were financial problems, type of health insurance (especially in the low educational group), lacking social participation, insufficient control beliefs, lacking physical activity and a BMI ≥ 30. The reasons for the importance of these variables are less clear and warrant further research.

The existing studies on the underlying mechanisms of educational inequalities in health with a special focus on the older population are of limited comparability due to differences in the statistical methods, health indicators and the included explanatory pathways. Studies that take into account material, behavioral and psychosocial factors rely on other health indicators than SRH and other measures of social inequality than education. For instance, Ploubidis et al.²⁵ showed that material and behavioral factors are most important in explaining socioeconomic inequalities (measured by an index of occupation, education, income and wealth) in somatic health, depression and well-being, whereas psychosocial factors exert most of their influence on depression and well-being. In contrast, Stolz et al.³⁸ showed that the impact of income and poverty on frailty was mediated by material and especially by psychosocial factors, whereas the contribution of behavioral factors was only marginal. Thus, further investigations should examine if the contribution of material, behavioral and psychosocial pathways to health inequalities in old age differs depending on the indicators of health and social inequality under study. It might be reasonable that the importance of psychosocial variables is higher in mental health problems, whereas material and behavioral factors might be more relevant for inequalities in somatic health. Furthermore, different national contexts and age groups should be taken into account as the relevance of health influences might differ between countries and life stages.

Limitations of this study

First, this study has a cross-sectional design so that no causal conclusions can be drawn. Unfortunately, several of our explanatory variables were not available in prior study waves so that we could not rely on longitudinal data. Second, we did not consider lifetime exposure to health relevant factors, although health in later life is related to living conditions in earlier adulthood³⁷ and even childhood.^35,36 Besides influences from prior life stages, there may be health risks especially relevant for the old aged such as care-giving for relatives or an inadequate intake of prescribed drugs. Such age-specific health influences are neither theoretically conceptualized nor empirically examined in health inequalities research, so that further studies are needed. Third, several studies have shown that the relevance of certain determinants for SRH differs between women and men^39,40 but due to small sample sizes separate analyses for women and men were not possible.

Acknowledgements

This article uses data from SHARE Waves 1, 2, 4 and 5 (DOIs:10.6103/SHARE.w1.600, 10.6103/SHARE.w2.600, 10.6103/SHARE.w4.600, 10.6103/SHARE.w5.600), see Börsch-Supan et al.¹⁵ for methodological details.

Funding

The SHARE data collection has been primarily funded by the European Commission through FP5 (QLK6-CT-2001-00360), FP6 (SHARE-I3: RII-CT-2006-062193, COMPARE: CIT5-CT-2005-028857, SHARELIFE: CIT4-CT-2006-028812) and FP7 (SHARE-PREP: N°211909, SHARE-LEAP: N°227822, SHARE M4: N°261982). Additional funding from the German Ministry of Education and Research, the Max Planck Society for the Advancement of Science, the U.S. National Institute on Aging (U01_AG09740-13S2, P01_AG005842, P01_AG08291, P30_AG12815, R21_AG025169, Y1-AG-4553-01, IAG_BSR06-11, OGHA_04-064, HHSN271201300071C) and from various national funding sources is gratefully acknowledged (see www.share-project.org).