Explanations of socioeconomic differences in changes in physical function in older adults: results from the Longitudinal Aging Study Amsterdam

Low socioeconomic status (SES) is related to poor physical function and the development of physical disabilities in older adults [1–6]. Disability, usually defined as the inability to carry out the usual tasks of daily living, is established as a powerful measure of health status in old age [7]. In addition, physical disability in older persons is related to many diseases and has been shown to be a predictor of mortality in older adults. Therefore, it is important to understand what factors to intervene upon in order to reduce physical disabilities and the SES differences therein.

Several factors, including biomedical, behavioral, and psychosocial factors, may be important in explaining these SES differences in physical function. Biomedical factors, such as cardiovascular disease, stroke, and diabetes, are related to both low SES and adverse functional outcomes [8–12]. A recent study shows that biomedical factors, including a wide range of diseases and biological risk factors, explained part of the SES differences in mobility limitation incidence in older adults [5]. Low SES is also related to many adverse behavioral factors, such as smoking, excessive alcohol consumption, and decreasing physical activity, that in turn are related to poor functional outcomes [13–16]. Finally, psychosocial factors may be important in explaining SES differences in poor functional outcomes because people with low SES generally have fewer psychosocial resources than people with high SES [17–19]. Psychosocial factors, such as control beliefs and social support, are also linked to ill health and poor functional outcomes [20–22].

The present study examines the association between SES and longitudinal changes in physical function in a representative sample of Dutch older adults, who participated in the Longitudinal Aging Study Amsterdam. Although SES differences in health occur in different age groups, these differences have been less examined in older age groups than in younger age groups. Some studies report that SES differences attenuate in old age [23–26], although evidence is contradictory [27]. The present study gives the opportunity to examine whether the effect of SES on physical function is different for younger-old (aged 55–70 years) and older-old (aged 70–85 years) adults and how these differences develop over nine years of follow-up. The contribution of a number of potential explanatory factors to explain SES differences in physical function has been studied before [5, 13, 28, 29]. In this study, however, we will integrate several explanatory factors, including diseases, behavioral, and psychosocial factors in order to determine the relative contribution of these factors to the explanation of SES inequalities in changes in physical function.

Table 3 shows the results of the multilevel analyses for the group younger than 70 years and table 4 for those 70 years and older. The tables present the intercept and slope of the variable time for each model. The intercept is the parameter estimate (baseline score) of the middle and low SES group compared to the highest SES group. The slope is the coefficient of the SES*longitudinal time interaction term, indicating the effect of longitudinal time on physical function. The fist model is graphically presented in figure 1. People younger than 70 years old in the lowest SES group not only had a significantly worse physical function score at baseline and follow-up, they were also declining significantly more over time compared to people in the highest SES group (Table 3: Model 1, Figure 1). Baseline differences between the lowest and the highest education group were larger in the oldest age group. SES differences did, however, not further widen over time in people of 70 years and older (Table 4: Model 1, figure 1).

	Model 1 * sociodemographics				Model 2 † model 1+ diseases				Model 3 ‡ model 1+ behavioral factors				Model 4 § model 1+ psychosocial factors				Model 5 ¶ all variables previous models
	Intercept	p	Slope	p	Intercept	p	Slope	p	Intercept	p	Slope	p	Intercept	p	Slope	p	Intercept	p	Slope	p
Education
High	Ref.
Medium	-0.16	0.45	-0.04	0.59	-0.08	0.69	-0.04	0.62	-0.07	0.73	-0.04	0.59	-0.14	0.49	-0.04	0.59	0.01	0.98	-0.04	0.62
Low	-0.43	0.05	-0.24	<0.01	-0.24	0.26	-0.24	<0.01	-0.07	0.74	-0.24	<0.01	-0.35	0.11	-0.24	<0.01	0.07	0.73	-0.24	<0.01
Income
High	Ref.
Medium	-0.22	0.26	-0.17	0.02	-0.05	0.79	-0.17	0.03	0.03	0.87	-0.17	0.02	-0.15	0.44	-0.17	0.02	0.18	0.35	-0.16	0.03
Low	-0.83	<0.01	-0.36	<0.01	-0.62	0.01	-0.36	<0.01	-0.46	0.06	-0.36	<0.01	-0.56	0.02	-0.36	<0.01	-0.16	0.50	-0.35	<0.01

* Model 1: adjusted for age, sex, and interactions between age and longitudinal time, and sex and longitudinal time

† Model 2: adjusted for age, sex, interactions between age and longitudinal time and sex and longitudinal time, cardiac disease, peripheral arterial disease, stroke, diabetes mellitus, lung disease, cancer, arthritis

‡ Model 3: adjusted for age, sex, interactions between age and longitudinal time and sex and longitudinal time, smoking, drinking, BMI, and physical activity

§Model 4: adjusted for age, sex, interactions between age and longitudinal time and sex and longitudinal time, partner status, network size, instrumental social support, emotional social support, mastery, and self-efficacy

¶Model 5: all variables of previous models

	Model 1 * sociodemographics				Model 2 † model 1+ diseases				Model 3 ‡ model 1+ behavioral factors				Model 4 § model 1+ psychosocial factors				Model 5 ¶ all variables previous models
	Intercept	p	Slope	p	Intercept	p	Slope	p	Intercept	p	Slope	p	Intercept	p	Slope	p	Intercept	p	Slope	p
Education
High	Ref.
Medium	-0.74	0.12	0.16	0.39	-0.66	0.15	0.15	0.41	-0.76	0.10	0.16	0.39	-0.69	0.14	0.16	0.39	-0.68	0.12	0.15	0.41
Low	-1.36	<0.01	0.01	0.94	-1.17	<0.01	-0.00	0.99	-1.16	<0.01	0.01	0.96	-1.04	0.02	0.01	0.95	-0.83	0.04	-0.01	0.97
Income
High	Ref.
Medium	0.30	0.52	-0.26	0.16	0.29	0.52	-0.25	0.17	0.18	0.68	-0.26	0.16	0.41	0.38	-0.26	0.15	0.22	0.60	-0.25	0.17
Low	-0.86	0.06	-0.10	0.57	-0.67	0.12	-0.13	0.47	-0.80	0.06	-0.10	0.58	-0.59	0.19	-0.10	0.56	-0.47	0.26	-0.12	0.48

* Model 1: adjusted for age, sex, and interactions between age and longitudinal time, and sex and longitudinal time

† Model 2: adjusted for age, sex, interactions between age and longitudinal time and sex and longitudinal time, cardiac disease, peripheral arterial disease, stroke, diabetes mellitus, lung disease, cancer, arthritis

‡ Model 3: adjusted for age, sex, interactions between age and longitudinal time and sex and longitudinal time, smoking, drinking, BMI, and physical activity

§Model 4: adjusted for age, sex, interactions between age and longitudinal time and sex and longitudinal time, partner status, network size, instrumental social support, emotional social support, mastery, and self-efficacy

¶Model 5: all variables of previous models

In model 2, 3, and 4 diseases, behavioral factors, and psychosocial factors were introduced separately into the first model. In the youngest age group (Table 3), the largest reduction in intercepts was found in model 3, due to behavioral factors, for both education and income. A percentage reduction in intercept from model 1 was computed by (intercept _{model 1} - intercept _{model 3})/(intercept _{model 1}). The average reduction due to behavioral factors in the lowest SES groups was 65%. BMI and physical activity most reduced the regression coefficients; smoking and drinking did not further decrease the regression coefficients (not tabulated). The average reduction in the lowest SES groups due to diseases was 35% and 26% due to psychosocial factors. None of the explanatory factors could explain the SES-differential decline. The SES differences in baseline physical function were completely explained when all factors of the previous models were added to the fifth model. The differences in slope, however, remained statistically significant.

In the oldest age group, the large differences in baseline physical function between the lowest and the highest SES group could not be fully explained by diseases, behavioral or psychosocial factors (Table 4). The largest reduction in intercept differences between the lowest and the highest SES group in this age group was found in model 4, due to psychosocial factors (on average 28%). In additional analyses, we determined which psychosocial factors were most important in reducing the regression coefficients. None of the psychosocial factors, however, had a large individual contribution. In the full model (Model 5), the intercept of low education still was significantly different from that of higher education.

Because the significant differences in slope between SES groups, that were found in the lowest age group, could not be explained by any of the variables that were added in model 2 to 4, we determined, in additional analyses, whether interactions between each explanatory factor and longitudinal time would reduce the slope differences. However, the coefficients were difficult to interpret which may be due to the complexity and potential over-fitting of the models and, therefore, did not add to the explanation of the slope differences.

In an older population, low SES was associated with poor physical function during nine years of follow-up. In contrast to people 70 years and older, where SES differentials in physical function did not further increase over time, SES differentials in physical function significantly increased over nine years in people who were younger than 70 years. Behavioral factors, mainly BMI and physical activity, explained a large part of the SES differentials in baseline physical function in the youngest age group. Behavioral factors could, however, not explain the SES-differential decline in physical function. In people 70 years and older, psychosocial factors reduced the SES differences most.

An important strength of our study is that we could examine SES differences in physical function in both younger-old and older-old adults. Furthermore, we were able to incorporate a wide range of explanatory factors in the explanation of SES differences in physical function in both age groups. Relatively few studies in older people have investigated the relationship between SES and functional outcomes [1, 2, 4, 5]. Studies have found that health outcomes were weaker in people aged 65 or older compared to younger people. A recent study in women showed that SES inequalities in physical health may attenuate by the age of 70 [24]. This latter study also found a strong cross-sectional association between SES and physical function, but no association with a decline in physical function [24]. The present study shows that even in old age, SES differences in physical function were still present. In people of 70 years and older, SES differences in physical function did not increase over time, but still existed after nine years of follow-up. The lowest SES group probably consists of "healthy survivors" because of selection prior to baseline and attrition due to mortality during the follow-up. It may be due to this healthy survivor effect that SES differentials in physical function did not widen over time in this age group. A recent study showed that mortality selection and cohort effects could not fully explain shrinking educational disparities in functional health in old age and that further research is necessary to explore explanation for diminishing SES differences in health [39].

In the youngest age group, a high percentage had no physical function problems at baseline (not tabulated). The decline in physical function that was found in this age group was mainly due to the onset of physical function problems rather than to a further functional decline related to an already existing disability. This may underlie the age-specificity that was found in both the pattern and the explanation of SES differences in physical function. Behavioral factors, in this study particularly high BMI and low physical activity, may be important in predicting the onset of physical function problems in the youngest age group. Adverse psychosocial factors, such as low social support and low self-efficacy, may cause a further decline in physical function in the oldest age group. Existing physical function problems may negatively influence someone's psychosocial profile, such as lower self-efficacy, which may further accelerate the decline in physical function as in a vicious circle [40]. Such pathways and reciprocal associations need further detailed examination.

In this study, a wide range of different explanatory factors was considered in order to get more insight into the explanation of SES inequalities in physical function. A few drawbacks, however, have to be considered. First, the SES differential decline that was found in the youngest age group could not be explained. This suggests that other factors that were not measured in this study may have caused the widening over time in this age group. Second, the contribution of the explanatory factors in reducing the SES differences and in particular the SES-differential decline in physical function could have been larger if these factors had also been considered longitudinally. However, there is evidence that behavioral factors, mainly smoking and alcohol consumption remain rather stable over time; physical activity showed greater variability over time. These changes in health behavior were, however, not related to SES [41]. Psychosocial factors, such as network size and social support, probably also remained rather stable during the follow-up [35, 42]. The prevalence of diseases, however, was more likely to have changed during the nine years of follow-up which may have led to an underestimation of the contribution of these factors. Third, we assumed that the three groups of explanatory factors had an independent contribution to the explanation of SES differences in physical function. The mechanism are probably interrelated, indicating that some mechanisms work through others rather than work independently from each other. This may have had consequences for the estimation of the exact contribution of each group of explanatory factors in the explanation of SES differences in physical function. However, the contribution of diseases, behavioral, and psychosocial factors together is not affected by whether these factors are interrelated or not. Fourth, missing values for psychosocial factors were replaced by group means, which could have led to an attenuation of the effect of these factors in explaining the SES differentials. In additional analyses, however, in which subjects with missing values on psychosocial factors were excluded, the contribution of these factors in the explanation of SES differences in physical function was very similar. Because of a large number of missing data on behavioral factors, a missing category was created and included in the main analyses. The contribution of behavioral factors was also very similar when we excluded subjects with missing data. Finally, we must acknowledge that, had more objective information on diseases been available, the reduction in the strength of the SES effect might have been larger.

Loss to follow-up was a limitation of this study. Attrition was mainly due to mortality (74% of the total loss to follow-up). People who were lost to follow-up for other reasons had worse physical function scores at baseline, compared to those who remained in the study (data not shown). In addition, those who were lost to follow-up also had a significantly lower SES compared to our study population. The association between SES and physical function may therefore have been underestimated. Whether the relative contribution of explanatory variables was equally underestimated remains unknown.

This study contributes to the understanding of the mechanisms underlying the association between SES and physical function in older adults. In order to reduce physical disabilities in older adults and in particular the socioeconomic differences therein, it may be helpful to improve health-related behaviors in younger-old adults and to intervene upon psychosocial characteristics in older-old adults. Further study is needed to confirm the age-specificity that was found in both the pattern and the explanation of socioeconomic differences in function in older adults.