The importance of age, sex and place in understanding socioeconomic inequalities in allostatic load: Evidence from the Scottish Health Survey (2008–2011)

Throughout the world individuals with lower socioeconomic position (SEP) experience lower levels of health and reduced life expectancy [1]. This trend is continuous and exhibits no threshold limit; even those of relatively high SEP experience reduced levels of health in comparison to those with a slightly greater SEP [2], known as the social gradient in health. Socioeconomic differences exist for almost every major contemporary and historical cause of mortality and morbidity, highly suggestive of a common physiological pathway [3, 4]. Mechanistically, chronic and/or high levels of stress evoke over-activation of the hypothalamic-pituitary-adrenal (HPA) and sympathetic-adrenal-medullary (SAM) axes and prolong the secretion of neuroendocrine measures (e.g. cortisol) as primary mediators, before later effects on secondary (e.g. blood pressure) and tertiary (e.g. cellular ageing) outcomes [5]. Sustained exposure results in dysregulation and desensitisation to the stress response (reviewed by McEwen, 2002 [6]). These exposures manifest in a reduced capacity to adapt [7], inducing a recalibration process to restore the normative environment, known as allostasis [8]. Although allostasis is beneficial in the short-term, chronic, high levels result in cumulative damage, known as allostatic load [9, 10]. The measure used most often is a simple count score of the number of biomarkers that fall into the highest quartile of risk of the sample distribution [11]. Additional common constructs include using clinical cut-offs or z-scores, although there is no agreed ‘preferred’ construct currently [12]. High allostatic load, which is typically measured using a number of the body’s physiological biomarkers (primarily constructed using secondary outcomes of the stress pathway, as described above) as indicators of regulation/dysregulation [13], is predictive of an increased risk of a number of adverse medical conditions [14‐16], and all-cause mortality [13]. Subsequent reduction in allostatic load is also associated with a reduction in risk of dying [17].

A growing literature has identified a consistent association between lower SEP and higher allostatic load across multiple cohorts and population samples [18‐24]. However, less attention has been focused on how this association differs within populations given different individual characteristics such as age, sex and place (urbanity and geographic location. Understanding how the relationship between SEP and allostatic load differs according to such individual characteristics across the lifecourse will be an important step in ensuring that the allostatic load concept can be used effectively across the entire population and in developing policies and programmes to reduce inequalities that are effective across these characteristics. The aim of this study was to acquire a better understanding of the social patterning of allostatic load given these demographic/contextual factors using data from a large, nationally representative health survey, an important consideration in the development of more effective and targeted interventions to equitably reduce allostatic load and health inequalities in the general population. We hypothesised that allostatic load would be higher in those with lower SEP, with these inequalities widening with age (given the accumulated nature of the SEP-allostatic load relationship [22]) and then narrowing at older ages (matching results for other health outcomes). We did not predict that the SEP-allostatic load relationship would change between men and women or by urbanity. With geographic location, we hypothesised that Greater Glasgow and Clyde would have the widest inequalities, given the city’s well-publicised poor health record in relation to the rest of Scotland, the UK and Europe [25‐28].

Of the 3,580 respondents who consented to having bloods and anthropometric measures taken, 1,918 had results for all 10 biomarkers and full data for sex, age, place and SEP and were included in the analysis, although 81 were excluded given high CRP values (>10 mg/dL) (Fig. 1). This sample was 44.3 % male, with ages ranging from 18–94 (mean age = 48.7; SD = 15.8) (Table 1). Considerable socioeconomic diversity was evident; one third of the sample attained at least a University degree, whilst 16.5 % had no qualifications. Mean allostatic load was 2.52 and scores ranged from 0–9 (Table 2). The distribution of scores had a positive skew, with 55 % of individuals having an allostatic load score ≤2. No respondents had an allostatic load of 10.

Increasing age was associated with higher allostatic load (b = 0.051, 95 % CI 0.036, 0.067, p < 0.001), representing a 1-unit increase in allostatic load with every 19.4 years. There was no evidence of a non-linear age effect (using an age² and an age³ transformation) (age²: b = −0.080, 95 % CI −0.286, 0.446, p = 0.668; age³: b = −0.001, 95 % CI −0.001, 0.001, p = 0.185). Sex differences were present, with female respondents significantly more likely to have lower allostatic load than males (b = −0.368, 95 % CI −0.507, −0.169, p < 0.001). Urban/rural place of residence was also associated with allostatic load, with residents in all places excluding small remote towns and accessible rural areas showing a higher allostatic load when in compared to primary cities. Compared to Lothian, only respondents from Ayrshire & Arran (b = 0.495, 95 % CI 0.058, 0.931, p = 0.026) and Fife (b = 0.503, 95 % CI 0.061, 0.946, p = 0.026) had significantly higher allostatic load scores.

There was a significant inequality in allostatic load by educational attainment, with those with higher SEP having significantly lower allostatic load scores compared to those with lower SEP (b = −0.959, 95 % CI −1.179, −0.739 p < 0.001). This disparity is the equivalent of a 18.6-year age difference in physiological health, based on the results of the age-allostatic load association described above. This association remained after adjustment for age, but the magnitude of effect was reduced to the equivalent of a 11.6-year difference in physiological burden (b = −0.593, 95 % CI −0.798, −0.387, p < 0.001). Adjusting for sex, urban/rural location and geographical location did not attenuate the associations (Table 3). Next we tested for any evidence of effect moderation by any of these variables (using interaction terms with SEP). However, for each of the variables, there was no significant interaction effect (p > 0.05), with higher allostatic load scores in those with lower SEP consistent across all analyses (Table 4). Although there was no effect moderation, we did examine the SEP inequality by each factor in stratified analyses to visualise these associations in closer detail. For both men and women, there was the expected social gradient in allostatic load, with higher allostatic load with lower SEP (Fig. 2, Table 4). For age, there were significantly higher allostatic load scores in those with lower SEP from ages 35 and above, but this pattern was not seen between 18 and 34 years of age. There was also evidence that this inequality had begun to narrow at older ages (75+) (Fig. 3, Table 5). For urban/rural, the social patterning of allostatic load was only significant in the primary cities, urban environments and accessible rural environments (Fig. 4, Table 5). Finally for geographical location, allostatic load was only socially patterned and statistically significant in 4 of the 10 locations, with the largest inequality seen in Lothian (which includes the city of Edinburgh) (Fig. 5, Table 5).

This study has explored in further detail the relationship between lower SEP and higher allostatic load across the lifecourse using a large, nationally representative population sample, finding that inequalities widen with age, before narrowing in later life through stratified analysis (although there was no statistically significant interaction effect between age and SEP). The social patterning of allostatic load appears to be consistent with sex, while there are more mixed findings with regards place (urbanity or geographical location).

While other studies have typically adjust for age or sex, the small number of studies that have considered the more nuanced patterns of inequalities in allostatic load have also shown some similar results. Geronimus (2006) showed inequalities in higher risk allostatic load (using poverty:income ratio) increase with age [18]. This study used data from the US National Health and Nutrition Examination Survey (NHANES IV, 1999–2002), but did not look at ages above 64 and focused on a binary measure considering the risk of having an allostatic load score of 4 or more. Additional analyses of the NHANES study (using NHANES III) has shown that these inequalities in allostatic load occur across adult life (ages 20-80+), also weakening with older age [47]. In analyses of socioeconomic inequalities using lifecourse measures of SEP (with allostatic load measured at age 43), Gustafsson et al. (2010) found that the inequalities were similar in men and women (although had potentially different mediators) [20]. Analyses of place effects on the relationship between SEP and allostatic load have typically focused on using neighbourhood measures of SEP (showing a potentially independent effect over and above individual SEP) [38]. However there is no other known evidence for place effects in terms of urbanity or geographic location.

Widening socioeconomic inequalities with age have been seen with various morbidities and mortality [48‐50], including a typical pattern of these inequalities narrowing at older ages [51‐54]. However, evidence from longitudinal studies has begun to find that these narrowing inequalities may be more of an artefact given survival bias (those with lowest SEP have the poorest health and therefore become underrepresented at older ages) and through using less proximal SEP measures [55]. The narrow inequalities at younger ages are also theorised to be due to the accumulated nature of disadvantage (and allostatic load) requiring time to show physiological and health consequences, with various potential mediators [22, 56]. Other studies have also found that the effect of SEP on health, rather than physiological dysregulation per se, increases with younger birth cohorts [57]. It has been suggested that this is the result of the changing contexts for the SEP-health association. For example, life expectancy has increased with younger cohorts; the differences in the meaning of different SEP measures have changed for different cohorts (e.g. the growing importance of education in people’s lives with younger birth cohorts); and the pattern of diseases has also altered across cohorts. While it is not possible in this study to unravel age and cohort effects, the possibility that there are cohort effects, rather than age effects present, may be a factor to consider. Given that our allostatic load measure contained pre- and post-clinical cut-offs, it is also telling that we witness the same patterns as seen with other morbidities and mortalities. However, the lack of statistical significance in the effect moderation analysis (compared to stratified analysis) suggests that the inequality seen at different ages is not as strong as perhaps seen with other more clinical outcomes. However, there remains a case for measures such as allostatic load to be considered as pre-clinical markers of ill health, and potential alternative outcomes/predictors when assessing health inequalities and the effects of interventions/programmes to improve health and reduce inequalities. What remains less known is if the allostatic load concept accurately predicts disease/death across the lifecourse (there is evidence that it does at older ages [13, 58, 59]) and if it offers any additional power over using single biomarkers or biomarkers across single physiological systems.

In this study we also considered the importance of place in the social patterning of allostatic load. There were no significant interaction effects, with allostatic load typically higher in lower versus higher SEP groups across all ten areas and urban/rural settings (although not always statistically significant). In terms of urbanity, it was not simply a case that the inequalities were present in the larger, more urban areas. Smaller sample sizes were present in the areas that were statistically non-significant, reducing the statistical power available. A similar inconsistent pattern was also seen with geographical location, where it was not the case that the areas with larger cities displayed the widest and significant inequalities. While the areas housing the major cities of Glasgow and Edinburgh (Greater Glasgow and Clyde and Lothian, respectively) did show significant inequalities, Grampian (Aberdeen city) and Tayside (Dundee city) did not. We did not consider area-level deprivation in this study as we viewed it a separate research question (does area-level SEP remain after adjusting for individual SEP), but it is possible that these larger area measures are less relevant to people’s health compared to more local-level measures (such as those derived from postcodes). In those studies that have considered area-level deprivation, it does appear there is an independent effect over and above individual SEP [21, 38].

As a sensitivity measure, we did consider the clinical cut-off operationalisation of allostatic load, finding that it mirrored the patterns seen in the ‘quartile’ distribution score. While arguments remain regarding the best operationalisation of allostatic load, the use of the distribution score measure is perhaps the most common [11]. This score is typically favoured given it matches the original allostatic load concept of a ‘sub-clinical dysregulation state’ more closely than a clinical-cut off score [11, 60]. However, it makes cross-sample comparisons more difficult as the cut-off points vary from one population to another. Clinical cut-offs do offer greater comparability though, important when considering the results of allostatic load studies (and perhaps more importantly, testing interventions involving allostatic load) in different populations and settings. However, it is relatively easy to measure allostatic load in more than one way to improve the reliability and comparability of results.

This study used a large, nationally representative sample of the Scottish population, a country with high-profile health inequalities compared to other regions of the UK and Europe [26, 27, 61]. This study also builds on the limited evidence that specifically focuses on how the socioeconomic patterning of allostatic load may differ given other demographic/contextual factors. Despite these strengths, we must note some potential limitations. First, the sample was cross-sectional in nature and would be improved upon by including a longitudinal component, especially in the progression of allostatic load over time. In addition, we have only included one population sample in this study. This study of the Scottish population may not be generalizable to other populations and, given the high allostatic load cut-offs (sometimes above clinical cut-offs), may also further highlight Scotland’s poor health record [62]. Third, although the overall sample size of the study was large, within some groupings small sample sizes were inevitable (e.g. small areas or locations). Looking at effect moderation using interaction terms revealed that the differences seen in the stratified analysis might be influenced by such small sample sizes (present in the youngest and oldest ages). A related issue is that of missing data, especially given the need for all 10 biomarkers being required to produce the allostatic load score. As well as reducing the sample size, there is the potential for the analysis sample to be substantially different to the original sample. However, comparisons between the whole ‘blood’ sample and the analysis sample for age, sex, urbanity, location and SEP revealed no substantive differences in frequencies (unpublished). The only category that perhaps showed underrepresentation was those age 75+ with lower SEP. The reasons for this were typically invalid/low blood samples or missing results from at least one of the anthropometric or blood measures. This may be relevant to the issues of survival bias discussed above with regards narrowing inequalities at these older ages. All analyses were weighted to be representative of the Scottish population using weights to correct for different selection probabilities and for non-response. However, while these weights make the ‘blood’ and ‘main’ sample comparable to the population in question, they cannot address issues of item missingness from respondents who did give samples.

Our allostatic load construct did not contain any markers from the hypothalamic pituitary adrenal (HPA) axis that forms part of the neuroendocrine system (stress response). The stress response is believed to play a key role in allostasis and subsequent allostatic load, with a cascade of events that starts with primary stress mediators, such as cortisol, before initial stress responses (‘primary effects’ such as rapid increases in blood pressure and sugars/fats that supply the body with extra energy) and then to secondary and tertiary outcomes (measured in our allostatic load model). These stress markers can be difficult to measure in large surveys where direct examination of the stress response (e.g. measuring cortisol) is difficult due to the circadian rhythms shown in these stress hormones (e.g. cortisol and DHEA-S) and the rapid sampling required in order to measure baseline versus activated levels (e.g. cortisol). This has made inclusion of these measures in large surveys less common than traditional clinical markers such as blood pressure or cholesterol. However, inclusion of measures such as cortisol or DHEA-S could improve the power of allostatic load as an earlier risk predictor for disease, but their exclusion does not invalidate this allostatic load construct as the subsequent outcomes of cortisol release are still being included. Finally there are potential limitations in the operationalisation of both allostatic load and SEP, and the subsequent interpretations that can be made (discussed above). We did perform sensitivity analysis with an alternative allostatic load construct and found no substantive differences. Although there remains no firm agreement on the best operationalisation of allostatic load, the quartile construct remains commonly used and relatively easy to interpret (compared to z-scores or more complex methods such as canonical correlation) [12, 63]. We did carry out sensitivity work using an alternative education cut-off (low SEP = no qualifications) to take into account changing educational attainment between the younger and older cohorts (not published). However, this measure produced a relatively small lower SEP group (emphasised in some stratified analyses) and meant the higher SEP group contained a broader grouping of educational attainment levels that may not best represent what ‘higher’ SEP is attempting to measure. Hence, the magnitude of the inequality and the statistical significance was reduced, despite demonstrating the same patterns. It could be possible to use other measures of SEP in addition to education, although in this sample this would have resulted in greater levels of missing data (e.g. using occupational-based social class, which also suffers from different meanings at different stages of the lifecourse). Education is “an indicator of SEP at the onset of adult life that sets an individual’s socioeconomic trajectory for the future” [64]. If follows that effects of SEP on cumulative physiological burden (such as allostatic load) may take many years to accumulate, so education may provide a more robust marker of SEP through, in particular, early adult life and middle age, compared to measures simply taken at the time of the sampling (e.g. housing status or occupational status/social class).

This study has found that socioeconomic inequalities (measured using education) in allostatic load are present in a large, nationally representative study. These inequalities widen with age, before narrowing in later life, although there was no significant interaction effect between age and SEP. These findings are important in the case being made for allostatic load to be considered a pre-clinical risk marker that may help explain some of the mechanisms linking SEP and health. However, individuals’ demographic and contextual circumstances can play a role in the precise nature of this socioeconomic patterning and thereby alter the assumptions and effects of subsequent interventions to reduce such health inequalities. In addition, this evidence is important in our understanding of how socioeconomic factors can influence health across the lifecourse, not just in individual physiological systems, but throughout the body.