Differential vulnerability and susceptibility: how to make use of recent development in our understanding of mediation and interaction to tackle health inequalities

Abstract

This paper discusses the concepts of vulnerability and susceptibility and their relevance for understanding and tackling health inequalities. Tackling socioeconomic inequalities in health is based on an understanding of how an individual’s social position influences disease risk. Conceptually, there are two possible mechanisms (not mutually exclusive): there is either some cause(s) of disease that are unevenly distributed across socioeconomic groups (differential exposure) or the effect of some cause(s) of disease differs across groups (differential effect). Since differential vulnerability and susceptibility are often used to denote the latter, we discuss these concepts and their current use and suggest an epidemiologically relevant distinction. The effect of social position can thus be mediated by causes that are unevenly distributed across social groups and/or interact with social position. Recent improvements in the methodology to estimate mediation and interaction have made it possible to calculate measures of relevance for setting targets and priorities in policy for health equity which include both mechanisms, i.e. equalize exposure or equalize effects. We finally discuss the importance of differential susceptibility and vulnerability for the choice of preventive strategies, including approaches that target high-risk individuals, whole populations and vulnerable groups.

Introduction

Tackling socioeconomic inequalities in health is based on an understanding of how an individual’s socioeconomic position (SEP) influences risk of disease and consequences of disease. Whereas the latter is strongly influenced by the health care system, the former is generated by exposure to causes of disease. There are here, in theory, two possible mechanisms (not mutually exclusive): there are either some cause(s) of disease that are unevenly distributed across socioeconomic groups (differential exposure); or the effect of some cause(s) differs across groups (differential effect, often called differential vulnerability or susceptibility, as discussed below). The first mechanism has been extensively studied, but the second has been subject to much less theoretical analysis and empirical research. Yet it might play an important role and have distinct implications for preventive health policies.

Let us illustrate with an example from alcohol epidemiology. It has been found that mortality rates from alcohol-related conditions in many countries are higher in more disadvantaged groups.¹ That is surprising, since high alcohol consumption in many of these countries is more prevalent in more advantaged groups.² The question is then whether there exists a differential effect of alcohol in different socioeconomic groups. Recently a Danish cohort study³ found evidence of such a differential effect. Whereas the rate difference among men for drinking >28 drinks per week compared with 0–14 drinks was 577 cases per 100 000 person-years of alcohol-related disease among the well-educated, the rate difference among those with shorter education was 866 per 100 000. This means that the differential effect can be expressed as the difference in effect: 866–577 = 289 [95% confidence interval (CI) = 123–457).³ Studies from Finland and England have similar findings.^4,5 Some studies on the role of differential exposure and differential effect have been carried out on cardiovascular and mental health outcomes,^6–12 and in particular in the past few years more papers have been published. But studies on differential effects across socioeconomic groups are still scarce, and the applied methodologies vary greatly. Assumptions are often made that effects, at least in relative terms, are the same across socioeconomic groups¹³ (which means that they might differ in absolute terms).

The potential relevance of differential effect for understanding health inequalities and for making the policies to tackle them was raised several years ago^14,15 and was pointed out by the World Health Organization (WHO) in the work on social determinants of health.¹⁶ There exists, however, in the literature a certain confusion about both conceptual issues and the methods used to estimate these mechanisms. The aim of this paper is therefore to contribute to the discussion of both the theoretical and the methodological issues involved, to suggest the use of new methodologies developed on how to estimate mediation and interaction, and to discuss the implications for public health policy.

Conceptual issues: vulnerability and susceptibility

Most of the current studies in social epidemiology that analyse differential effects use the term differential vulnerability. The concept of vulnerability is however also used by many other very different disciplines, ranging from bioethics to environmental science, psychology and genetics. Vulnerability was a key concept in an early version of the international bioethical guidelines for medical research, there used in the sense of lack of individual autonomy.¹⁷ Henk ten Have has recently proposed a more political analysis and a contextual definition, where humans are seen as vulnerable since they are dependent on other people. As we live in a context where resources and power are unequally distributed, some people become more dependent and vulnerable than others.¹⁸ Researchers within bioethics, environmental sciences and some areas of epidemiology have now adopted a functional definition of vulnerability that covers three dimensions: exposure to hazard; susceptibility i.e. effect of exposure; and capacity of response by coping and adaptability.^19–21 This definition has recently been used by the US Environmental Protection Agency in their analysis of health effects of climate change.²² Here vulnerability not only refers to individuals but also to communities and systems. From an epidemiological perspective, this definition is problematic since it tends to conflate exposure and susceptibility. Capacity of response is, however, important as a separate dimension, as it reflects power and resources to change exposures and to cope with, adapt to and recover from their effects. It raises—from an inequality perspective—interesting research questions as to what determines people’s options and capabilities to respond and act, and therefore has relevance for health promotion.^23,24 To avoid confusion it might therefore be preferable in epidemiology (as we will do in the rest of this paper) to use the term differential susceptibility when referring to differential effects. Differential vulnerability should then be used when it is relevant to include all three dimensions: exposure, susceptibility and capacity of response.

In epidemiology the definition of susceptibility is closely linked to Rothman’s²⁵ sufficient-component-cause model, where component causes complement each other to generate a sufficient cause. The effect of one cause depends on the exposure to other—interacting—component causes of the same disease. Susceptibility to the health effects of one specific cause can then be defined as the set of complementing genetic or environmental causes sufficient to make a person contract a disease after being exposed to the specific cause.²⁶ This definition provides an understanding of susceptibility as conditional causation and causal interaction. Whereas interaction is a clear empirical criterion for differential susceptibility, the estimation of mediation not only reflects differential exposure but will also be influenced by differential susceptibility (as it is often estimated by comparing the effects of exposure before and after adjusting for the potential mediator).

Measurement issues: interaction and mediation

For priority and target setting in policies aiming at tackling health inequalities, different estimates are relevant. It is important to be able to estimate how much of the effect of SEP on health would be removed if a mediating exposure is removed—what has been called the ‘proportion eliminated’²⁷—or if the social distribution of the mediator is changed. But it might also be important to estimate how much the inequality would be reduced if an interaction between socioeconomic position (SEP) and the mediator is removed, for example by eliminating another interacting mediator. Achieving unbiased estimates of mediated (indirect) effects, direct effects and effects due to interaction between SEP and mediators has, however, turned out to be difficult. It is only now, 40 years after the first efforts in social epidemiology, that VanderWeele has presented an elegant solution on how to decompose the health effect of an exposure (e.g. SEP) into its components created by mediation and interaction. Four different pathways are involved, each representing a mechanistic alternative.^27,28

• SEP has a direct effect on disease even among those who are not exposed to the mediator (‘controlled direct effect’).

• The effect of SEP on disease is dependent on the exposure to the mediator and vice versa; the effect of the mediator is dependent on SEP, i.e. they interact, but SEP does not influence exposure to the mediator (‘reference interaction’).

• The effect of SEP is, as in (ii), dependent on exposure to the mediator (and vice versa), but here SEP has an influence on the exposure level of the mediator (‘mediated interaction’).

• The effect of SEP on disease is entirely mediated by differential exposure to the mediator (‘pure indirect effect’).

The health effect of SEP mediated by what we have called differential exposure to a mediating cause is expressed by the sum of components (iii) and (iv), and differential susceptibility is expressed by the sum of components (ii) and (iii) —i.e. ‘portion attributable to interaction’.²⁷ The portion eliminated by removing the mediator is the sum of (ii) + (iii) + (iv).

The statistical analysis of interaction still builds on some critical assumptions such as the functional relationship or dose-response relationship between exposure and disease risk.²⁹ The importance in mediation analysis of controlling not only for exposure-outcome confounding but also for mediator-outcome confounding has been emphasized earlier,²⁸ but the fact that many mediator-outcome confounders might be influenced by SEP might be less of a problem since the decomposition includes controlled direct effect and not natural direct and indirect effects.²⁸ A very simple calculation of the relative importance of differential exposure and differential susceptibility and the decomposition of effects is made in Box 1.

Interaction analysis demands much statistical power since it depends on the number of double-exposed cases. Interaction analysis is in addition very sensitive to misclassification of exposures, in particular when the misclassification of one exposure is dependent on the other. In social epidemiology it might not be unusual that a mediator is differentially misclassified across SEPs. The interaction effect will then often be underestimated.³⁰

Empirical examples and mechanisms

Social epidemiology

In social epidemiology the issue of differential susceptibility was raised already in the early 1970 s. Dohrenwend found in 1973 that differential exposure to stressful life events could only partly explain social inequalities in distress,³¹ and that the correlation between stressor and distress was stronger among lower status groups. Syme and Berkman³² noted in 1976 that the same social patterning was found for many (albeit not all) diseases with very different aetiology, and suggested the existence of a generalized susceptibility as an explanation. Kessler³³ and later Grzywacz et al.⁷ analysed more systematically both differential exposure and differential susceptibility to stressors. None of these early studies applied an understanding of susceptibility as causal interaction. That was later done by Hallqvist et al.,⁶ and recent studies have analysed departure from additivity as criterion for differential susceptibility,¹¹ and some of them have applied additive hazard models for survival analysis.^3,8,9 Many studies still compare relative risks across socioeconomic strata.^4,5,10,12,13

The findings on cardiovascular outcomes are heterogeneous. Some find a clear differential susceptibility to the effect of smoking, whereas findings for hypertension and body mass index (BMI) are mixed. The methodologies applied are, however, still very different, which might explain some of the heterogeneities. None has so far applied VanderWeele’s decomposition and, as a result, they cannot fully separate the effects of differential exposure and differential susceptibility.

Susceptibility at the molecular level

Individual variation in susceptibility to health effects of many exposures might often be genetically determined. If genotypes associated with diseases are unequally distributed across SEPs, they might have relevance for socioeconomically differential susceptibility. The relevance of this for health inequalities is however still unclear,³⁴ and the few population-based studies that exist have not shown any association between, for example, diabetes-related polymorphisms and SEP.³⁵ But even equally distributed genes are obviously of relevance if they interact with unequally distributed exposures.³⁶

The growing insights of epigenetics have, however, shifted the focus from gene sequence to gene expression. Environmental epigenetics has shown that a broad range of physical and social exposures may influence how genes are regulated and modify their influence on disease aetiology.³⁷ Studies have, for example, shown that early childhood SEP is associated with differential methylation of several gene promoter regions.^38,39 Even during adulthood, gene expression can be modified by SES in ways that influence inflammatory reactions of importance for susceptibility to causes of both chronic disorders and infections.⁴⁰ So even if disease-related genotypes are not unequally distributed across socioeconomic groups, epigenetically modified gene function might be so. This leads to the hypothesis that epigenetic changes might mediate the effects on health of SEP. An exposure-generated epigenetic change might also modify the effect of another exposure if its effect depends on the expressed gene.^37,41 The ability of a cell to respond to a specific exposure such as social stress may thus be dependent upon the underlying epigenetic state, i.e. whether the cell is methylated in the region of the gene involved in responding to stress.⁴¹ If that response is silenced, then the organism might not react appropriately to stress exposure, and the effect of repeated or long-term exposure might then cause allostatic load.⁴²

While allostasis and allostatic load might be both a cause and an effect of epigenetic changes, they might also be a mechanism in their own right of relevance for differential susceptibility. Allostasis refers to the multiple adaptive responses to stress, including neuroendocrine, autonomic, immunologic and metabolic mediators as well as health behaviours. These responses might initially be adaptive but, repeated over a long time, they might create allostatic load that in itself increases the susceptibility to further stressor exposure.^36,42 With allostatic load, the normal adaptive responses to stress are worn out or otherwise dysregulated. Increased susceptibility to stress then occurs, not as a result of interaction between different mediators, but as an interaction between earlier and later exposure to the same or similar stressors.

Vulnerability at the community level

Many exposures, such as environmental air pollution and climate change, infectious agents and social contexts, are characterized by being non-differential in the sense that everybody in the population are equally exposed. Their health consequences are, however, sometimes still very unequally distributed across communities.^22,43 The question of what makes communities vulnerable to environmental exposures has stimulated much research. Models of both Turner et al.²⁰ in the USA and Birkmann et al.²¹ in Europe apply the concept of vulnerability at the community level, covering the three dimensions: exposure, susceptibility and capability of response, including the options and ability to change exposure or susceptibility in the population. This aspect of capability was in focus in the United Nations Development Program (UNDP)’s annual Human Development Report in 2014, which focused on vulnerability.⁴⁴ The dimension of capability is, according to these models, what primarily makes vulnerability different from susceptibility in its policy relevance. Vulnerability has been operationalized into a mapping technology and applied in epidemiological studies that, for example, aim to understand why water-related and vector-borne diseases, such as Dengue fever, show a very unequal distribution between similar equally exposed areas.⁴⁵ Measures of vulnerability then include different items that represent each of the three dimensions, but interactions between them have not been studied. A similar approach has been suggested in studies of the recent, alarming (and so far poorly understood) case of geographical and social variations in susceptibility to the teratogenic effects of Zika virus. Cases of Congenital Zika Syndrome including microcephaly have accumulated in poor urban areas of North-Eastern Brazil, while cases of Zika virus infections are spread over most of Latin America.⁴⁶

Policy implications

The existence of differential susceptibility and vulnerability influences the choice of preventive strategies to tackle health inequalities. How different preventive programmes actually impact on health inequalities depends on at least four aspects:¹⁵ differential implementation, i.e. how programmes are implemented and reach different population groups; differential effectiveness in how an intervention influences exposure to risk factors in different population groups all reached by the same intervention; differential susceptibility, i.e. how a certain change in exposure levels translates into changing incidence of disease in different groups; and there might finally also be differential capability of how different groups actually can change exposures, and cope with them.

A key question in preventive policies is the balance between three options:^23,47 (i) the high-risk strategy of identifying and treating high-risk individuals; (ii) the population strategy, moving the whole distribution of exposure; and (iii) the ‘vulnerable population approach’, targeting population groups with high levels of vulnerability including at least one of the dimensions of exposure, susceptibility and capability.²⁴

The first option—the high-risk strategy—aims at identifying individuals with a high level of exposure and then treating them. If such identification is based on SCORE-charts⁴⁸ or similar instruments estimating total risk of a combination of often clustering and interacting risk factors, it can be argued that this approach takes into account the existence of differential susceptibility. Recent analysis has shown, however, that combining SCORE estimates with data on educational level significantly improves the discriminatory power.⁴⁹ The main questions relating to equity effects in clinical prevention is about differential implementation and differential effectiveness of screening, treatment and follow-up. Individual behavioural interventions require mobilization of an individual’s resources, and will thus often primarily benefit those with more capabilities.^23,50

The second option—the population strategy—is by definition reaching the whole population, but the differential effectiveness will depend on what intervention methods are chosen. Broad information campaigns on smoking, physical activity and diet have been shown to be less effective in changing behaviour among more disadvantaged groups, contributing to increased health inequalities.^23,47 In contrast, more ‘structural’ universal measures, such as increased tobacco tax and environmental legislation, may have differential effectiveness in the opposite direction, i.e. being more effective with low-income groups.⁴⁷ One important conclusion is, however, that when differential susceptibility exists, then also preventive interventions with equal impact on exposure across groups will have a stronger health effect among the more susceptible—which often will be the disadvantaged. That does not change the fact that vulnerable groups might still suffer larger health effects than others from exposure to the same reference dose level, and differential susceptibility might therefore be an argument for having stricter reference dose levels when heterogeneous populations include more vulnerable segments.⁴³ Schwarz⁴³ has, for example, shown that the effect of lead exposure on child development is stronger among children living in poverty.

Because some universal population measures due to differential effectiveness may widen inequalities, it has been argued that such measures should be combined with a strategy that targets vulnerable groups.²³ Estimates made by modelling have, however, shown that programmes aiming at empowering populations in deprived areas may not succeed in reducing health inequalities, when fundamental contextual causes such as neighbourhood economical segregation is not addressed.⁵⁰ This illustrates the importance of viewing vulnerability as a contextual phenomenon.¹⁸ The net result will clearly depend on what resources are addressed. The theoretical understanding favours a policy that focuses on the capability dimension of vulnerability by increasing contextual, and not only individual, resources that widen people’s range of options and capabilities.⁴⁵

Conclusions

Estimating both differential exposure and differential susceptibility to causes mediating the effect of social position is relevant in health inequality research. Recent methodological developments have made it possible to decompose the effect of social position on health into four components of mediation and interaction, and to estimate absolute effects based on different study designs. Knowledge of biological mechanisms from epigenetics and stress research indicate that differential susceptibility might be highly relevant in social epidemiology. So too is the concept of differential vulnerability, though the empirical evidence is still sparse and needs to identify for which exposures differential vulnerability is particularly important.

Conflict of interest: None declared.

References