Causal mechanisms proposed for the alcohol harm paradox—a systematic review

Background and Aims The alcohol harm paradox (AHP) posits that disadvantaged groups suffer from higher rates of alcohol-related harm compared with advantaged groups, despite reporting similar or lower levels of consumption on average. The causes of this relationship remain unclear. This study aimed to identify explanations proposed for the AHP. Secondary aims were to review the existing evidence for those explanations and investigate whether authors linked explanations to one another. Methods This was a systematic review. We searched MEDLINE (1946–January 2021), EMBASE (1974–January 2021) and PsycINFO (1967–January 2021), supplemented with manual searching of grey literature. Included papers either explored the causes of the AHP or investigated the relationship between alcohol consumption, alcohol-related harm and socio-economic position. Papers were set in Organization for Economic Cooperation and Development high-income countries. Explanations extracted for analysis could be evidenced in the empirical results or suggested by researchers in their narrative. Inductive thematic analysis was applied to group explanations. Results Seventy-nine papers met the inclusion criteria and initial coding revealed that these papers contained 41 distinct explanations for the AHP. Following inductive thematic analysis, these explanations were grouped into 16 themes within six broad domains: individual, life-style, contextual, disadvantage, upstream and artefactual. Explanations related to risk behaviours, which fitted within the life-style domain, were the most frequently proposed (n = 51) and analysed (n = 21). Conclusions While there are many potential explanations for the alcohol harm paradox, most research focuses on risk behaviours while other explanations lack empirical testing.


INTRODUCTION
Alcohol accounts for 5.3% of deaths and 5.1% of the burden of disease and injury globally [1]. However, alcohol-related harms (e.g. deaths, illnesses and hospitalizations due partly or wholly to alcohol) are not equally distributed across socio-economic positions (SEP)-the social and economic factors that determine an individual's position in society [2]. Disadvantaged groups suffer from higher rates of alcohol-related hospital admissions and deaths compared with advantaged groups, despite reporting similar or lower average levels of consumption [3,4]. For example, in the United Kingdom, the proportion of people in the highest SEP group drinking more than 4/3 (45%) or 8/6 (23%) units per day is almost double compared to the lowest SEP (22 and 10%, respectively) [5]. Despite this, the alcoholspecific mortality rate among the most deprived is 5.5 times higher [6]. This relationship, termed the alcohol harm paradox (AHP), is found internationally, including in the United Kingdom [4], Australia [7], the Netherlands [8] and Finland [9] and across measures of SEP (e.g. social grade, income, education, car ownership, employment and housing tenure) [10]. Prior to 1980, findings suggest a clear dose-response relationship between alcohol consumption and alcohol-related hospitalization and mortality, irrespective of SEP [11][12][13]. However, in the last 40 years the AHP has become a consistent and long-standing finding [14]. Despite this, there is a paucity of research attempting to understand the underlying causes of the AHP. Several reviews and meta-analyses have described socio-economic differences in alcoholrelated harms based on existing evidence or available survey data [3,[15][16][17][18][19]. However, only a subset also focused upon the contribution of alcohol consumption to this relationship, measured as average consumption (e.g. grams or units weekly, monthly or yearly) or drinking patterns (how often and how much people drink) [3,17,18]. This evidence highlights that neither average alcohol consumption nor heavy drinking patterns can explain differences in alcohol-attributable outcomes between SEP groups. At best, heavy drinking occasions partially attenuate the link between SEP and hospitalizations or mortality by 15-30% [3]. Put simply, the most disadvantaged consistently suffer disproportionate risks of harm from their alcohol consumption when compared to their advantaged counterparts, which is not only a health burden on society but contributes to increasingly widening health inequalities [20].
Empirical studies of the AHP have largely focused upon proximal individual-level factors as potential explanations. The role of unrecorded alcohol consumption has, to an extent, been investigated, and results suggest that under-reporting is similar across socio-economic groups [21]. Cross-sectional studies have also tested differences in drinking patterns, behavioural clustering and drinking histories [21,22]. Although there is evidence that low SEP groups tend to have heavier drinking patterns [21,22] and engage in multiple risky health-related behaviours [21], fewer studies go on to test the degree to which life-style risk factors explain differences in alcohol-related harm. One study highlighted that the rate of alcohol-attributable mortality and hospital admissions was three times higher for the most disadvantaged compared with the most advantaged; this association remained after adjusting for weekly consumption and heavy drinking occasions, and it was only slightly attenuated after further adjusting for body mass index (BMI) and smoking [4]. While investigation of life-style factors is prominent, other potentially fruitful avenues of explanation, such as social and economic causes (e.g. social support, housing and employment), have been neglected.
Substantial socio-economic gradients in health exist across countries and contexts [23,24]. There is a critical need for evidence to support public health policies that tackle not only behaviour, but also the broader social determinants of health to mitigate the AHP. This study aimed to review explanations for the paradox put forward in relevant scientific literature. Secondary aims were to review the existing evidence for or against these explanations, and to explore how authors combine different explanations to shed light upon potential relationships between different causal factors. To our knowledge, this is the first review to collate explanations for the AHP.

Search strategy
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (PRISMA) [25]. The protocol for this study can be found at: http://doi.org/ 10.13140/RG.2.2.25606.60489. MEDLINE (1946 -January 2021), EMBASE (1974-January 2021) and PsycINFO (1967-January 2021) were searched to identify peer-reviewed literature on the topic of the AHP or studies that investigated the relationship between alcohol-related harm, socio-economic position and alcohol consumption. An extensive list of search terms was used (see Supporting information, Table S1) to capture the themes of alcohol (e.g. alcohol adj3 drink*) and socioeconomic factors (e.g. disadvantage*). Given the large number of results returned during test searches, further specifications were made by focusing upon papers with alcohol in the title, and some exclusory terms were included (e.g. NOT therapeutics). Terms were tailored dependent upon database requirements. For grey literature, Google and Google Scholar were searched, and this was supplemented via expert identification of relevant reports (C.A.).
in English and (ii) explicitly explored the AHP OR investigated the relationship between: alcohol-related harm, socio-economic position and alcohol consumption (Table 1). We focused upon high-income Office for Economic Cooperation and Development (OECD) countries as classified by the World Bank [26], primarily due to differences in alcohol environments between high-and low-middle-income countries, e.g. greater availability of informally produced alcohol in low-middle-income countries [27]. A range of study designs were eligible for inclusion. Systematic reviews and meta-analyses were included, as it is equally possible to extract 'explanations' for the paradox from these studies. However, intervention and treatment studies were outside the scope of this review. Additionally, empirical studies which analysed data exclusively collected pre-1980s were excluded.

Screening
All records were imported to EndNote Online and duplicates were removed. Titles and abstracts were screened to identify papers matching the inclusion criteria. Full-text versions of the papers were then screened to determine inclusion. Initial screening was carried out by one reviewer (J.B.). A second reviewer (O.S.) then randomly screened a sample of the included studies (n = 20) to validate that papers were correctly included. There was no disagreement between reviewers regarding inclusion.

Data extraction
Data from the papers were extracted by one reviewer (J.B.). A second reviewer (O.S.) independently assessed the accuracy of data extraction for a sample of the included studies (n = 20). In the case of disagreement both reviewers referred to the paper in question, and a consensus was reached. A data extraction matrix was developed, which included characteristics of the studies (design, year of data collection and location), participants (age, target population and sample size), measures (unit of analyses, SEP, alcohol consumption and alcohol harm measures) and outcomes (main findings and explanations for the AHP). Both tested and hypothetical explanations were extracted. 'Explanations' were any reasons identified from the empirical results or proposed by the authors which explain why alcoholrelated harm outcomes were worse for those of a low SEP. Explanations were commonly taken from the results and discussion sections of empirical papers or the main body of other types of included paper. Hypothetical explanations were extracted verbatim. The evidence for these explanations was also extracted from included primary research or from authors citing other research findings when proposing an explanation.

Quality assessment
Quality appraisal of the included studies was conducted by one researcher (J.B.) to assess risk of bias. The AXIS critical appraisal tool [28], CASP qualitative, CASP systematic review, CASP cohort study and CASP case-control study checklists [29] were used depending upon the study design. Commentaries, author replies, discussion papers and reports were not critically appraised. Overall, the quality of included papers was assessed as good. More information on critical appraisal can be found in the Supporting information, Table S2.

Evidence of the AHP
Only three of the included empirical studies found that those of a lower SEP had higher alcohol consumption which then led to increased harm, two of which were specifically focused upon pancreatitis [30][31][32]. Therefore, the evidence base generally supported the existence of the AHP (n = 36, including three meta-analyses of a total of 72 studies); excess harm among those of lower SEP could not be explained by the volume of alcohol consumed.

Thematic analysis
Initial coding revealed 41 explanations for the AHP. The explanations were often presented in discussion sections, did not draw upon existing theory and often appeared to be post-hoc explanations for findings. Following inductive thematic analysis of the 41 explanations, we identified 16 themes and then grouped these themes into six domains: individual, lifestyle, contextual, disadvantage, upstream and artefactual. Domains, themes and explanation definitions are shown in Table 3. The number of papers suggesting each theme as an explanation is presented; however, it should be noted that this is a metric of popularity rather than merit. There was no obvious connection between study design or population and the type of explanation given (Table 3). Themes were not mutually exclusive, and authors often combined or indicated interactions between explanations. These relationships are highlighted in a network diagram (Fig. 2).
Individual-Individual explanations consisted of processes which take place within individuals that could increase their susceptibility to alcohol-related harm. Themes within this domain included biological (n = 7), psychological (n = 22) and health and wellbeing (n = 19) (Table 3). Explanations within the individual domain were often not amenable to human intervention (e.g. genetic make-up or a pre-existing physical health condition).
Individual explanations for the AHP were only hypothesized and had not been tested within any causal or correlational analyses. In related areas, one author has used the tension reduction model to explain alcohol consumption (the idea that alcohol is consumed as a coping strategy to achieve tension reduction) [33]. There was also some evidence to suggest coping strategies more broadly [8,34], and abstention due to pre-existing health conditions [34,35] differed by SEP. Another paper highlighted that the biological effects of social inequality which leads to higher mortality of lower social classes has been observed in primates [34]. However, given the lack of evidence it is unclear whether these explanations contribute to the AHP.
Life-style-The life-style domain focused upon health behaviour of individuals and groups. These were distinct from individual explanations, as they involved an element of choice. Themes were risk behaviour (n = 51), drinking practices (n = 11) and health-consciousness (n = 10) ( Table 3). One paper explicitly referred to theories of social practice (the context, how and why of drinking) when discussing how drinking practices at the group level could contribute to the paradox [36]. Another discussed diffusion of innovation theory: the idea that higher SEP groups are faster to adopt new and healthier behaviours [37].
Several papers (n = 21) investigated the role of risk behaviour in explaining the AHP. One study highlighted higher rates of hazardous behaviour (e.g. creating a public disturbance or physically abusing someone) among the socio-economically advantaged rather than the disadvantaged [38]. Another study also highlighted that, for young adults, risky alcohol consumption and heavy drinking was more prevalent in the employed compared to the unemployed, while alcohol-related problems were greater for the unemployed [42]. Otherwise, there was evidence to suggest that drinking patterns and clustering of health behaviours may play some role, as several cross-sectional studies highlighted that those of a low SEP tend to engage in heavier drinking patterns and multiple unhealthy behaviours [8,21,22,30,[39][40][41]. Those testing the causal role of risk behaviour (n = 13) found that these factors partially attenuate the AHP but could not fully explain excess harm experienced by lower SEP groups [3,4,9,18,[43][44][45][46][47][48]. For example, one record linkage study revealed that when adjusting for alcohol consumption, heavy drinking, BMI and smoking, the hazard ratio for the most deprived group compared to the least deprived was 2.71 [95% confidence interval (CI) = 2.01-3.64] [4]. However, two studies found that controlling for drinking pattern completely accounted for differences in alcohol-related problems in an adult and young adult population [49,50]. In contrast, there was no evidence on the impact of drinking practices or the protective effects of health-consciousness.
Contextual-Contextual factors were those in the individual's immediate environment which may contribute to the AHP. Themes included social (n = 20), drinking context (n = 11) and place (n = 18) ( Table 3).
Although widely discussed, contextual explanations lacked empirical testing. One study, using a within-and between-subjects design, found that when individuals live in neighbourhoods with higher levels of poverty they report 5% more negative alcohol consequences compared to when they lived in a wealthier area [credible interval (CR) = 1.05; 95% CI = 1.00, 1.11; P = 0.045] and those who, on average, reside in more impoverished areas also report more negative alcohol consequences (CR = 1.27; 95% CI = 1.10, 1.46; P = 0.001) [51]. Some studies provided evidence that social factors (e.g. marital status) provide a protective effect [9,52]. However, the limited evidence on other contextual factors, including the relationship between outlet density, consumption and harm, was mixed [53,54].
Disadvantage-Explanations in the disadvantage domain tended to focus upon the lived experience of those in poverty and how different facets of this may contribute to the AHP. Themes included intersectionality (n = 8), life-course (n = 14), material (n = 10) and neo-materialist (n = 21) ( Table 3).
Despite repeatedly appearing in the discussion sections of included papers, only a few explanations associated with disadvantage were empirically tested. Adjusting for material and behavioural factors [45] or cumulative behaviours during the life-course [37] attenuated the relationship between SEP and harm by 18-31% and 38-77%, respectively. There was also evidence that early SEP, disadvantage during adulthood and negative prenatal factors (e.g. maternal heavy drinking) all increased the risk of developing a comorbid mental health and alcohol use disorder, which was not attenuated when controlling for own adolescent drinking [55].
Upstream-The upstream domain captured explanations at the macro-level which were hypothesized to have effects on alcohol-related harm. Themes included economic (n = 11), socio-political (n = 7), alcohol policy (n = 5), corporate influence (n = 1), employment (n = 8), power (n = 1) and broad determinants (n = 4) ( Table 3). These explanations focused upon the structure of society rather than factors associated with belonging to SEP groups. However, the pathways between these societal structures and alcohol harm were not well explained.
None of the included papers attempted to empirically assess whether structural factors can account for the AHP. There was evidence to suggest that economic stressors are more closely associated with mortality in the lowest SEP groups [33,56]. There is also mixed evidence that negative health effects associated with job loss are concentrated in those already at risk due to pre-existing alcohol problems [57], and that SEP overlaps with harmful occupational exposures [43]. However, the extent to which these contribute to the AHP is unknown.
There was evidence which opposed artefactual explanations for the AHP. Although downward drift was commonly discussed, the only study to test it found that it could not account for the AHP [4]. Record linkage and longitudinal studies also support the existence of the paradox [4,9,37,40,41,45,52,55,[57][58][59][60][61][62][63], and therefore diminished concerns of underrepresentation of low-income heavy drinkers in the alcohol consumption data. Another study highlighted that adjusting for alcohol biomarkers only slightly attenuated socio-economic differences in alcohol mortality (1.0-12.1%), suggesting that measurement error is not a probable explanation for the AHP [64]. There was a lack of evidence investigating the impact of often unmeasured factors (e.g. type of cigarette).

Relationships between the thematic explanations-
The relationships between all themes (colour-coded for domain) are shown in Fig. 2. The connections represent where authors have combined themes within a single explanation. For example, the methodology theme is connected to risk behaviour, as one explanation argues that lower SEP groups drink more than they self-report and their heavy consumption leads to greater harm [60].
It is clear that risk behaviour is central to explanations for the AHP, with the greatest number of connections to other themes (n = 10) and links with every other domain (Fig. 2). This is unsurprising, given that health risk behaviours have been the focus of empirical efforts to understand the causes of the AHP.
Other themes, specifically within the upstream and disadvantage domains, were also well connected, possessing connections to four of the five domains. Despite this, they lacked empirical testing.
However, some themes-biological, intersectionality, drinking context and those in the artefactual domain-only had one or two connections. This could reflect the characteristics of the explanation; for example, one of the methodological explanations suggests that, due to the use of self-report measures, research has failed to capture accurate levels of alcohol consumption for low SEP groups: they consume more than they report. Alternatively, the lack of connectivity could reflect value in terms of what researchers think are important explanations for the paradox.

DISCUSSION
This review examined explanations for the AHP to identify potential pathways and mechanisms which result in differential risk of harm between SEP groups. This is a new approach, and goes beyond previous systematic reviews and meta-analyses which have so far established the existence of the AHP and the contribution of alcohol to this relationship [3,18]. We identified 16 themes within six domains used to explain the AHP. Risk behaviours were the most prevalent explanations. This finding, paired with the dominance of the behavioural paradigm in empirical work, suggests that there has been a reliance upon using risk behaviour to understand the AHP. Evidence found in this review opposed the idea that the AHP was an artefact. There were many other, mainly hypothetical, explanations for the AHP proposed in the literature. This included individuallevel mechanisms (e.g. biological or psychological), contextual factors (e.g. place-based factors), the lived experience of disadvantage and upstream structural factors (e.g. the economy and politics). In part, this reflects an awareness that the AHP is complex; there is no simple explanation, and researchers do not view causes in isolation. However, it remains unclear why other re-occurring explanations (e.g. social support or access to health care) have been neglected, while researchers frequently return to risk behaviours. This is particularly puzzling, given that quantitative evidence suggests that risk behaviours only play a partial role [4,47].
There are two potential reasons for this: theoretical and methodological. Study of the AHP is rooted in alcohol epidemiology, which singularly focuses upon the causes and effects of alcohol consumption [65]. More broadly, the field of epidemiology has faced criticism regarding its approach to understand population health. One of the earliest critiques by Krieger points to fundamental errors in developing epidemiological methods rather than theory, with greater weight given to proximal risk factors and a focus upon causes without context [66]. These limitations have led to an emphasis upon individual disease susceptibility and individual-level interventions. Instead, Krieger argues that the eco-social perspective (the idea that biology and biological changes are shaped by the social environment) should be used to understand health [66]. Concerns regarding how causation is viewed in epidemiology have persisted in contemporary public health, with similar criticisms raised more recently [67]. These concerns continue, despite efforts to raise the profile of theories such as the eco-social perspective and calls to adopt pluralist approaches to causality in epidemiology, which stipulate that causation is not a single connection between two things, but the context in which a causal relationship is observed plays a role [67]. Adopting such an approach would change the way alcohol researchers conceptualize and investigate the AHP.
The lack of clear theoretical structuring in epidemiology, which is argued to have led to a focus upon proximal risk factors (e.g. risk behaviours), could also be a symptom of a lack of methods to carry out more complex analyses of distal factors. Possible solutions to this include the use of complex system modelling methods, which have gained traction within public health and are now being implemented in a UK-based project to gain insight into the causal relationships between policy and health-related outcomes [68]. Software architecture has also recently been devised to address how theory can be systematically incorporated into individual-level and agent-based computer simulations to understand health and health behaviours [69]. Applying these computer simulation methods to the AHP could provide the opportunity to shift the empirical focus from risk behaviours to wider determinants, as they can capture complexity and are mechanism-based rather than focused upon testing relationships between variables.

Strengths and limitations
This is the first review, to our knowledge, to catalogue explanations provided for the AHP across a breadth of literature. In taking a broad approach to literature searching and inclusion criteria it was possible to review work from multiple disciplines employing varied methodologies. This led to the identification of a varied set of explanations. However, it is possible that some explanations are more appropriate, depending upon the study design, population and measure of harm. As the primary aim of this review was to collate and review explanations more generally, we did not conduct an in-depth exploration of this issue. However, upon examination there was no evidence that study design or population influenced which explanations were presented. In terms of measures, we found one clear example of an explanation only applicable when using a subjective measure of alcohol harm -those in low SEP groups who drink may feel their outcomes are worse because their peers are more likely to be abstainers [8]. This issue awaits further examination.
This review was restricted to high-income countries. The results and conclusions are therefore only applicable to this context. Furthermore, most papers focused upon the United Kingdom, which may limit generalizability. This was justified, given substantial differences in alcohol environments. However, given that alcohol is a global issue [1], future research should gain insight into how alcohol affects the disadvantaged in low-middle-income countries to help address the deepening of local and global health inequalities.
Another limitation is that only one reviewer screened and extracted data from the papers. We recruited an independent researcher to re-assess a sample of papers for inclusion and extraction. Cross-checking between the two reviewers demonstrated good reliability.

Research and policy implications
The lack of explicit theory used to present explanations is a barrier to understanding the causes of the AHP. The development or application of theory may be fundamental to identify the true causal mechanisms which create and sustain the AHP. Several explanations have been proposed which align with the vast literature detailing theories of health inequality more generally. The eco-social perspective, among those more commonly discussed [e.g. the materialist (the link between wealth and resources and health) or political economy perspective (the idea that risk factors for health inequalities are rooted in structures)] [70], are just some examples of health inequality theory which could be applied to understand the AHP.
The AHP is well-evidenced, and behavioural-related explanations play a partial role. However, these explanations fall short in understanding the complex causes of inequalities in alcohol-related harm. There is a current lack of evidence investigating other explanations found in this review, which makes it difficult to suggest potential interventions to mitigate the AHP. Future research should empirically investigate these alternative explanations for the AHP. Computer simulations models offer one potential way of achieving this aim in the short term and for relatively low cost.
Based on the evidence from this review, the key policy implication is that tackling drinking alone will not reduce inequalities in alcohol-related harm. While there is some evidence that improving multiple health behaviours may attenuate the risk of alcohol-related harm, it is critical that policymakers look to policies outside the scope of public health to mitigate the inequality produced by the paradox.

Conclusions
There are many proposed explanations for the AHP; however, efforts thus far have revolved around risk behaviours as the main cause. Other potentially promising explanations associated within the individual, contextual, disadvantage and upstream domains have remained hypothetical and understudied. Implementation of health inequality theory and complex modelling techniques could provide the opportunity to explore the role of wider determinants in creating and sustaining the AHP.
Additional supporting information may be found online in the Supporting Information section at the end of the article.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.     Table 3 Thematic Lower SEP groups experience more labelling and discrediting which leads to social rejection and exclusion. This could result in a self-fulfilling prophecy, whereby members of that group enact the behaviours they are expected to possess. This could also increase group and individual tensions which find an outlet via harmful drinking. This may also lead to fewer social resources, increasing psychological vulnerability Discussion paper, cross-sectional, cohort, review General population Attribution [8,83] There are a higher number of abstainers in low SEP groups, therefore the alcohol problems faced by those who do drink in this group may seem worse by comparison. This only holds true for subjective measures of alcohol-related harm

Cross-sectional
General population Health and wellbeing Physical health [8,9,32,34,35,41,47,62,63,74,76,93,94,96,98] There is a higher prevalence of pre-existing physical health conditions, poorer general health, multimorbidities or being overweight/obese in low SEP groups which could explain disproportionate effects of alcohol The mutually beneficial economic relationship between the state and the alcohol industry shapes policy decisions. Although it is hoped that this is counterbalanced by 'helping professions' it is also in their interest to continue the expansion of treatment and this is deflected by each entity casting blame on the another. Additionally, a lack of policy that aims to reduce harmful consumption, alcohol availability, pricing and promotion, and global market liberalization (changes in affordability), production, importation, distribution, and pricing of alcohol were hypothesized to contribute to the AHP Cohort, discussion paper, report, review Adults with long-term conditions Corporate influence [61] The alcohol industry funds alcohol research which may misinform policy decision-making. Privately owned media was also argued to play a role via diffusing true or false information Discussion paper NA Employment [9,14,43,52,74,75,81,96] There were several mechanisms through which employment could worsen alcohol-related harms for low SEP groups. This included the working conditions or occupational exposures faced by low SEP individuals. Job type, low wages and inflexible employment, and job alienation, stress and low satisfaction are all thought to negatively impact harm outcomes. Those from more deprived backgrounds with insecure employment may also be less able to take time off work when they become ill, compounding the problem. This contrasts with the idea that high SEP individuals may get more support from their employers, whereby employers are more willing to invest energy in solving their alcohol problems. Relatedly issues of unemployment were also discussed including the issue of receiving additional help of benefits related to a long-term condition or disability which may discourage some people from getting better as they would lose this additional help as a result The heaviest drinkers in deprived areas are often under-represented in studies. This is a potential confounder for cross-sectional studies using aggregate data, as once the heaviest drinkers are accounted for higher rates of harm are no longer paradoxical Systematic review, metaanalysis, report, commentary NA SEP = socio-economic position; LGBT = lesbian, gay, bisexual, transgender; NA = not applicable; TB = tuberculosis.