WHAT IS ALREADY KNOWN ON THIS TOPIC

WHAT THIS STUDY ADDS

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

We conducted a retrospective cohort study among respondents living in two Canadian provinces, Ontario and Alberta, pooling three 2-year cross-sectional Canadian Community Health Survey (CCHS) cycles (2003, 2005 and 2007–2008) linked to administrative ED visits.12 Conducted by Statistics Canada, the CCHS is a cross-sectional survey representative of >96% of the Canadian population,12 capturing sociodemographic, health status and healthcare utilisation information on individuals aged 12+ living in private dwellings. The CCHS does not include Indigenous peoples living on reserve, Canadian Armed Forces members, people living in institutions or in certain remote areas.12 Response rates for the 2003–2008 CCHS varied between 76% and 81%. The CCHS (CCHS linked: 2000–2017) was linked to the National Ambulatory Care Reporting System (NACRS: 2002-2017), the Discharge Abstract Database (DAD: 1999–2017), the Ontario Mental Health Reporting System (OMHRS: 2006–2017), the Canadian Vital Statistics-Death Database (CVSD: 2000–2017) and the T1 Family File (T1FF, 2003–2008) by Statistics Canada. Outcome data derived from NACRS includes event-based hospital and community care ED visits across Canada. However, only Ontario (12.2 million residents in 2003) and Alberta (3.2 million residents in 2003) report data from all facilities. ED visits were followed up until 31 December 2017, the most recent linkage available. DAD and OMHRS, which include all Canadian hospital discharges, were used to identify 100% alcohol-attributable hospitalisation prior to the CCHS survey. The CVSD includes all deaths in Canada, which are used to account for all-cause mortality as a competing risk. The T1FF included the annual income estimates derived from tax returns and was the source of income data.

Overall, 85% of CCHS participants agreed to provide their data for linkage.12 Of these, 96% were successfully linked probabilistically to Statistics Canada’s Social Data Linkage Environment using given and family names, birth date, sex and postal code. Using unique identifiers in health administrative datasets, 95.3% of ED visits were deterministically linked (to this environment).12 Full details on microdata linkage can be found elsewhere.12 Our study received ethical clearance from Public Health Ontario’s Research Ethics Board (file number: 2021–034.01).

We pooled 326 600 respondents from the 2003, 2005 and 2007–2008 CCHS surveys who were linked to ED visits through 2017 (figure 1). We excluded respondents living outside of Ontario and Alberta (n=192 700), in regions that did not include the optional alcohol use module (in 2007/2008, n=9600), who prior to their CCHS interview had a healthcare encounter for a 100% alcohol-attributable condition (n=700) and were aged less than 15 or 65+ (n=28 000). We excluded individuals aged 65+ as their reported alcohol use is often lower and therefore does not align with their lifetime alcohol use patterns.13 Additionally, social assistance programmes available to older adults may mitigate socioeconomic inequities in this population. Furthermore, we excluded women who were pregnant or breastfeeding (n=2200), whose alcohol use may not represent normal levels; those with missing covariate data (n=8000) or lifetime alcohol abstainers who could not experience an alcohol-attributable ED visit (n=8900).

We examined incidents of 100% alcohol-attributable ED visits identified using the International Classification of Diseases 10th revision and Diagnostic and Statistical Manual of Mental Disorders-fifth revision diagnostic codes listed as the underlying or contributing cause based on the Canadian Institute for Health Information’s indicator ‘Conditions Entirely Caused by Alcohol’14(online supplemental eAppendix 1). This definition has been used to examine alcohol-attributable ED visits in Ontario.10 15 16 Follow-up time was measured in person-years from the CCHS survey date to incident alcohol-attributable ED visit, death or end of follow-up.

We operationalised SEP using two measures. Self-reported educational attainment was categorised as less than high school, high school diploma/some postsecondary, trades or certificates below Bachelor’s degree and Bachelor’s degree or above. For individuals aged 30+, the respondent’s education was used. For those under age 30, the highest household educational attainment was used, as respondents may not yet have had the opportunity to achieve the highest education group.6 This approach is consistent with previous Canadian work on educational inequities in alcohol-attributable harm.6 We also examined equivalised after-tax household income quintiles (Q1: lowest and Q5: highest), calculated by dividing after-tax income from T1FF by the square root of the number of individuals living in the household.

Two self-reported alcohol use measures were explored. We dichotomised heavy episodic drinking (yes/no), based on whether respondents reported binge drinking, consuming ≥5 standard drinks (13.45 g of ethanol) on a single occasion, at least monthly in the past year.17 Volume of alcohol use was measured using four risk zones based on the number of standard drinks consumed in the past 7 days based on the continuum of risk described in Canada’s Guidance on Alcohol and Health18: former consumer (no alcohol use in the past year), low (≤2 drinks/week), medium (3–6 drinks/week), high (7–15 drinks/week) and excess (>15 drinks/week) risk. We selected low risk as the reference due to the heterogeneity of alcohol risk among former consumers, some of whom may have stopped using alcohol due to poor health.

We used self-reported measures of biological sex (male/female). We interpret our findings with the recognition that both sex (eg, females experience greater alcohol harm at similar volumes of alcohol use) and gender (eg, socially constructed roles, attitudes and expectations) play a role in generating alcohol harm,19 factors that are further complicated across SEP. Therefore, we use the terms sex/gender, women and men.

Age (in years), marital status (married/common-law, windowed/separated/divorced, single/never married), immigrant status (immigrant and non-immigrant), province (Ontario and Alberta), rurality (urban/rural defined as population concentration ≥1000 and a population density ≥400 km²) and survey cycle (categorical) were included as confounders.

We assessed alcohol-attributable ED visit rates by baseline sociodemographic characteristics by sex/gender. We used Fine and Gray subdistribution hazard models with competing risk (all-cause mortality) to estimate the sex-/gender-specific associations between SEP and incident alcohol-attributable ED visits, including models adjusted for CCHS cycle and all confounders (CCHS cycle, age and age-squared, marital status, immigrant status, province and rurality). We ensured the model’s assumptions were satisfied using a modified version of the log-rank test to test the proportional hazards assumption and Martingale residuals to assess linearity between the log hazard and linear covariates such as age. We investigated a potential joint effect by including an interaction term between SEP and alcohol use. In our joint effect models, we further dichotomised education (low: less than high school or high school diploma/some postsecondary; high: trades/certificate below Bachelor’s, Bachelor’s or above) and income (low: Q1–Q3; high: Q4–Q5) to improve the precision of our estimates. We estimated 95% CIs using bootstrapping (500 repetitions) and applied Statistics Canada linkage survey weights to all models, scaled by a third to account for combining three CCHS cycles, to produce population-representative estimates adjusted for non-response and respondents that were unable or did not consent to link their responses.12

We tested the joint effects of SEP and alcohol use on the additive scale using the Synergy Index (S). Identifying additive interactions is crucial for public health, as they highlight subpopulations that could gain the most from targeted interventions.20 S represents the ratio of the joint relative effect of two exposures exceeding 1 to the sum of their independent relative effects exceeding 1. S greater than 1 (superadditive) or less than 1 (subadditive) indicates an interaction between exposures. We use S to estimate additive interactions because it remains consistent across different strata defined by covariates, making it ideal for use in multivariate regression models.21 We estimated 95% CIs for S using the Hosmer-Lemeshow delta method22 using the postestimated bootstrapped variances. Statistical analyses were conducted using SAS Enterprise Guide V.8.1.

Table 1

Sex/gender-specific rates of alcohol-attributable emergency department visits by sociodemographic characteristics, Ontario and Alberta, Canada (men=36 900, women=39 700)

Table 2

The associations between socioeconomic position and alcohol-attributable emergency department visits

Table 3

Joint effects of SEP and heavy episodic drinking on alcohol-attributable emergency department visits

Table 4

Joint effect of SEP and volume of alcohol use on alcohol-attributable emergency department visits

In a large population-representative cohort study from Ontario and Alberta, Canada, we found socioeconomic inequities in alcohol-attributable ED visits, consistent with previous Canadian studies using individual11 and area-level SEP measures.10 15 Furthermore, we identified joint effects of low SEP and both heavy episodic drinking and volume of alcohol use on alcohol-attributable ED visits, indicating greater vulnerability at different levels of alcohol use among individuals with low SEP. Our findings underscore the need for population-level interventions that account for both SEP and alcohol use to reduce the high burden and social inequities in alcohol-attributable harm.

Our study adds to the international literature on the joint effects of SEP and alcohol use on alcohol-attributable harm by demonstrating this effect for the first time in alcohol-attributable ED visits. Previous longitudinal studies from Canada,6 Denmark,7 Finland8 and Scotland23 have found a joint effect of SEP and alcohol use on 100% alcohol-attributable hospitalisations or deaths. Our results extend these findings by demonstrating increased vulnerability from acute alcohol use, which accounted for 78% of alcohol-attributable ED visits. This insight suggests that the mechanisms increasing vulnerability to alcohol among individuals with low SEP affect both acute and chronic outcomes, such as ED visits and hospitalisations/mortality. This highlights the broader influence of cumulative socioeconomic disadvantage and the role of structural determinants of health in shaping alcohol-attributable risks.

We also identified sex/gender differences in the strength of the joint effect of SEP and alcohol use on alcohol-attributable ED visits across SEP indicators. The joint effect was stronger in women with low income for both alcohol use indicators. This may be explained by individuals with fewer resources and lower SEP being less able to shield themselves from the negative consequences of acute alcohol use.24 Furthermore, these effects may interact with sex/gender-related risk, where women experience increased stigma from alcohol use and intoxication24 and have greater biological susceptibility to alcohol’s detrimental effects.19 Among men, education was a stronger indicator of vulnerability, which may be linked to men with lower education often working in occupations with stronger drinking cultures and higher alcohol use and harm (eg, trades, transport, equipment operations, primary industry and manufacturing).25 Future research is needed to confirm our findings and to explore potential sex/gender differences leading to differential vulnerability to (acute) alcohol-attributable ED visits.

Several potential mechanisms, beyond alcohol use patterns, have been proposed to contribute to the differential vulnerability to alcohol-attributable harm across SEP. For example, greater exposure and vulnerability to lifestyle risk factors (eg, smoking, overweight/obesity, low physical activity and poor diet) partially explain the disproportionate alcohol-attributable harm.8 23 Despite Canada’s universal healthcare system, individuals with lower SEP or living in rural communities (where lower SEP is more common)26 have reduced healthcare access27 and may lack extended health benefits that cover substance use-related programmes and prescriptions.28

Our findings highlight the need for population-level interventions that target the social determinants of health, including low income and education, to mitigate alcohol harm inequities.5 These policies address the root causes of differential vulnerability to alcohol use experienced by individuals with lower SEP. Furthermore, real-world experimental and modelling studies evaluating population-level alcohol interventions have shown promise for disproportionately reducing alcohol use and harm in individuals with low SEP.29–31 For example, the introduction of minimum alcohol unit pricing in Scotland in 2018 reduced the quantity of alcohol sold overall, with greater reductions in alcohol purchases in households with lower incomes30 and alcohol-attributable hospitalisations/deaths in neighbourhoods with the highest deprivation.31 Additional research is needed to understand the unintended, yet potentially disproportionate, health equity impacts of population-level alcohol interventions in contexts where they are eroding. For example, in Ontario, where the ongoing expansion of alcohol sales into convenience and grocery stores is increasing alcohol availability in up to 8500 new outlets,32 there may be disproportionate impacts on low SEP groups.33 34

Our study has several limitations. Our results may be conservative, due to the presence of non-response bias (higher among individuals with heavy episodic drinking)35 and selection bias (due to the exclusion of hard-to-reach populations with higher alcohol use)36 inherent in population-representative surveys. The exclusion of individuals with prior alcohol-attributable healthcare encounters, including individuals with alcohol use disorder at baseline, resulted in a healthier study population than the population of alcohol users. As such, the true socioeconomic inequities in alcohol harm are underestimated. However, this exclusion is required to examine incident alcohol-attributable ED visits, minimising reverse causality of alcohol use on SEP. Additionally, self-reported alcohol use is likely underestimated, estimated to account for 30%–60% of per capita alcohol sales.37 However, evidence suggests alcohol use is not differentially reported by sociodemographic factors38 and is an unlikely explanation of the alcohol harm paradox.39 Furthermore, SEP and alcohol use were measured at baseline only, not accounting for lifetime alcohol use and its time-varying relationship with SEP. Given the long follow-up between baseline and alcohol-attributable ED visits in our study, measured alcohol use cannot directly account for these ED visits. We also likely underestimated the true extent of socioeconomic inequities in alcohol-attributable harm, as partially alcohol-attributable harm and non-health harm due to alcohol were accounted for. Finally, our findings may not be generalisable to jurisdictions without universal healthcare.

Our study has several strengths. We take advantage of a large cohort study that includes measures of individual-level SEP, alcohol use and up to 15 years of follow-up for alcohol-attributable ED visits. Our study is among the first to examine the joint effect of education and alcohol use on alcohol-attributable ED visits, a finding that was robust across SEP indicators. This is important as SEP indicators represent distinct yet interrelated pathways that impact alcohol-attributable harm, for example, through knowledge-related assets (ie, education) or material resources (ie, income).

Not applicable.