The aim of the present study was to test whether (a) the previously observed immediate effects of changes in MAP on acute AA hospitalizations in British Columbia (BC) vary by regional annual average family income level (< Cn$65 000, Cn$65000–< 75 000 and Cn$75000+) and (b) the previously observed lagged effects 2 years after minimum price changes on chronic AA hospitalizations also vary by income. Following Holmes et al . 16 , it was predicted that the observed relationships would be larger in lower‐income regions. We also explored whether such relationships applied for both 100% and partially AA conditions.

Several alcohol policies and interventions have been developed, with the aim of reducing the harmful use of alcohol and the alcohol‐attributable (AA) health and social burden in a population and in society 4 . Pricing is a policy with a substantial body of literature indicating that higher prices and taxes can be highly effective in reducing the harmful use of alcohol. Studies show that as the price of alcohol increases, AA morbidity and mortality decline 5 - 7 . In recent years, many attempts have been made to develop and test more specific varieties of pricing policies such as minimum prices 8 . Minimum alcohol pricing (MAP) involves setting a floor price below which a specified quantity of ethanol or alcoholic beverage cannot be sold. Minimum prices in BC are set by the government alcohol distribution monopoly and apply to the retail prices of alcohol in both government and privately owned liquor stores. MAP has been used in some Canadian provinces since the 1920s and it is now utilized in all 10 provinces, whether to set floor prices for alcohol sold for on premise and‐/or off‐premise consumption or both 9 , 10 . Research conducted in Canada has found that minimum prices are associated with reduced alcohol consumption 11 , 12 , fewer offences for impaired driving 13 and less AA mortality and morbidity 14 , 15 . UK modelers predict greater impacts for low‐income drinkers 16 . It was predicted, therefore, that low‐income drinkers would reduce their alcohol consumption the most and, as a group, demonstrate the greatest reductions in alcohol related harms. However, no studies have been conducted to confirm whether this occurs in practice in a jurisdiction where MAP has been introduced or varied over time, both by regulation and by value in relation to the cost of living. This question is of substantial policy significance, as concerns are often expressed that raising the price of cheap alcohol has adverse effects on people on low incomes 17 . The concern is amplified by evidence that although individuals on lower incomes are more likely to be abstainers, those who do drink are more likely to experience harm 18 .

Alcohol is a psychoactive substance with dependence‐producing properties. The harmful use of alcohol ranks among the top five risk factors for disease, disability and death throughout the world 1 , 2 . It is a causal factor in more than 200 disease and injury conditions, as described in Statistical Classification of Diseases and Related Health Problems, 10th revision 3 . Globally, harmful use of alcohol is estimated to cause approximately 3.3 million deaths every year, or 5.9% of all deaths, and 5.1% of the global burden of disease is attributable to alcohol consumption 4 .

Lagged effects on chronic AA hospital admissions were explored at approximately 1–3years (lags 5–12) after the pricing interventions to replicate time‐periods observed to be associated with significant province‐wide effects in our earlier paper 14 . We conducted all statistical analyses using SAS version 9.3 and the SAS MIXED procedure was used to perform mixed‐effect regression analyses 29 . All significance tests assumed two‐tailed P ‐values or 95% confidence intervals (CI).

We used multi‐level models to model the AA hospitalization data (total and by type) for all 89 LHAs and by regions defined by the above three income levels 27 . Multi‐level models 27 provide straightforward but flexible methods for assessing spatial and temporal dynamics of data in cross‐sectional versus time–series designs. Log‐transformations were applied when necessary to correct for significantly skewed distributions and make the variance stationary for dependent variables. Adjustments for temporal autocorrelation were made if this was detected by the Durbin–Watson (DW) statistic. A seasonal index method 28 was used to adjust for the significant seasonal variations in AA hospitalization rates. Other covariates included in models were: the density of restaurants and bars, government stores and private stores; and percentages of the population that were aboriginal, visible minorities, high school completers, annual family income (only in pooled data analyses) and population density.

We first describe the characteristics of the data and present preparatory analyses on the whole data set in order to examine potential confounding effects of covariates on the relationship between AA morbidity and MAP 14 , 26 . We then tested for an interaction effect between minimum price and income level to investigate whether the association between the minimum prices and hospitalizations was modified by income. If an interaction effect was present, we conducted stratified analyses to estimate the relationships between the minimum prices and admissions within each income group. The analyses were conducted on both total and different types of AA hospitalizations: chronic versus acute and partial versus 100% AA. Regions were categorized into three approximately equal groups according to annual family income reported in the 2006 census using $65 000 and $75 000 as cut‐off.

We obtained BC government data for LHA administrative areal units from DataBC 25 . The geometric center (centroid) of each area was computed and included in the models. We used the centroids from the LHAs to incorporate spatial dependence in the model by using the distance between centroids to develop the spatial covariance structure using a semivariogram function 25 .

Socio‐demographic variables likely to confound the relationships of interest were included in the models 23 , 24 . We took percentages of individuals who were aboriginal, were visible minorities, did not complete high school and were in different family income brackets for each area from the 2006 Canadian census. Population densities were calculated as total population divided by land area (km 2 ).

We obtained minimum prices data from the BC Liquor Distribution Branch (LDB) 12 . During the study, spirit minimum prices increased in five increments from Cn$25.91 to Cn$31.66 per liter of beverage (including sales taxes), and packaged and draft beer prices each increased in increments from Cn$3.00 and Cn$2.05 to Cn$3.54 and Cn$2.22, respectively. Other minimum prices were unchanged. We then re‐calculated mean minimum prices as dollar values per standard drink (= 13.45 g or 17.05 ml ethanol) using estimates of mean percentage of alcohol content for each main beverage type, adjusted by the CPI 12 .

We estimated AA completed hospital admissions (i.e. those that had resulted in a discharge) by applying population AA fractions (AAFs) to admission data for 60 categories of disease and injury 1 . For non‐100% AA diseases, we calculated AAFs from the level of exposure to alcohol and the risk relations between consumption and different disease categories (Table S1 in Supporting information, Appendix S1 ). AAFs were multiplied with overall numbers of completed hospital admissions for each International Classification of Diseases–10 (ICD‐10) code 20 and summed to obtain the total burden of disease from alcohol by age and gender for each of the 89 LHAs in each quarter of years 2002–13. Rates of AA admissions were age and sex‐standardized with reference to the 2001 BC population 21 .

We conducted a cross‐sectional versus time–series design study 19 to test for an interaction effect of regional average annual family income in the relationship between minimum alcohol prices and AA hospitalization rates. Where interaction effects were present, we estimated the association between prices and admissions for each income group separately. The units of analyses were 89 Local Health Areas (LHAs) tracked across 48 annual quarters (2002–13). Adjustments were made for LHA density of different liquor outlets, underlying long‐term linear trend, seasonality, LHA socio‐demographic characteristics and regional and temporal autocorrelation. The LHAs are nested within 16 larger health service delivery areas (HSDAs). The analyses also measured and corrected for the effect of correlation between LHA‐level values of dependent variables within each HSDA.

Table 5 presents the percentage change in the rate of chronic partial AA hospital admissions with a 1% change in minimum prices by regional mean family income. Multi‐level regression analysis found that a 1% increase in minimum prices was associated with a significant 1.073% reduction ( t ‐test P s < 0.001) in chronic partial AA hospital admissions 2 years later for the province as a whole. The effect estimates varied by the regions with different incomes; a 1% increase in minimum prices was associated with a significant 2.474% reduction (95% CI: −3.937, −1.011; t ‐test P < 0.001) in chronic partial AA hospital admissions 2 years later in regions with low average annual family income and a significant 1.491% reduction in regions with high mean family income ( t‐ test, P < 0.001). No significant association was found in the regions with medium average annual family income ( t ‐test P > 0.05).

Table 4 presents the percentage change in rate of chronic 100% AA hospital admissions with a 1% change in minimum prices by regional mean family income at quarterly lags following MAP increases. Multi‐level regression analyses did not show a significant effect for the province as a whole when all the 89 LHAs were included. However, the analyses by regions with different income levels showed that a 1% increase in minimum prices was associated with a significant 2.242% reduction (95% CI: −4.097, −0.388; t ‐test P < 0.05) in chronic 100% AA hospital admissions 2 years later in the regions with low average annual family income, while no significant association was found in the regions with medium or high average annual family income ( t ‐test P > 0.05 in each case).

Table 3 presents the percentage change in rate of chronic AA hospital admissions with a 1% change in minimum prices for the LHAs in BC, 2002–13, by mean family income at quarterly lags following minimum price increases. Multi‐level regression analysis found that a 1% increase in minimum prices was associated with a significant 1.303% reduction ( t ‐test P s < 0.001) in total chronic AA hospital admissions 2 years later (i.e. lag quarter 9) for the province as a whole. The effect estimates varied by the regions with different incomes; a 1% increase in minimum prices was associated with a significant 2.518% reduction ( t ‐test P < 0.001) in total AA hospital admissions 2 years later in regions with low average annual family income and a significant 1.069% reduction in regions with high mean family income ( t‐ test P < 0.001). No significant association was found in the regions with medium mean family income ( t ‐test P > 0.05).

Table 2 presents the percent change in the rate of simultaneous (quarterly lag = 0) AA hospital admissions with a 1% change in minimum prices for the LHAs in BC, 2002–13, by regional mean family income. Multi‐level modeling shows that a 1% increase in minimum prices was associated with an immediate and significant 1.653% decreases ( t ‐test P < 0.01) in the rate of acute 100% AA hospital admissions in all the health regions (income was included as a covariate in the models). However, the analyses by income shows the effects differed by regional income levels. Multi‐level modeling shows that a 1% increase in minimum prices was associated with an immediate and significant 3.547% decrease (95% CI: −5.719, −1.377; t ‐test P < 0.01) in the rate of acute 100% AA hospital admissions in the health regions with low average annual family income; a 0.855% decrease ( t ‐test P > 0.05) in regions with medium average annual family income and a 0.557% decrease ( t ‐test P > 0.05) was found in the regions with high average annual family income.

Table 1 presents the mean age‐ and sex‐standardized rate per annual quarter of acute and chronic 100% and partial AA hospitalizations per 100 000 population in the LHAs with three levels of income in 2002–13. The interaction terms of minimum price and income (MAP × INC) suggested a modified effect of income on the association of AA morbidity with MAP in fully adjusted immediate effect models of acute and chronic 100% AA hospitalization ( F ‐test P = 0.0332 and 0.0486) and in fully adjusted lag effect models of chronic 100%, partial and total AA hospitalization (lag quarter 9: F ‐test P = 0.0439, 0.0441 and 0.0043). The analyses by regions with three level incomes are also presented in the following as well as the analyses on the pooled data.

During the study period, 239 022 AA admissions were estimated, 48.05% of which were acute and 51.95% were chronic. Some factors related to AA hospitalizations may have affected the relationship between MAP and AA hospitalizations and we first examined these factors and AA hospitalization rates. Figure 1 presents the trend in mean age‐ and sex‐standardized rates of AA hospitalizations by types. The analysis of variance identified significant variance in these rates by both HSDA and LHA ( F ‐test P < 0.001 in each case). Age‐ and sex‐standardized acute partial and chronic 100% and partial AA admission rates each showed significantly increasing trends over the study period ( t ‐test P s < 0.001) and age‐ and sex‐standardized acute 100% AA admission rates showed significantly decreasing trends during the study period ( t ‐test P < 0.001). There were also significant seasonal differences in these rates ( F ‐test P s < 0.05), being lower in January–March and higher in April–June of each year. Rates of AA admissions were associated significantly with the percentage of population that was aboriginal, was a visible minority and had not completed high school. Lower population density and lower family income were also significant predictors ( t ‐test P < 0.001 in each case). Outlet density of restaurants and bars, government and private liquor stores were associated significantly and positively with four types of AA admission rates ( t ‐test P < 0.001 in each case. Moran's I autocorrelation analyses identified significant spatial autocorrelation on rates at each centroid for each type of AA admission rates on the basis of a distance‐defined neighborhood ( P < 0.01 in each case). DW tests for rates of acute partial, chronic 100% and partial AA admissions were significant (DW‐test P < 0.05), confirming the presence of fourth‐order temporal autocorrelation, corresponding to consistent seasonal variation ( P < 0.01).

Discussion

These analyses only partially confirm the overall pattern of findings in our earlier 14 publication regarding associations between MAP and AA hospitalizations. Using these same data, but for the shorter time‐span of 2002–09, we reported previously that a 1% increase in minimum prices for all alcohol products was generally associated with a significant and immediate 0.89% decrease for acute (100% and partial combined) AA hospitalizations. In the present study, a significant decreasing effect on acute hospital admissions of higher minimum prices was observed only for those deemed 100% alcohol‐caused. The observation of delayed reductions in chronic AA admissions approximately 2 years after a minimum price increase was still confirmed. The closer analysis of the present paper of mean regional family income levels, however, reveals important variations in the overall pattern of results.

While we observed a similar pattern of results for areas classified as having low, medium or high average family income, there was also a tendency for effect sizes to be larger in the ‘low’ income strata, especially for 100% AA hospitalizations. Thus, these new analyses suggest that the regions of BC with lower income appear to experience the greatest health benefits when minimum prices increase. Conversely, low‐income areas are likely to experience the greatest increments in acute 100% AA hospitalizations when minimum prices decrease in value. The significance of this effect is magnified by the other main finding, that low‐income areas already had significantly higher overall rates of AA morbidity compared with both the medium‐ and high‐income regions studied.

In the present study, we observed that a 1% change in minimum prices was associated significantly with an immediate 1.642% reduction in the rate of acute 100% AA morbidity (i.e. injuries, poisoning, acute illnesses) in the province; this effect was large and significant in the low‐income (−3.547%) in comparison to the medium‐ (−0.855%) and higher‐income (−0.557%) regions of the province.

As in our earlier study 14, significant delayed negative effects on total chronic conditions after approximately 2 years were observed in the present study. However, the effects varied by regions with different income levels and for type of chronic morbidity (100% versus partial AA). This finding is consistent with the plausible theory that minimum prices are especially effective among low‐income heavy drinkers 16. These effect sizes also tended to be greater in magnitude in the low‐income regions. However, there were also significant reductions in chronic AA admissions at several lagged time‐points for higher‐income regions. Nonetheless, the significant effect sizes at lags 8, 9 and 10 were larger for low‐ than high‐income regions.

MAP is likely to reduce rates of AA hospitalizations 14 by reducing population levels of alcohol consumption 12. Minimum prices may have larger impacts in the regions with lowest income, because alcohol is already less affordable in these communities. While it is known that there are more abstainers among people on low income 16, studies also show that expenditure on alcoholic beverages as a proportion of household income is twice as large in low‐income households as in high‐income households in countries such as the United States and the United Kingdom 30. However, Holmes et al. 16 have modeled the impacts of minimum unit prices and concluded that for low‐income drinkers with heavy consumption not only are there large reductions in their alcohol intake, there is also no overall increase in expenditure when minimum prices are introduced; i.e. in traditional economic parlance, minimum prices may not be ‘regressive’. Similar to Holmes et al. 16, the present analyses suggest that even if MAP is in fact ‘regressive’ to some degree (i.e. they have greater impacts on available income of low‐ versus high‐income groups), these concerns may be offset by some disproportionately large reductions in AA morbidity for low‐ versus high‐income groups when minimum prices are introduced or increased.

MAP aims to increase the price of very cheap alcohol, therefore limiting its affordability. Compared to taxation (e.g. sales and excise taxes), setting a minimum price may produce large effects on drinkers with heavier alcohol consumption patterns as they tend to purchase cheaper alcoholic beverages 30. Meier et al. 31 have shown that minimum prices have a more targeted effect on heavier drinkers, especially those on low incomes, than does an across‐the‐board alcohol excise tax increase.

It is noteworthy that we tended to find larger and more significant effects for 100% AA diagnoses compared with those considered to be partially AA. It is possible that the methodology we applied to estimate partially AA conditions results in less accurate estimates of trends than obtained from 100% AA conditions. This difference could also be due to differences in other unmeasured risk factors confounding the relationships of interest. This observation applied to both acute conditions (e.g. injuries, poisonings) and chronic diseases (e.g. liver disease, cancers).

A number of limitations of this study need to be borne in mind. First, these are observational data and the associations identified, however statistically significant, could be caused by other unmeasured factors simultaneously influencing minimum prices and AA hospital admissions to change in opposite directions. The effect sizes calculated mainly have quite wide confidence intervals. Thus, there is greater certainty regarding the statistical significance of the direction of the observed associations (i.e. that some are negative) than in their precise size. It also needs to be acknowledged that this study examines ecological relationships at a broad aggregate level across quite large populations. In areas with a ‘low’ mean, a family income of less than Cn$65 000 per annum will incorporate a wide range of family incomes from very low to very high. The intention of the study was to create broadly different areas which, on average, have contrasting overall levels of family income in order to explore whether changes to MAP have different impacts on these broadly different regions. This is the first study, to our knowledge, to observe and test the income effect on associations between changes in minimum prices and in AA hospital admissions. Other such studies conducted in other jurisdictions are recommended to confirm or disconfirm these findings.