Effects of minimum unit pricing for alcohol on different income and socioeconomic groups: a modelling study

Summary Background Several countries are considering a minimum price policy for alcohol, but concerns exist about the potential effects on drinkers with low incomes. We aimed to assess the effect of a £0·45 minimum unit price (1 unit is 8 g/10 mL ethanol) in England across the income and socioeconomic distributions. Methods We used the Sheffield Alcohol Policy Model (SAPM) version 2.6, a causal, deterministic, epidemiological model, to assess effects of a minimum unit price policy. SAPM accounts for alcohol purchasing and consumption preferences for population subgroups including income and socioeconomic groups. Purchasing preferences are regarded as the types and volumes of alcohol beverages, prices paid, and the balance between on-trade (eg, bars) and off-trade (eg, shops). We estimated price elasticities from 9 years of survey data and did sensitivity analyses with alternative elasticities. We assessed effects of the policy on moderate, hazardous, and harmful drinkers, split into three socioeconomic groups (living in routine or manual households, intermediate households, and managerial or professional households). We examined policy effects on alcohol consumption, spending, rates of alcohol-related health harm, and opportunity costs associated with that harm. Rates of harm and costs were estimated for a 10 year period after policy implementation. We adjusted baseline rates of mortality and morbidity to account for differential risk between socioeconomic groups. Findings Overall, a minimum unit price of £0·45 led to an immediate reduction in consumption of 1·6% (−11·7 units per drinker per year) in our model. Moderate drinkers were least affected in terms of consumption (−3·8 units per drinker per year for the lowest income quintile vs 0·8 units increase for the highest income quintile) and spending (increase in spending of £0·04 vs £1·86 per year). The greatest behavioural changes occurred in harmful drinkers (change in consumption of −3·7% or −138·2 units per drinker per year, with a decrease in spending of £4·01), especially in the lowest income quintile (−7·6% or −299·8 units per drinker per year, with a decrease in spending of £34·63) compared with the highest income quintile (−1·0% or −34·3 units, with an increase in spending of £16·35). Estimated health benefits from the policy were also unequally distributed. Individuals in the lowest socioeconomic group (living in routine or manual worker households and comprising 41·7% of the sample population) would accrue 81·8% of reductions in premature deaths and 87·1% of gains in terms of quality-adjusted life-years. Interpretation Irrespective of income, moderate drinkers were little affected by a minimum unit price of £0·45 in our model, with the greatest effects noted for harmful drinkers. Because harmful drinkers on low incomes purchase more alcohol at less than the minimum unit price threshold compared with other groups, they would be affected most by this policy. Large reductions in consumption in this group would however coincide with substantial health gains in terms of morbidity and mortality related to reduced alcohol consumption. Funding UK Medical Research Council and Economic and Social Research Council (grant G1000043).


Deriving socioeconomic classification-specific mortality and morbidity rates
Investigation and adjustment of SAPM's subgroup-specific mortality and morbidity rates to account for variation in risk across socioeconomic (SEC) groups was undertaken in five steps, as detailed below.

Step 1: Baseline age, gender and health condition-specific mortality and morbidity are extracted from a previous report for use as the basis for adjustments
Tables A1-A4 show the age and gender-specific mortality and morbidity rates for the 48 alcoholrelated health conditions modelled in SAPM. These values are calculated using absolute England mortality and person-specific hospital admissions figures for 2005 and 2005/6 respectively as reported in Tables 15-16 of Jones et al. 1 These are the most up-to-date age, gender and conditionspecific data available. Rates were obtained by dividing the absolute numbers by corresponding population statistics for England in 2005. 2 Health conditions are categorised as either wholly alcohol-attributable (e.g. alcohol poisoning) or partially alcohol-attributable (e.g. ischaemic heart disease). For partially attributable conditions, the rates in Tables A1-A4 are overall rates, not alcohol-attributable rates.           Tables A5-A6). 3 These are the most recent data of this kind available. As the age groups defined by Siegler et al. are different to those used in SAPM, ONS mid-2010 population estimates are used as weights to allow estimation of rates for the modelled age groups (Tables A7-A8). 4 Next, the alcoholrelated mortality rate for each SEC group within an age and sex group is expressed as a ratio of the total alcohol-related mortality rate for that age and sex group (Tables A9-A10).      Step 3: Estimation of SEC-specific alcohol-related mortality rates using SAPM.
SAPM already implicitly accounts for differences in drinking patterns between SEC groups; therefore, an assessment is required of the extent to which the differences in alcohol-related mortality between SEC groups seen in Tables A9-A10 are solely due to drinking patterns. This assessment is undertaken by modelling within SAPM a scenario where everybody stops drinking and using the resulting estimated harm reductions to derive alcohol-related mortality rates for each age, gender and SEC group. If the estimated alcohol-related mortality rates are not comparable to the patterns in Tables A5-A10, then this indicates the differential mortality rates are not solely attributable to differences in drinking patterns across SEC groups. Adjustments to the underlying absolute alcohol-related mortality risks used in SAPM would then be required. Table A11 presents the estimated alcohol-related deaths and mortality rates under the 'everybody stops drinking scenario'. It also presents ratios of the alcohol-related mortality rate for each SEC group to the population alcohol-related mortality rate. Unlike Tables A5-A10, the results from SAPM show higher alcohol-related mortality rates for higher SEC groups. Although contrary to the evidence from Siegler et al., this is expected as higher SEC groups drink more on average than lower SEC groups and, therefore, SAPM would be expected to estimate a higher alcohol-related mortality rate for higher SEC groups. Adjustments to SAPM to account for this are required.

Step 4: Modifying adjustment factors to account for SEC-related differences in drinking patterns already modelled in SAPM.
Adjustment factors are calculated by dividing the estimated age, gender and SEC group ratios in Step 2 by the corresponding SEC group ratio in Step 3. For example, Table A9 gives a ratio of 0.24 for 25-34 year-old men in NS-SEC Group 1.1 based on evidence from Siegler et al. 3 Tables A1-A2 to derive age, gender and SEC group-specific mortality rates. For example, the original mortality rate for alcoholic liver disease for 25-34 year-old men is 2.6 per 100,000. Applying the adjustment factors gives a mortality rate of 0.5 per 100,000 for 25-34 year-old men in NS-SEC group 1.1 (2.6 * 0.18 = 0.5) and 6.0 per 100,000 for counterparts in NS-SEC group 7 (2.6 * 2.30 = 6.0).
Given the data reported by Siegler et al., it is not possible to derive adjustment factors for the following age groups modelled in SAPM: 16-17, 18-24, 65-74 and 75+. Therefore, it is assumed that the adjustment factors for the younger age groups are the same as for 25-34 year-olds and for the older age groups are the same as 55-64 year-olds. It is also not possible to derive adjustment factors from Siegler et al. for NS-SEC Group 8 (never worked and long-term unemployed); therefore, it is assumed that NS-SEC Group 8 has the same mortality rates as the general population (i.e. they are assigned the mortality rates shown in Tables A1-A2). A sensitivity analysis (reported below) tests the assumption that NS-SEC Group 8 has the same mortality rate as NS-SEC Group 7. The adjusted mortality rates for NS-SEC Group 1.1 are presented in Tables A14-15. Results for other NS-SEC groups are available on request.

1.1.5.
Step 5: Applying mortality adjustment factors to morbidity data.
As Siegler et al. only examined alcohol-related mortality and no comparable evidence is available relating to alcohol-related morbidity, the mortality adjustment factors in Tables A12-A13 and the method described above are used to adjust alcohol-related morbidity rates for the different SEC groups. The adjusted morbidity rates for NS-SEC group 1.1 are presented in Tables A16-17. Results for other NS-SEC groups are available on request.

Probabilistic sensitivity analysis
In this analysis, probability distributions are fitted to the base case econometric modelling parameters that drive the impact of MUP policies. The pseudo-panel approach used to estimate price elasticities produces variance-covariance matrices for each of the beveragespecific models. For example, Table A18 shows the variance-covariance matrix for the offtrade beer base case model. Assuming conditions of multivariate normality, Cholesky decomposition can be used to sample alternative parameter estimates (from which ownprice and cross-price elasticities can directly be derived). The model is then re-run with the new parameter estimates to generate fresh outcomes. The process is repeated a large number of times (30 here due to the time required for model runs, but ideally more) to produce a distribution of outcomes. From this, estimates of the 95% confidence interval around consumption reductions are obtained. The estimated confidence intervals are shown in Figure 2 of the main article.

Alternative price elasticity matrices
The price elasticity matrix is the key driver of policy impact; therefore, we investigate the robustness of our results to alternative methods of defining and deriving the matric. Four alternative sets of price elasticity matrices were used in sensitivity analyses. These are: 1. Own-price elasticities only (Table A19) 2. Significant own-price and cross-price elasticities only (Table A20) 3. Separate elasticity matrices for low vs. higher income groups (Tables A20 and A21) 4. Separate elasticity matrices for moderate vs. hazardous and harmful drinkers (Table  A22 and A23) All elasticity matrices were calculated using the pseudo-panel method outlined in the main article and described in more detail elsewhere. 5 For the income and consumption groupspecific matrices, separate analyses are conducted for each group within the LCF and the matrices are then applied to individuals in the corresponding group within the GLF. Households are categorised as low income or higher income based on whether their equivalised household incomes are above or below 60% of the median equivalised household income. This threshold is the standard definition of relative poverty in the UK and uses equivalised household income to account for differences in levels of disposable income based on household composition. Figures A1-A2 and support the conclusions of the main article which seem very robust to the use of alternative price elasticities. In each sensitivity analysis, consumption among moderate drinkers in all income groups is largely unaffected by the policy whereas an income trend is consistently seen in the effects for hazardous and, particularly, harmful drinkers. Similarly, as in the base case, spending changes remain small for moderate drinkers while for hazardous and harmful drinkers spending tends to decreases in lower income groups and increase in higher income groups. The main differences in results across the sensitivity analyses reflect that, compared to the base case elasticities, the low income and hazardous and harmful drinker groups have larger price elasticities and the higher income and moderate drinker groups have smaller price elasticities. As such, under the income group-specific elasticities, the lowest income quintile reduce their consumption and decrease their spending by greater amounts compared to the base case while the higher income groups reduce their consumption by less and increase their spending by more. Similarly, under the consumption group-specific elasticities, moderate drinkers' reduce their consumption by less compared to the base case while hazardous and harmful drinkers reduce their consumption and consumption by more.

Results of the sensitivity analyses are presented in
The results for the income and consumption group-specific elasticity matrices should be treated with caution as the underlying analyses violate assumptions of the pseudo-panel method and may not be robust. Specifically, it is recommended that each population subgroup, or panel member in the pseudo-panel, contains at least 100 survey respondents at each wave and this was not the case when deriving elasticity matrices for the low income or hazardous and harmful drinker populations.
Previous versions of SAPM used separate price elasticities for moderate vs. hazardous and harmful drinkers in the base case. These elasticities were calculated by pooling the crosssectional waves of LCF data and conducting a three-stage least squares regression. 6 The pseudo-panel approach addresses several limitations identified with our previous approach (e.g. it analyses the data longitudinally and better addresses non-purchasers). Therefore, despite using a single elasticity matrix, we consider it to provide a more robust estimate of consumers' price responsiveness in England.

Alternative price thresholds
Previous analyses have demonstrated that the scale of impact from minimum unit pricing (MUP) policies increases as the price threshold is raised. To examine whether our conclusions were robust to higher and lower price thresholds, we repeated the income group analysis for MUP thresholds of 40p and 50p. The results of the consumption and spending impacts are presented in Table A24 and show that, as expected, the size of the impact from the policy increases as the price threshold increases; however, the patterning of the results across groups and the conclusions which can be drawn from this remain the same. For all price thresholds, impacts on spending and consumption among moderate drinkers are small in each income group. Income gradients for consumption and spending impacts remain for hazardous and harmful drinkers and policy impacts become more substantive for higher income hazardous and harmful drinkers at higher price thresholds. In general, fewer spending reductions and more spending increases are seen for higher price thresholds. The impact of alternative MUP threshold health harms are presented in Table A25 and similarly show policy impacts increase across all socioeconomic groups as the price threshold increases. However, the pattern of results remains the same. Although the distribution of harm reductions is marginally more equal across groups at higher price thresholds, the vast majority of harm reductions remain in the routine and manual worker household group.

Never worked and long-term unemployed -NS-SEC Group 8
The NS-SEC socioeconomic status classificatory system has eight groups, the last of which is the never worked and long-term unemployed (NS-SEC8). Our source data for alcoholrelated mortality estimates by NS-SEC group provides no data for NS-SEC8. 3 After examining the relationship between income quintile and NS-SEC group membership (see Table A26), we concluded that the NS-SEC8 group typically has a low income. Therefore, in the base case, they are assigned to the next lowest NS-SEC group where mortality data is available (NS-SEC7). In this sensitivity analysis, we re-run the analysis under the more conservative assumption that NS-SEC8 have the average alcohol-related mortality rate for the population.
Model results comparing estimated impacts on health harms in the base case and the sensitivity analysis are shown in Table A27 and suggest reallocation of NS-SEC8 makes only marginal differences to the results.

Previous sensitivity analyses
SAPM was originally developed in 2008 and has been regularly updated and developed since as well as being adapted to appraise policy impacts in other countries. During this period, a range of sensitivity analyses have been conducted on key aspects of uncertainty in the underlying evidence base and datasets. Two of these are reproduced here from earlier technical reports as they are instructive for understanding the implications of uncertainties in the present analysis.
3.1. Effects of accounting for underestimation of alcohol consumption in self-report population surveys (source: Scottish adaptation of SAPM, second update) 7 Alcohol consumption data from self-report population surveys are known to substantially underestimate a population's total consumption compared to more reliable tax or sales data. 8 However, survey data are essential for the types of subgroup analyses conducted using SAPM. We have developed a method for adjusting survey-based estimates of a population's consumption distribution to account for various known biases. The full method is described elsewhere; 8 in brief, the main steps are: • Simulating additional survey respondents for missing populations (e.g. the homeless, prisoners); • Adjustment of survey weights to account for key under-represented populations (e.g. students and dependent drinkers); • Adjustments to individual consumption data to account for other biases (e.g. assumed size of self-poured drinks); • Calibration of the revised consumption distribution to aggregated sales data using the gamma function.
When applied to the Scottish adaptation of SAPM, this method increased estimates of alcohol consumption derived from the Scottish Health Survey, the underlying consumption survey, from an average of 11.5 units per person per week to 19.6 units. The largest increases were due to underrepresentation of dependent drinkers within the survey (increased mean consumption by 4.7 units) and accounting for evidence showing drinkers typically under-estimate the size of self-poured spirits consumed in the off-trade (increased mean consumption by 3.1 units).
The impact of using the adjusted consumption distribution to model the effects of a 45p MUP in Scotland are summarised in Table A28 (full results available in the original report 7 ). The results show accounting for underestimation of consumption leads to substantially larger estimated reductions in consumption, deaths, hospital admissions and associated costs.

Effects of making alternative assumption regarding who consumes purchased alcohol (source: version 2.0 of English SAPM) 9
Estimates of alcohol purchasing and alcohol consumption for the English population are not available in a single survey and thus two surveys are combined to estimate mean and peak consumption levels and preferences for price point, beverage type and purchase location for each population subgroup. Differences in survey design mean we assume purchasing in the two week LCF diary is equivalent to mean consumption by the purchaser. This assumption may introduce multiple errors and a detailed comparison of the LCF purchasing volumes for different beverage types compared to GLF consumption of those beverages suggests discrepancies between purchase and consumption, particularly for females aged 35+. To examine the impact of this, a sensitivity analysis was performed using version 2.0 of the English SAPM. Methods and results are reported in full in the original report to NICE. 9 In brief, alcohol purchases in the LCF were reallocated using a stochastic heuristic which yields an improved match between the LCF and GLF consumption data: "For women whose beer or spirit consumption exceeds 30% in the LCF data, 70% of their off-trade beer transactions and 40% of their off-trade spirit transactions are randomly reallocated to men. For older women (age 25 and older) whose beer or spirit consumption exceeds 30%, 4% of their off-trade wine transactions are reallocated to younger women" The results of this sensitivity analysis are summarised in Table A29 and suggest reallocations of consumption lead to larger estimates of policy impact on alcohol-related harm and costs which are primarily driven by substantially bigger reductions in consumption among harmful drinkers.