Potential effects of minimum unit pricing at local authority level on alcohol-attributed harms in North West and North East England: a modelling study

Project development phase

The idea for the project arose through conversations between stakeholders in the North West of England and the research team. From as early as 2010, a number of local authority directors of public health in the region, along with a public health advocacy organisation, Drinkwise (later renamed Healthier Futures), were keenly interested in measures to reduce alcohol-related harm in their communities and identified minimum unit pricing as a desirable policy option. Stakeholders initially anticipated the national-level introduction of a MUP under the 2012 Alcohol Strategy. 5 However, once it became apparent that minimum unit pricing was not going to be introduced nationally, the stakeholders formed a coalition to consider local action. This coalition, the Tackling Cheap Alcohol Group (TCAG), identified the Sustainable Communities Act14 as a route by which minimum unit pricing might be implemented. 14,17,18 As this route required local-level evidence regarding alcohol consumption, related harms, and the estimated effects of a MUP, stakeholders discussed with the University of Sheffield research team whether or not and how such evidence could be generated. Although this was of interest to stakeholders and the research team alike, there was no opportunity to undertake the work until the 2014 NIHR research call for local interventions to reduce intake and harm from alcohol (https://njl-admin.nihr.ac.uk/document/download/2025655; accessed 30 September 2019).

In preparing the NIHR funding bid, the University of Sheffield research team had several meetings with the chairperson of the TCAG and a representative of Healthier Futures to better understand what local evidence would be required to make a submission under the Sustainable Communities Act14 for the introduction of a MUP. Discussions were also held regarding preferences for ongoing stakeholder engagement during the research (see Project delivery phase). These evidence and engagement requirements were directly reflected in the funding bid to NIHR. During these conversations, it became apparent that stakeholders in the North East region (i.e. directors of public health and a public health advocacy organisation, Balance) shared the concerns of their counterparts in the North West regarding the scale of alcohol-related harms in their communities, and that local authorities in the North East would also benefit from being provided with the local-level evidence. Accordingly, contact was made with the alcohol lead among the North East directors of public health and with the Director of Balance (a third-sector organisation focused on the harms caused by alcohol and the benefits of reducing alcohol consumption), both of whom confirmed the strong interest of the region in the project. Letters of support for the project were provided by the above-named organisations.

Project delivery phase

As had been agreed with stakeholders during the project development phase, the research team engaged with North West stakeholders during the project via three meetings of the TCAG and one meeting with North East stakeholders. Major dissemination events were also held in each region. It had been planned that Healthier Futures would have a role in the delivery of these; however, that organisation ceased operation as the project began. Some administrative assistance in booking venues and sending invitations for the North West events was therefore provided by Wirral Council, and some event planning and facilitation services were provided by an external consultant. These functions were fulfilled by Balance for the North East. Table 1 contains a summary of each event held, including the number of people attending; the local authorities, other organisation and areas of expertise represented; the meeting agenda items; and key outcomes.

How stakeholder engagement influenced the project

In addition to providing an opportunity for the research team to update stakeholders on project progress, the feedback of stakeholders influenced the project in two important ways. First, it became clear that stakeholders were interested in a wider set of outcomes than were originally proposed. In particular, stakeholders in both the North West and North East regions felt that information regarding the impact of a MUP on both crime and social care costs would be valuable. The decision was therefore made to expand the scope of the project to include crime (at no additional cost to the NIHR). However, it was not possible, within the project resources, to model the effects on social care, although baseline data on alcohol-related social care were included. Second, in the original project proposal, we had indicated that static, lay-friendly project summaries would be produced by a professional graphic design company. However, through discussion with stakeholders, it became apparent that a more flexible set of communication materials than originally envisaged would be valued to meet their need to tailor the information for a range of potential audiences. A social marketing agency, Hitch Marketing Ltd (Wirral, UK), was therefore engaged to work with the research team to co-produce a set of dynamic evidence ‘assets’. This included a slide deck linked to a spreadsheet of all estimates for each local authority and English region, as well as national-level results. This will allow end users to produce a bespoke presentation highlighting the results for their local authority. A document of responses to frequently asked questions was also produced.

Report on two major stakeholder meetings held in November 2018

Two major stakeholder meetings were held in Warrington and Durham during September 2018. For each meeting, ≈ 80–100 people from local authorities with an interest in alcohol were invited. A summary of the attendance/invited participants, along with the agenda for each meeting, is given in Report Supplementary Material 1. A detailed presentation of materials very similar to those shown in Chapter 3, The evidence assets developed in the project: some illustrations for one exemplar local authority (Sefton), was given by members of the research team.

Feedback was obtained from the participants by plenary question-and-answer sessions and by notes made during breakout sessions for the participants in groups of four to eight at their own tables. A formal process for the feedback was used (see Report Supplementary Material 1). In addition, more informal reflective feedback was obtained by researchers sitting with participants and observing their discussion of the assets, being asked questions about the assets by participants and being told directly by participants which parts of the evidence asset pack worked (and which perhaps would not) for different potential audiences.

The main ‘take-home message’ from this process was that the key findings of the research would be extremely interesting to a wide range of stakeholders in local authorities. Participants were keen to share the findings with officers, elected members and a wider set of stakeholders in the community. For this reason, a range of evidence assets that could be tailored to different audiences would be necessary (depending on how much time would be available for presentation on an agenda within meetings, e.g. of the health and well-being board). This feedback, together with detailed comments on individual Microsoft PowerPoint^® (Microsoft Corporation, Redmond, WA, USA) slides and presentation topics, was helpful in finalising and refining the evidence assets that have been produced from the project, including the two-sided summary evidence briefing [see Chapter 3, High-level two-sided summary evidence briefing results for one exemplar local authority (Sefton) and for the North West region] and the detailed set of evidence assets PowerPoint slides [see Chapter 3, The evidence assets developed in the project: some illustrations for one exemplar local authority (Sefton)].

Background

Recent estimates from the Global Burden of Disease study suggest that 17% of the total burden of ill health in England is due to behavioural risk factors, and that there is significant variation across the country’s nine regions in both the scale and pattern of associated harms. 19 These variations and those in the associated risky health behaviours are likely to be even greater at smaller levels of geography because predictors of both behaviour and harm, including sociodemographic characteristics,20,21 availability of harmful commodities22–24 and regional cultural differences,25,26 have been shown to vary markedly across such geographies. Set against this background, there has been increasing devolution of responsibility for public health policy decisions to local authorities in England, driving a need for local-level data on health behaviours and harms.

Although harm data are often available at local level from routinely collected records on deaths and hospital admissions, data on health behaviours usually come from government-funded large-scale surveys, which are representative only at the national, or some other large geographical, level. The implication of this is that, given the small samples, direct estimation of small-area characteristics is not possible for each area, which poses a challenge to policy-makers wanting to know the pattern of health behaviours in their locality. This has further implications for modelling effects of policy on a small geographical scale. A common method for small-area characteristics is to produce point estimates of a variable of interest, for example for smoking rates,27,28 poverty29 and multimorbidities. 30 Alcohol consumption has previously been estimated at the local level by Beynon et al. ,31 who produced synthetic estimates of the proportion of the population who are abstainers and the proportions who are lower-risk, increasing-risk or higher-risk drinkers in English local authorities, but did not break this down by the age or sex of deprivation groups.

The method we have developed goes beyond creating synthetic point estimates of alcohol consumption; instead, it re-weights individual-level survey data to make it representative of the local area’s sociodemographic characteristics and expected alcohol consumption. The re-weighted data can then be used to produce an estimate of the complete distribution of drinking in an area, and, by comparing across different re-weighted data sets, demonstrate variation across areas. This method is valuable because understanding the distribution of drinking across the population of a given area is key to estimating both the overall and distributional effects of public health interventions in that area. 32 Detailed local estimates are also more informative to local policy-makers seeking to identify the relative magnitude of public health problems and their distribution across society, and the potential of policies that they may enact to address these. Recent studies have sought to estimate the effects of such local policy approaches, for example on the impact of licensing restrictions,33 and such investigations could be enhanced by data on local alcohol consumption patterns.

The re-weighting method we describe here combines population characteristics with local area characteristics to estimate new weights, so that we move from a national survey to a survey that is representative of the local population with reference to key characteristics, for example to create a synthetic ‘Health Survey for Sheffield’ from the HSE.

In the following section, we present the method of re-weighting survey data to generate a locally representative version, which is potentially useful for a variety of purposes. Second, in the calculation of new weights, we provide updated estimates of the proportion of the population drinking at different levels. The estimates compare well with direct comparison with the original data at region level. Third, we create a re-weighted HSE for each of the 151 UTLAs (throughout this work, the Isles of Scilly are included with Cornwall), which we then use for modelling local authority-level policy effects.

Methods

The re-weighting method involved three steps.

First, the probability of an individual belonging to one of seven alcohol consumption bands was estimated using statistical modelling of the HSE, and was adjusted for individual sociodemographic factors and for local area-level factors (in this case using local authority alcohol-attributable hospital admission rates and mortality rates). These probabilities were calculated for every combination of demographic characteristics and local factors.

Second, the probabilities were multiplied by the corresponding number of individuals in a local authority to provide estimates of the number of people in each of the seven alcohol consumption bands in the local authority’s population. These two steps are identical to the method employed by Beynon et al. 31

The third step went beyond small-area point estimation by re-weighting the survey data. This was done by dividing the number in the population with certain characteristics by the number of survey respondents with the same characteristics. This produced the re-weighted survey that is locally representative and can be used directly for statistical analysis or incorporated into more complex modelling work to produce locally representative policy effect estimates.

The underlying data set to be re-weighted was the HSE, which is a nationally representative, repeated cross-sectional survey of ≈ 8000 individuals in private households per year, covering health and health-related behaviours. For more information on the HSE, see http://content.digital.nhs.uk/healthsurveyengland (accessed 30 September 2019). The HSE contains information about each household member, including age, sex, ethnicity and alcohol consumption. We also received information on each respondent’s UTLA of residence. To avoid disclosure issues, UTLAs in London are listed as either Inner or Outer London. The HSE also provides survey weights to make the survey representative at the national level, but we did not use these to create the new, UTLA-level weights because they correct only for national-level sample representativeness.

To create a large enough sample, the HSE data from 2011–13 were pooled to give a sample of 25,086 adults aged ≥ 18 years. This was reduced to 24,685 for the final analysis because of missing information regarding ethnicity or alcohol consumption for 401 respondents. A sample size of 24,685 and a total of 151 UTLAs means that there was an average of just over 162 respondents in the survey per UTLA, meaning that direct estimation of drinking patterns at the UTLA level would be negatively affected by small-sample problems.

Respondents in the HSE are asked questions about their frequency of consuming, and typical consumption quantities for, several alcoholic beverages. This allows a total mean weekly alcohol consumption variable to be constructed that is measured in units of alcohol. A UK unit is 10 ml, or 8 g, of pure alcohol. The distribution of alcohol consumption in the HSE is shown in Figure 1. Seventeen per cent of the sample did not drink alcohol in the previous year, and roughly three-quarters of drinkers drank moderately (< 14 units per week). Summary statistics of the sample are presented in Table 2.

FIGURE 1.

Alcohol consumption distribution from the HSE 2011–13. Note that it is truncated at the 99th percentile.

Each respondent’s weekly alcohol consumption was assigned to one of seven consumption bands (abstainer, or 1–10, 11–20, 21–30, 31–40, 41–50 or ≥ 51 units per week) and the probability of drinking at each consumption band was estimated using a logistic regression to predict abstention and a multinomial logistic regression to predict positive consumption bands. The multinomial logistic regression was preferred to the ordered logistic regression because it allows the most flexibility and does not impose the proportional odds assumption (for more on this, see Greene and Hensher34). These are estimated as a function of age band (18–24, 25–34, 35–54 and ≥ 55 years), sex, ethnicity (white, Asian, other), IMD quintile (the IMD is a composite measure of deprivation covering seven domains: income, employment, health, education, housing, crime and the living environment), the GOR and the alcohol-attributable hospital admissions rate and alcohol-related mortality rate for the respondent’s UTLA. 35 The admissions rates are taken from the LAPE (for more information, see www.lape.org.uk/), and are applied according to the age band and sex of the respondent. These two statistical models can be written as in Equations 1 and 2:

where iar denotes individual i in age–sex group a in UTLA r, and IMDq is the IMD quintile. Variables were treated as categorical except for HES_ar (the rate of alcohol-attributable hospital admission episodes per 1000 population) and MORT_ar (the rate of alcohol-attributable mortality per 1000 population), which were modelled as continuous variables. Using a separate logistic regression for abstention from drinking allows the direction of the coefficients to vary. For example, for a population in an area with high hospital admissions rates there may be a higher proportion likely to abstain, but also a higher proportion who drink heavily, conditional on not being an abstainer. Such divergent patterns between abstention and heavy consumption are seen in some papers in the international literature. 36 These explanatory variables are chosen because they have previously been shown to be significant predictors of alcohol consumption37 and because there are known population data for each UTLA that can be used to calculate population sizes in the subgroups defined by the combination of age, sex, ethnicity and IMD quintile. For robustness, alternative specifications of the regressions are tested, including changing the number of consumption bands (from six predefined bands to 20 equal bands, i.e. 5% in each band), and removing GOR as a predictor. These make very minor differences to the estimates in terms of mean consumption, which are shown in Appendix 1.

Once the regression parameters are estimated, these are applied for each combination of characteristics and UTLAs. For example, we calculated that the probability that a male, aged 18–24 years, of white ethnicity, in IMD quintile 3, in Sheffield, drinks 11–20 units per week is 19.6%. These probabilities are then applied to the known population data for each UTLA. The population data come from the ONS mid-year population estimates for 2013. 38 For example, there are 5196 males aged 18–24 years of white ethnicity in IMD quintile 3 in Sheffield; so we estimate that there are 1017 males aged 18–24 years of white ethnicity in IMD quintile 3 in Sheffield drinking 11–20 units per week. This calculation is performed for all combinations of characteristics and UTLAs.

These population subgroup estimates by drinker level are then used to create a new survey weight for each individual in the HSE – a survey weight specific to each UTLA. This is done by dividing the number in the UTLA population with a set of demographic characteristics and consumption band by the number of respondents with the same set of demographic characteristics and consumption band. This can be written mathematically as:

where w_idcr is the weight given to an individual i with demographic characteristics d and consumption band c for UTLA r; N_dcr is the number in the population with demographic characteristics d, with estimated consumption band c, in UTLA r; and n_dcr is the number of HSE respondents with demographic characteristics d and consumption band c. The denominator in Equation 3 is the same for every UTLA because the number of HSE respondents by subgroup and drinker level does not differ. This now means that we can calculate 151 different weights for each individual HSE respondent – one for each UTLA. This can be used to make the HSE representative for any UTLA, and any statistic of interest on alcohol consumption can be estimated for any UTLA.

Results

Local authority variation

Four consumption metrics for each UTLA are shown in Figure 2. There is variation in the estimates of mean weekly consumption across UTLAs, with twofold variation between the lowest estimates (around 7 units per week) and the highest (around 14 units per week). Abstention estimates again show large variation across UTLAs, from as low as 11% to as high as 42%, which is probably driven by variation in ethnicity. Because the x-axis is sorted by mean consumption estimate, it shows that there is correlation between abstention and mean consumption, but that some areas have high abstention and high mean consumption. The estimates for the proportion of people drinking more than the recent Chief Medical Officers’ guidelines40,41 of 14 units vary from around 13% of the population to > 30%. Estimates for those drinking at a harmful rate (high-risk drinking is classified as drinking > 35 units per week for females and > 50 units per week for males) vary from around 2% to almost 8% of the population, and both high-risk drinking and increasing-risk drinking (over the Chief Medical Officer’s guidelines) are strongly correlated with the mean consumption estimates.

FIGURE 2.

Estimated consumption metrics based on method to produce synthetic re-weighted survey for each UTLA in England, sorted by mean consumption. (a) Mean consumption; (b) abstention; (c) more than the guidelines recommend; and (d) harmful drinking.

Comparison with observed data

One method of validation is to compare results generated from the re-weighting method with directly measured statistics at GOR level, as the HSE is designed to be representative at this level. Four scatterplots comparing re-weighted estimates with direct measures are presented in Figure 3. The model performs very well at predicting GOR-level estimates, with all estimates lying within the 95% confidence intervals calculated from the observed data. The correlation coefficient between re-weighted estimates and GOR-level direct measures are 0.95, 0.98, 0.82, and 0.99 for mean consumption, abstention, more than the guidelines recommend and harmful drinking, respectively.

FIGURE 3.

Comparison between re-weighted estimates and HSE data at region level. (a) Mean consumption; (b) abstention; (c) more than the guidelines recommend; and (d) harmful drinking.

Discussion

This section has presented a method of re-weighting nationally representative data so that they are representative at the local authority level, in this case constructing 151 locally representative versions of the HSE. This was done by estimating the population of each local authority according to age band, sex, ethnicity, IMD and alcohol consumption. Dividing the number of people with these characteristics in each local authority population by the number of respondents with the same characteristics in the survey gives a new survey weight. To our knowledge, this is the first paper to re-weight the HSE data to obtain local survey data on alcohol use, and, to our knowledge, the first to use such an approach to develop local estimates of alcohol consumption distributions anywhere in the world. The findings show substantial variation in estimated alcohol consumption and abstention rates across UTLAs with a 4.5-fold variation in the estimated abstention rates and a twofold variation in the estimated mean consumption. The results are stable across the alternative specifications of statistical models, as shown in Appendix 1.

Local authorities need to be aware of the variation in estimated drinking volumes, given that there are several policy options, such as licensing decisions and provision of screening and brief interventions, which are decided at the local level. Despite the limitations of the models, the results have clear potential to be used by local decision-makers. The model fit, compared with direct estimates at the GOR level, is excellent. A potentially useful feature of the re-weighting method is that an estimate of the dependent variable (i.e. consumption) can be obtained for any cut-off point – so that policy-makers could estimate how many people drink more than any number of units in their area. Individual areas may also wish to use these estimates for benchmarking and perhaps plan or prioritise services accordingly. The re-weighted HSE data can be used, in conjunction with other local data sources, to model local policy interventions. Given the heterogeneous effects of policies across population subgroups, the capability of the new methods to enable local-level modelling of outcomes for subgroups stratified by age, sex and social deprivation is especially important.

There are several limitations to the method presented, as well as some assumptions that will carry through to any modelling work. Perhaps the most important assumption is that the statistical relationship between the left- and right-hand side variables is constant across the UTLAs; the effect on consumption of being male, for example, is assumed to be similar across all UTLAs. However, this is always implicitly the case with survey weights more generally, in that survey respondents are representative of their sample frame. Furthermore, caution is required when looking at variables not included in the analysis, such as, in this case, smoking habits. These wider attributes of the respondents have not been modelled here, and may differ by UTLA. In addition, the statistical re-weighting analysis method presented here examines the level of mean weekly consumption and does not disentangle drinking patterns and beverage type when undertaking the re-weighting. That is not to say that analysis of this type is not feasible, simply that further work could address this issue through the inclusion of exogenous local data on beverage preferences. Further work could also look at why people in some regions of England drink more than people in other regions, even when controlling for explanatory factors including demographics and hospitalisation rates. This has been noted in the existing literature. 37

There are several implications for future related research. First, more detailed validation against external data would require locally representative data to be collected. This is not easy. Public Health England has conducted surveys in 25 UTLAs to get a measure of local consumption. 42 The re-weighted method correlates moderately well, but the sample size of the local surveys is not large enough to be sufficient. Extensions of this work could provide updates when new data become available, or look at other health risk factors such as smoking or obesity. Combinations of behaviours, to allow multibehaviour modelling, could be analysed. Harm risks are particularly acute when individuals have multiple unhealthy behaviours, as the risks are multiplicative, and unhealthy behaviours tend to cluster within individuals. 43 Furthermore, the method of re-weighting presented in this paper is not unique to either alcohol or small geographical areas, and can be applied to a whole host of outcomes, estimates of which are not directly available for small populations. None of this analysis has looked at geography in the UTLA boundary, for example at electoral ward level or even finer geographies that could relate to specific licensing decisions for on-trade or off-trade outlets.

In conclusion, this section finds that re-weighting nationally representative surveys to make them representative at the local level is possible and also finds large variation in alcohol abstention, mean consumption and measures of heavy drinking across UTLAs. The results of the estimation when aggregated up to provide GOR estimates align closely with directly observed data. This method could be used in any country where national survey data are available and could be applied to many other outcomes of public health interest to inform local priorities and decisions.

Introduction

A fundamental requirement for modelling the local effect of minimum unit pricing is estimates of the price distribution at the local level. This is because there is good reason to suspect that prices paid for alcoholic beverages differ across localities, partly for demand-side reasons (differing population characteristics, different drink preferences) and partly for supply-side reasons (differing rental prices, differing levels of competition among suppliers).

Preferences are also likely to vary across the country because of differing population characteristics and some underlying regional cultural preferences; for example, cider is typically consumed at higher levels in the South West. Preferences are important for modelling the local effect of minimum unit pricing because different drinks will be affected at different levels. Perhaps most important is the split between on-premise (pubs, bars and restaurants) and off-premise (shops and supermarkets), because the on-premise market is barely affected by minimum unit pricing.

For local alcohol price estimates, we estimated the prices currently paid for 10 beverage categories: beer, cider, wine, spirits and RTDs, split by off-trade (supermarkets and shops) and on-trade (pubs, bars etc.). This required synthesis of evidence from the LCFS and market research data. The market research data were provided by CGA (CGA Strategy Ltd, Stockport, UK) and Nielsen (Nielsen Holdings plc, New York, NY, USA) at GOR level. We needed these data because it is known that the amount of alcohol bought in each price band as reported by respondents to the LCFS is slightly different from the evidence available from retail sales data. Typically, the LCFS underestimates the amount of cheap alcohol being sold, and so the methods developed and applied here allow us to adjust for this underestimation for each UTLA.

In terms of subgroups, for both price distributions and preferences, we estimate results for 120 different population subgroups. These are based on two sexes, four age groups (18–24, 25–34, 35–54 and ≥ 55 years), five quintiles of IMD and three groups based on the amount of alcohol purchased in the 2-week diary window (moderate, increasing risk and high risk).

Data sources: the Living Costs and Food Survey calibrated to market research company sales data

The price distributions are estimated using the LCFS 2012–14, accessed via the UK Data Service Secure Lab. The LCFS is a nationally representative survey that requires respondents to complete a 2-week purchasing diary of all items, including alcohol. We analysed 10,065 individuals with 57,581 alcohol transactions in the years 2012–14 in England. The data are available at transaction level, meaning that we see the price paid for every item. The LCFS also records the demographic characteristics of the respondent, including their age and sex, IMD quintile and income quintile. The secure version of the data allows us to see the UTLA of residence, which means characteristics of the local area can be merged in as explanatory variables in the statistical model.

Data on UTLA area characteristics were merged with the UK Data Service Secure Lab transaction records from three sources. The first was data on hospital admission rates by UTLA, taken from the LAPE. 4 The second source was data on outlet density rates, provided by the market research company CGA for each UTLA. The outlet density is defined as the mean number of outlets selling alcohol within 1 km of a postcode in the UTLA, based on data from 2013. 24 The third source was data on the average house prices in the UTLA, which are taken from the Land Registry. 44

The estimated price distributions are calibrated using market research data from CGA (on-premise) and Nielsen (off-premise). The market research data do not provide any information on who is buying alcohol at certain price points; instead, they provide information on the total volume of sales at different price bands (i.e. the price distribution of alcohol prices in each GOR for each of the 10 beverage categories). The data were provided at TV region level, which broadly maps on to GOR. The off-trade data from Nielsen provide price distributions, in bands of 5p, up to £1.20 per unit of alcohol, whereas the on-trade data from CGA are in 5p bands up to £1.20 per unit of alcohol and in 10p bands from £1.20 to £2.60 per unit of alcohol. Both sets of data include an open category for sales above the highest price band and relate to alcohol sold in 2016.

The preferences for different categories of alcohol for each UTLA and population subgroup are estimated using just the LCFS data set. This is because, unlike price distributions, there was no local authority variable that could easily be justified to predict preferences. This is compounded by the fact that preferences are non-ordinal, meaning that patterns in the data cannot be used as easily. The preference vectors are taken empirically from the LCFS 2010–15. This provided more observations – a total of 25,186 individuals with 116,408 alcohol transactions – although it did not include information on the UTLA of residence. The data are, again, available at transaction level, meaning that we can get information on the amount, in ml, purchased. These volumes are converted to units of alcohol using standard assumptions about alcoholic strength, which are set out in Appendix 2.

Method for estimating preferences for the 10 beverage categories

The method employed to estimate preference vectors for each subgroup can be broken down into three stages. The preference vectors are taken empirically from the data.

The aim was to get a preference vector (the proportion of purchased alcohol that is in each of the 10 beverage categories) for each of 120 population subgroups in each UTLA. For example, for males aged 35–54 years, in IMD quintile 3, drinking at high-risk levels, in the Sefton UTLA. If we wanted to calculate the proportion of alcohol purchased in each of the 10 categories, we might find that it was as follows: off-trade beer, 25%; off-trade cider, 5%; off-trade wine, 10%; off-trade spirits, 20%; off-trade RTDs, 1%; on-trade beer, 20%; on-trade cider, 3%; on-trade wine, 5%; on-trade spirits, 10%; and on-trade RTDs, 1%, which adds up to 100%.

To do this, we followed these steps:

1. Transaction-level LCFS data from 2010–15 were cleaned and processed. Purchases were split into the 10 beverage categories (beer, cider, wine, spirits, RTDs, both on- and off-premise). We use this to calculate a preference vector (the 10 percentages) for the whole adult population of England.

2. Next, we divided the sample population into subgroups to get preferences for each subgroup. This was done in stages. The sample was first divided by the nine GORs, as this was chosen as the most important driver of preferences. Thus, we have nine preference vectors, one for each GOR for all adults. The sample is then divided by three drinker types (moderate, increasing risk, high risk), as different drinker types are likely to have different preferences. For example, heavier drinkers tend to drink proportionally more cider and spirits, and less on-premise alcohol. Therefore, we now have 9 × 3 preference vectors. The sample is then divided by sex, as men and women are likely to drink different beverages; for example, women drink a higher proportion of wine. We now have the proportion of alcohol purchased split into the 10 beverage categories for 54 subgroups (nine regions, by three drinker types, by two sexes). The data are disaggregated further only if subgroups have ≥ 50 transactions, as fewer than this would not give a robust preference split. Some subgroups are able to be disaggregated by age band to give region–drinker–sex–age preference vectors.

3. Finally, the preference vectors are applied to the re-weighted consumption estimates (using a method detailed in Method to combine the preference and price distribution modelling with the re-weighted consumption estimates) to give local UTLA consumption estimates broken down into the 10 beverage categories.

Methods for statistical modelling to estimate price distributions

The method employed to estimate price distributions for each subgroup in each UTLA for each beverage type can be broken down into the following stages.

Method to combine the preference and price distribution modelling with the re-weighted consumption estimates

Once the statistical models were estimated, the resulting proportions were applied to the re-weighted consumption data with beverage splits, to give local consumption estimates broken down into 10 beverage categories and 50 price bands.

This was done by selecting a subgroup to analyse, and then adding the total re-weighted consumption (calculated as shown in Local authority variation) of all the individuals in that subgroup in the UTLA. We then applied the proportions in each of the 10 beverage categories (see Method for estimating preferences for the 10 beverage categories) to that total consumption, giving the number of units consumed in 1 year in each beverage category by that subgroup in that UTLA. Next, we applied the price distribution estimate for the subgroup for each beverage category (see Ordinal logistic regression models for the probability that a transaction falls within each of 50 price bands, defined by price per unit of alcohol), so that we had the estimated number of units of each beverage category of alcohol consumed that were purchased in each of the 50 price bands. This gave us, for example, an estimate of how many units of off-trade beer were consumed by males aged 35–54 years, in IMD quintile 3, who are moderate drinkers, that were purchased at a price of between 40p and 45p per unit of alcohol in the Sefton UTLA.

This process was repeated for each subgroup.

We then generated a total for the UTLA population, showing the number of units of each of the 10 beverage categories consumed, split by the 50 price bands, for example to estimate how many units of off-trade beer were consumed that were purchased at a price of between 40p and 45p per unit of alcohol in the Sefton UTLA.

Method to calibrate to market research company price distributions

Finally, the price distributions for each beverage category in each of the UTLAs in the North West region were aggregated up to GOR level and compared with the market research data supplied by CGA and Nielsen. This comparison was used to calibrate, at the beverage level, to the market research data.

The process involved redefining the 50 price band boundaries such that the LCFS price distributions matched the market research data. This has the principal advantage that it corrects the price distribution to observed levels without distorting the relative price distributions across individual-level characteristics.

Consider the following example: imagine that we have estimated that 5.0% of all off-premise beer purchases in the North West fall into the LCFS price band 11, corresponding to 50–55p per unit of alcohol. Moreover, imagine that we have also estimated that 40.0% of all the off-premise beer in the North West is sold at < 50p per unit of alcohol; thus, 45.0% of all off-premise beer is sold at < 55p per unit. To calibrate this, we take the price point corresponding to the 40th percentile of Nielsen’s price distribution for the North West (imagine that this is actually 47.5p), and the price point corresponding to the 45th percentile (imagine that this is 53p per unit). These are the price points to which the data are calibrated. What this means, in practice, is that every transaction that, in the LCFS, falls into price band 11 is actually priced at between 47.5p and 53p.

Results of price distributions analyses

Figure 4 shows the mean price per unit of alcohol in each region, split by on-trade and off-trade, from the raw LCFS secure data. It shows that there is a large difference in prices between on-premise and off-premise beverages. It also shows that there is more across-GOR variation in on-premise prices than in off-premise prices. The most expensive mean price per unit is in London for both on-premise and off-premise, whereas the cheapest mean price per unit is in the North East for both on-premise and off-premise. The variation in the mean price per unit arises from both preference (e.g. Londoners may drink more champagne, which is expensive) and price differences (e.g. Londoners purchase more expensive wine when they purchase wine).

FIGURE 4.

The mean price per unit in each region, split by on-trade and off-trade, from raw LCFS data. EM, East Midlands; EoE, East of England; LON, London; NE, North East; NW, North West; SE, South East; SW, South West; Y&H, Yorkshire and the Humber; WM, West Midlands.

The regression results are presented in Appendix 3, Tables 12 and 13, for on-premise and off-premise, respectively. The results are regression parameters, so their magnitude cannot be directly assessed from the table, but the direction can.

For on-premise alcohol, the results in Appendix 3, Table 12, show that the youngest age group tend to pay less per unit than their elder counterparts. Similarly, and perhaps as expected, price paid per unit increases as income increases. High-risk drinkers tend to spend significantly less per unit on on-premise alcohol. The outlet density in the UTLA has a mixed effect on price paid. In contrast, higher hospital admission rates tend to reduce the amount paid per unit of on-premise alcohol. Higher average house prices increase the price paid per unit of on-premise alcohol, reflecting higher on-premise property costs being passed on to consumers.

For off-premise alcohol, sales of which are affected more by a MUP than on-premise alcohol sales, Appendix 3, Table 13, shows a similar pattern in terms of age and income: younger people and those who are less well-off tend to pay lower prices. Most importantly, high-risk and increasing-risk drinkers spend significantly less per unit on alcohol, after controlling for beverage choice. The outlet density tends to be an insignificant explanatory variable in the off-premise category, except for off-premise cider. Hospital admission rates and average house prices are significant predictors of the price paid for off-premise alcohol, with people in areas with higher hospital admission rates and lower house prices tending to pay less for their shop-bought alcohol.

Figure 5 shows the variation in the percentage of units purchased at < 50p in the uncalibrated estimates. It shows that there is substantial variation across UTLAs, and that the North East and North West of England have some of the highest proportions of alcohol purchased that comes under the potential 50p MUP. Although the price distribution estimates match the raw LCFS data fairly well at GOR level, there is concern that the LCFS data may routinely under- or over-report purchases of cheap alcohol as a result of measurement error.

FIGURE 5.

Uncalibrated price distribution estimates.

Figures 6 and 7 show kernel density graphs of the cumulative price distributions from the statistical model and the market research data at the GOR level for the North West and North East, respectively. The estimates from the statistical model are then calibrated so that the cumulative distributions match.

FIGURE 6.

Comparison of price estimates with market research data: North West. (a) On-premise beer; (b) on-premise cider; (c) on-premise wine; (d) on-premise sprits; (e) on-premise RTDs; (f) off-premise beer; (g) off-premise cider; (h) off-premise wine; (i) off-premise sprits; and (j) off-premise RTDs.

FIGURE 7.

Comparison of price estimates with market research data: North East. (a) On-premise beer; (b) on-premise cider; (c) on-premise wine; (d) on-premise sprits; (e) on-premise RTDs; (f) off-premise beer; (g) off-premise cider; (h) off-premise wine; (i) off-premise sprits; and (j) off-premise RTDs.

Figure 8 shows the variation in the percentage of units purchased at < 50p in the calibrated estimates. It shows that there remains substantial variation across UTLAs.

FIGURE 8.

Calibrated price distributions.

A further example of price distributions is shown in Figure 9, which shows the price distributions for a male, aged 35–54 years, in IMD quintile 3, for off-trade beer in Sefton. The lines show different distributions for moderate, increasing-risk and high-risk drinkers in this subgroup. The reason for showing this distribution is to give an idea of the granularity of the estimates. These graphs could be replicated for any subgroup of the population.

FIGURE 9.

Example price distributions: males, aged 35–54 years, IMD quintile 3, for off-trade beer in Sefton, split by moderate, increasing-risk and high-risk drinkers. The price bands are equal-width, 5p bands; the lowest band is ‘< £0.05 per unit’ and the top band is ‘≥ £2.45 per unit.

Results: preferences

Figure 10 shows how preferences vary across England by GOR, whereby the top of the band represents the GOR with the highest proportion of alcohol coming from that beverage type and the bottom of the band represents the GOR with the lowest proportion of alcohol coming from that beverage type. The North West region is coloured turquoise and is in the middle of the rankings by GOR.

FIGURE 10.

Preferences by GOR.

Similar to the price distributions, a graph like Figure 10 could be drawn for any subgroup.

An example of the detailed breakdown for preferences by subgroup is provided in Table 4.

Overview of Sheffield Alcohol Policy Model version 4.0

The aim of the SAPM version 4.0 is to appraise alcohol policy options via cost-effectiveness and cost–benefit analyses. This is achieved by breaking the policy impact into a series of linked effects to model the effects of:

• the policy on the distribution of prices for different types of alcohol

• changes in price distributions on patterns of both on-trade and off-trade alcohol consumption

• changes in alcohol consumption patterns on consumer spending on alcohol

• changes in alcohol consumption patterns on retailer and exchequer revenue from alcohol sales

• changes in alcohol consumption patterns on levels of alcohol-related health harms

• changes in alcohol consumption patterns on levels of alcohol-related crime.

To estimate the range of these effects, the SAPM version 4.0 consists of two connected models:

1. A model of the relationship between alcohol policies and alcohol consumption that accounts for the relationship between average weekly alcohol consumption, the patterns in which that alcohol is drunk and how these are distributed within the population considering gender, age, income/socioeconomic status and consumption level.

2. A model of the relationship between: (1) both average level and patterns of alcohol consumption and (2) harms related to health and workplace absenteeism and the costs associated with these harms.

Details on the modelling are also available in previously published works. 45–47

Baseline data for modelling the relationship between minimum unit pricing policy and alcohol consumption

As described in Estimating alcohol consumption for each local authority by population subgroup, alcohol consumption data are derived from the HSE, using a novel statistical modelling approach to adjust the original data to be representative of each UTLA in England. Unlike previous versions of the SAPM, we do not use the beverage-specific consumption information for each HSE individual, as the re-weighting procedure described in Estimating alcohol consumption for each local authority by population subgroup accounts for total alcohol consumption only. Therefore, these data take the form of individual-level data describing the demographic characteristics of each individual (age, sex, IMD quintile, mean alcohol consumption, peak alcohol consumption in the preceding week) and a weight that represents the number of individuals in the UTLA that they represent (i.e. the sum of the weights in each UTLA model is equal to the adult population).

As described in Methods for statistical modelling to estimate price distributions, pricing data are derived from the LCFS, using a variation on the same approach as that used for the consumption data. In previous national analysis versions of the SAPM, we have directly used transaction-level data from the LCFS. We do not have transaction-level data at UTLA level. Therefore, in this version, we use the statistical modelling described in Methods for statistical modelling to estimate price distributions to derive a data table. This data table takes the form of 60,000 rows of data, one for each unique combination of age (18–24, 25–34, 35–54 and ≥ 55 years), sex, IMD quintile, alcohol consumption category (moderate, increasing risk, high risk), drink type (beer, cider, wine, spirits and RTDs), purchase location (on- or off-trade) and price band (50 bands from 0–5p per unit to ≥ £2.45 per unit). For each row, the statistical models described in Estimating beverage preferences and the distribution of prices paid for each local authority by beverage category and population subgroup are used to derive the predicted probability that a unit of alcohol purchased by somebody of the relevant age, sex, IMD and drinker group characteristics will be that type of alcohol, in that setting and at that price.

The final data used are the population structure of each UTLA, taken from ONS mid-year population estimates. These are used to derive the population count by age, sex, IMD quintile and single year of age for each UTLA.

Modelling the impact of minimum unit pricing on prices

For each of the 10 types of alcohol included in the SAPM (five beverage types by two purchase locations), the first step in the modelling process is to estimate the impact of a price-based intervention, such as a MUP, on the price distribution at which alcohol can be purchased.

In line with previous iterations of the SAPM, we make the assumption that all prices below the MUP threshold are raised to the level of the threshold, whereas prices above the threshold are unaffected. This is operationalised in the model by reducing the probabilities of purchasing all types of alcohol at prices of < 50p to zero, and increasing the probabilities of purchasing the same types of alcohol at prices of 50–55p by the same amount. This assumption is highly likely to be conservative, as it is likely that, in reality, the supply-side response to a MUP policy would be to increase the prices of some items beyond the threshold, while also increasing the prices of other products that are not currently being sold below the threshold in order to maintain some degree of price differentiation. As a result, the observed price changes are likely to exceed those modelled here, with correspondingly larger reductions in consumption and impacts on model outcomes, such as reductions in the numbers of deaths, hospital admissions and crimes.

Modelling the impact of price changes on consumption

For each of the 10 types of alcohol, we calculated the mean price paid by each of the 120 age–sex–IMD–drinker subgroups in the model before and after the intervention. This gave us 1200 individual changes in price following the introduction of a MUP. The 10 price changes for each subgroup are combined with a 10 × 10 matrix of price elasticities (Table 5), which contains the own- and cross-price elasticities for each beverage type taken from a previously published and used study of price elasticities by Meng et al. 48 For each of the 120 subgroups, the 10 price changes are combined with these elasticities to estimate the percentage change in consumption of each beverage, using the following equation:

where:

• %ΔC_i is the estimated percentage change in consumption for beverage i

• e_ii is the own-price elasticity for beverage i

• %Δp_i is the percentage change in price for beverage i

• e_ij is the cross-price elasticities for the consumption of beverage i due to a change in the price of beverage j

• %Δp_j is the percentage change in price for beverage j.

For each subgroup, the cumulative probability of any unit purchased falling into each of the 10 categories was calculated and applied to the baseline consumption of each individual in the model to calculate the baseline weekly consumption of each beverage type. This was then adjusted using the percentage changes calculated using Equation 1 to obtain the estimated post-intervention consumption for each individual.

Model structure for modelling the relationship between consumption and harm

An epidemiological approach was used to model the relationship between consumption and harm, relating changes in alcohol consumption to changes in the risk of experiencing harmful outcomes. Risk functions relating consumption (however described) to level of risk are a fundamental component of the model.

The ‘consumption to harm’ model considers the impact of consumption on harms in two domains: health (including the impact on both mortality and morbidity) and crime.

In the following sections, we describe how the health component of the model estimates the impact of changes in alcohol consumption on 45 different health conditions, and how the crime component of the model estimates the impact on 14 different categories of offence. Both components use the same basic approach: the potential impact fraction (PIF).

The potential impact fraction methodology

The methodology is similar to that used in Gunning-Schepers’ Prevent model,49 being based on the notion of the alcohol-attributable fraction (AAF) and its more general form, the PIF.

The AAF of a disease can be defined as the difference between the overall average risk (or incidence rate) of the disease in the entire population (drinkers and never-drinkers) and the average risk in those without the exposure factor under investigation (never-drinkers), expressed as a fraction of the overall average risk. For example, the AAF for female breast cancer is simply the risk of breast cancer in the total female population minus the risk of breast cancer in women who have never consumed alcohol, divided by the breast cancer risk for the total female population. Thus, AAFs are used as a measure of the proportion of the disease that is attributable to alcohol. Although this approach has traditionally been used for chronic health-related outcomes, it can, in principle, be applied to other harms (including those outside the health domain).

The AAF can be calculated using the following formula:

where:

• RR_i is the relative risk due to exposure to alcohol at consumption state i

• p_i is the proportion of the population exposed to alcohol at consumption state i

• n is the number of consumption states.

If the reference category is abstention from alcohol, then the AAF describes the proportion of outcomes that would not have occurred if everyone in the population had abstained from drinking. Thus, the numerator is essentially the excess expected cases due to alcohol exposure and the denominator is the total expected cases. In situations in which certain levels of alcohol consumption reduce the risk of an outcome (e.g. coronary heart disease), the AAF can be negative and would describe the additional cases that would have occurred if everyone were an abstainer.

Note that there are methodological difficulties with AAF studies. One problem is in defining the non-exposed group: in one sense ‘never-drinkers’ are the only correct non-exposed group, but they are rare and usually quite different from the general population in various respects. However, current non-drinkers include those who were heavy drinkers in the past (and these remain a high-risk group, especially if they have given up because of alcohol-related health problems). Several studies show that findings of avoided coronary heart disease risk may be based on systematic errors in the way abstainers were defined in the underlying studies. 50

The PIF is a generalisation of the AAF based on arbitrary changes to the prevalence of alcohol consumption (rather than assuming that all drinkers become abstainers). Note that a lag may exist between the exposure to alcohol and the resulting change in risk. The PIF can be calculated using the following formula:

where pi¯ is the modified prevalence for consumption state i, and state 0 corresponds to abstention.

In the model, alcohol consumption in a population subgroup is described non-parametrically by the associated observations from the HSE. For any harmful outcome, risk levels are associated with consumption level for each of the observations (note that these are not person-level risk functions). The associated prevalence for the observation is simply defined by its sample weight from the survey. Therefore, the PIF is implemented in the model as:

where:

• w_i is the weight for observation i

• RRi¯ is the modified risk for the new consumption level

• N is the number of samples.

Applying potential impact fractions for health outcomes

The impact of changes in alcohol consumption on health harms was examined using three different implementations of the PIF, depending on the health condition type:

1. direct application of consumption measures to calculate PIFs for wholly attributable acute and chronic conditions

2. relative risk functions from the published literature for partial chronic conditions

3. occasion-level relative risk functions from the published literature converted to annualised risks for partially attributable acute conditions (i.e. injuries).

Wholly attributable conditions

As wholly attributable chronic and acute conditions, by definition, have an AAF of 1 and, therefore, the relative risk cannot be defined, the relative risk term in Equation 7 is replaced with alcohol consumption that is likely to lead to increased risk for the health condition, denoted by RiskAlc_i, in the revised Equation 8:

For wholly attributable chronic conditions, RiskAlc_i is defined as the difference between mean daily consumption and a lower threshold below which risk is assumed to be equivalent to that of abstainers. In line with previous studies, these thresholds were assumed to be four and three units per day for men and women, respectively. 51 Below these thresholds, RiskAlc_i is assumed to be 0. An equivalent approach is used for wholly attributable acute conditions, for which RiskAlc_i is defined as the difference between peak day consumption (the response that individual HSE respondents give to the question ‘on the day in the last week when you drank the most, how much did you drink?’) and the cut-off thresholds of four/three units for men/women at which we assume the acute risk starts to increase, or 0 if peak day consumption is below the threshold.

Partially attributable chronic conditions

The relative risk functions for all chronic conditions that are partially attributable to alcohol are taken from published meta-analyses and are used in Equation 7. The sources for these risk functions and their corresponding curves are illustrated in Angus et al. 52

Ischaemic heart disease represents a special case in SAPM version 4.0 as this is the only condition for which a literature-based risk function is adjusted to reflect additional evidence. Although many studies have found that drinking at low levels reduces the risk of ischaemic heart disease relative to abstainers,53 it has been demonstrated that this reduced risk is substantially attenuated or eliminated for those engaging in heavy episodic drinking (defined as consuming at least 7.5 units on a single day) at least once a month. 54 In the SAPM, this evidence is incorporated using a method employed by Shield et al. ,55 whereby any protective effect is removed (i.e. the relative risk is assumed to be at least 1) for any drinkers who consume > 7.5 units per day, on average (52.5 units per week). This limited adjustment means that cardioprotective effects are likely to be overestimated in the model, as many individuals with a mean consumption of < 7.5 units per day probably drink this amount at least once a month. The most recent epidemiological risk functions for mortality suggest that protective effects disappear at below 7.5 units per day56 and, therefore, this adjustment has no effect. However, the evidence for morbidity outcomes is older and suggests that protective effects exist above a mean consumption level of 7.5 units per day;53 these are therefore removed in line with the newer evidence on heavy episodic drinking.

Partially attributable acute conditions

Partially attributable acute conditions include various traffic and non-traffic injuries. The identified relative risk functions for these conditions57 are different from the relative risk functions for partially attributable chronic conditions and cannot be used directly in Equation 4. The input and outcome of the relative risk functions for partially attributable chronic conditions are usual alcohol consumption and relative risk over a certain period of time, respectively; however, the input and outcome of the identified relative risk functions for traffic and non-traffic injuries are levels of drinking on the occasion prior to the injury and the relative risk for the drinking occasion, respectively. 58 As the SAPM version 4.0 works on annual cycles, relative risk in Equation 7 is defined as annual relative risk. Therefore, to apply Equation 7, single drinking occasion-based relative risk needs to be converted to long-term (e.g. annual) relative risk of an individual in the survey.

A method to estimate annualised relative risk of alcohol-attributable traffic- and non-traffic injuries has been developed. A detailed description of the method can be found elsewhere. 59,60 Briefly, three measures are defined to represent drinking pattern based on single drinking occasions, which are the frequency of drinking occasions (defined as n, or number of drinking occasions per week), the mean level of alcohol consumption for a given drinking occasion (defined as µ, or units of alcohol) and the variability of alcohol consumption for a given drinking occasion (defined as σ, or standard deviation of units of alcohol consumed on drinking occasions). Using the ONS National Diet and Nutrition Survey,60 regression models were fitted to relate the three measures with mean consumption and a range of independent variables (e.g. age, gender, education and ethnicity). These regression models are used to impute the three measures for each individual in the HSE. For each individual, alcohol consumption on a given drinking occasion is assumed to follow a normal distribution, with mean of µ and standard deviation of σ; and the duration of intoxication for a given drinking occasion is calculated by applying the equation for estimating blood alcohol content. Finally, a series of integrations were performed to calculate the annualised relative risk for traffic and non-traffic accidents.

The annualised relative risk obtained from this calculation process is used in Equation 7 to estimate the PIF for partially attributable acute conditions.

Conditions defined in the ICD-10

Table 6 lists the conditions that are in the SAPM version 4.0 that are defined in the ICD-10.

Mortality data

For each of the 45 health conditions listed in Table 6, we obtained mortality counts, further stratified by age (four categories: 18–24, 25–34, 35–54 and 55–89 years), sex and socioeconomic status (five quintiles of the IMD based on the deceased’s home postcode), for every one of the 151 UTLAs in England from the ONS. To address year-on-year variance in small counts, we pooled data across the years 2012–16 and converted to mortality rates per 100,000, using population estimates for the relevant years from the ONS. 61

Hospitalisation data

For each of the 45 health conditions listed in Table 6, we calculated admission rates, stratified by age, sex, socioeconomic status and UTLA, from individual patient records taken from HES data for England. As each individual admission can have multiple ICD-10 diagnostic codes, each admission was coded to a single condition using the ‘broad measure’ approach, which is recommended by Public Health England (PHE) as the most appropriate measure ‘of the total burden that alcohol has on community and health services’62 (contains public sector information licensed under the Open Government Licence v3.0). To prevent double-counting, admissions within 1 year for the same individual were linked; if there were multiple admissions coded to different conditions in a single year, only the admissions for one condition were retained. For a full description of this process, see Jones et al. 63 From this process, two outcomes were produced:

1. Morbidity rates per 100,000 (derived using ONS population estimates) for each health condition and age–sex–socioeconomic status subgroup, after removing all repeat admissions. That is, these rates represent the rate of individuals with each condition in each subgroup. As with the mortality data, year-on-year variance in small counts was addressed by calculating rates based on pooled data for 2012/13–2016/17.

2. The average number of admissions per year for somebody with at least one admission by condition and age group. These figures were used to estimate the number of hospital admissions that were associated with the modelled morbidity prevalence and changes in this prevalence.

Mortality model structure

A simplified version of the model structure for mortality is presented in Figure 11. The model is developed to represent the population of England in a life table. Separate life tables have been implemented for males and females.

FIGURE 11.

Simplified mortality model structure.

The life table is implemented as a linked set of simple Markov models with individuals of age a transitioning between two states – alive or dead – at model time step t. A Markov model is a state-transition model in which individuals can exist in a set number of states at any time period and transition between states using a set of transition probabilities, which are conditional on the current state of the individual. Those of age a still alive after the transition then form the initial population for age a + 1 at time t + 1, and the sequence repeats.

The transition probabilities from the alive to dead state are broken down by condition and are individually modified via PIFs over time t, where the PIF essentially varies with consumption over time:

where:

• PIF_t is the PIF relating to consumption at time t

• i = survey sample number

• N = number of samples in subgroup i

• RR_i,t is the risk relating to the consumption of SHeS sample i at time t

• RR_i,0 is the risk at baseline

• w_i is the weight of sample i.

Note that the PIF can be decomposed to enable different population groups at baseline (e.g. moderate, increasing-risk and high-risk drinkers or individuals in poverty and not in poverty) to be followed separately over the course of the model.

The model computes mortality results for two scenarios (a baseline – implemented as ‘no change to consumption’ in the analysis herein – and an intervention). The effect of the intervention is then calculated as the difference between the life table for the intervention scenario and the life table for the baseline scenario, enabling the change in the total expected deaths attributable to alcohol as a result of the policy to be estimated.

Outcomes from mortality modelling are expressed in terms of life-years saved. Morbidity valuation is the purpose of a second model described below.

Morbidity model structure

A simplified schematic of the morbidity model is shown in Figure 12. The model focuses on the expected disease prevalence for population cohorts. Note that if an incidence-based approach were used instead, then much more detailed modelling of survival time, cure rates, death rates and, possibly, disease progression for each disease for each population subgroup would be required.

FIGURE 12.

Simplified morbidity model structure. QALY, quality-adjusted life-year.

The morbidity model works by partitioning the alive population at time t, rather than using a transition approach between states, as previously described for the mortality model. Alive individuals are partitioned between all 45 alcohol-related conditions (and a 46th condition representing overall population health, not attributable to alcohol).

As in the mortality model, the PIF is calculated based on the consumption distribution at times 0 and t. The PIF is then used to modify the partition rate (i.e. the distribution of the 45 conditions for alive individuals) to produce person-specific sickness volumes. These volumes form the basis for estimating both health service costs and health-related quality of life.

Time lag effects for chronic harms

When modelling the link between consumption and harm, one important input is the assumption surrounding the ‘time lag’: the time needed to achieve the full benefit (reduction in harms) associated with a reduction in consumption. Such data are necessary for chronic conditions whereby the development of diseases often occurs over many years.

Following a 2012 systematic review,64 the SAPM version 4.0 incorporates lag structures for all chronic harms based on the best available published evidence to estimate the temporal relationship between changes in consumption and changes in risk of harm (see table 2 in Holmes et al. 64 for full details of these relationships as implemented in the model). In line with the findings of this review, health outcomes are reported at 20 years as ‘full effect’, as this is the time at which the full impact of a change in consumption on health will have occurred. Note that this 20-year time horizon for the model means that all outcomes are reported over the 20 years subsequent to the model baseline year (2017).

NHS costs used

Estimates of the costs associated with morbidity for each health condition were taken from previously published estimates of the full NHS cost of each condition,51 inflated to 2017 prices using the Hospital and Community Health Service index. 65 The final costs used in the models are presented in Table 7.

Data and methods used to model crime

Time horizon and discounting

In view of evidence suggesting that there may be a lag of up to 20 years between changes in alcohol consumption and changes in the risk of health harms,64 the time horizon of the model is 20 years. This allows the results presented to reflect the full effect of the policy on health outcomes. In line with the NICE guidelines on economic evaluations,71 all costs are discounted at 3.5%.

The assumptions on the impact of a minimum unit price intervention on the price distributions

We assume that prices of products currently sold at < 50p per unit rise to exactly 50p per unit. We assume that other prices remain unchanged.

Geographies analysed

We used the UTLA-level data described previously to produce the SAPMLA versions for each UTLA in the North West (23 UTLAs) and North East (12 UTLAs) of England. The process used to generate UTLA-level inputs was repeated at GOR level, and these inputs were used to create further models for each of the nine GORs.

Analysis of local burden of harm (comparison of current harm versus estimated harm if there were zero alcohol consumption)

For each geography, the current level of harm that is attributable to alcohol is estimated by comparing the data on current total level of harm (due to alcohol and other causes) with the results of a model run assuming that the population stopped drinking entirely.

Appraising the impact of 30p, 40p, 50p, 60p and 70p per unit minimum unit pricing policies

The models were then used to appraise the impact of 30p, 40p, 50p, 60p and 70p per unit minimum unit pricing policies. Finally, results for the nine GORs were combined to estimate the overall impact of each policy at the national level. The outputs from each model were the changes in alcohol consumption, changes in spending on alcohol, resulting changes in revenue to retailers and the exchequer, changes in alcohol-related hospital admissions and deaths, reductions in NHS costs, changes in annual volumes of crime and the associated costs.

Analysis of the slope index of inequality

All outputs were stratified by drinker group (moderate, increasing risk and high risk) and IMD quintile. The slope index of inequality16 (a regression approach to estimate the absolute difference between the highest and lowest percentiles in the population) was calculated for alcohol-attributable mortality both before and after each policy.

Two-sided evidence briefing for Sefton

The University of Sheffield logo has been reproduced with permission (University of Sheffield, 2021, personal communication).

Two-sided evidence briefing for North West Government Office Region as a whole