Impact of minimum unit pricing on alcohol-related emergency department attendances in Scotland: a natural experiment study

Background Minimum Unit Pricing (MUP) was introduced in Scotland at 50p per unit (8g) of alcohol on 1st May 2018 to reduce alcohol consumption and associated harms. We assessed its impact on alcohol-related emergency department (ED) attendances, drinking patterns, and having an alcohol-related diagnosis amongst ED attendees. Methods We used a natural experiment approach to compare outcomes between Scotland (intervention group) and England (comparison group). Two EDs in Scotland and two in Northern England were recruited for one baseline and two post-intervention waves during selected weekday and weekend hours. Research nurses considered all attendees for interview, and recorded reasons for not interviewing attendees. The primary outcome was alcohol-related attendances among all recorded attendees. Secondary outcomes included alcohol-related diagnosis, binge drinking and high-risk drinking, and tested for differential effects across socioeconomic groups. Differencein-difference regression models adjusted for age, sex and baseline covariates. Findings 12,207 participants were recruited in Scotland and 11,248 in Northern England. The odds ratio for an alcohol-related attendance was 1.14 (95% CI 0.90-1.44) after the introduction of MUP in Scotland relative to Northern England, after controlling for covariates. It is estimated that an additional 1.0% (95% CI -0.7% to 2.7%) of the ED attendances were alcohol-related than would have been the case in the absence of MUP. Meanwhile, the odds for an attendee having at least one alcohol-related diagnosis increased after MUP (OR=1.25, 95%CI 1.00-1.57). There was no evidence of substantive differences in the majority of other secondary outcomes after the introduction of MUP in Scotland, or of differential effects across socioeconomic groups. Interpretation We found no evidence that MUP impacted on alcohol-related ED attendances, suggesting that the underlying price may not have been high enough. Funding NIHR, MRC, CSO Word limits: 293/250 This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Pr ep rin t n ot pe r r ev ie w ed Research in context Evidence before this study Excessive alcohol consumption is a major cause of disease and death across the world. In the European context, Scotland, where the real price of alcohol has reduced over recent years, is particularly badly affected. There is a dose-response relationship between the alcohol price and the amount consumed. We carried out an initial narrative literature review in 2012 when Minimum Unit Pricing was first considered in Scotland, and updated our review in 2020. We searched Medline, Psychinfo and Google Scholar for papers on alcohol and minimum unit price. Although similar interventions have been implemented elsewhere (e.g. Canada, Russia) the evidence for MUP’s impacts on health specifically, as opposed to minimum pricing policies in general (which have often set differing minimum prices based on beverage type), came only from modelling studies, and these showed MUP was the most effective pricing policy for public health. The only empirical study to date has shown a fall in consumption following MUP in Scotland. The level for MUP at 50p per unit of alcohol was set in 2012 based on the modelling, and retained without adjustment for inflation after consultation in 2017. Added value of this study This is the first evaluation of the national implementation of MUP based on pure alcohol content to evaluate its impacts on alcohol-related emergency department (ED) attendances, drinking patterns, and alcohol-related diagnosis amongst ED attendees. We found no clear evidence in the ED setting that MUP at a level of 50p per unit of alcohol reduced alcohol-related attendances. Similarly, there was no evidence for a consistent effect on different age, sex and socioeconomic population subgroups. Implications of all the available evidence We found no evidence in the ED context that a 50p MUP provides health benefits or harm in Scotland after a one year period. Despite that, if other forthcoming evidence shows MUP improves health in other settings, in combination with recent evidence of reductions in alcohol sales following MUP in Scotland, it would suggest MUP may be worth retaining. We consider our findings to likely reflect the nature of harms within the ED setting and during the relatively short time period studied. Therefore, there may indeed be no effect on ED attendances for MUP at the 50p per unit level. The implication is that the price per unit for MUP should be raised and then further evaluated. Modelling certainly suggested greater effect at an increased price level, so it would be logical to test whether that holds in the real world. Finally, there may be further lessons here for the design of policy and associated evaluations to maximise their chances of finding the clearest results and answers. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Pr ep rin t n ot p ee r r ev ie w ed Introduction Alcohol accounts for 2.8 million deaths every year, approximately 10% of all deaths worldwide 2016.1 Alcohol misuse not only affects public health, but also contributes to socioeconomic inequalities in health. The lowest socioeconomic groups are considered more likely to report extreme heavy drinking, and to experience greater alcohol harm for similar levels of alcohol consumption in comparison to higher socioeconomic groups.2 3 The Scottish Government has been implementing a range of strategies to reduce alcohol consumption, alcohol-related harms, and health inequalities.4 Minimum Unit Pricing (MUP) of alcohol was an innovative and highprofile component of a comprehensive alcohol strategy. There is considerable evidence of an inverse alcohol price-consumption relationship.5-7 These studies show that pricing policies are one of the most effective strategies to reduce alcohol consumption and the associated health harms.5-9 Data modelling suggested that MUP would be an effective policy for reducing alcohol consumption and associated health harms.5 6 10 Recent findings also show that the introduction of MUP has significantly reduced alcohol-related harms in the Northern Territory, Australia.11 The Sheffield Alcohol Policy Model10, in particular, indicated that MUP would be effective in targeting heavier drinkers with lower incomes. Therefore, health inequalities are likely to be reduced by the introduction of MUP. The Alcohol (Minimum Pricing) (Scotland) Bill was first introduced to the Scottish Parliament on 31 October 2011 and passed in May 2012. After a series of legal challenges from the alcohol industry, the UK Supreme Court confirmed that the MUP legislation was lawful and proportionate in November 2017. On 1 May 2018, Scotland became the first country to carry out a national implementation of a MUP for alcohol. Under the new legislation, the minimum price is set to be 50p per UK unit of pure alcohol (1 unit is 8g/10mL ethanol). Unlike the MUP policies in some Canadian provinces that introduced a minimum price for selling specific beverages or the policies in the Northern Territories in Australia that introduced a minimum price for per standard drink, the minimum price in Scotland is based purely on alcohol content without reference to beverage type. In 2016, the Sheffield Alcohol Policy Model estimated a 50p minimum unit price would be effective in reducing alcohol consumption in Scotland by 3.5% per year, particularly among harmful (7.0%) and hazardous drinkers (2.5%) who are at greatest risk of alcohol harms.10 Recent evidence suggests MUP has reduced population alcohol consumption in Scotland,12 but the impact on particular groups and on alcohol-related harm is less conclusive.13 14 Emergency Department (ED) attendances are likely to be sensitive to changes in alcohol-related harms as they reflect both acute and chronic health problems. There is only limited evidence regarding alcohol use disorder on ED, or the sensitivity of ED to detect policy changes. Therefore, we assessed the impact of MUP on alcohol-related ED attendances and drinking patterns amongst the ED attendees, and whether this varied by age, sex, and socioeconomic group. Methods Study design Since alcohol-related attendances to EDs that do not result in admission are not routinely, we collected primary data in EDs to examine changes in alcohol-related attendance and in patterns of alcohol consumption among attendees that occur as a result of MUP. We employed a repeated cross-sectional design to compare outcome measures between Scotland and Northern England as a natural experiment. Northern England was chosen as a comparison group as alcohol consumption levels, and culture are more similar to Scotland.15 16 The natural experiment was This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Pr ep rin t n ot p e r r ev i w ed the introduction of MUP in Scotland and we used Northern England as a control using regression analysis to compare changes since baseline. Setting We recruited one large hospital with an ED in each of four cities of comparable population size, two exposed to MUP in Scotland (Glasgow and Edinburgh), and two unexposed in Northern England (Liverpool and Sheffield). Data collection took place over three, three-week waves. Following the decision to implement MUP, there was time for a single baseline, taken as quickly as possible (February 2018) to minimise behaviour changes in anticipation of implementation. There were two postimplementation follow-ups, in September to October 2018 and February 2019. In each wave, data collection took place from 20:00 until 03:30 the following day from Thursday to Sunday, and from 09:00 to 16:30 on Monday to Wednesday. We also requested anonymised information (sex, age group, and diagnoses) collected routinely on all attendees over the three-week collection periods for each wave. Participants Trained research nurses considered all attendees for approach. Attendees who were clearly clinically inappropriate or unavailable were not approached, and therefore ineligible for the study. Research nurses used iPad to record the reasons for not approaching, sex and age group for those who were not approached. Attendees who were approached by research nurses were then given written information about the study and had up to 40 minutes to decide whether to take part. Face-to-face structured interviews were carried out by research nurses using iPad. There was a formal screening where the approached attendees were asked eligibility questions before consent was taken. The eligibility criteria were: age ≥ 16 years, able to speak English or interpreter available, a new ED presentation during that shift, conscious, well enough (physically and mentally), sober enough (alcohol or drugs), still in the department for interview (i.e. had not left or been admitted), and safe for staff to approach. Eligible respondents were then asked to sign their consent on an iPad, and whether they further consented to linkage of their hospital notes to the interview data. For respondents who consented to the data linkage, we requested date of birth, full postcode, and diagnoses. More detail about reasons for not being approached, interviews being terminated, and failing the inclusion criteria can be found in Appendix 1. Variables Exposures and outcomes Exposure to MUP was defined as living in Scotland after the introduction of MUP. We, therefore, considered attendees in Scottish EDs were exposed to MUP from Wave 2 onwards and not in Wave 1. On the other hand, attendees in Northern England were not exposed to MUP at any wave. The primary outcome of interest was alcohol-related attendances among attendees who were recorded by research nurses through either observation or interview. An attendance was alcohol-related if the attendee was not eligible for interview owing to alcohol intoxication (for those who were not approached by research nurses or those who terminated the interview), or if the respondents reported binge-drinking (≥ 6/8 units for women/men) in the last 24 hours, or self-reported the attendance was alcohol-related due to their own or another’s drinking. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Pr ep ri t ot e r r ev i w ed We analysed alcohol-related diagnosis as a secondary outcome. The anonymised data requested from hospitals allowed us to examine all attendees during the three study periods. A diagnosis was alcohol-related if attributable to alcohol consumption according to the definition used by NHS Health Scotland.17 Appendix 2 lists the alcohol-related conditions which are based on International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10) codes. 18 We further examined secondary outcomes among respondents who completed the interviews. Three dichotomous outcomes were assessed: current alcohol use, binge-drinking in the past week, and binge-drinking in the past 24 hours. Then, we examined three other secondary outcomes among respondents who were current drinkers. These outcomes were FAST 19 20 (FAST Alcohol Screening Test) score as a continuous measure, alcohol misuse (FAST score 3+), and increased alcohol use in the past year as two dichotomous outcomes. Covariates Our primary outcome focuses on attendees who were recorded by research nurses through either observation or interview. Research nurses recorded sex and age group for attendees based on their observation. This information allowed us to adjust for sex and age group in the analysis of the primary outcome. The anonymised data from the hospitals contained information about sex and age group of all attendees. Therefore, we adjusted for sex and age group in the analysis of alcohol-related diagnosis. The questionnaire covered sociodemographic data, including sex, age, ethnicity, employment status, marital status and housing ownership. Area-based deprivation scores were assigned to each interviewee based on their postcode of residence. We used 2011 Carstairs area deprivation scores21 calculated for wards in England and postcode sectors in Scotland.22 This gave geographies with similarly sized populations and so a measure of deprivation comparable across all four EDs and the two countries. In Scotland, postcode sectors were sometimes split between two Carstairs deciles where a postcode covered two councils. We used a population weighting method to assign a Carstairs score to the whole postcode dependent on the population split between the councils. These variables were used as covariates when we analysed secondary outcomes. Statistical analysis We evaluated the impact of the implementation of MUP by fitting fixed-effect multivariate regression models. For our primary analysis, we fitted the following models: Model 1: y = β0 + β1MUP + β2country + β3time + ε Model 2: y = β0 + β1MUP + β2hospital + β3wave + ε Model 3: y = β0 + β1MUP + β2hospital + β3wave + β4covariates + ε where is the outcome variable, is the residual, and is a dichotomous indicator with the y ε MUP value 1 for attendees who attended Scotland EDs after the implementation of MUP, and 0 otherwise. Our coefficient of interest is , the difference-in-difference (DID) estimate, which is β1 defined as the differences in outcome between Scotland and England before and after the introduction of MUP. We used logistic regression for binary outcomes, and linear regression for continuous outcome. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Pr ep rin t n ot p er r vi ew ed Model 1 is the unadjusted model with only the DID estimate, and fixed-effects for country (0 = England and 1 = Scotland) and time (0 = before the introduction of MUP and 1 = after the introduction of MUP). The country fixed-effects control for all unobserved country-specific factors that are time-invariant, while the time fixed-effects account for seasonal effects over time. In Model 2, we further adjusted for hospital (0 = Edinburgh ED, 1 = Glasgow ED, 2 = Liverpool ED and 3 = Sheffield ED) and wave (0 = Wave 1, 1 = Wave 2 and 3 = Wave 3) fixedeffects. Since the country and time fixed-effects in the unadjusted model were confounded with the newly included hospital and wave, we omitted them from Model 2. In Model 3, the final model, we further included a set of covariates: sex, age group, ethnicity, employment status, marital status, housing ownership, and Carstairs. Appendices 3 and 4 give the percentage of missing data for each demographic and outcome variable by country and wave. We imputed all variables in the dataset (except the anonymised dataset requested from hospitals) using multiple imputation. A total of 20 imputed datasets were created and analysed in R using the MICE package.23 The parameters of interest were estimated in each imputed dataset separately, and combined using Rubin’s rules. We included non-response weights in the imputation process and regression models. Using the anonymised information for all attendees from the hospitals, we calculated inverse probability weights to account for the differences in distribution of sex and age group between interviewees and attendees. We undertook various sensitivity analyses to investigate whether our results were sensitive to the model specification. To examine whether our findings were sensitive to the FAST cut-off score, we also analysed the effect of MUP against FAST cut-offs of 2+ (hazardous drinker), 4+ (harmful drinker) and 6+ (dependent drinker). These cut-offs were validated using data from the Adult Psychiatric Morbidity Survey 2007.24 We replicated the analyses on alcohol-related attendance (primary outcome), and alcohol-related diagnosis (secondary outcome) using the sample based on all interviewees by including ethnicity, employment status, marital status, housing ownership, and Carstairs as covariates. Finally, we also performed the weighted and unweighted analysis on the complete cases. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. VS and ADM had full access to all the data in the study. All authors had final responsibility for the decision to submit for publication. Results Descriptive of sample A total of 26,969 attendees aged at least 16 years visited the EDs during the three study periods, and 23,455 (87.0%) of them were recorded by research nurses. Among those who were recorded, 14,047 (59.9%) of them were approached and 12,249 were identified to be eligible to participate in the study, of whom 8,746 (71.4%) completed the interview. Figure 1 illustrates the flowchart which summarises the study participants in all four EDs and three waves. We calculated two response rates: the realistic response rate uses a denominator of all eligible attendees, and the absolute response rate uses all recorded attendees as the denominator. Table 1 presents both response rates by wave and hospital. The response rates in Scotland were generally higher than those in England. The overall realistic response rates decreased over the three waves from 78.0% in Wave 1 to 71.6% in Wave 2, and 66.5% in Wave 3. Across three This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Pr ep rin t n ot p er r vi ew ed waves, Liverpool had the lowest realistic response rate (60.8%) among four hospitals. Meanwhile, Sheffield had the lowest absolute response rate (27.9%). Table 1 Summary of response rates by wave and hospital Wave 1 Wave 2 Wave 3 Overall Edinburgh Realistic response rate 81.1% 72.9% 73.9% 75.6% Absolute response rate 44.2% 44.2% 45.5% 44.6% Glasgow Realistic response rate 81.3% 77.5% 77.7% 78.7% Absolute response rate 40.3% 38.1% 36.1% 38.0% Liverpool Realistic response rate 72.3% 63.2% 53.0% 60.8% Absolute response rate 36.7% 42.6% 39.5% 39.9% Sheffield Realistic response rate 74.4% 73.6% 61.0% 69.1% Absolute response rate 27.8% 30.2% 25.9% 27.9% Overall Realistic response rate 78.0% 71.6% 66.5% 71.4% Absolute response rate 37.0% 38.4% 36.4% 37.3% We performed Pearson’s chi-square test to compare the sex and age differences between respondents (those who completed the interview) and all attendees (see Table 2). The differences between waves were small for sex but there were greater differences for age groups. Despite these differences, inverse probability weights were applied in all analysis models. Table 2 Summary of Pearson’s chi-square test between survey respondents and sampling frame Wave 1 Wave 2 Wave 3 Overall χ2 pvalue χ2 pvalue χ2 pvalue χ2 pvalue Edinburgh Sex 4.8 0.028 6.6 0.010 1.0 0.315 11.0 0.001 Age 13.1 0.005 27.9 0.000 7.6 0.054 43.7 0.000 Glasgow Sex 1.2 0.267 0.0 0.992 0.7 0.419 1.3 0.251 Age 69.1 0.000 43.5 0.000 29.3 0.000 132.5 0.000 Liverpool Sex 1.1 0.295 1.1 0.298 0.0 0.945 1.2 0.267 Age 3.7 0.295 23.1 0.000 10.2 0.017 29.8 0.000 Sheffield Sex 0.7 0.390 1.2 0.277 1.9 0.168 0.1 0.724 Age 7.7 0.052 15.7 0.001 21.8 0.000 37.2 0.000 Overall Sex 3.5 0.060 2.1 0.143 0.0 0.847 4.2 0.041 Age 53.9 0.000 82.9 0.000 55.6 0.000 189.1 0.000 Descriptive statistics The demographic characteristics of all attendees, attendees who were recorded by nurse interviewers, and those who completed the interview are shown in Table 3. The analysis for the primary outcome focused on the sample of recorded attendees (n=23,455). Meanwhile, the analytic sample for alcohol-related diagnosis was based on all attendees. A total of 8,746 attendees completed the interview. We excluded those who lived outside Scotland and England (n=20) and non-UK residents (n=39). As a result, 8,687 respondents were included in the analytic sample for the following secondary outcomes: current alcohol use, binge-drinking in the past week, and binge-drinking in the past 24 hours. The remaining three secondary outcomes (FAST score, alcohol misuse, binge-drinking at least weekly, and increased alcohol use in the past year) were based on respondents who were current drinkers (N=6,991). Although there are some slight differences in the demographic distribution between the Scottish and English samples, we accounted for these in our difference-in-difference analysis. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Pr ep rin t n ot p ee r r vi ew ed Table 3 Demographic characteristics of samples All attendees Attendees recorded by research nurses Respondents Scotland (N=14,051) England (N=12,918) Scotland (N=12,207) England (N=11,248) Scotland (N=5,059) England (N=3,628) Sex Female 7,212 (51.3%) 6,552 (50.7%) 6,131 (50.2%) 5,634 (50.1%) 2,483 (49.1%) 1,854 (51.1%) Male 6,837 (48.7%) 6,366 (49.3%) 6,015 (49.3%) 5,499 (48.9%) 2,574 (50.9%) 1,774 (48.9%) Non-binary 2 (0.0%) 0 (0.0%) 2 (0.0%) 0 (0.0%) 2 (0.0%) 0 (0.0%) Missing 0 (0.0%) 0 (0.0%) 59 (0.5%) 115 (1.0%) 0 (0.0%) 0 (0.0%) Age 16-25 2,509 (17.9%) 2,725 (21.1%) 2,450 (20.1%) 2,210 (19.6%) 1,137 (22.5%) 861 (23.7%) 26-45 4,211 (30.0%) 3,830 (29.6%) 3,769 (30.9%) 3,119 (27.7%) 1,613 (31.9%) 1,146 (31.6%) 46-65 3,832 (27.3%) 3,081 (23.9%) 3,155 (25.8%) 2,571 (22.9%) 1,352 (26.7%) 901 (24.8%) 66+ 3,499 (24.9%) 3,251 (25.2%) 2,762 (22.6%) 2,846 (25.3%) 957 (18.9%) 720 (19.8%) Missing 0 (0.0%) 31 (0.2%) 71 (0.6%) 502 (4.5%) 0 (0.0%) 0 (0.0%) Ethnicity White 4,717 (93.2%) 3,172 (87.4%) Non-white 325 (6.4%) 438 (12.1%) Missing 17 (0.3%) 18 (0.5%) Employ status Employed 2,590 (51.2%) 1,690 (46.6%) Economically inactive 1,938 (38.3%) 1,479 (40.8%) Unemployed 498 (9.8%) 431 (11.9%) Missing 33 (0.7%) 28 (0.8%) Marital status Married/Co-habiting 2,116 (41.8%) 1,453 (40.0%) Separated/Divorced/Widowed 770 (15.2%) 547 (15.1%) Single 2,097 (41.5%) 1,588 (43.8%) Missing 76 (1.5%) 40 (1.1%) Housing ownership Owner Occupied 1,917 (37.9%) 1,285 (35.4%) Rented 1,306 (25.8%) 1,207 (33.3%) Housing Association/Council 888 (17.6%) 446 (12.3%) This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Pre pri nt no t p ee r re vie we d All attendees Attendees recorded by research nurses Respondents Scotland (N=14,051) England (N=12,918) Scotland (N=12,207) England (N=11,248) Scotland (N=5,059) England (N=3,628) Other 881 (17.4%) 627 (17.3%) Missing 67 (1.3%) 63 (1.7%) Carstairs Mean (SD) 7.06 (2.60) 7.37 (2.54) Median [Min, Max] 8.00 [1.00, 10.0] 8.00 [1.00, 10.0] Missing 54 (1.1%) 166 (4.6%) This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Pre pri nt no t p ee r re vie we Figure 1 Flowchart of study participants


Findings
,207 participants were recruited in Scotland and 11,248 in Northern England. The odds ratio for an alcohol-related attendance was 1.14 (95% CI 0.90-1.44) after the introduction of MUP in Scotland relative to Northern England, after controlling for covariates. It is estimated that an additional 1.0% (95% CI -0.7% to 2.7%) of the ED attendances were alcohol-related than would have been the case in the absence of MUP. Meanwhile, the odds for an attendee having at least one alcohol-related diagnosis increased after MUP (OR=1.25, 95%CI 1.00-1.57). There was no evidence of substantive differences in the majority of other secondary outcomes after the introduction of MUP in Scotland, or of differential effects across socioeconomic groups.

Interpretation
We found no evidence that MUP impacted on alcohol-related ED attendances, suggesting that the underlying price may not have been high enough.

Introduction
Alcohol accounts for 2.8 million deaths every year, approximately 10% of all deaths worldwide 2016. 1 Alcohol misuse not only affects public health, but also contributes to socioeconomic inequalities in health. The lowest socioeconomic groups are considered more likely to report extreme heavy drinking, and to experience greater alcohol harm for similar levels of alcohol consumption in comparison to higher socioeconomic groups. 2 3 The Scottish Government has been implementing a range of strategies to reduce alcohol consumption, alcohol-related harms, and health inequalities. 4 Minimum Unit Pricing (MUP) of alcohol was an innovative and highprofile component of a comprehensive alcohol strategy.
There is considerable evidence of an inverse alcohol price-consumption relationship. [5][6][7] These studies show that pricing policies are one of the most effective strategies to reduce alcohol consumption and the associated health harms. [5][6][7][8][9] Data modelling suggested that MUP would be an effective policy for reducing alcohol consumption and associated health harms. 5 6 10 Recent findings also show that the introduction of MUP has significantly reduced alcohol-related harms in the Northern Territory, Australia. 11 The Sheffield Alcohol Policy Model 10 , in particular, indicated that MUP would be effective in targeting heavier drinkers with lower incomes. Therefore, health inequalities are likely to be reduced by the introduction of MUP.
The Alcohol (Minimum Pricing) (Scotland) Bill was first introduced to the Scottish Parliament on 31 October 2011 and passed in May 2012. After a series of legal challenges from the alcohol industry, the UK Supreme Court confirmed that the MUP legislation was lawful and proportionate in November 2017. On 1 May 2018, Scotland became the first country to carry out a national implementation of a MUP for alcohol. Under the new legislation, the minimum price is set to be 50p per UK unit of pure alcohol (1 unit is 8g/10mL ethanol). Unlike the MUP policies in some Canadian provinces that introduced a minimum price for selling specific beverages or the policies in the Northern Territories in Australia that introduced a minimum price for per standard drink, the minimum price in Scotland is based purely on alcohol content without reference to beverage type. In 2016, the Sheffield Alcohol Policy Model estimated a 50p minimum unit price would be effective in reducing alcohol consumption in Scotland by 3.5% per year, particularly among harmful (7.0%) and hazardous drinkers (2.5%) who are at greatest risk of alcohol harms. 10 Recent evidence suggests MUP has reduced population alcohol consumption in Scotland, 12 but the impact on particular groups and on alcohol-related harm is less conclusive. 13 14 Emergency Department (ED) attendances are likely to be sensitive to changes in alcohol-related harms as they reflect both acute and chronic health problems. There is only limited evidence regarding alcohol use disorder on ED, or the sensitivity of ED to detect policy changes. Therefore, we assessed the impact of MUP on alcohol-related ED attendances and drinking patterns amongst the ED attendees, and whether this varied by age, sex, and socioeconomic group.

Study design
Since alcohol-related attendances to EDs that do not result in admission are not routinely, we collected primary data in EDs to examine changes in alcohol-related attendance and in patterns of alcohol consumption among attendees that occur as a result of MUP. We employed a repeated cross-sectional design to compare outcome measures between Scotland and Northern England as a natural experiment. Northern England was chosen as a comparison group as alcohol consumption levels, and culture are more similar to Scotland. 15 16 The natural experiment was This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d the introduction of MUP in Scotland and we used Northern England as a control using regression analysis to compare changes since baseline.

Setting
We recruited one large hospital with an ED in each of four cities of comparable population size, two exposed to MUP in Scotland (Glasgow and Edinburgh), and two unexposed in Northern England (Liverpool and Sheffield).
Data collection took place over three, three-week waves. Following the decision to implement MUP, there was time for a single baseline, taken as quickly as possible (February 2018) to minimise behaviour changes in anticipation of implementation. There were two postimplementation follow-ups, in September to October 2018 and February 2019. In each wave, data collection took place from 20:00 until 03:30 the following day from Thursday to Sunday, and from 09:00 to 16:30 on Monday to Wednesday.
We also requested anonymised information (sex, age group, and diagnoses) collected routinely on all attendees over the three-week collection periods for each wave.

Participants
Trained research nurses considered all attendees for approach. Attendees who were clearly clinically inappropriate or unavailable were not approached, and therefore ineligible for the study. Research nurses used iPad to record the reasons for not approaching, sex and age group for those who were not approached.
Attendees who were approached by research nurses were then given written information about the study and had up to 40 minutes to decide whether to take part. Face-to-face structured interviews were carried out by research nurses using iPad. There was a formal screening where the approached attendees were asked eligibility questions before consent was taken. The eligibility criteria were: age ≥ 16 years, able to speak English or interpreter available, a new ED presentation during that shift, conscious, well enough (physically and mentally), sober enough (alcohol or drugs), still in the department for interview (i.e. had not left or been admitted), and safe for staff to approach. Eligible respondents were then asked to sign their consent on an iPad, and whether they further consented to linkage of their hospital notes to the interview data. For respondents who consented to the data linkage, we requested date of birth, full postcode, and diagnoses. More detail about reasons for not being approached, interviews being terminated, and failing the inclusion criteria can be found in Appendix 1.

Variables Exposures and outcomes
Exposure to MUP was defined as living in Scotland after the introduction of MUP. We, therefore, considered attendees in Scottish EDs were exposed to MUP from Wave 2 onwards and not in Wave 1. On the other hand, attendees in Northern England were not exposed to MUP at any wave.
The primary outcome of interest was alcohol-related attendances among attendees who were recorded by research nurses through either observation or interview. An attendance was alcohol-related if the attendee was not eligible for interview owing to alcohol intoxication (for those who were not approached by research nurses or those who terminated the interview), or if the respondents reported binge-drinking (≥ 6/8 units for women/men) in the last 24 hours, or self-reported the attendance was alcohol-related due to their own or another's drinking.
We analysed alcohol-related diagnosis as a secondary outcome. The anonymised data requested from hospitals allowed us to examine all attendees during the three study periods. A diagnosis was alcohol-related if attributable to alcohol consumption according to the definition used by NHS Health Scotland. 17 Appendix 2 lists the alcohol-related conditions which are based on International Statistical Classification of Diseases and Related Health Problems 10 th Revision (ICD-10) codes. 18 We further examined secondary outcomes among respondents who completed the interviews. Three dichotomous outcomes were assessed: current alcohol use, binge-drinking in the past week, and binge-drinking in the past 24 hours. Then, we examined three other secondary outcomes among respondents who were current drinkers. These outcomes were FAST 19 20 (FAST Alcohol Screening Test) score as a continuous measure, alcohol misuse (FAST score 3+), and increased alcohol use in the past year as two dichotomous outcomes.

Covariates
Our primary outcome focuses on attendees who were recorded by research nurses through either observation or interview. Research nurses recorded sex and age group for attendees based on their observation. This information allowed us to adjust for sex and age group in the analysis of the primary outcome.
The anonymised data from the hospitals contained information about sex and age group of all attendees. Therefore, we adjusted for sex and age group in the analysis of alcohol-related diagnosis.
The questionnaire covered sociodemographic data, including sex, age, ethnicity, employment status, marital status and housing ownership. Area-based deprivation scores were assigned to each interviewee based on their postcode of residence. We used 2011 Carstairs area deprivation scores 21 calculated for wards in England and postcode sectors in Scotland. 22 This gave geographies with similarly sized populations and so a measure of deprivation comparable across all four EDs and the two countries. In Scotland, postcode sectors were sometimes split between two Carstairs deciles where a postcode covered two councils. We used a population weighting method to assign a Carstairs score to the whole postcode dependent on the population split between the councils. These variables were used as covariates when we analysed secondary outcomes.

Statistical analysis
We evaluated the impact of the implementation of MUP by fitting fixed-effect multivariate regression models. For our primary analysis, we fitted the following models: where is the outcome variable, is the residual, and is a dichotomous indicator with the value 1 for attendees who attended Scotland EDs after the implementation of MUP, and 0 otherwise. Our coefficient of interest is , the difference-in-difference (DID) estimate, which is 1 defined as the differences in outcome between Scotland and England before and after the introduction of MUP. We used logistic regression for binary outcomes, and linear regression for continuous outcome.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 Model 1 is the unadjusted model with only the DID estimate, and fixed-effects for country (0 = England and 1 = Scotland) and time (0 = before the introduction of MUP and 1 = after the introduction of MUP). The country fixed-effects control for all unobserved country-specific factors that are time-invariant, while the time fixed-effects account for seasonal effects over time. In Model 2, we further adjusted for hospital (0 = Edinburgh ED, 1 = Glasgow ED, 2 = Liverpool ED and 3 = Sheffield ED) and wave (0 = Wave 1, 1 = Wave 2 and 3 = Wave 3) fixedeffects. Since the country and time fixed-effects in the unadjusted model were confounded with the newly included hospital and wave, we omitted them from Model 2. In Model 3, the final model, we further included a set of covariates: sex, age group, ethnicity, employment status, marital status, housing ownership, and Carstairs.
Appendices 3 and 4 give the percentage of missing data for each demographic and outcome variable by country and wave. We imputed all variables in the dataset (except the anonymised dataset requested from hospitals) using multiple imputation. A total of 20 imputed datasets were created and analysed in R using the MICE package. 23 The parameters of interest were estimated in each imputed dataset separately, and combined using Rubin's rules.
We included non-response weights in the imputation process and regression models. Using the anonymised information for all attendees from the hospitals, we calculated inverse probability weights to account for the differences in distribution of sex and age group between interviewees and attendees.
We undertook various sensitivity analyses to investigate whether our results were sensitive to the model specification. To examine whether our findings were sensitive to the FAST cut-off score, we also analysed the effect of MUP against FAST cut-offs of 2+ (hazardous drinker), 4+ (harmful drinker) and 6+ (dependent drinker). These cut-offs were validated using data from the Adult Psychiatric Morbidity Survey 2007. 24 We replicated the analyses on alcohol-related attendance (primary outcome), and alcohol-related diagnosis (secondary outcome) using the sample based on all interviewees by including ethnicity, employment status, marital status, housing ownership, and Carstairs as covariates. Finally, we also performed the weighted and unweighted analysis on the complete cases.

Role of the funding source
The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. VS and ADM had full access to all the data in the study. All authors had final responsibility for the decision to submit for publication.

Results
Descriptive of sample A total of 26,969 attendees aged at least 16 years visited the EDs during the three study periods, and 23,455 (87.0%) of them were recorded by research nurses. Among those who were recorded, 14,047 (59.9%) of them were approached and 12,249 were identified to be eligible to participate in the study, of whom 8,746 (71.4%) completed the interview. Figure 1 illustrates the flowchart which summarises the study participants in all four EDs and three waves.
We calculated two response rates: the realistic response rate uses a denominator of all eligible attendees, and the absolute response rate uses all recorded attendees as the denominator. Table  1 presents both response rates by wave and hospital. The response rates in Scotland were generally higher than those in England. The overall realistic response rates decreased over the three waves from 78.0% in Wave 1 to 71.6% in Wave 2, and 66.5% in Wave 3. Across three This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d waves, Liverpool had the lowest realistic response rate (60.8%) among four hospitals. Meanwhile, Sheffield had the lowest absolute response rate (27.9%). We performed Pearson's chi-square test to compare the sex and age differences between respondents (those who completed the interview) and all attendees (see Table 2). The differences between waves were small for sex but there were greater differences for age groups. Despite these differences, inverse probability weights were applied in all analysis models.

Descriptive statistics
The demographic characteristics of all attendees, attendees who were recorded by nurse interviewers, and those who completed the interview are shown in Table 3. The analysis for the primary outcome focused on the sample of recorded attendees (n=23,455). Meanwhile, the analytic sample for alcohol-related diagnosis was based on all attendees.
A total of 8,746 attendees completed the interview. We excluded those who lived outside Scotland and England (n=20) and non-UK residents (n=39). As a result, 8,687 respondents were included in the analytic sample for the following secondary outcomes: current alcohol use, binge-drinking in the past week, and binge-drinking in the past 24 hours. The remaining three secondary outcomes (FAST score, alcohol misuse, binge-drinking at least weekly, and increased alcohol use in the past year) were based on respondents who were current drinkers (N=6,991). Although there are some slight differences in the demographic distribution between the Scottish and English samples, we accounted for these in our difference-in-difference analysis. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d Figure 2A shows the changes in the proportion of attendees with alcohol-related attendance in Scotland and England before and after the introduction of the MUP. On average, Scotland had a higher proportion of attendances that were alcohol-related than England. Scotland had a stable trend, while there was a decreasing trend in England. In contrast, England had a higher prevalence of alcohol-related diagnosis than Scotland ( Figure 2B). The proportion of attendees with at least one alcohol-related condition rose slightly in Scotland but fell in England.
Across waves, there was a slightly increasing trend in being a current alcohol drinker in both countries ( Figure 2C). Binge-drinking in the past week among all respondents increased slightly in Scotland but decreased in England ( Figure 2D). However, both countries showed a slight increase in binge-drinking in the past 24 hours across waves ( Figure 2E). The mean FAST score among drinkers increased in both Scotland and England ( Figure 2F). The proportion of alcohol misuse (FAST score 3+) increased in England, while Scotland had a relatively stable trend ( Figure 2G). Meanwhile, the proportion of drinkers who reported an increase in alcohol use in the past 12 months also had a stable trend in both countries ( Figure 2H). Figure 3 shows the DID estimates from the final regression models for our primary outcome and seven secondary outcomes (see Appendix 5 for the full regression models). There was no evidence of substantive differences in most outcomes after the introduction of MUP in Scotland. The odds ratio of an alcohol-related attendance was 1.14 (95% CI 0.90 to 1.44, p=0.272), indicating that there was little difference between Scotland and England before and after MUP was implemented in Scotland. Based on marginal analysis, it is estimated that an additional 1.0% (95% CI -0.7% to 2.7%) of the ED attendances were alcohol-related than would have been the case in the absence of MUP. We estimated that approximately 258 attendances at ED were alcohol-related as a result of the introduction of MUP (95% CI -191 to 707).
However, the DID estimates show that among all attendees, the odds for an attendee having at least one alcohol-related diagnosis increased by 25% relative to change observed in England after MUP (OR=1.25, 95%CI 1.00 to 1.57, p=0.046). Nevertheless, there was no effect on other secondary outcomes, suggesting that the introduction of MUP in Scotland did not substantially alter these outcomes in the population studied.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d Changes in primary and secondary outcomes across waves This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d Figure 3 Difference-in-difference estimates of the overall effects of MUP We further investigated the outcomes by sex, age group, ethnicity, employment status, marital status, and housing ownership. A Bonferroni correction was used to adjust the p-values for multiple comparison. Figure 4 shows the stratified results for the primary outcome. There was no evidence to show MUP had any differential effect across sex and age group. Full results for other secondary outcomes are given in Appendix 6. The stratified analysis shows the introduction of MUP in Scotland was associated with increased odds of alcohol-related diagnosis among men who attended the EDs (OR=1.56, 95% CI 1.16 to 2.11, p=0.004, Bonferroni-corrected p=0.021, Figure 5). Meanwhile, the stratified analyses on the remaining secondary outcomes did not show any differential effect across sociodemographic groups, after Bonferroni correction.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d Stratified analysis for primary outcome: Alcohol-related attendance Testing the robustness of our analysis, we analysed the effect of MUP against FAST cut-offs of 2+, 4+, and 6+, repeated the analysis on primary outcome using the sample based on survey respondents, and replicated the analysis using unweighted and weighted complete cases. All these analyses produced similar results (see Appendix 6). We also performed sensitivity analysis on alcohol-related diagnosis based on survey respondents who consented to data linkage. Results from the sensitivity analysis showed that the DID estimate was not significant at 5% level, whereas the main analysis showed a significant difference. Since the main analysis was based on all attendees while sensitivity analysis was based on respondents who consented to data linkage, we were confident that the main analysis was not subjected to any selection bias, and therefore, our results were also robust.

Key results
We examined the impact of MUP on alcohol-related ED visits and alcohol-related diagnosis among attendees. We also studied patterns of alcohol use among those who participated in the interview. Our results showed that MUP was only marginally associated with alcohol-related diagnosis and This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d there was no evidence that MUP had any effect on primary and other secondary outcomes. We similarly found no evidence of MUP having differential effects across socioeconomic groups, except for alcohol-related diagnosis. After correcting for multiple comparisons, we found that MUP was associated with increased odds of alcohol-related diagnosis only among male attendees.
There is limited research on the association between hospital admissions and the introduction of MUP. Existing studies have suggested MUP led to a reduction in alcohol-attributable hospital admissions and alcohol-related deaths, [25][26][27] but there was no immediate effect on ED attendances for alcohol-related injury in Canada. 28 These studies focus on alcohol-related admissions and attendances based on patients' diagnosed diseases and injuries. Our study, on the other hand, examined alcohol-related attendances based on the nurse interviewers' observations and attendees' self-reported behaviours. Given that the definitions of alcohol-related attendance were not comparable between our study and the Canadian studies, it may explain why our results were different from theirs. Our study also examined alcohol-attributable diagnosis among all ED attendees. We combined all chronic and acute alcohol-related conditions as a whole rather than acute injuries covered elsewhere. 28 Our data show that less than 1% of all attendees in both Scotland and England were diagnosed with partially acute injuries; as such, it was not possible to analyse acute partially alcohol-related diagnosis separately. However, our research provides further evidence that there was no immediate association between MUP and changes in the prevalence of alcohol-related diagnosis among ED attendees.
Our analyses have several important strengths. This is the first study to examine the association between MUP and alcohol-related attendances and alcohol-related diagnosis within the ED setting in Scotland. Diagnostic data on alcohol-related attendances that do not result in admissions are not routinely captured in administrative health data in both Scotland and England. In contrast to research which relies on hospitalisations data, our study is more sensitive in detecting alcoholrelated harms which result in ED attendance, including injury-related harms that are common among young people. Although we found a weak significant immediate association between MUP and alcohol-related diagnosis, the result echoes another study which examined alcohol-related injury ED visits in Canada. 28 It provides more evidence that MUP may be less likely to impact on harms that most commonly present to EDs, including those related to acute consumption among young people. However, the consequences on the broader range of alcohol-related harms remain unknown and it is therefore important to monitor how alcohol-related diagnosis would change in the longer run.
This study has some limitations. First, the definition of alcohol-related attendances for unapproached or ineligible attendees was based on nurse interviewers' observations only. Attendances were considered as alcohol-related only if the nurses recorded them as alcohol intoxicated, on the basis of the interview or for non-participants, interviewers' observation. As a result, we may have misclassified some survey non-participants who attended the ED because of another's drinking or underlying alcohol-related conditions that were not observable to interviewers. Our analysis is therefore likely to underestimate the association between alcoholrelated attendances and the introduction of MUP. Second, we were unable to test the parallel trend assumption when difference-in-difference analysis was used. The Scottish Government announced on 21 November 2017 that MUP would be implemented on 1 May 2018. It gave us limited time for data collection, therefore only one pre-MUP data time point was possible. However, other data suggest the prior trends in alcohol-specific deaths 29 , and alcohol-related hospital admission 15 16 in Scotland and Northern England since 2012 were broadly similar (see Appendix 7). These data provide some proxy information on alcohol-related ED attendances and alcohol drinking patterns in both countries to validate the parallel trend assumption. Third, we excluded one hospital from England when we analysed alcohol-related diagnosis among respondents who consented for data linkage. The data provided from this hospital did not allow us to convert to the ICD-10 diagnostic coding system on which the alcohol-attributable diagnosis was based. Therefore, we lost 1,368 cases (around 17.5% of total sample size) when we performed this analysis. That may have affected the statistical power.
The study protocol identified a number of potential risks arising from MUP: 30 1) displacement effects where reductions in alcohol-related harms may be accompanied by increases in other drug related harms; 2) increased alcohol-related harm through substitution or changed drinking patterns; 3) consumers may switch to alternative sources of alcohol not subject to MUP such that the price paid does not increase; and 4) MUP could unfairly penalise poorer drinkers who may be less able to absorb the additional costs and may also forgo other essentials such as food. Our results show that there was no evidence that alcohol-related harms increased within the ED setting as a result of the implementation of MUP which echoes the results from a previous Canadian study. 28 There was a six-year delay before the legislation was passed after the Alcohol (Minimum Pricing) (Scotland) Bill was first introduced. The gap between legislation being first introduced and its implementation has meant the magnitude of price changes has been relatively small. It might also explain why we were unable to detect any significant effects of MUP on alcohol-related harms and drinking patterns as it may not have been implemented at an adequate level. The underlying inflation rates may also devalue the 50p potentially and hence limiting the intended impact. MUP might have also increased public awareness of health harms relating to alcohol, and much of that could have happened around the time of legislation and during the legal challenges from the alcohol industry. Our study would not pick up such an effect due to the research design.
In summary, we did not find evidence for the introduction of MUP in Scotland impacting on alcoholrelated harms within the ED setting. However, the broader evidence base is more consistent with an effect of MUP on both alcohol consumption and harms. This study is part of a wider evaluation programme coordinated by Public Health Scotland to inform the decision by the Scottish Parliament as to whether they will vote for MUP to continue following the sixth year of implementation. Therefore, we should interpret the results with caution and should not draw conclusions regarding the wider societal impact of MUP on alcohol harm purely based on this study.

Demographics of survey respondents across three waves in Scotland and England
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d * One hospital from England was omitted from analysis as the hospital data provided by that hospital did not allow us to convert to the ICD-10 diagnostic coding system which the alcohol-related diagnoses are based on.
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3697993 P r e p r i n t n o t p e e r r e v i e w e d P r e p r i n t n o t p e e r r e v i e w e d