select search filters
briefings
roundups & rapid reactions
before the headlines
Fiona fox's blog
An observational study published in the Lancet looks at alcohol-related deaths and hospitalisations in Scotland since the minimum unit pricing for alcohol policy (MUP) was brought in by the Scottish government in 2018
This Roundup accompanied an SMC Briefing.
Dr Sadie Boniface, Head of Research, Institute of Alcohol Studies; and Visiting Researcher, King’s College London, said:
“The Lancet and PHS press releases both accurately reflect the findings of this important study, which adds to a large and consistent evidence base on the effectiveness of alcohol pricing policies.
“This is high quality research using official data on deaths and hospitalisations, and the main finding of a 13% reduction in deaths from alcohol is highly significant, in both health and statistical terms. The authors appropriately use England as a control group and have addressed other factors such as the impact of the COVID-19 pandemic and associated restrictions.
“This research is part of the ‘minimum unit pricing evaluation portfolio’, meaning this research was a pre-planned part of the evaluation of this policy. A study protocol and analysis plan were published in advance, which is good practice, and these are further reasons to be confident in the findings.
“The results fit with earlier real-world evidence from Scotland that alcohol consumption decreased following the introduction of minimum unit pricing, and they are in line with findings of modelling studies from before the policy was introduced. The decrease in deaths was bigger in the more disadvantaged group, which again fits with the findings from modelling studies that minimum pricing helps to narrow health inequalities.
“This study only looks at the first 32 months of minimum pricing in Scotland. Previous evidence suggests these health gains should continue into the future, although high levels of inflation risk watering down the impact of the policy as it currently stands.”
Dr Adam Jacobs, Senior Director, Biostatistical Sciences at Premier Research, said:
“The paper claims to have found a statistically significant decrease in deaths that are due to drinking alcohol – a decrease of 13.4%. I have two problems with that number. First, there may be some uncertainty in attributing which deaths were ‘wholly attributable’ to alcohol consumption, and the paper would be more convincing to me if they presented statistics on all-cause mortality.
“Second, figure 1 in the paper appears to show the trend for mortality increasing since the MUP was implemented. There may be reasons why despite this there is statistically a decrease in deaths due to alcohol, but I would find it more convincing if you could see the death rate actually decreasing in the raw data.
“It’s important to note that the decrease in hospitalisations did not reach conventional levels of statistical significance.
“It is plausible that the MUP policy would bring down deaths and hospitalisations due to alcohol consumption, but I don’t think this paper shows it convincingly.”
Prof Petra Meier, Professor of Public Health and MRC Investigator, University of Glasgow, said:
“The authors present a timely, high-quality evaluation of the effect of Scotland’s Minimum Unit Pricing policy on wholly alcohol-attributable hospital admissions and deaths, that is hospital stays and deaths that are directly caused by alcohol consumption. They find reductions in both outcomes, with larger reductions in the most deprived areas, thereby reducing the stark health inequalities in alcohol-related deaths in Scotland. Hospital stays for these conditions decreased by 4% and deaths by 13% on average. The effects of the policy are in line with pre-implementation expectations.
“We need to be aware that wholly attributable outcomes represent only a small part of the likely effect of Minimum Unit Pricing on overall health and wellbeing, but one that can be estimated with far greater certainty than where alcohol is just one of a number of contributory factors. So we are missing here potentially large additional effects on alcohol-related deaths from cancer, liver cirrhosis, accidents, violence or suicide, but as the authors explain, there is no easy way to robustly estimate these effects yet, as several of these outcomes result from drinking over a number of years, so recent consumption changes would not yet show up in deaths statistics.
“The authors choose a robust, “gold standard” evaluation method for public health policies, something called a controlled interrupted timeseries analysis. This method capitalizes on the fact that Minimum Unit Pricing was introduced in Scotland but not England, and uses this to estimate what would have happened in Scotland had the policy not been introduced. The basic idea here is that you compare trends in alcohol-specific health outcomes before and after the introduction of Minimum Unit Pricing. If changes happen in both countries, they cannot be caused by the policy which only affected Scotland, but Scotland-only changes are highly likely to be caused by Minimum Unit Pricing, given there were no other major alcohol-related changes that occurred in only one of the countries in the meantime. In this way, the authors’ analyses can discount alternative explanations for the changes in deaths and hospital admissions, for example changes in alcohol taxation, disposable incomes or beverage market developments.”
Prof John Holmes, Director of the Sheffield Alcohol Research Group, ScHARR, University of Sheffield, said:
“This study provides the clearest evidence to date that minimum unit pricing (MUP) has reduced the harm caused by alcohol in Scotland. Previous work has demonstrated that MUP reduced the amount of alcohol sold in Scotland, but it has provided inconsistent evidence on whether heavier drinkers are consuming less alcohol. This study provides much clearer evidence. This is because alcohol-specific deaths are concentrated among people with very high levels of alcohol consumption. It is unlikely we would see the large effects reported in this study if heavier drinkers had not reduced their consumption.
“The study uses robust methods and the best available data. In particular, it uses data from England to help rule out explanations for the findings that are unrelated to MUP. It also finds very similar results when excluding data from during the COVID-19 pandemic – so these findings are unlikely to be due to things going on during the pandemic. Analyses of this kind are always reliant on a number of assumptions, but this study is very thorough in stress-testing its results and should therefore be seen as strong evidence that MUP is having its intended effects.
“This report is from the last major study in Public Health Scotland’s examination of the effects of MUP. Although some individual studies have raised questions about the effectiveness of MUP, we can now look at all of the available evidence together. This suggests that MUP has had largely positive effects including reducing alcohol consumption and alcohol-related harm.”
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“Statistically this paper is generally okay – the authors have carried out a quite complicated but appropriate time series analysis. However, my main concern is that there is a lot of causal interpretation here. This is an observational study, and no matter how well other factors are controlled for, it can never prove conclusively that the changes observed in deaths were due to the minimum unit pricing policy. In my view there hasn’t been enough caution given around assuming this relationship is causal. For example, in the paper the authors say “Our findings indicate the implementation of alcohol minimum unit pricing legislation in May 2018 in Scotland, led to significant reductions in deaths, and reductions in hospitalisations, wholly attributable to alcohol consumption” – the main problem here is “led to”, which is surely causal.
“I also think there are some questions to ask around what would be expected to happen in terms of a time lag between the policy being brought in and deaths being averted – would we expect large numbers of deaths to have been prevented already given that many deaths due to alcohol are due to chronic conditions?
“On the point about whether this study provides evidence that MUP caused the changes they observe. The authors’ analysis indicates that there were decreases in the numbers of deaths from the chronic conditions they analysed, after the introduction of MUP. That decrease was greater than the changes that happened in England in the same time period. (This is also supported by other sensitivity analyses they did including that these specific deaths in Scotland were greater in number than that in two English regions as a control instead of the whole of England, using alternative methods of analysis and only pre-pandemic data; and also using Scottish data for a different set of diseases (genitourinary) rather than English data as the control (“non-geographical control group”).) The decreases in deaths were much clearer than any decrease in hospitalisations – which was not statistically significant.
“However, all this just shows that there were decreases in deaths after the MUP policy was brought in – but that doesn’t necessarily imply that the decrease in deaths was caused by MUP. There are two aspects to think about here.
“First, it’s possible the decrease in deaths was caused by something other than changes in alcohol consumption. The analysis didn’t control for other factors apart from the Oxford Covid-19 Government Response Tracker (though of course it did take into account the trends and seasonality they estimated from their time series models). That can’t rule out confounding by something else – meaning some factor other than alcohol consumption could be behind the changes seen in deaths. However, the comparisons with England, and with the non-geographical controls, do give us more confidence that restricts the kind of confounders that might matter here – as does, in particular, the fact that they analysed only deaths and hospitalisations from conditions that are wholly attributable to alcohol consumption. So I actually think the risk of confounding of this sort is actually pretty remote (though not zero).
“Second, though, there’s a question of whether the decreases could indeed be caused by changes in alcohol consumption but that those consumption changes were not caused (or not totally caused) by MUP. The authors mention this briefly in the paper. If this is what was going on, that would be compatible with not observing changes in the non-geographical controls (which is what they found), assuming (as they do) that changes in deaths or hospitalisations from genitourinary conditions wouldn’t be expected to happen as a result of changes in alcohol consumption. Because of the comparison with the English controls, if there is some reason other than MUP for changes in alcohol consumption, it would have to be something that only happened in Scotland, or happened in Scotland much more than in England. There may be little reason to suspect that that happened, but the authors don’t say much about whether there are potential Scotland-only reasons, other than MUP, for a reduction in alcohol consumption.
“The authors did carry out a ‘falsification test’ – in this they simulated that the MUP was brought in six months earlier than it actually was. That indicated only very small decreases in deaths and hospitalisations (nothing near statistical significance). The idea is that, if MUP actually was the reason for reductions in deaths, then there would not have been any reduction in the first six months of the falsification test period, because in truth MUP had not been introduced then. There still would have been some reduction after MUP was really brought in, if MUP was the cause of the reductions, but putting that together with the initial six months would make any effect pretty well undetectable, statistically. This can add some degree of confidence that it might be the MUP that is behind the changes – but it still can’t prove it and it doesn’t take away all reasonable possibilities of there being some other factor behind changes in alcohol consumption. So, we can’t say that MUP definitely led to a 13.4% reduction in deaths, though that does clearly remain an important possibility.
“Coming to the point of a time lag and whether we would reasonably have expected to have already seen an impact on deaths given the policy was only brought in in 2018. The authors suggest that long lags are more likely to be associated with health outcomes that are partially but not wholly associated with alcohol, and that they believe from previous evidence that immediate effects of wholly attributable outcomes are plausible. That previous evidence they give as reference 23, a paper by Holmes et al. However, I have looked at that paper and I’m not convinced it provides strong evidence of that at all. That Holmes paper gives some specifications for what Holmes et al think is an appropriate pattern of lags for some (though not all) of the wholly attributable chronic outcomes that are looked at in this new paper – they give a time for there to be any effect at all, a time to get the full effect, and a specification of how the percentage of the full effect goes up over time. For all the outcomes in the Holmes et al. paper that are listed as wholly-attributable chronic outcomes in the new paper, the time till the effect starts is indeed given as “Immediate”. (The only ones in the Holmes paper that don’t start for some time are cancers where alcohol increases the risk, and they aren’t wholly-attributable to alcohol because there are other risk factors involved.) The new authors seem to be taking this to indicate that you’ll be able to see the effects immediately. But a diagram in the Holmes paper indicates that this isn’t necessarily the case – although the effect starts to occur immediately, at what may well be a very low level, it might take a long time for the effects to get to be large enough to detect. The time to full effect for the outcomes in the new paper is given by Holmes et al. as 20 years for most of them, and 10 only for alcoholic gastritis.
“So my interpretation is that, while it’s possible that the deaths or hospitalisations would have decreased enough to be detectable in the follow-up period here of 32 months after MUP, it’s also possible (given the 20-year period to full effect in the Holmes paper for most of the outcomes) that they aren’t clearly detectable on that time scale, though (if they really exist) the effect should show up, and indeed be much larger, later. And given what the time lag specifications look like in the Holmes paper, in another 7 or 8 years the reductions in deaths would be immense, implausibly immense indeed, given the size of the estimate after just over 2.5 years. Or it’s possible that what is being picked up in the new study is an effect of a change in alcohol consumption that occurred considerably earlier than MUP, so couldn’t have been caused directly by MUP – and I don’t think the falsification test can rule that out entirely, given that the Holmes paper indicates that any effects would take time to build up and might be too small to detect reliably to begin with. The Holmes et al. paper doesn’t deal with acute outcomes – but actually both deaths and hospitalisations from acute causes increased after MUP in the current study, though nowhere near statistically significantly (and the present authors give reasons why that might have happened).
“So overall, in my view, there remains some doubt about whether MUP definitely caused the alcohol consumption change and therefore whether it is responsible for reductions in deaths.
“One final point is about what ‘wholly attributable’ means, as it is a technical term and not necessarily very understandable. The authors use the phrase “wholly attributable to alcohol consumption” – this is referring to a death that has been categorised as having been caused solely by drinking alcohol. It doesn’t mean that the entire relative reduction in deaths the authors have reported – 13.4% – has been caused by the MUP policy. The study can’t prove that, because it’s observational.”
‘Evaluating the impact of alcohol minimum unit pricing on deaths and hospitalisations in Scotland: a controlled interrupted time series study’ by Grant M A Wyper et al. was published in the Lancet at 00:01 UK time on Tuesday 21 March 2023.
DOI: 10.1016/S0140-6736(23)00497-X
Declared interests
Dr Sadie Boniface: “I work at the Institute of Alcohol Studies which receives funding from the Alliance House Foundation. In 2016 I published a review (unfunded) of evidence on minimum unit pricing.”
Dr Adam Jacobs: “None.”
Prof Petra Meier: “Prof Jim Lewsey, one of the authors of the Lancet paper, works in the same academic institution as me. However, we work in different departments and have not collaborated on research to date.”
Prof John Holmes: “I am the Director of the Sheffield Alcohol Research Group, which played a substantial role in producing, disseminating and debating key evidence that informed the introduction of minimum unit pricing in Scotland and has played a similar role in other jurisdictions in the UK and internationally.”
Prof Kevin McConway: “I am a Trustee of the SMC and a member of its Advisory Committee. My quote above is in my capacity as an independent professional statistician.”