Nature and Potential Impact of Alcohol Health Warning Labels: A Scoping Review

Alcohol is toxic to human health. In addition to providing nutritional information, labels on alcohol products can be used to communicate warnings on alcohol-related harms to consumers. This scoping review examined novel or enhanced health warning labels to assess the current state of the research and the key studied characteristics of labels, along with their impact on the studied outcomes. Four databases (Web of Science, MEDLINE, PsycInfo, CINAHL) were searched between January 2010 and April 2021, and 27 papers were included in the review. The results found that most studies were undertaken in English-speaking populations, with the majority conducted online or in the laboratory setting as opposed to the real world. Seventy percent of the papers included at least one cancer-related message, in most instances referring either to cancer in general or to bowel cancer. Evidence from the only real-world long-term labelling intervention demonstrated that alcohol health warning labels designed to be visible and contain novel and specific information have the potential to be part of an effective labelling strategy. Alcohol health warning labels should be seen as tools to raise awareness on alcohol-related risks, being part of wider alcohol policy approaches.

Despite alcohol's detrimental effect on a variety of health outcomes, public awareness about some health risks associated with alcohol consumption remains relatively low. A 2018 review of 32 studies found that while awareness of alcohol as a risk factor for cancer varies by country, in most studies, less than half of the respondents correctly identified the alcohol-cancer link [15]. This finding is confirmed in more recent studies [16][17][18], which additionally found that awareness of the link between alcohol and cancer differed by type of cancer (e.g., the link between alcohol and breast cancer tended to be the least well known).
One of the approaches to raise awareness of the health risks associated with alcohol consumption can be product labelling. A review from 2013 on enhanced labelling [19] dependent variables, the outcomes of interest, as well as label content and format), and (2) what is the impact of the labels and key label characteristics on the studied outcomes?

Design
To answer the research questions, we used scoping review methodology, guided by the methodological framework of Arksey and O'Malley [47]. Scoping reviews are used to map the main concepts in the research area and present a broad overview of the existing evidence, including identification of the gaps, regardless of the study quality [48]. The protocol was not published in advance, and the results are reported in accordance with PRISMA-ScR checklist [49].

Eligibility Criteria
To be included in the review, studies had to evaluate new or enhanced alcohol health warning labels in any population. By "new", we considered labels developed for the purpose of the research and not previously implemented in practice, and by "enhanced", we considered improved versions of existing labels already implemented in practice. Studies had to include experimental or quasi-experimental methodology with clearly delineated independent and dependent variables, in order to be able to synthesise results based on the manipulated variables and to identify the key characteristics of effective labels. Only peer-reviewed articles in the English language published in 2010 or after were included, as in the previous reviews, there were no older experimental studies of new messages. Studies that did not include health warning messages, that focused on qualitative data or were only cross-sectional evaluations were excluded.

Information Sources and Search Strategy
Four databases (Web of Science, MEDLINE, PsycInfo, CINAHL) were searched with the abovementioned restrictions on language and dates (English language, January 2010-April 2021). Additionally, we complemented the database searches with inspection of reference lists of relevant articles and reviews, and we undertook a Google Scholar search. The search strategy contained a combination of terms related to alcohol labelling: [alcohol AND (label OR label*) AND (message OR information OR warning) AND (experiment* OR eval* OR effect)].

Study Selection and Summary
Study selection was performed by one researcher according to the abovementioned inclusion and exclusion criteria, with ambiguous cases being resolved through consultation with the second researcher. After the removal of duplicates, the studies were first screened by title and abstract, followed by full text examination and final article selection. Next, the data were charted to answer the main research questions. Next to general information on the papers' aim and methodology, information on label development and implementation was charted, as well as the independent studied variables and the results for each of the outcomes. Papers rather than studies were taken as the units of interest, as the focus of the review was to appraise the scope of the published literature and to identify results for all the studied outcomes.

Results
In total, we identified 27 papers that fit our inclusion criteria (see Appendix A for PRISMA diagram) . One paper reported on two studies using different methodologies to study the same variables [50], and four papers reported on different aspects of one intervention [51][52][53][54]. The remaining papers were individual studies (with a complete or partial focus on alcohol HWL labelling), although some of them were part of the same research line of label development and testing (e.g., [55][56][57][58][59][60]) or studied the same labels (e.g., [61,62]). Table 1 provides an overview of the papers with their aim and methodology. Twothirds (18) of the included papers were published in 2018 or later, indicating an increasing research interest in the topic. All of the research was undertaken in Western countries: most papers reported on research conducted in the UK (nine), Canada (six) and Australia (five). The remaining papers reported on research conducted in the US (three), Germany (two) or were multi-country (twice Luxembourg/Germany, and once Italy/France). Only four papers described real-world interventions. The remainder of the studies were either conducted online (sixteen) or in a laboratory (eight, of which one was in a naturalistic shopping laboratory). Six studies had under 100 participants, eleven studies had above 1000 participants and one study relied on sales data. The majority (80%) of the studies with a lower number of participants were conducted in the laboratory and before 2018, indicating that in recent years, larger sample sizes were accessible in an easier manner through online panels. Two-thirds of the papers focused on the general adult (of legal drinking age in the country) drinking population, although the exact inclusion criteria differed and ranged from people drinking at least once in the past 12 months, to people consuming alcohol weekly. Eight of the papers focused on the younger population/students (some by convenience rather than by inclusion criteria) and one paper focused on secondary school students.

Research Scope
To answer the first research question, we looked at the aim of the papers and the variables studied (both independent variables and the outcomes), as well as at the content and format of the studied labels.
The tested labels were predominantly developed by the researchers themselves, based either on other research (e.g., related to tobacco and alcohol labels, alcohol harm, effective messages) or on own development and piloting (e.g., through qualitative studies), or both. In the four papers related to a real-world intervention [51][52][53][54], the labels were additionally (next to own and other research) developed in consultation with local stakeholders. In two cases, the labels used in the research were developed by a non-governmental organisation [67,75], and in one case [50], existing voluntary labels were enhanced based on theory. To examine whether enhancing self-affirmation among a population of drinkers, prior to viewing threatening alcohol pictorial health warning labels, would reduce defensive reactions and promote reactions related to behaviour change, and whether there is an interaction between self-affirmation and severity of warning.
In terms of label format, the majority of the papers formatted the labels to be clearly visible. This meant putting labels on the front of the pack, with a predetermined size and clear contrast between text and background and separation from the rest of the label (e.g., black text on white background with black or red border, such as [55][56][57]61,63] or black text on a bright yellow background and red border, such as [51][52][53][54]). A complete description of the studied labels content and format is available in Supplementary Table S1.

Key Label Characteristics-Impact on the Outcomes
Results for all the studied outcomes from each of the papers are presented in Table 3. Below, the results are summarised based on the study aims identified through the two research questions: (1) studying overall label effectiveness by comparing image and text labels and (2) studying other label message characteristics and label formats.

Label Effectiveness
A large online UK study [63] did not find that adding a text warning to alcohol drink label changed perceived personal risk, motivation to drink less or perception of damaging drinking among adult drinkers. On the other hand, the Canadian real-world study, which included putting three types of labels (one of them a cancer warning) on the majority of alcohol drinks in the participating liquor stores in the studied region, found that these enhanced labels led to increased recall, cognitive processing and self-reported impact of drinking [51,54], as well as decreased alcohol sales [53] in the intervention municipality compared to the control municipality. While the whole intervention included both cancer warnings as well as low-risk guidelines/standard drink labels, it is not possible to completely disentangle the impact of solely the cancer warning labels from the results, although some results pointed to effects being driven predominantly by the cancer label. The one paper that looked specifically at the cancer labels [52] found that cancer labels increased knowledge of alcohol as a carcinogen, and that knowledge increased with time. Additionally, the authors found that knowledge about alcohol causing cancer was associated with a greater likelihood of supporting health warning labels.

Label Content-Image vs. Text
Among the eight papers that compared image and text labels, six had control group comparison [55,61,62,64,67] and in two [65,66], only text and image were compared. Six of the papers used graphic images (e.g., diseased body part); in one, the images were neutral, and one contained both graphic and neutral images.
In the studies where graphic images were used, comparison between text-only and image-and-text labels showed higher negative emotional arousal [55,56], fear [55,56,61,65] and reactance [65] for the image-and-text labels among adults compared to text only labels, as well as lower product appeal [64,65], label acceptability [55,56] and believability [65]. Moreover, in the study using neutral images [67], there were higher negative emotions for image-and-text labels compared to text-only were found among the secondary school students, but only for some messages. The same study also found that both types of labels are better than no label for increasing knowledge about alcohol-related risks when knowledge is low at baseline. While an online experiment showed a selection of alcoholic drinks in the given scenario to be lower after seeing image-and-text labels compared to text-only labels [55], this finding did not replicate in a naturalistic lab shopping setting [56] among adult regular drinkers, where none of the labels affected purchasing behaviour. In another study, both text-only and image-and-text labels decreased the speed of consumption compared to no label, with no significant difference between them [62].  Between subjects: • substance: cigarette, alcohol, sugar sweetened beverages • causal language: "causes", "contributes to", "can contribute to" and "may contribute to".      [53] Between subjects: • label: enhanced vs. regular Alcohol sales Behaviour Enhanced labels < Regular labels * Significant results column describes the impact of independent variables (tested groups) on the dependent variable (outcome) and should be read together with the outcome column, for example, "plain packaging group had significantly lower product based perception compared to standard packaging group".
The two studies actively looking at the differences between graphic and neutral images did not find any difference between the two in attention [66,75], although the latter found higher avoidance, reactance, perceived warning effectiveness and motivation to drink less in graphic compared to neutral images.

Label Content-Message Characteristics
Several papers also compared different messages or their characteristics. In terms of message content, in Australia, one study among adult drinkers found increased intention to reduce drinking for cancer, diabetes and mental illness messages, but not for heart disease and liver damage-related messages [59]. Another study found that a bowel cancer message had the highest believability, convincingness and personal relevance [60]. A bowel cancer message also had the highest negative emotional arousal, lowest desire to consume the labelled product and lowest acceptability in a large-scale UK online study of adult drinkers [57], but in this study, all the labels included graphic photos, which might have influenced the outcome. Another large scale online experiment in the UK among adult drinkers [71] compared a cancer message to a mental health message and found the former led to a higher motivation to drink less and higher avoidance, but there was no difference in reactance, believability, self-efficacy and response efficacy. Another smaller study found that positive expectancy-related messages lowered drinking intentions compared to health-related messages among students from Germany and Luxembourg [72]. In a German sample of secondary school students, cancer and liver cirrhosis messages, but not a pregnancy label, increased knowledge about alcohol-related risks [67].
In terms of framing, the discrete choice experiments gave mixed results [70,74]; in the first study, no message at all was more preferred than any message in a sample of Italian and French young wine drinkers, and in the latter, the negative message was less preferred than the neutral message among Australian students. In an online study of British adult drinkers [71], a negative framing led to a higher motivation to drink less, with increased reactance and avoidance compared to a positive framing, but did not influence believability, self-efficacy and response efficacy. This study also looked at specificity and found that specific messages led to lower reactance, higher response efficacy and higher believability compared to general messages, but there was no difference in motivation to drink less, avoidance and self-efficacy.
One study looked at whether framing labels as questions or statements impacts attitude and intentions, and found no difference [68]. Another study compared narrative and non-narrative labels and found narrative labels led to higher worry about developing alcohol-related cancer and higher feelings of risk of developing alcohol-related cancer, but no difference in intentions to reduce alcohol use [69]. The study examining causal language [73] found that using the word "causes" had higher perceived message effectiveness, and "may contribute to" had the highest perceived ineffectiveness, but there was no difference in support for any of the labels. Finally, a study of Australian adult drinkers found that receiving the message from multiple sources is better than receiving a message from one single source [58].

Label Format
Finally, four studies examined label format: branding/plain packaging [64,76], warning size [50,70,76] and warning position [70]. Two smaller-scale Canadian studies found that plain packaging (removing any branding) decreased product-based and consumerbased perception [64,76]. The latter study [76] also found that after a certain size of the warning (50% of the bottle), there is no further increased difference in outcomes. In a discrete choice experiment, young Italian and French wine drinkers preferred to have warning sizes as little as possible, and on the back of the label rather than on the front of the label [70]. Finally, one study [50] found that an increase in size and change in prominent colour was effective in self-reported measures of attention among the Australian drinkers, but not when measured directly with eye-tracking.

Discussion
This review aimed to examine the scope of the (quasi-)experimental research on novel and enhanced alcohol health warning labels and to identify the key characteristics of the effective warning labels. Twenty-seven papers were included, the majority of which were published in 2018 or later, indicating an uptake in interest surrounding the topic from the research community (recognised also in other broader reviews [45]). This is aligned with the increase in policy-related labelling discussions, such as in Codex Alimentarius, at the EU or national level (e.g., [77,78]). All of the research has been undertaken in Western countries, where alcohol labelling has either not been updated since its implementation decades ago (e.g., Canada, US), carried out on a voluntary basis (e.g., Australia, UK) or not mandated at all (e.g., many of the EU countries). Given high alcohol consumption in those regions [79], this may reflect the recognition of labelling of potentially effective means of reducing alcohol-related harm-although it is debatable if this focus adds to or replaces focus on "best buys" policies [80], such as taxation, pricing and advertising restrictions. In addition, the majority of the studies focused on general population, which reflects a shift from findings of previous review [43], where less than a third of studies were targeting the broader population, and the remaining were targeting younger populations and (pregnant) women.
In terms of content, most commonly, the studied messages were referring to cancer-70% of papers included at least one cancer-related message, out of which the most commonly mentioned cancers were bowel, breast and liver cancer. This was followed by liver cirrhosis or liver disease messages, used in a third of papers. Other topics of messages were heart disease, mental illness, brain damage, pregnancy, drinking and driving and other (e.g., positive expectancies surrounding alcohol). This brings us to the question on how to select the content of the message to be included on the labels. In the papers included in this review, researchers have commonly either developed messages based on "evidence on alcohol-related harm" or have taken messages used in other studies. We would argue that the message first has to reflect evidence-based risk (e.g., that causal relationship between alcohol and the outcome has been clearly demonstrated [3]), and then the message should be tailored to the context in which the labels will be displayed, based on the burden of disease. Another factor that has been found to be important for the selection of the message is also its novelty for the given audience [16,53]. For example, in the EU, 29.4% of all alcohol-attributable deaths and 19.4% of all attributable DALYs are due to cancers [81], while awareness of the risk between alcohol and cancer is low [15]. Based on these guidelines we can conclude that cancer warnings should be a priority when mandating health warning labels in this region.
Next, the focus of research implies the researcher's theory of change-the mechanisms by which they consider labelling to work [82]-and this will influence the targeted outcomes and how effectiveness is measured. This is relevant because the policy debate will inevitably discuss the merit of the health warning labels based on their "effectiveness"previously, one of the arguments in the World Trade Organisation's discussions was that labels are considered unnecessary interference with international trade because there is no scientific evidence to support such warnings as an effective public health measure [83]. Based on the scope of the existing research, we propose that two possible theories of how labelling brings about change are currently being investigated. The first one is focused on persuasion and short-term individual behaviour change, and optimises using health warning labels that aim to elicit reactance, avoidance, fear and other negative emotional reactions as means of changing behaviour, drawing heavily on the experience from tobacco warning labels. For this line of research, there is no real-world evidence from the alcohol field, and mixed evidence from lab studies regarding (short-term) behaviour [56,62]. The (online and lab) studies that focused on studying the graphic images on labels found that they elicited higher negative emotional arousal, fear and reactance, and were not well accepted by the participants. These results are also corroborated by the metanalysis investigating the impact of HWL in both alcohol and food products, and found that while health warning labels reduce product selection compared to no labels, the difference between image-and-text HWLs and text-only HWLs was not statistically significant (although in the former, the effect was slightly larger) [44].
The second theory of change sees labelling as a tool to raise awareness of alcoholrelated risks as a part of a comprehensive strategy to reduce alcohol-related harm (which should include other alcohol policies and interventions), and emphasises optimising labels to increase attention to and cognitive processing of the information on the labels, leading to increased knowledge (e.g., on alcohol health risks). This approach is more in line with system perspectives [84] and its focus is to use health warning labels to arrive at a better informed population that could in turn lead to more support for other alcohol policies. Individual short-term behaviour change might be a welcome by-product of this approach, but in the long term, labels can be seen as a tool to help with changing societal conversation and norms surrounding alcohol (as suggested also by O'Brien [85]). In terms of evidence, the best current evidence points to the second theory of change as having greater potential to be explored in future research and practice. The most comprehensive study to date and the only one measuring long-term effects in the real world [86] showed that simple, specific and well-visible messages led to an increase in knowledge (with associated support for stronger alcohol policies [87]) and change in the amount of alcohol consumed, partially mediated by the consumer attention to and processing of label messages [54]. The labels were also well accepted and supported by the population, as well as the drinkers [51].
It is important to note that this study [86] was implemented in context with existing relatively strict other policy measures-e.g., national monopoly on the alcohol stores (alcohol being sold separately in the government-owned liquor stores), and enhanced label replacing existing label (rather than no label). Nevertheless, a key approach to label development and implementation in this study is a good practice example and should be used in any future attempts to study alcohol HWLs in different contexts. This means developing labels in consultation with local stakeholders, including a range of different messages (not only health but also related to standard drinks and lower-risk guidelines) and including evidence-based messages for which there was low awareness at baseline and are relevant to the population. This study [83] also points to the unintended consequences that can be anticipated-originally, the study was planned to run for eight months with all three labels, but the alcohol industry intervened and threatened with a lawsuit if cancer warning labels were to continue being applied on the products [83]. Thus, cancer labels were only applied to the alcohol products for one instead of eight months. This shows that in any attempts to mandate evidence-based alcohol health warning labels, strong opposition from the industry can be expected, similar to what happened in the tobacco field [88].
Thus, while there should be some lessons from tobacco taken for alcohol health warning labels, we must be careful to still consider alcohol in its own context, and note that the evolution of alcohol legislation lags behind the tobacco field by a couple of decades. There is evidence for the effectiveness of pictorial labels for tobacco [36], but even here, the argument can be made that labels have been part of a broader approach that involved international collaboration and changed societal norms around tobacco [89], and their purpose was perhaps less to deter smokers from smoking than to indirectly prevent younger populations from taking up smoking, as tobacco warning labels replaced packaging as the last venue of marketing after other forms of advertising had been prohibited [38]. The value of adding images on alcohol labels might be to help with attracting attention and take away space from the branding of the product, but not enough studies with only neutral images or symbols have been conducted to make any firm conclusions.
The main implications for research and policy based on the results of this review are summarised in Table 4. We would like to note that the same theory of change useful for health warnings could also be useful to develop and test for optimal nutrition information on alcohol labels (not direct focus on changing behaviour, but focus on providing the information), although as previously mentioned, the issue with nutritional labels is to first align requirements for alcoholic beverages with existing labelling requirements for food and non-alcoholic beverages.

Future Labelling Research
Aim to be explicit about the theory of change underpinning the research and adjust the studied outcomes accordingly. Develop labels to be tested with own qualitative research and preferably co-created with local stakeholders/populations-focus on both content and format. The information provided in the labels should be clear and reflect the up-to-date literature on alcohol-attributable burden of disease. Online surveys with large samples and experimental design can be a good lower-cost approach to help with tailoring the messages to new contexts before testing them in the real world. Real-world studies with longer-term outcomes are preferable in the context a lot of existing research (e.g., online and lab surveys). When reporting, describe the context in which the intervention was conducted; frameworks such as Medical Research Councils' process evaluation guidance [82] can be applied to support design and reporting.

Development of mandatory labels
As currently implemented mandatory and voluntary labels are often suboptimal, existing research on new health warning labels points to the following characteristics: In terms of format: good visibility of the label is essential: this is achieved by large size, contrasting colours, large text, prominent position on the product In terms of content: it is important to have messages that convey novel information to the target population messages should be evidence-based-for example, take into account alcohol-attributable risk, morbidity and mortality wording should be simple, messages should be specific rather than general current evidence suggests that graphic images are not well accepted, and does not point to a superiority of their inclusion use rotating health warning labels presented in combination with other messages information on health warning labels should be presented also through other sources of information (e.g., information campaigns), and labelling should preferably also be introduced as part of a broader package of alcohol policy measures

Limitations
Our review is subject to general limitations of scoping reviews: we did not appraise studies for quality, and the conclusions are still somewhat broad and qualitative. In terms of this specific review, we narrowed the scope of our investigation to health warning labels and overall outcomes and we did not focus on other labels, such as lower-risk drinking guidelines or standard drinks, even though some research suggests it might be worth taking them into account simultaneously. We also did not investigate differential impact of labels on different subgroups. Evidence from Canada points to some subgroup differences in cognitive processing of the labels, although concludes that vulnerable populations still attended to the labels [52]. Our search was also focused only on peer-reviewed English language papers; therefore, we might have missed peer-reviewed papers in other languages and non-peer-reviewed research.

Conclusions
The majority of the current evidence regarding alcohol health warning labels comes from very short interventions in the online or the laboratory setting as opposed to the real world, and there is rather large diversity in the content and format of labels being evaluated. Nevertheless, evidence from real-world, long-term interventions shows that alcohol health warning labels designed to be visible and contain novel and specific information have the potential to be part of an effective labelling strategy. Labelling should be seen as tool to raise awareness and a small part of wider alcohol prevention and policy approach. Future research could focus on importance of labels in a broader societal context, and go beyond individual behaviour change, recognising that change in social norms can be a complex, multifaceted and long-term process.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/nu13093065/s1, Table S1. Label development, content and format. Acknowledgments: Part of this research was produced under the service contract AlHaMBRA Project (Alcohol Harm-Measuring and Building Capacity for Policy Response and Action, Contract No. 20197105). The information and views presented are those of the authors, and hence represent their sole responsibility. Accordingly, the information and views presented cannot be considered to reflect the views of all AlHaMBRA Project consortium members, the European Commission and/or the Health and Digital Executive Agency or any other body of the European Union. The European Commission and the Agency do not accept any responsibility for use that may be made of the information contained herein.

Conflicts of Interest:
The authors declare no conflict of interest. Appendix A Figure A1. PRISMA Diagram.