The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach will be utilized to formulate conclusions on how alcohol impacts health (20–22). Project oversight will be performed by the U.S. Department of Agriculture and the U.S. Department of Health and Human Services (who formulated the project’s “Terms of Reference”). Furthermore, to achieve the project’s “Terms of Reference”, a Technical Advisory Panel has been formulated to oversee the project.
The project will determine the health impacts caused by ethanol in alcoholic beverages, and, therefore, will not distinguish between harms caused by beer, wine, spirits, and other alcoholic beverages. Harms caused by beer, wine, spirits, and other alcoholic beverages are based mainly on ethanol content, regardless of the form in which the ethanol is consumed, with the exception of alcohol poisonings which are caused predominately by the consumption of spirits (23) and some other forms of unintentional injury where there is evidence of a higher risk due to the consumption of spirits (24). A systematic review of existing guidelines on alcohol and health as well as individual modelling projects will be performed to generate conclusions on (i) weekly thresholds to minimize long-term and short-term risks of morbidity and mortality, (ii) per occassion thresholds to minimize the short-term risks of injury or acute illness due to per occasion drinking, (iii) alcohol use among vulnerable populations, and (iv) situations and individual circumstances that are hazardous. An overview of the proposed methodology and data sources are outlined in Table 1 and the supplemental material.
Table 1
Methodology and data sources
Conclusions on alcohol’s impact on health | Methodology used to reach conclusions | Underlying data sources |
Weekly / daily thresholds to minimize long-term and short-term risks of morbidity and mortality | Conclusions based on nominal group interviews | Lifetime risk of alcohol-attributable mortality and morbidity at different drinking levels |
Per occasion thresholds to minimize the short-term risks of injury or acute illness due to per occasion drinking | Conclusions based on nominal group interviews | Systematic review of meta-analyses which summarize how drinking patterns affect the risk of disease and injury occurrence |
Results from a reanalysis of United States’ emergency room case-cross-over studies |
Results from roadside survey studies |
Alcohol use among vulnerable populations | Conclusions based on nominal group interviews | Systematic review of previously published guidelines |
Situations and individual circumstances that are hazardous |
< Insert Fig. 1 and Table 1 about here >
Systematic review of previous guidelines on alcohol and health
A 2016 systematic review highlighted differences in alcohol guidelines across 37 countries (8), but did not assess their scientific quality or the methodologies used to produce these guidelines. Therefore, this review is of limited value for the development of future guidelines. Accordingly, a systematic literature search of country-level guidance on alcohol and health will be performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), and will be preregistered with PROSPERO (25, 26). Guidance on alcohol and health will be identified through websites of countries' health ministries using an internet search (i.e., ‘[country] AND health ministry’), and/or using the World Health Organization’s Global Information System on Alcohol and Health (27). A list of all country names will be obtained from the World Health Organization (https://www.who.int/countries). In addition to searching websites of countries' health ministries, a peer-reviewed literature search for guidance documents will be performed. Articles will be identified through a systematic search of PubMed, PsycInfo, and Web of Science based on keywords and subject headings. As an example, the search strategy for PubMed is outlined in Table S7 (see supplemental material). Searches will be piloted and refined with the aid of a research librarian. No publication year or language restrictions will be applied. References contained in the articles included will be examined for relevant publications. The processes used for the screening of articles, data extraction, and standardization are outlined in the supplemental material (see Table S8).
Each guideline on alcohol and health will be evaluated according to its methodological rigor; the Appraisal of Guidelines for Research and Evaluation (AGREE) II instrument is the most commonly used guideline appraisal and consistency evaluation tool in the GRADE protocol and will be used here (28, 29). The review will summarize advice on (i) weekly thresholds to minimize long-term and short-term risks of morbidity and mortality, (ii) daily thresholds to minimize the short-term risks of injury or acute illness due to per occasion drinking, (iii) vulnerable populations, and (iv) situations and individual circumstances that are hazardous.
Collecting and generating evidence on weekly thresholds to minimize health risks
Two methods can be used to generate evidence on weekly thresholds to minimize health risks associated with different levels of average alcohol consumption: (i) the examination of all-cause absolute risk curves from cohort studies (which include deaths from all causes, including conditions from which a causal relationship with alcohol use has not been determined), and (ii) the modelling of lifetime all-cause absolute alcohol-attributable mortality and morbidity risk curves by combining data on cause-specific relative risks with corresponding data on absolute mortality risks for lifetime abstainers. Based on the positives and negatives of each method, it was decided to proceed with the modelling of lifetime all-cause absolute alcohol-attributable mortality and morbidity risk curves (i.e., method (ii)) to align with the current practices of the Centers for Disease Control and Prevention, the World Health Organization, and the Institute for Health Metrics and Evaluation for estimating the burden of disease attributable to alcohol use (30–33), and with other recently released guidance documents on alcohol and health (11–13, 34).
Mathematical modelling will be used to estimate the lifetime risk of death and disability for different levels of average alcohol consumption. The risk curves generated will be based on the average amount of pure alcohol (i.e., ethanol). “People who consumed alcohol” will be defined as those individuals who consumed at least one standard drink (14 grams of alcohol) in the past year. The mathematical modelling will specifically examine the following outcomes: deaths, premature deaths (i.e., deaths that occur among those < 70 years of age), years of life lost (YLL), years lived with disability (YLD: a measure of disease occurrence and the disability caused by the disease), disease and injury incidence, and disability adjusted years of life lost (a combination of both YLL and YLD).
The inclusion of diseases and injuries in the modelling of alcohol-attributable deaths and disability will be based on three criteria: (i) the disease or injury has to be causally related to alcohol use, (ii) a dose-response risk function needs to be available for the risk relationship between alcohol consumption (measured in grams per day) and the disease or injury of interest, and (iii) either death or disability needs to be measured specifically for the disease or injury causally related to alcohol use. Multiple data sources will be used to establish causality for this review. Causality can be assumed in all causes of death that are solely (100%) attributable to alcohol use (e.g., alcoholic cardiomyopathy). Furthermore, this review will use information from the World Health Organization’s Global Status Report on Alcohol and Health (GSRAH) (30), the Institute for Health Metrics and Evaluation’s Global Burden of Disease study (28), and the World Cancer Research Fund’s (WCRF’s) Continuous Update Project (35–37), all of which have assessed which diseases and injuries are causally related to alcohol consumption. A sensitivity analysis will be performed in which causality will be based on conditions included in the Alcohol-Related Disease Impact application of the Centers for Disease Control and Prevention (38).
Systematic Review of Meta-analyses
This study also aims to conduct a systematic review of all relevant meta-analyses that have investigated diseases causally associated with alcohol use. Relative Risk (RR) estimates for each included disease category and injury will be obtained from systematic reviews identified in the evidence review process. The RRs in the reviews will have used people who are lifetime abstainers as a reference group. In cases where lifetime abstainers were not the reference group, these RRs will be rescaled so that the lifetime abstainers are the reference group.
PubMed (including MEDLINE and Life Science Journals) and Web of Science will be searched for relevant articles based on keywords and subject headings. The search strategy for MEDLINE, as an example, is outlined in Table S9 in the supplemental matrial. Searches will be piloted and refined with the aid of a research librarian. No language restrictions will be applied. A publication restriction of 2010 and later will be applied to ensure that only recent summaries of the epidemiological evidence are included in the review. References contained in the articles included will be examined for relevant publications. Subject experts will be contacted to identify relevant publications. Previous systematic reviews, the International Agency for Research on Cancer Monographs, and the WCRF’s Continuous Update Project reports will be searched for relevant publications.
All systematic reviews will be evaluated for quality using the criteria from A Measurement Tool to Assess Systematic Reviews (AMSTAR 2) and from the Risk of Bias in Systematic Reviews (ROBIS) tool (39, 40). Quality assessments will consider study design and adjustments for confounding factors, among other criteria. In addition these quality assessments will take into account factors specific to alcohol epidemiological studies. For example, biases in exposure and outcome measurements, such as the use of current abstainers instead of lifetime abstainers as the risk reference group (i.e., allowing the misclassification of former drinkers as lifetime abstainers), and how representative the cohorts are of the US general population (41).
The systematic review will be pre-registered with PROSPERO, and conducted in accordance with the PRISMA guidelines (42).
Selection of Meta-analyses
Meta-analyses used in the modelling process will be selected by panels of nominal group interview experts in the distinct areas of (i) cancer, (ii) cardiovascular diseases, (iii) digestive conditions, (iv) neurological disorders, (v) infectious diseases, and (vi) injuries who will be consulted to determine the RR estimates for each condition considered to be causally related to alcohol use. The sampling frame of experts will be decided based on the authors who have published the largest number of first and last author publications concerning the above-noted disease areas (as determined by performing a PubMed Search) in the past 10 years. These authors will be asked to participate in the nominal group interview panels. Quota sampling will also be used to establish nominal group interview panels which ensure diversity and representation based on geographic location, sex, race and ethnicity, which will be reported by the potential participants before they are included in the nominal group interview panels. Specifically, the following parameters will be established: (i) a minimum of 90% of the experts will be United States-based, (ii) a minimum of 40% of the experts will identify as female, with a goal of 50%, and (iii) no more than 80% of the experts will be non-Hispanic White, with a goal of no more than 60% of participants being non-Hispanic White. Experts will be required to have no potential conflicts of interest.
Once the panels of experts have been established, a nominal group process will be used to determine the most appropriate meta-analyses to use in the modelling study. The nominal group interview allows for the selection of meta-analyses avoiding group think (as compared to group interviews), and reduces random error in decision making by increasing the number of people whose opinions are considered (e.g., (43)). All panels will also provide input on the magnitude of the underreporting of alcohol consumption in medical observation studies, as well as input on vulnerable populations and situations and individual circumstances where the consumption of alcohol is hazardous. In the first round, each expert will be provided with the results of the preregistered systematic scoping review, and will be asked to provide free-text comments on the interpretation of these studies. When necessary, experts will be contacted to clarify their answers. The project team will analyze answers to identify potential underlying patterns and prepare a draft consensus position. Each expert will also be asked the degree to which they hypothesize alcohol use to be misestimated in population surveys (what fraction is due to spillage and wastage, and what fraction is due to survey and response biases) and in cohort studies.
In the second round of interviews, based on feedback regarding the results of round 1, the consolidated information will be provided to the expert participants. During these interviews, experts will interact and clarify their positions in small group virtual meetings. At the end of the second round of interviews, all experts will be asked to complete a questionnaire which will be used to build a consensus on which meta-analyses to include in the modelling study.
In the event that this process proves too time consuming, selection of meta-analysis will be modified and streamlined. The streamlined approach will involve selecting the meta-analysis for modeling based on its performance on the AMSTAR 2 and ROBIS tools, prioritizing the study with the least potential risk of bias.
Data sources: Mortality and Alcohol Exposure
A comparative risk method (i.e., the systematic evaluation of changes in population health that would be avoided in the absence of a risk factor) will be used to estimate the alcohol-attributable burden of disease. For this method, the reference group of lifetime abstainers will be utilized − thus determining the number of deaths and the burden of disability that would not occur if alcohol consumption was eliminated (see: (11–13, 44)). Lifetime risk curves will be based on an exposure of average grams of pure alcohol (i.e., ethanol) consumed per week.
Data on the number of deaths that occurred in the United States by age, sex, and cause will be obtained from the National Vital Statistics System (45, 46). Data on population by age and sex will be obtained from the Census Bureau of the United States (47). Data on disease incidence and YLD will be obtained from the Institute for Health Metrics and Evaluation’s Global Burden of Disease study (48). Life tables by sex will be obtained from the Centers for Disease Control and Prevention, National Center for Health Statistics (49).
Relative risks from the systematic review of meta-analyses will be combined with estimates of the risk of mortality and morbidity for lifetime abstainers to estimate lifetime risk curves. A comparative risk assessment methodology will be used to estimate the risk of mortality and morbidity for lifetime abstainers (see supplemental material). Lifetime risk curves will be calculated by estimating alcohol-attributable mortality and morbidity risk by cause, age, and sex. The life-year specific alcohol-attributable mortality risks will be multiplied by the YLL for each cause of death. The lifetime risk curves will also account for competing causes of death (i.e., deaths which occur that are not attributable to alcohol use). Seesupplemental material for further details regarding the lifetime risk methodology.
< Insert Fig. 2 about here >
The lifetime risk curves for alcohol-attributable mortality and morbidity risks will be provided to members of all nominal group interview panels. We will then use the nominal group interview process, including all members from all panels, to establish conclusions on weekly and/or daily alcohol use thresholds.
Collecting and generating evidence on per occasion alcohol use thresholds to minimize short-term health risks
Per occasion alcohol use is based on consumption during particular times or occasions, (as compared to average consumption which is determined by dividing total consumption by time). As such, per occasion consumption is related to consumption patterns, and provides a proxy measure for blood alcohol concentrations that might be achieved based on a particular level of per occasion consumption. High per occasion consumption resulting in high blood alcohol concentrations is strongly associated with “acute” disease processes such as injuries (50), but also is associated with the risk of infectious diseases (51), breast cancer (52), ischemic heart disease and ischemic stroke (53), diabetes (54), epilepsy (55, 56), and liver cirrhosis (57).
Multiple sources of data will be used to reach conclusions on the health impacts of per occasion alcohol use, namely: (i) results from the systematic review of meta-analyses and systematic reviews which summarize how drinking patterns affect the risk of disease and injury occurrence, (ii) results from a reanalysis of United States’ emergency room case-cross-over studies which examined the relationship between alcohol use and injury occurrence (50), and (iii) results from roadside survey studies where the blood alcohol content of road injury decedents was compared to that of drivers randomly selected from the same road on the same day and at the same hour as the fatal crash case (58, 59). Based on these data sources, the expert groups will reach public health conclusions on the maximum number of drinks people should consume during one drinking occasion. The nominal group interviews will follow a two round structure that will also be used to determine the best RR functions to be used in the modelling of the health impacts of alcohol use.
Collecting and generating evidence on the effects of alcohol use among vulnerable populations, and situations and individual circumstances where consuming alcohol is hazardous
The planned study will build upon previous guidelines (identified through the previously noted systematic review) using nominal group interviews to provide conclusions with respect to evidence on the effects of alcohol use among vulnerable populations (e.g., pregnant persons, underage youth), and situations and individual circumstances where consuming alcohol is hazardous. The experts selected from the nominal group surveys will review evidence of individuals in vulnerable populations or who are subject to situations and individual circumstances where consuming alcohol is hazardous to determine if these considerations should be included in the conclusions of the study. In addition, experts who are part of the nominal group interviews will be invited to identify vulnerable populations and situations and individual circumstances which have not been previously noted in any other guidelines, but where current research supports that these vulnerable populations and/or situations and individual circumstances should be added to alcohol and health guidance documents, as well as any relevant peer-reviewed research documents which support their addition. The structure of these nominal group interviews will follow a two round structure that will also be used to determine the best RR functions to be used in the modelling of the health impacts of alcohol use.
Collecting and generating evidence on the public’s understanding and reaction to conclusions on alcohol consumption and health
To understand how conclusions regarding the health impacts of alcohol use align with the alcohol consumption of the general public, we will examine alcohol use in the general population using data from the National Alcohol Survey, National Survey on Drug Use and Health (NSDUH), National Health Interview Survey (NHIS), The National Epidemiologic Survey on Alcohol and Related Conditions - III (NESARC-III), and Behavioral Risk Factor Surveillance System (BRFSS) in order to obtain data on alcohol exposure in the United States (see Tables S1 to S6 in supplemental materal). These population surveys, which have differing sampling frames, response rates, and validation of self-reporting of alcohol use, provide comprehensive data on alcohol consumption patterns and allow estimations of the prevalences of different levels of alcohol use in the population.
To gain an understanding of the public’s understanding and reaction to the conclusions on alcohol and health produced by this project, we will engage in public consultations using a mixed methods (quantitative and quantitative) approach. Public consultation recruitment will be performed using a convenience sampling methodology. Specifically, the consultation process will be promoted through institutional websites, mass emails, social media, and national and regional digital news media advertisements. All residents of the United States will be eligible to participate. Individuals who do participate will be asked categorical and open-ended questions (in English) about:
their demographics, including geographic location;
their profession;
their current sources of information on health and, more specifically, on the impacts of alcohol consumption on health;
current and previous health conditions;
their awareness of the Dietary Guidelines for Americans, 2020–2025;
their alcohol use and whether they are trying to follow the alcohol consumption guidance in the Dietary Guidelines for Americans, 2020–2025 and their reasons for following or not following the guidelines;
their requirements for and expectations of the Dietary Guidelines for Americans, 2026–2030;
what information they found useful about the project’s conclusions on alcohol and health; and
what challenges they had in interpreting the project’s conclusions on alcohol and health.
Aggregate information on the demographics of participants will be examined, published and compared to United States Census data to assess whether participants in the consultation process are representative of the general population (47). A thematic analysis will be performed on responses to open-ended questions to identify common themes between respondents’ answers to each of the questions. As the information obtained from an analysis of the findings of the consultation process is to be used to revise and refine the research conclusions, both a semantic and latent coding approach and analysis will be employed. Furthermore, to analyse the qualitative data, Braun and Clarke's six phases of thematic analysis approach will be utilized (60).