Review of Evidence on Alcohol and Health (2025)
Chapter: 2 Approach to the Task
2
Approach to the Task
This chapter describes the methods the committee used to perform systematic evidence scans based on (1) the eight questions posed in the Statement of Task (Box 1-3), (2) the committee’s criteria for assessing the results of the evidence scans to determine whether there was sufficient evidence to support an updated or de novo systematic review of the more recent evidence, and (3) the committee’s framework for assessing the certainty of conclusions as well as the process used for making conclusions. Box 2-1 provides definitions for key terminology related to systematic reviews discussed in this report.
The committee used a systematic approach to gather evidence that included evidence scans, systematic reviews (SRs) and a systematic narrative review. Additional evidence from SRs, meta-analyses, and other primary research publications were submitted to the committee for consideration. From among these, those that met the committee’s defined criteria were included, while those that fell outside of the criteria were not.
EVIDENCE SCAN AND OUTCOMES
The committee developed search strategies for evidence scans based on the search terms and inclusion and exclusion criteria as outlined by the U.S. Department of Agriculture (USDA) Nutrition Evidence Systematic Review (NESR) center for the Dietary Guidelines for Americans (DGA). Search terms for alcohol consumption were derived from the NESR systematic review on alcohol and all-cause mortality for the DGA 2020–2025 and were applied to each of the literature searches (see Appendix E–I for search
terms). Except for the all-cause mortality outcome, for which a systematic review was completed for the DGA 2020–2025, inclusion and exclusion criteria for the searches for other health outcomes were derived and refined based on the criteria used in the most recent NESR reviews on dietary patterns, which included some alcohol criteria. After reviewing these search terms and inclusion/exclusion criteria, the committee made minor revisions to the search strategy by adding terms and editing criteria based on the expertise of committee members with the intent of being more inclusive of data relevant to the Statement of Task (Table 2-1).
Eligibility Criteria
The committee developed study inclusion and exclusion criteria and sought studies primarily in humans who were at least 21 years of age and at risk for chronic disease. Studies on interventions had to include at least 30 participants per arm or a power calculation, and observational studies had to include at least 1,000 individuals (comparisons with never and former drinks combined as nondrinker groups were excluded). Primary literature was used, therefore systematic reviews or meta-analyses were excluded. Table 2-1 lists the outcomes considered across the eight research questions
TABLE 2-1 Inclusion and Exclusion Criteria
| Category | Inclusion Criteria | Exclusion Criteria |
|---|---|---|
| Population | Human participants (females, males) | Nonhuman participants (e.g., animal or in vitro models) Women during pregnancy |
| Studies conducted in countries ranked as high or higher human development. | Studies conducted in countries ranked as medium or lower human development | |
Age of study participants (at intervention or exposure):
|
Age of study participants (at intervention or exposure):
|
|
Health status:
|
Health status:
|
|
Size of study groups:
|
Size of study groups:
|
| Category | Inclusion Criteria | Exclusion Criteria |
|---|---|---|
| Exposure |
|
|
| Comparator |
Primary
|
|
Secondary
|
||
| Outcome | All-Cause Mortality Outcomes All-cause mortality (i.e., total mortality): the total number of deaths from all causes during a specific time period (ideally stratified by sex) |
All-Cause Mortality Outcomes Studies that only report cause-specific mortality (total number of deaths from a specific disease, such as cardiovascular disease or cancer) |
| Weight Outcomes Weight BMI (body mass index) Waist circumference Incidence of overweight and obesity Body composition |
Weight Outcomes Gestational weight gain |
|
| Cancer Outcomes Breast (female), oral, pharyngeal, laryngeal, esophageal, colon, and rectal |
Cancer Outcomes Studies that exclusively examine cancer-related mortality, prevalence, survivorship, or recurrence of cancer |
|
| CVD Outcomes Nonfatal myocardial infarction Nonfatal stroke CVD-related mortality |
CVD Outcomes Hypertension disorders during pregnancy and/or lactation |
|
| Neurocognitive Outcomes Total dementia, Alzheimer’s disease, Word recall, verbal fluency, MMSE |
N/A |
| Category | Inclusion Criteria | Exclusion Criteria |
|---|---|---|
| Study Design |
|
|
| Peer-reviewed articles published in English Publication date range: 2019 to current (all-cause mortality) Publication date range: 2010 to current (weight, cancer, CVD, neurocognition) |
Articles not published in peer-reviewed journals, including unpublished data, manuscripts, reports, abstracts, preprints, and conference proceedings; non-English publications |
NOTES: CVD = cardiovascular disease; MMSE = Mini-Mental State Examination; N = number.
in the Statement of Task. Studies had to evaluate levels or patterns of alcohol consumption. The committee notes that two exclusion criteria warrant specific mention: first, studies were excluded if the exposure measurement (amount of alcohol consumption) did not allow for evaluating associations for moderate drinking distinct from greater consumption amounts (i.e., when all consumption amounts were combined as the exposure), and second, studies were excluded if persons who never consumed alcohol were combined with persons who formerly consumed alcohol to avoid the influence of abstainer bias.
Literature Search
The literature search approach was iterative based on search results and ongoing committee discussion. Databases searched included Med-line, Embase, and the Cochrane Register of Controlled Trials (CENTRAL). The search comprised terms for alcohol and the eight specific outcomes. The search terms for the SRs are in Appendixes E–I.1 Search
___________________
1 Appendixes E through J are available online at https://nap.nationalacademies.org/catalog/28582.
terms for literature searches that did not result in an SR are provided in Appendix J.
Because the DGA were supported by a new systematic review for all-cause mortality, the search frame included studies published in January 2019 or later. The other seven questions in the Statement of Task included studies published in January 2010 (the date of the previous edition of the DGA that covered alcohol and health) or later. The search dates and periods for each topic are described in Box 2-2, and different iterations and committee discussions and decisions are provided in Appendix C. As a result of this process, two additional studies for cardiovascular disease and neurocognition were deemed to be eligible.
Screening
Title and abstract screening were conducted in two phases. In the first stage, OpenAI’s GPT-4.0 in the PICO Portal evidence synthesis platform removed titles and abstracts using natural language processing and machine learning based on initial screening by humans. The second stage of title/abstract screening was conducted in duplicate by independent researchers based on the eligibility criteria (Table 2-1). Subsequently, full-text articles of potentially relevant abstracts were reviewed in duplicate by independent researchers. All discrepancies were resolved by a third researcher. Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow charts were created for each of the topics and are included in each chapter. The search results for the topics identified in the Statement of Task are summarized in Table 2-2.
Data Extraction
Data extraction was completed by consultants at the Academy for Nutrition and Dietetics (AND). The information extracted from each study included author(s), year of publication, country where the study was conducted, source of funding, follow-up time, sample sizes, years of data collection, description of alcohol intake and assessment tool, description of comparison group (i.e., age, sex, race/ethnicity), confounders accounted for in analysis, and results for specific analysis. One researcher extracted data from each study, where they were verified by a second researcher. Discrepancies were resolved through discussion or a third researcher.
TABLE 2-2 Search Details by Statement of Task Questions
| Search question/topic | Number of articles identified by search criteria | Number of articles eligible by title and abstract screening | Number of articles eligible by full-text screening | Number of articles included in review for data extraction |
|---|---|---|---|---|
| All-cause mortality (2019–2023 only) | 17,404 | 320 | 34 | 34 |
| Weight changes | 4,458 | 64 | 7 | 7 |
| Cancer | 20,190 | 382 | 25 | 25 |
| Cardiovascular disease | 20,227 | 423 | 26 | 26 |
| Neurocognition | 19,997 | 364 | 24 | 24 |
| Lactation | 4,714 | 17 | 7 | 0 |
Risk-of-Bias Assessment
All included studies were cohort studies, and risk of bias was assessed using the Risk of Bias in Nonrandomized Studies of Exposures (ROBINS-E) tool (Higgins et al., 2024). Domains evaluated had bias caused by confounding, bias arising from measurement of the exposure, bias in selection of participants, bias owing to post-exposure interventions, bias caused by missing data, bias arising from measurement of outcomes, and bias in selection of the reported result. For the five domains on bias due to confounding, important confounding variables considered were age, sex, smoking status, socioeconomic status, diet, physical activity, weight/body mass index (BMI), and comorbidities. Overall (study-level) ratings for risk of bias were as follows: low, some concerns, high, or very high (Higgins et al., 2024). Risk-of-bias assessments were conducted independently by two researchers, and discrepancies in domain-specific and/or overall assessments were resolved by a third researcher. Risk of bias was reported using a figure created in R using the robvis function (McGuinness and Higgins, 2021).2
Data Synthesis
Comparative results used mean differences for continuous outcomes, hazard ratios (HR), risk ratios (RR), or odds ratios (OR) for binary outcomes. Fully adjusted effect estimates were used when determining impact from nonrandomized studies. When studies did not report results that could be pooled, results were summarized narratively. When results from at least three studies with comparable exposures were available, meta-analysis were conducted. Two overarching meta-analyses were conducted: one with those never consuming alcohol as the comparator group, and one with those consuming alcohol. RRs were transformed to natural logs to address skewness, and studies were weighted by the inverse of the estimated variance of the natural log of the RR (Alavi et al., 2020). A restricted maximum likelihood random-effects model was used for meta-analyses. Heterogeneity was assessed using the I2 statistic that refers to the percentage of total variability in study results caused by between-study variability. A p-value <0.05 indicated statistical significance.
When information was available, subgroup analyses were conducted according to sex, age (<60 or ≥60 years), race/ethnicity, and smoking status. A sensitivity analysis was conducted by examining results after dropping studies with high risk of bias. Another sensitivity analysis was conducted
___________________
2 Robvis (Risk-Of-Bias-VISualization) is a tool for visualizing risk-of-bias assessments in a systematic review.
using the five categories from Zhao et al. (2023) and a “one drink/day” limit for both females and males.
All meta-analyses were conducted using Stata 16 and OpenMeta.3 Results were reported in a study characteristics table, forest plots for meta-analysis are included in Chapters 3–7 (Box 2-3), and a summary of findings tables. For analyses with at least 10 studies, publication bias was determined using visual examination of funnel plots.
Assessment of Certainty of Evidence
Certainty of evidence was rated by the AND consultants using the GRADE method, which considers study design, risk of bias, directness, inconsistency in results between studies, precision of the findings, and other factors (GRADE Working Group, 2013). Evidence certainty was initially rated as high, moderate, low, or very low by the consultants who conducted the systematic reviews (Appendixes E–I). Although these systematic reviews examined observational studies, evidence certainty started with a high rating and then downgraded because risk of bias was assessed using ROBINS-E, which is a stricter assessment of observational studies (Higgins
___________________
3 OpenMeta is an open-sourced software platform used for meta-analyses. http://www.cebm.brown.edu/openmeta/ (accessed November 15, 2024).
et al., 2024). A GRADE table was created using GRADEPro Guideline Development Tool.4
Decision Process for Undertaking De Novo Systematic Reviews
To inform the DGA, the committee decided to undertake de novo systematic reviews rather than perform updates and reanalysis for past reviews. To decide whether to send a topic for a systematic review of studies published since the last review, the committee established a process based on the review of articles published in the search time frames by committee members expert in the specific topics. For all-cause mortality, this included articles published between January 2019 and September 2023. For CVD, overweight/obesity, cancer, neurocognitive health, and lactation, this included articles published between January 2010 and September 2023. Additional articles for cancers other than the seven specified as related to alcohol by the U.S. National Cancer Institute and the American Cancer Society were also reviewed (ACS, 2020; NCI, 2015).
Topics Without Systematic Reviews
Based on the low number of studies, small sample sizes, and methodological challenges related to exposure and outcome measurement, other/emerging cancer sites (see Chapter 5) and lactation (see Chapter 6) were not submitted for evidence synthesis. Conclusions for these topics were based on a review of the individual study results by the committee.
Similarly, if there were fewer than three studies meeting inclusion criteria, results from a full evidence review with meta-analysis were not conducted; rather, the committee summarized the literature, specifically evaluating whether the results were congruent with or different from previous reviews used to develop prior DGA (Table 2-3). The committee decided to base conclusions on systematic reviews and narrative synthesis of individual study results for overweight/obesity and neurocognitive health.
Process for Committee Conclusions
When there were at least three studies included in the meta-analysis, the committee included forest plots in the report to support conclusions (Box 2-3). To determine the certainty of its conclusions, the committee
___________________
4 GRADEPro Guideline Development Tood is an evidence synthesis tool used to create summary and findings tables for Cochrane systematic reviews. https://methods.cochrane.org/gradeing/gradepro-gdt (accessed November 15, 2024).
TABLE 2-3 Dietary Guidelines Advisory Committee Links with Alcohol Sections
| Category | 2010 | 2015 | 2020 |
|---|---|---|---|
| Report | 2010 DGAC Reporta | 2015 DGAC Reportb | 2020 DGAC Reportc |
| Methodology | 2010 SR Methodologyd | 2015 SRse | 2020 SR Methodologyf |
| Systematic Reviews | 2010 SRg | N/A | 2020 SRh |
NOTES: DGAC = Dietary Guidelines Advisory Committee; N/A = not available; SR = systematic review.
a See https://www.dietaryguidelines.gov/sites/default/files/2019-05/2010DGACReport-camera-ready-Jan11-11.pdf (accessed September 19, 2024).
b See https://health.gov/sites/default/files/2019-09/Scientific-Report-of-the-2015-Dietary-Guidelines-Advisory-Committee.pdf (accessed September 19, 2024).
c See https://www.dietaryguidelines.gov/sites/default/files/2020-07/ScientificReport_of_the_2020DietaryGuidelinesAdvisoryCommittee_first-print.pdf (accessed September 19, 2024).
d See https://nesr.usda.gov/sites/default/files/2019-04/2010DGAC-SR-Methods.pdf (accessed October 10, 2024).
e See https://nesr.usda.gov/sites/default/files/2019-04/2015DGAC-SR-Methods.pdf (accessed October 10, 2024).
f See https://nesr.usda.gov/sites/default/files/2020-07/NESR Systematic Review Methodology for the 2020 Advisory Committee_0.pdf (accessed October 10, 2024).
g See https://nesr.usda.gov/sites/default/files/2019-04/2010DGAC-SR-Alcohol.pdf (accessed September 19, 2024).
h See https://nesr.usda.gov/2020-dietary-guidelines-advisory-committee-systematic-reviews/beverages-and-added-sugars-subcommittee/alcohol-all-cause-mortality (accessed October 10, 2024).
used a framework based on methods from the U.S. Preventive Services Task Force (USPSTF, 2018).5
- High certainty: Evidence includes consistent results from good-quality studies in relevant populations assessing effects on health outcomes; the conclusion is unlikely to be affected by future studies. (Note that it is unlikely to be rated as high certainty without a randomized controlled trial).
- Moderate certainty: Evidence is sufficient to determine effects on health outcomes but is constrained by issues raised in the quality assessment of the evidence; additional information from future studies could change the conclusion.
___________________
5 For the systematic reviews commissioned by AND that assigned “very low” certainty, the committee used the phrase “insufficient evidence” to reflect a lower level of certainty of the evidence using the USPSTF framework.
- Low certainty: Evidence is insufficient to assess effects on health outcomes; additional information from future studies may allow for assessment.
Low certainty was concluded when the results of eligible studies were inconsistent or the data were too sparse. When the level of certainty could not be assigned, the committee determined that no conclusion could be drawn. This determination was made when there was a statistically nonsignificant meta-analysis result or there were no eligible studies.
REFERENCES
ACS (American Cancer Society). 2020. Alcohol use and cancer. https://www.cancer.org/cancer/risk-prevention/diet-physical-activity/alcohol-use-and-cancer.html (accessed September 20, 2024).
Alavi, M., G. E. Hunt, D. C. Visentin, R. Watson, D. K. Thapa, and M. Cleary. 2020. Using risk and odds ratios to assess effect size for meta-analysis outcome measures. Journal of Advanced Nursing 76(12):3231–3234.
GRADE Working Group. 2013. GRADE handbook. https://gdt.gradepro.org/app/handbook/handbook.html (accessed July 3, 2024).
Higgins, J. P. T., R. L. Morgan, A. A. Rooney, K. W. Taylor, K. A. Thayer, R. A. Silva, C. Lemeris, E. A. Akl, T. F. Bateson, N. D. Berkman, B. S. Glenn, A. Hrobjartsson, J. S. LaKind, A. McAleenan, J. J. Meerpohl, R. M. Nachman, J. E. Obbagy, A. O’Connor, E. G. Radke, J. Savovic, H. J. Schunemann, B. Shea, K. Tilling, J. Verbeek, M. Viswanathan, and J. A. C. Sterne. 2024. A tool to assess risk of bias in non-randomized follow-up studies of exposure effects (ROBINS-E). Environment International 186:108602.
McGuinness, L. A., and J. P. T. Higgins. 2021. Risk-of-bias VISualization (robvis): An R package and shiny web app for visualizing risk-of-bias assessments. Research Synthesis Methods 12(1):55-61.
NCI (National Cancer Institute). 2015. Alcohol. https://www.cancer.gov/about-cancer/causesprevention/risk/alcohol#:~:text=Drinking%20alcohol%20can%20increase%20your%20risk%20of%20cancer,who%20drink%20to%20do%20so%20in%20moderate%20amounts (accessed September 20, 2024).
USPSTF (U.S. Preventative Services Task Force). 2018. Grade definitions. https://www.uspreventiveservicestaskforce.org/uspstf/about-uspstf/methods-and-processes/grade-definitions (accessed September 20, 2024).
Zhao, J., T. Stockwell, T. Naimi, S. Churchill, J. Clay, and A. Sherk. 2023. Association between daily alcohol intake and risk of all-cause mortality: A systematic review and meta-analyses. JAMA Network Open 6(3):e236185.
