Fermented Beverages, Ethanol and Health: A Critical Appraisal of Meta-Analytical Studies

Abstract

The effect of alcohol on health is a controversial topic when it comes to the moderate or conscious consumption of fermented beverages. The recent claim by the World Health Organisation (WHO) and the European Heart Network (EHN) that the safe level of alcohol consumption is zero has compromised the efforts of the fermentation scientific community in developing healthier and more sustainable beverages. Therefore, the objective of this review was to assess the scientific background for such a claim that appears to be the result of recent scientific evidence. Using the meta-analytic data supporting WHO and EHN guidelines, it was possible to demonstrate that fermented beverages (e.g., wine and beer) have lower effects compared to spirits, that some population ethnicities have higher sensitivity to alcohol, and that drinking patterns influence the outcomes. Moreover, higher relative risks associated with younger individuals are mostly related to injuries (e.g., car accidents, self-inflicted injuries) and not with diseases. Sequential WHO studies produced significantly higher limits and emphasized that preventive policies should be tailored to populations at higher risk. In conclusion, the statement that “all alcohol is hazardous” has no scientific background and should be understood under the perspective that “one drink is too many and one thousand is never enough” used in alcoholism prevention. Fermentation researchers should continue their efforts on the promotion of healthier lifestyles, sustainable development and on the preservation of cultural heritage under the responsible drinking perspective.

1. Introduction

The influence of fermented foods and beverages on health has been widely debated, with varying opinions on their effects ranging from beneficial [1] to harmful [2]. There appears to be a consensus that fermented products are globally a healthy constituent of the human daily diet [3]. These foods and beverages are part of the so-called Mediterranean diet, which UNESCO has recognized as part of the Intangible Heritage of Humanity since 2010 [4]. Even the undeniable intoxicating effect of ethanol is dealt with under the assumption of moderate consumption or responsible drinking. However, the World Health Organisation (WHO) challenged this status quo by altering its advisory message from “the unhealthy alcohol effect over a certain limit” to “all alcohol is hazardous” [5]. The same claim has also been made by the European Heart Network [6]. This shift ultimately questions the endeavor of fermentation and nutrition researchers to contribute to healthier lifestyles and to environmental and heritage sustainability [7] and turns them into unconscious contributors of unhealthy social behaviors.

The conceptual frameworks of scientists dealing with fermented beverages, nutrition and healthcare may not be coincident, and, certainly, the word “alcohol” has different implications in the mind of each one. First, for the former, alcohol is a vague word, being ethanol, the precise product of alcoholic fermentation, together with other secondary metabolites (e.g., glycerol, methanol, acetic acid, fatty acids). Second, fermented beverages are not restricted to wine and beer, but include other forms from lesser-known raw materials with lower ethanol levels (e.g., kombucha, cider, fruit “wines”, milk, honey) [8,9]. Nevertheless, healthcare scholars use the word alcohol to describe it as an addictive drug irrespective of its origin [10]. Even when recognizing that moderate consumption may not be harmful, their overwhelming message is that “application of these findings to humans should be approached with caution because the health effects of alcohol are complex and vary based on individual factors such as genetics and lifestyle” [11]. This issue must be approached critically to find the source of what seems to be now the mainstream opinion among health officials and institutions. Otherwise, all alcohol fermentation scientists could be responsible for harming people’s health.

Fermentation primary research is mainly driven by the assessment of cause-and-effect relationships, while clinical epidemiologists are largely dependent on correlational analysis obtained through an approach known as meta-analysis.

A meta-analysis (MA) is a quantitative methodology that combines results from various studies addressing the same research question, and intends to derive a pooled effect size estimate that provides a more precise estimate of the intervention or exposure effect [12]. This methodology is particularly advantageous to (a) enhance statistical power and the capacity to detect true associations or intervention effects; (b) provide a more representative view of the existing evidence by combining results from various studies; (c) identify recurring patterns and trends across multiple studies; and (d) resolve contradictory results among individual studies [12].

MA is presently a widespread tool among many research fields that owes much of its success to the low cost and high publication rate among scientific journals [13,14]. However, this high volume appears to serve primarily as a marketing instrument, as it is often misleading and does not contribute to evidence-based medicine [15]. Indeed, a robust finding from a well-designed randomized controlled trial (RCT) is far more reliable than the inferences drawn from most meta-analyses [16].

The objective of this review is to address the effects of ethanol/alcohol on health as tackled by epidemiological studies based on MA that underlie the claim that “all alcohol is hazardous”. The focus is on explaining how meta-analytic studies derive the results that appear to be the primary support for the WHO’s and EHN’s current guidelines. A detailed analysis of the methodological issues or indications on safe consumption levels are outside the scope of this review. Our goal is to assess whether the meta-analytic outputs are consistent with the knowledge acquired by causal relationships studied by primary empirical research. Hopefully, this review will help bridge the knowledge gap between fermentation experts and those committed to healthcare research, as evidenced by meta-analyses.

2. Understanding the Basic Theoretical Background of Meta-Analysis

2.1. Conceptual Definitions and Production Pipeline

MAs are usually presented within the framework of systematic reviews (SRs), which summarize existing evidence to answer a specific clinical question and include a thorough, unbiased search of the relevant literature, explicit criteria for assessing studies, and a structured presentation of the results. Meta-syntheses (MS) concern qualitative data, while MA incorporates a quantitative combination of several similar studies to produce an overall summary of treatment effects amenable to statistical analysis [17].

Meta-regressions are obtained by pooling the results of several MAs [18,19]. A meta-regression aims to relate effect size to study characteristics in a manner analogous to regression in primary studies. Associations derived from meta-regressions are observational and have a weaker interpretation than causal relationships derived from randomized studies [20]. In addition, meta-reviews are obtained by gathering the results of several MSs or MAs and are also known as umbrella reviews [21].

Logically, these approaches could only be efficiently handled with the advent of powerful software solutions designed to analyze big data. Indeed, meta-analytic software has driven the exponential growth of SRs, as evidenced by the number of citations in scientific databases [22]. Therefore, it is a very attractive field of research guaranteeing high and swift publication rates at low cost and large dissemination over the community [15,23]. However, as pointed out by Ioannidis [24], “it is clear that doing so is no guarantee of a rigorous and trustworthy SR”. Indeed, conducting and publishing a SR with poor methodology may actually worsen human welfare [24]. Hence, it is not surprising the declining citation number of MAs in medical studies since the beginning of 21st century [14].

The pipeline process necessary to produce a MA has been extensively described in the literature (see for instance [12,20]). Briefly, the pipeline develops through: (i) formulating the research question, (ii) literature search and study selection; (iii) data extraction; (iv) assessing study quality; (v) statistical analysis; (vi) heterogeneity assessment; (vii) publication bias assessment; and (viii) interpretation and reporting.

Statistical analysis could appear as the central and most rigorous of all the processes. However, MA results can be very misleading if suitable attention has not been previously given to formulating the review question; specifying eligibility criteria; identifying and selecting studies; collecting appropriate data; considering risk of bias; planning intervention comparisons; and deciding what data would be meaningful to analyze [25]. In other words, meta-analytic design does not guarantee the highest attainable level of evidence. Nonetheless, rigorous SRs are in the best position to identify subjects for which the evidence is thinnest and, at their best, to help focus resources on the most needed new research [24].

2.2. The Significance of Effects in Meta-Analytical Determinations

The outputs of MAs are loosely defined as “effects”, although the correct term should be “outcome”, since there are no cause–effect assessments [26]. The statistical models are at the core of MA, and quality guidelines accept a wide variety of approaches (see [27] for a detailed description). Even with the best practices, incorrect conclusions are not uncommon [28,29]. Indeed, Nelson et al. [30] clearly stated that “…meta-analytic thinking not only fails to solve the problems of p-hacking, reporting errors, and fraud, it dramatically exacerbates them…, … meta-analytic thinking would make the false-positives problem worse, not better”. Therefore, the results should be interpreted with caution, particularly given the often-low statistical power, which may limit the ability to detect true relationships or significant differences [31]. To establish significance, the p-value is widely regarded as the gold standard in science. However, the notion that a treatment is effective for a particular outcome if p < 0.05 and ineffective if that threshold is not met is a reductionist view of medicine that lacks ecological validity [32]. Ioannidis [33] summarized these warnings: “There is increasing concern that most current published research findings are false… simulations show that for most study designs and settings, it is more likely for a research claim to be false than true…claimed research findings may often be simply accurate measures of the prevailing bias”. Based on statistical interpretation, Ioannidis [33] listed a series of useful corollaries that illustrate how likely it is to find skewed MAs. Briefly, research findings are more likely true when

(i)

Large studies are produced (over several thousand subjects) over small studies (100-fold smaller);

(ii)

Large effects are obtained (relative risks 3–20) compared to small effects (relative risks 1.1–1.5);

(iii)

Smaller numbers of relationships are used, which allows the greater the selection of tested relationships;

(iv)

There is smaller the flexibility in designs, definitions, outcomes, and analytical models;

(v)

Financial and other interests and prejudices in a scientific field are minimized, including expert opinion.

In contrast, (vi) the hotter a scientific field (with more scientific teams involved), the less likely the research findings are to be true.

3. The Financial Biases and Societal Implications

The increasing involvement of industry in scientific research may compromise scientific integrity, leading to conflicting media messages, confusion, and, in extreme cases, public disbelief in science [34]. In the case of the alcohol industry, this involvement has been compared to the tobacco industry in its support for research on the health benefits [35,36]. Contrarily, Patel et al. [37] saw little funding from the alcohol industry in such studies. On the other hand, the anti-alcohol institutional bodies are also funded by large organizations and provide governments with justification to increase revenue by taxation or insurance companies to increase premiums [38].

The observation that the fermentation community is an instrument in the hands of large corporations that sell alcohol is as appropriate as saying that the funding stakeholders of clinical research limit healthcare researchers. Moreover, unlike in the tobacco industry, fermented beverages worldwide are primarily produced by numerous small businesses. Indeed, small and medium companies of family nature deeply rooted in cultural, heritage, and religious beliefs constitute the base of the fermentation industry, be it wine, beer, or traditional fermented products [39]. Fermented products are clear contributors to the United Nations (UN) goals 1 (No poverty), 2 (Zero hunger), 11 (Sustainable cities and communities) and 12 (Responsible consumption and production) (https://sdgs.un.org/goals, accessed on the 29 November 2025). The argument that preventing alcohol-induced diseases complies with the UN goal 3 (Good health and well-being) is obviously accepted either by healthcare professionals or fermentation researchers. However, the contemporary approach to health is not restricted to public health but also includes the holistic concept of “One Health” or “Planetary Health” [40] where fermented beverages play a role in biocultural heritage [41].

3.1. The Power to Influence Institutional and Public Opinions

The WHO and EHN, as partner of the World Heart Federation, are highly reputed global institutions that swiftly disseminate their recommendations. Indeed, many countries and institutions have presently stricter alcohol consumption guidelines. In particular, the EHN further recommends alcohol taxation and the end of European Union support for agriculture that produces alcohol [6].

The position of anti-alcohol (AA) policy makers claiming that all alcohol should be banned is understandable from the Alcoholics Anonymous (AA) perspective, where “one drink is too many, and a thousand is never enough”. The medical research groups dealing with addictions face the overwhelming societal burden of alcohol use disorders (AUD) where abstinence is the only way out. Griswold et al. [42], on behalf of Global Burden Disease (GBD) 2016 collaborators, definitely assumed that “Debates concerning whether the safest level of consumption is zero or near zero are missing the point. There is a major obligation of the public health community to address the massive disease burden due to alcohol”. With this justification, it is understandable that the safe level must be zero. Under this view, it should be difficult to recognize an alternative pathway for those whose daily lives include fermented beverages, responsible drinking, or drinking in moderation.

The two opposing perspectives contend that the other party benefits. According to Roerecke and Rehm [43], “We sense a desire by some in the field to apply tough standards on protective effects and more lenient standards on other effects”. Moreover, Stockwell et al. [35] stated that some studies promote health benefits and soften harms due to methodological flaws. The International Scientific Forum on Alcohol Research (ISFAR) is accused of minimizing evidence of harm, regardless of study quality, comparing its conduct to the strategies of the tobacco industry [35,36]. However, the swift dissemination of health harms across diverse societal levels contradicts these beliefs, as evidenced by changes in institutional guidelines and opinions published online and in the popular press.

The Quick Propagation Among Popular Media

The claimed new scientific evidence that supports both AA perspectives made the related research a hot scientific field among popular media. The public trusts scientific evidence, and debunking long-standing views increases sales and audience ratings [44]. Overall, the message is clear: “the present new scientific evidence demonstrates that alcohol harms start from the first drop”. As a corollary, those who do not believe in it are outdated, sold to the alcohol corporations and fit in the dangerous category of science denialists/negationists (Supplementary Table S1).

Reputed scientific divulgation magazines also propagate these views. A National Geographic article showed that reporters are more likely to convey the harms than the benefits of moderate or responsible consumption (Supplementary Table S2). Indeed, all interviewed specialists have a recognized career related to AUDs. As expected, the only response to reduce several diseases and overall mortality is to fully cut alcohol consumption. The appeal of moderate consumption is not advised since it would lead to abuse, and these researchers do not conceptually understand it, as is implicit in their answers.

In addition, online popular informative sources minimize moderate consumption and pronounce the scientific foundations for WHO and EHN teetotalism (e.g., https://en.wikipedia.org/wiki/Alcohol_consumption_recommendations, accessed 30 November 2025).

3.2. Examples of Contradictory Health Studies with Societal Impact

The existence of contradictory results in MAs has been widely acknowledged [12]. Two examples are presented here to illustrate the position of antagonistic researchers. Albeit unexpected for a primary scientist, this controversy is understandable when knowledge is based on associations easily skewed by methodological biases.

Autism in the offspring has been related to the maternal use of acetaminophen (paracetamol) during pregnancy [45,46]. Nevertheless, the Food and Drug Administration (FDA) stated that “It is important to note that while an association between acetaminophen and neurological conditions has been described in many studies, a causal relationship has not been established and there are contrary studies in the scientific literature” [47]. Therefore, the absence of causality was essential for the conclusive remarks. Overall, the FDA accepted that clinicians and patients could make a reasonable decision without institutional patronizing advice. The FDA guidance was further supported by a recent umbrella review that concluded with the absence of a clear link between maternal paracetamol use during pregnancy and autism [48]. These authors emphasized the low confidence and the weight of confounding factors (e.g., family factors, sibling control, maternal characteristics) in explaining the outcomes of the previous MAs.

Nordhagen et al. [49] examined a WHO meta-analytic review (EHC2023) that addressed the effect of radiofrequency electromagnetic radiation (RF-EMF) on pregnancy and birth outcomes in non-human mammals. The review concluded that the analyzed data did not provide sufficiently certain conclusions to inform regulatory decisions. The criticisms by Nordhagen et al. [49] did not question the article selection, the hazard thresholds, or the statistical methods employed. On the contrary, using the same data, these authors concluded that, when relevant studies were cited correctly, there were clear indications of detrimental non-thermal effects from RF-EMF exposure. The skewed methodology and the low quality of the work were considered to threaten and to undermine the WHO’s trustworthiness and professionalism in the area of human health hazards from man-made RF-EMF.

Overall, these two examples align with the perspective of Halevi and Pinotti [14], who attributed the decline in the number of SRs informing healthcare policymakers to their low quality and overall importance. On the contrary, regarding alcohol, their influence appeared to be key to the shift in WHO and other institutional recommendations.

4. Interim Summary

The previous descriptions provided the basic concepts required to understand and interpret the results of a SRMA by researchers without experience in this field. Peer-reviewed journals publish the arguments among research teams on specific methodological flaws. The controversy reaches the point of proposing retraction not by methodological flaws but by conclusion disagreements [50,51]. The discussed reasons are numerous (see for instance [34,35,52]), but scientific censorship is unwarranted [28].

The next sections will be devoted to understanding the research conducted by clinical epidemiologists. The focus will be put on WHO- and EHN-funded studies since they appear to be the sources for the present AA perspectives. The quality of data acquisition, processing, or statistical regression will not be addressed, as these topics were recently discussed [53,54].

5. Understanding Meta-Analytic Outcomes

5.1. The Heterogeneity of Outcomes in WHO Studies

The initial work supporting the WHO policy shift was developed under the frame of the Global Burden Disease report 2016 published in 2018 [55]. The overall effect of alcohol on all diseases/hazards was estimated using meta-regression (Figure 5 in [55]). The presented spline was obtained by the summation of 26 dose–response curves (figures in pages 53 to 140 of the Supplementary Appendix 1 in [55]). The authors did not present a forest plot, but the inspection of those curves enabled the determination of the average effect size and the upper confidence interval (CI) from the green-shaded area (the Python 3.13 code to obtain the effect and the CI from the graphs is shown in Supplementary Table S3). Next, the standard errors (SE) were calculated using the natural logarithms of the upper CI and of the effect size using the spreadsheet as provided by Neyeloff et al. [26]. This back-estimation process for relative ratio (RR) measures is described in the Cochrane manual (www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-06#section-6-3, accessed on 30 November 2025). A sample size calculation was not required, as it was already accounted for in the CI.

The MA results were obtained using the Comprehensive Meta Analysis (CMA) software (version 4.0.000, August 2022). The forest plot is presented in Figure 1 for the lowest alcohol concentration (10 g/day), and the remaining doses are listed in Supplementary Figure S1. The combined effect estimate and 95% CI for all diseases is shown at the bottom by a diamond. There are effects on the right side of the plot (higher risk) opposed to the studies on the left side (lower risk). The vertical line (mean difference of 1) indicates absence of alcohol effect on the disease. The process was repeated at 2–6 standard drinks per day (each standard drink equals 10 g alcohol/day). The calculations were made up to 60 g/day, which is equivalent to binge drinking, a pattern of drinking that brings blood alcohol concentration levels to 0.08 g/dL [56]. This typically occurs after four drinks (56 g alcohol) for women and five drinks (60 g alcohol) for men over approximately two hours [56].

The results clearly show that the summary effect was obtained by gathering very different outputs. Moreover, in meta-regressions, the number of studies relative to covariates (e.g., diseases) should be high. The suggested number is 10 studies for each covariate [18], which was not met for nine of the 26 studied effects.

The results showed that, under 10 g/L alcohol, cardiovascular diseases had relatively small effect sizes, while cancer had higher effect sizes. Diabetes’ effects were reduced, while the other diseases showed effects up to a RR of 1.473 for pharynx and nasopharynx cancer. With increasing alcohol concentrations, as expected, cirrhosis displayed the highest effect under 60 g/L (up to a RR of 9.492, Supplementary Figure S1e), easily explained by its direct relation with alcohol abuse. The injuries also had comparatively lower effect sizes (up to a RR of 1.934 for self-harms under 60 g/L, Supplementary Figure S1e), not linearly related to alcohol concentration. The effect sizes increased greatly after 60 g/L as also the corresponding CI’s did (Figure 5 in [55]).

5.1.1. True Effects and Prediction Intervals

The mean effect size in MA is insufficient to describe how populations react to a treatment. The variability in responses should be addressed using prediction intervals [57] given that CI’s only show the dispersion of the mean effect (i.e., precision) and not the dispersion of all responses. While the mean effect depends on the standard error (the higher number of studies, the lesser error), the prediction interval is based on the standard deviation of the effects. The results are shown in Figure 2. A 95% prediction interval of 0.82 to 1.26 means that the true effect size lies in the interval 0.82–1.26 for 10 g/day of alcohol. The same rationale applies to the remaining doses in Figure 2. The high variability in the results suggest that there are very different sensitivities of the populations under study. Even under 60 g/L, the RR is lower than 1 for a small proportion of the population. Therefore, the WHO study clearly indicated that further research was required for the populations in which stronger effects were observed and that zero tolerance had no scientific certainty.

The implications of the variability in responses are illustrated by the work of Patel et al. [37]. These authors preconize that, due to the lack of randomized control trials, routine recommendations to initiate alcohol consumption for cardioprotection should not be made, especially given its potential for addiction and abuse. Nevertheless, from their review, there was no significant justification to encourage abstinence in light to moderate drinkers (10 g to 20 g alcohol/day). The authors add that “if you are physically active, do not smoke, eat a healthy diet, are not overweight, and have no family history of heart disease, there is little added benefit from drinking alcohol to protect your heart. However, in patients at risk for cardiovascular events who already responsibly consume alcohol, there is little reason to encourage abstinence”.

5.1.2. The Enhanced WHO Reports

The continuation of the studies funded by WHO have clarified some sources for the observed heterogeneity. The GBD 2020 study [58] determined the theoretical minimum risk exposure level (TMREL), corresponding to the nadir point of the J-shaped curve (see Section 5.2.1), and the non-drinker equivalence (NDE), corresponding to the alcohol level equivalent to a relative risk of 1. These authors showed an increase in TMREL from 0 in the GBD 2016 [55] to 5.11 g alcohol/day (

Table 1. Theoretical minimum risk exposure level (TMREL) and non-drinker equivalence (NDE) estimates based on all-attributable cause alcohol relative risk curves. Values in standard drinks (10 g alcohol per day); UI, uncertainty intervals; data of 2020 [58].

	TMREL	UI (95%)	NDE	UI (95%)	Main Causes
Overall (GBD 2016 [55]) a	0	0–0.800	-	-	-
Overall (GBD 2020 [58]) a	0.511	0.4–0.700	1.72	0.80–3.30	-
	1.87	0.50–3.30			-
Males aged 15–39 years	0.136	0–0.400	0.249	0–1.00	66.3% all injuries b
Females aged 15–39 years	0.273	0–0.500	0.546	0–1.30	47.9% all injuries
40 years or older	0.114	0–0.403	0.193	0–0.900	-
	1.87	0.50–3.30	6.94	3.40–8.30	-
Males aged 40–64 years	0.527	0.400–1.00	1.69	0.800–3.20	34.4% CVD c, 23% all injuries
Females aged 40–64 years	0.562	0.400–0.800	1.82	1.00–3.10	30.85% CVD, 16.7% all injuries
Males 65 years or older	0.636	0.500–1.00	3.19	1.50–5.20	57.3% CVD
Females 65 years or older	0.656	0.50–1.00	3.51	1.70–5.50	56.6% CVD

^a Data retrieved from Table 2 in reference [58]. ^b All injuries: mainly transport, self-harm and interpersonal violence. ^c CVD, Cardiovascular diseases: mainly ischaemic heart disease, intracerebral hemorrhage and ischaemic stroke.

). The NDE value was 17.2 alcohol/day (uncertainty intervals 95%: 8–33 g/L alcohol). In addition, several limitations were properly acknowledged (e.g., the absence of drinking patterns, the absence of adjustment for measurement and selection biases, reliance on self-reported data, the lack of distinction by alcohol type or quality, and the lack of inclusion of health conditions). As a result, the authors correctly advised tailored guidelines and recommendations on alcohol consumption, prioritizing interventions targeted at more vulnerable demographic groups. Particular attention should be given to young males, who were the largest group with the highest level of harmful alcohol consumption, mostly due to injury hazards [58].

Nevertheless, despite increasing the number of diseases, Bryaska et al. [58] did not include other cancers that have not been associated with moderate alcohol consumption [59,60].

Shield et al. [61], in the most recent WHO work, used relative risks estimates to calculate the alcohol-attributable burden (deaths or DALYs lost) and the age-adjusted attributed burden (deaths or DALYs lost), yielding the Population-Attributable fraction (% of deaths or DALYs lost). The authors acknowledged that uncertainty intervals are likely to underestimate the true error. In addition, they observed that the reduction in health harms was accompanied by an increase in alcohol consumption from 2019 to 2020. This discrepancy was not associated with potential methodological limitations, but with a likely decrease in the underlying risk of diseases, conditions and injuries causally related to alcohol. However, the average alcohol per capita consumption (about 5 L alcohol) would only correspond to about 1 standard drink/day. Therefore, aggregate health harm experienced across the mean alcohol use spectrum may be used to help determine where alcohol policies should be targeted for highest efficacy [62]. Indeed, despite a drop in alcohol consumption, more people died due to alcohol-specific causes during the COVID-19 pandemic in Europe, probably by heavy drinking [63].

As far as we are aware, the conclusions from the most recent WHO studies have not been incorporated in the present guidelines. In fact, when zero tolerance is advised and understandable under the AA perspective, new scientific evidence is not really necessary.

5.2. EHN Studies on Cardiovascular Health

The EHN policy brief indicates that the research supporting the claim was conducted by the Stockwell research group (Supplementary Table S2). They state, based on new evidence, that the previously assumed protective effects against cardiovascular disease are not supported. The rationale is based on eliminating the J-shaped curves.

5.2.1. The Debate on J-Shaped Associations and Confounding Factors

The phenomenon underlying the J-shape is known as hormesis, in which a toxin shows a decreasing effect until a certain concentration (the nadir point) and increases thereafter [64]. The question in public health is relevant because it shows that a toxin/poison may have a protective effect. The effect of alcohol on cardiovascular diseases is a well-known example of this behavior [65], contrary to the effect of smoke [66].

With alcohol, the main issue is the definition of the baseline that should correspond to individuals where the absence of alcohol consumption is certain. Appropriate adjustment for confounding factors is mandatory when comparing prognostic factors. The comparison with true abstainers implies removing those that may have impaired health when compared with healthier drinking cohorts, so that alcohol affects the latter cohort relatively less (see [36,67] for a full discussion). These poorer health individuals include “sick quitters” (those who avoided drinking due to health reasons) and those who have unhealthier lifestyles (e.g., do not engage in physical activity). These are examples of reverse causation, in which poor health leads to drinking cessation, rather than the reverse. Moreover, in developed societies, drinkers are supposed to have better health because they belong to wealthier ranks, and the poor have lower health, so alcohol is not so harmful in the former segments. The inclusion of occasional drinkers in the abstainer group is also a confounder. The consideration of all these confounding factors enabled Stockwell et al. [36] to distinguish high- from low-quality studies as a function of baseline-appropriate corrections.

5.2.2. How to Make the J-Shape Disappear

The accepted combination of very different outcomes in MA makes it relatively easy to eliminate hormesis. The case may be exemplified by the report by Stockwell et al. [36], which abolished the J-shape by synthesizing results from six high-quality studies (Figure 3). Among these studies, three reported J-shaped associations and originated from the USA and Europe [68,69,70]. The absence of a J-shape was observed in one study of vodka consumption in Russia [71], one involving individuals from rural Northern Japan [72], and another conducted with a cohort of African Americans in the USA [73].

This absence may be easily explained simply by: (i) the higher harmful effect of spirits [74,75], (ii) higher sensitivity of Asian populations [76], and (iii) higher susceptibility of African Americans to cardiovascular diseases [77]. Interestingly, all six of these studies are not recent (2001–2014) and so the AA claims were not supported by new evidence. What seems to be new is the utilization of observational meta-analytic associations without stating its limitations, despite being an appropriate methodology when randomized control trials are not indicated or ethical to conduct [78].

Figure 3 shows a clear difference in the mean effect size (non-coincident CI limits) only for the high-volume drinkers. A higher difference between both curves was obtained by removing the Bergmann et al. [70] study (Supplementary Figure B5 in [36]). Clearly, the choice of these studies was essential to reduce alcohol’s mean effect and to show the importance of correct baseline definition. The J-shaped curve below RR of 1 vanished by using true abstainers as reference. Consequently, the NDE decreased from >45 g/day to between 25 and 44 g/day, considering the CI. Biddinger et al. [79] also showed this reduction in the J-shape due to cardiovascular diseases but, in this case, the NDE was similar irrespective of baseline correction, mainly for coronary heart disease (Figure 1 in [79]).

The relevant conclusion from these results is that the reduction in the J-shape did not support the idea that every drop of alcohol is hazardous. The results were skewed by merging studies using different alcohol sources and populations with different sensitivities. The likely publication bias [80] was not addressed by Stockwell et al. [36]. Moreover, fewer than 10 studies were included, which is insufficient [81].

5.2.3. The Persistence of J-Shaped Relationships

The transformation of meta-analytic correlations into cause-and-effect relations would be perfect to understand the role of ethanol on health. However, with alcohol-mediated effects, this endeavor is not simple given the complexity of the involved factors. To improve causal inference, Visontay et al. [82] reviewed several observational studies that employed improved approaches to minimize confounding factors and triangulated with biomarkers related to alcohol consumption. Not only was the J-shape kept, but the reverse J-shape and U-shape were also apparent, meaning that the abstinence may have higher relative risks than moderate or even heavy drinking for several outcomes (e.g., prostate cancer and related mortality, mental health).

Using the burden of proof approach, as described by Zheng et al. [83] and Carr et al. [84], it was found that alcohol, at average consumption levels, was inversely associated with ischemic heart disease (IHD) by pooling of wide diversity of observational studies.

Seemingly, the early statement of Rehm et al. [68] continues to be valid: “Overall, the J-shaped curve seems to be remarkably stable and independent of assessment measures”. Future studies should triangulate evidence across multiple research domains (e.g., Mendelian randomization, RCT’s) and not just rely on observational studies to estimate alcohol’s overall impact on health [36].

6. Recent Studies in Alcohol Related Meta-Analysis

The relationship between level of alcohol use and all-cause mortality based on meta-analytic cohort studies has been questioned [85]. Nevertheless, the approach continues to be widely used and subjected to notable improvements. In particular, the attempts to minimize the associated pitfalls have been mostly compelled by the wanted elimination of the J-shaped curve. From an AA perspective, this behavior may give the wrong impression that moderate alcohol consumption has beneficial effects (moderate drinkers having better health than abstainers). The previously discussed aspects related to “sick-quitters” and confounding factors are properly acknowledged in recent studies considering the following:

(a)

Occasional drinkers as reference category

The appropriate baseline level should be constituted by occasional drinkers instead of never-drinkers. Previous evidence showed that individuals who report no current alcohol use (e.g., AUDIT-C  =  0) are a heterogeneous group that comprises individuals who quit drinking after alcohol-associated or other health problems, those who stopped drinking for other reasons, those who misreport abstinence, and a relatively small proportion of lifetime abstainers [86]. Thus, AUDIT-C  =  1 (occasional drinkers) should be used as the referent group, especially for middle-aged and older adults. Otherwise, results may underestimate hazard ratios in relation to heavier drinking, because abstainers have also very different sociodemographic and lifestyle characteristics from the general population [87,88].

(b)

Self-reported levels and patterns of consumption

In observational studies alcohol consumption measurements rely on self-reported data that may lead to an underestimation of the observed associations for high consumption levels and an overestimation for moderate levels [88]. Despite this caveat, researchers continue to rely on questionnaires to assess alcohol consumption. The present improvements reside in the assessment of the frequency of alcohol consumption (drinking patterns) together with the usual consumption levels [89]. Moreover, drinking with food should also be considered in the drinking pattern [90].

(c)

The mandatory inclusion of a wide variety of confounding factors

The lower quality of earlier meta-analysis resulted from improper handling of confounders like age, sex, diet, smoking, weight or socioeconomic status and lifestyle [91]. Presently these aspects, besides conventional demography (e.g., age, sex, race or ethnicity, socioeconomic status), include lifestyle, dietary habits and disease history. However, although most studies report confounding bias, they rarely follow the cautionary principle when interpreting the results [91].

Table 2. Self-reported alcohol consumption using observational meta-analysis with large cohorts of different origins.

Study Cohorts	Sociodemographic and Clinical Covariates	Reference Cohort	Main Conclusions	Reference
22,091 individuals aged 30–79 from Chongqing, China	Age, sex, marital status, household income, education, smoking status, physical activity, red meat, fruits, vegetable intake, and spicy food intake, disease history (hypertension, diabetes)	Non-drinker (0 g/day)	Moderate drinking (13 to 36 g/day) and drinking 6–7 days per week were associated with a reduced risk of total stroke	[89]
43,071 individuals from Spain, aged 15 years or older	Sex, age, education, marital status, size of the residence municipality, lifestyle behaviors, diet style, body mass index, health status, disease number	Infrequent occasional drinkers (consumption ≤ once/month)	Low-volume drinkers (<20 g/day) had a comparable mortality risk Never-drinkers, former drinkers, regular drinkers (>20 g/day), and those engaging in weekly heavy episodic drinking (HED), experienced higher mortality risk	[87]
135,103 individuals of UK Biobank, with 60 years or older	Sex, age, race and ethnicity, education, drinking with meals, smoking, leisure-time, physical activity, time spent watching television, prevalent morbidities	Occasional drinkers (≤2.86 g/d)	No evidence of a beneficial association between low-risk (men: >2.86–20.00 g/d; women: >2.86–10.00 g/d) consumption and mortality Detrimental association of low-risk drinking in individuals with socioeconomic or health-related risk factors, especially for cancer deaths Preference for wine and drinking only during meals were associated with lower all-cause mortality	[92]
77,409 individuals of the Republic of Korea during 4 years	Age, sex, social income, body mass index, regular exercise, smoking status, several blood analyses, hypertension, diabetes mellitus, dyslipidemia	Persistent non-drinkers	Both within-the-guideline (<8 g/day for women and <16 g/day for men) and above-the-guideline drinkers had a lower major adverse cardiovascular and cerebrovascular events (MACCE) risk than the non-drinkers Lowest risk with once-per-week and a mild amount per body weight (≤0.33 g/kg/week)	[93]
531,851 health insured patients from Washington state (US) over 8 years	Age, sex, race and ethnicity, socioeconomic status, tobacco use, body mass index, substance use disorder and comorbidities	Low risk alcohol use	Patients with no use or very high-risk use had higher mortality Moderate-risk users had lower mortality Associations were stronger among young adults but not among sexes	[94]
353,834 individuals of UK Biobank, 40–69 years, from 2006 to 2010	Age, sex, education, household income, physical activity, smoking habits, comorbidities	Median of alcohol consumption (9.1 g/day)	Low and high consumption were associated with higher risk of contracting and dying of infectious diseases	[95]
181,607 individuals of New South Wales, Australia, with 45 years or older (2005–2009), over a median of 11.4 years	Smoking, physical activity, eating habits, marital status, annual income, medical history	Low-volume drinkers (≥10 to ≤35 g/week) Pattern of drinking: cut-point of 1–3 vs. 4–7 days/week	J-shape for all alcohol-related cancer and diabetes U-shape for lower respiratory infection J-shaped for ischemic heart disease mortality: decreased risk from ≥40 to ≤200 g/week, and increased risk from ≥1170 g/week No significant effect for dementia and external causes (e.g., accidents, suicide, fall) All-cause mortality higher for >300 g/week	[88]

illustrates the most recent studies based on large cohort observational studies where the above-mentioned issues have been addressed.

Mendelian Randomisation: The Ultimate Approach

A major breakthrough in alcohol studies was due to Mendelian randomization (MR), which is now frequently used under the expectation that it might definitely lead to the elimination of the J-shaped curve. The objective is to use genetic variation as a natural experiment to improve causal inferences from observational data [96]. The main advantage relies on the minimization of biases from reverse causation and confounding factors that are ubiquitous in observational research [97]. However, the first studies showed substantial heterogeneity in the chosen methodology and in the reporting of the methodological quality, hampering the obtention of strong conclusions [98]. Nonetheless, with improved methods for promoting causal inference, small benefits of low-level alcohol consumption appear to persist for selected health outcomes, but not others [99].

Concerning cardiometabolic diseases (coronary heart disease and type 2 diabetes mellitus), the observational J-shaped curve was not shown under MR outcomes [100]. Instead, these authors found a horizontal line and concluded that there was not a causal association, since there were not beneficial or harmful effects under heavy alcohol consumption (AUDIT-C level ≥ 8) [100]. Moreover, there has been no consistent evidence for a harmful effect of alcohol consumption on cancer risk [101]. This research team has just found inverse associations with certain cancer types and concluded that this approach did not provide evidence that alcohol consumption is a cause of all cancers [102].

In conclusion, MR studies should be used in conjunction with, rather than instead of, traditional longitudinal and epidemiological cohort studies to provide a more comprehensive picture of the relation between alcohol and health [103]. Alvarez-Mon et al. [104] reinforced the need for large randomized controlled trials in drinkers promoting cessation versus moderation to solve the continuing contradictions, despite their difficult or unethical implementation.

7. Limitations and Future Prospects

The critical evaluation of meta-analytic studies was limited to a small number of cases that served as the scientific support for global agencies to the claim “all alcohol is harmful”. The literature on this subject is vast, but the analyzed works were sufficient to show that the supporting articles do not substantiate the claim. The importance of these studies was not neglected. Indeed, they identified where further research should focus to address the harmful effects of ethanol on health, as long preconized by the WHO [105]. Several other medical societies have published lower limits, but total abstinence was not advised [106]. Furthermore, both benefits and harms should be juxtaposed to determine trade-offs between them [21].

Another aspect that deserves a deeper analysis is the alcohol origin. Given the observed different effects of spirits, beer and wine, in some studies, the acknowledgement of this variable should be mandatory, as recently demonstrated by Mayer and Fontelo [107]. In addition, patterns of consumption and drinking with food should also be treated as covariates. Therefore, studies should differentiate the drinking patterns and not use alcohol measures to determine the effects [69,73,108,109]. If these issues are not evaluated, authors should clearly state these limitations or be advised to do so by editorial reviewers.

A critical evaluation of the different effect measures was not performed to assess the significance of the results. In the case of observational studies, vibration of effects to provide insight regarding the stability of findings should be performed [110]. Concerning relative risks, false assumptions combined with poor reporting and misleading dissemination may even propagate absurdity [111]. Estimates of mortality and disability-adjusted life years (DALYs) attributable to risk factors may be rather variable, particularly for behavioral risks [112].

Fermentation scientists are not immune to all discussions on the risks of alcohol on health and are genuinely interested in finding the right answer. For them, alcoholism is not an issue; they are on the frontline of preventing alcohol abuse, as portrayed by the efforts put into moderation and responsibility [113]. New methodologies should be proposed to study the complex relationship between alcohol in moderation, disease burden and mortality. Observational epidemiology plays an important role in identifying relevant associations, but more definitive evidence on health benefits or harms requires long-term randomized controlled trials [34,106]. Mendelian randomization studies are supposed to provide a link with genetic causes but have shown an unexpected absence of alcohol’s effect on several cancers [85]. As a result, these authors enumerated a series of drawbacks of such studies that should be acknowledged in future research. Furthermore, alcohol exposure biomarkers [98] may be better indicators of consumption than self-reported questionnaires.

Global health agencies would benefit from including primary scientists, either from fermentation or nutrition, on research teams to improve the ecological validity of epidemiological nonrandomized observational studies. The planned WHO studies on the impact of processed foods and additives on health would provide an opportunity to demonstrate the utility of multidisciplinary teams in establishing advisory guidelines.

8. Conclusions

The WHO and other health agencies have used observational meta-analytic studies to provide scientific support for their present guidelines. However, these studies only offer associations that must be further tested to find cause–effect relationships. Moreover, observational studies showed a high heterogeneity in the outcomes, as demonstrated by relatively high prediction intervals. Therefore, we found no foundation for a generalized ban on alcohol consumption based on the claim “there is no safe amount that does not affect health”. This statement should be understood from the perspective of the prevention of alcohol abuse disorders (AUDs). Efforts must be strengthened in the prevention and treatment of this social threat. To achieve this goal, recent meta-analytic evidence justifies the establishment of guidelines tailored for specific diseases, or harms, and for sensitive populations.

Research by fermentation scientists on the development of healthier alcoholic beverages remains valid from the perspective of moderate and responsible consumption. Moreover, their work contributes to preserving cultural heritage and promoting rural sustainability.