Observational evidence does not necessarily imply causation

Although part [but not all; see below] of the observational evidence has linked vegetarian and vegan diets to a lower risk of overweight, obesity, and illnesses such as ischemic heart disease, diabetes, diverticular disease, eye cataract, and cancer [Appleby & Key 2016], as well as vulnerability to coronavirus infection [Kim et al. 2021], this is not a valid reason to assume that ASFs are unhealthy. Below we argue that such observational data are not a solid basis for causal interpretations or dietary recommendations, for a variety of reasons. 
 
The same is valid for the specific case against red and/or processed meats, despite numerous reported associations between their intake and mortality [Pan et al. 2012; Larsson & Orsini 2014; Etemadi et al. 2017; Schwingschackl et al. 2017; Zheng et al. 2019; Wang et al. 2020], even including 'accidental death' [Sinha et al. 2009], as well as a variety of illnesses, including intestinal disease [Cao et al. 2018; Papier et al. 2021], asthma [Adrianasolo et al. 2019], pneumonia [Papier et al. 2021], kidney disease [Kelly et al. 2017], type-2 diabetes and cardiometabolic disease [Pan et al. 2011; Kaluza et al. 2012; Chen et al. 2013; Feskens et al. 2013; Johnson et al. 2013; Abete et al. 2014; Rohrmann & Linseisen 2016; Yang et al. 2016; Wolk 2017; Ekmekcioglu et al. 2018; Kim & Je 2018, Al-Shaar et al. 2020; Papier et al. 2021a, 2021b], some cancers [Norat et al. 2002; Huncharek & Kupelnick 2004; Larsson et al. 2006; Bandera et al. 2007Faramawi et al. 2007; Larsson & Wolk 2012; Wang & Jiang 2012; Yang et al. 2012; Huang et al. 2013; Qu et al. 2013; Xu et al. 2013; Zhu et al. 2013; Fallahzadeh et al. 2014; Farvid et al. 2015; Caini et al. 2016; Carr et al. 2016; Vieira et al. 2017; Zheng et al. 2017], depression [Nucci et al. 2020].
 
Weak associations are only indicative
 
Taking into account that there can be an abundance of false-positive findings [Bofetta et al. 2008; Young & Karr, 2011], and because of the large bias and uncertainties [see below], the very low relative risk (RR) levels that are typically found in such studies would not be considered as strong evidence in most epidemiological research outside nutrition. Even if a threshold theoretically depends on the risk prevalence in the reference group [Sainani 2011], RR values far below 2 should be considered as poorly informative [Shapiro 2004]. They should not be used to infer strong causal claims, especially when there is a strong suspicion of confounding [McAfee et al. 2010; Alexander & Cushing, 2011; Alexander et al. 2015; Klurfeld 2015; Feinman 2018; Leroy et al. 2018; Händel et al. 2021]. Hazard ratios (HR), specifically, come with their own issues regarding causal inference [Hernán 2010; Uno et al. 2014].
 
To put things in perspective, the association of colorectal cancer with visceral fat is coupled to a RR of 5.9 for the highest compared to the lowest tertile [Yamamoto et al. 2010], overshadowing the value for meat eating (RR of 1.2 per 100g/d of red meat or 50g/d of processed meat). Likewise, presence of the metabolic syndrome at baseline was associated with increased risk of colorectal cancer (HR 2.2) in postmenopausal women [Kabat et al. 2012], acting as an important prognostic factor [Shen et al. 2010], probably driven by hyperinsulinemia (e.g., enhanced angiogenesis) [Liu et al. 2014]. 
 
An assessment using Bradford-Hill criteria concluded that the causality of the associations between red meat intake and cardiometabolic diseases cannot be stated with confidence due to 'weakness of associations' and a 'lack of coherence with short-term experimental evidence' [Hill et al. 2020]. The alleged link with colorectal cancer has been criticized for demonstrating serious or critical risks of bias (risk of confounding, missing data, selective outcome reporting bias) [Händel et al. 2020] and displaying an unclear dose-response effect and weakening evidence over time [Alexander et al. 2015]. After the most comprehensive assessment of the evidence, the observational evidence used to incriminate red and processed meats for increased mortality or a variety of morbidities was deemed to be of 'low-' to 'very-low-' certainty [Han et al. 2019; Johnston et al. 2019; Vernooij et al. 2019; Zeraatkar et al. 2019; see elsewhere]. In support of this finding, another GRADE-based approach concluded that the recommendation to reduce the consumption of processed meat and meat products in the general population to lower the risk of cancer 'seems to be based on evidence that is not methodologically strong' [Händel et al. 2021]. Likewise, an umbrella review classified evidence for the link between red and/or processed meats and colorectal cancer as merely 'suggestive' (i.e., above 'weak' but below 'highly suggestive' and 'strong' [Papadimitriou et al. 2021].
 
Problematic input data: questionnaires and biases

The input information for epidemiological assessments is usually obtained from surveys and food frequency questionnaires, providing an imperfect and, arguably, 'fatally flawed' way of data collection [Schatzkin et al. 2003; Archer et al. 2013, 2018; Händel et al. 2021]. In the case of meat, for instance, the categories and descriptors used for muscle foods display heterogeneity and often do not match those of public regulatory definitions [O'Connor et al. 2020]. Also, processed and (red) meats are often grouped together in 'all meat' even though they may lead to very different results [Männistö et al. 2010]. Furthermore, memory-based reporting issues undermine data robustness, while social desirability bias may cause underreported intake of meat in self-defined vegetarians and other health-conscious groups [Haddad & Tanzman 2003]. 
 
In a UK survey, only one on four respondents who considered themselves to be reducing meat consumption had actually cut down on a variety of meats over the year [Richardson 1993]. Moreover, some of the studies claiming to look at 'vegetarian diets' are in reality combining true vegetarians with semi-vegetarians to avoid low sample sizes. Therefore, the term 'vegetarian' in self-reports requires caution [Juan et al. 2015]. In the Oxford Vegetarian Study, 23% of the 'non-meat-eaters' ate meat occasionally but less than once a week, or ate fish, or both [Appleby et al. 1999]. 
 
Survivorship bias is also likely, since many vegans and vegetarians (up to 70-80%) rapidly revert to eating ASFs, a third even within three months of their change of diet [Faunalytics 2014, 2015]. Only 12- 24% of current vegans may be in the diet for >5 years [Kerschke-Risch 2015; FCN 2018; VOMAD 2019], 7% for >10 years, and 3% for >20 years [VOMAD 2019]. Studies looking into 'plant-based' diets may have a selection bias for the minority of subjects that do well on such diets.
 
Lifestyle confounding and healthy-user bias
 
It is primordial yet very difficult to unambiguously disentangle specific dietary effects from the overall complexity of lifestyles within observational data sets. For example, higher red meat intake increases cancer risk at low levels of fruit and vegetable intake, yet may show a neutral to protective association in people who also report an increased consumption of fruits and vegetables [Maximova et al., 2020]. This may or may not point to protective interactions, but can as well be indicative of more intricate confounding; in any case, such findings demonstrate that context is primordial. 
 
Similarly, the association between meat eating and inflammatory response or disease becomes non-significant or is at least substantially attenuated after adjustment for excess body weight [Montonen et al., 2013; Chai et al. 2017; Papier et al. 2021]. Yet, whereas such well-characterized factors as obesity can be statistically accounted for as such, at least to some degree, it is impossible to sufficiently correct for all lifestyle confounders. Even anxiety has recently been mentioned as a potential confounder of the associations between meat eating and colorectal cancer [Beslay et al. 2020]. Socio-demographic elements, in particular, are particularly difficult to account for.
 
In observational studies, the problem of healthy user and related biases (e.g., healthy adherer) is a fundamental one [cf. Shrank et al. 2011]. It explains at least part of the deviations between the results of observational studies and randomized controlled trials [see also elsewhere]. Western upper-middle classes are inclined to eat less meat, as they are susceptible to what is perceived as virtuous eating, obediently following official advice [Leroy & Hite 2020]. Self-declared vegetarians tend to be more health conscious and socioeconomically advantaged [Bedford & Barr 2005], while plebeian eating patterns are less prone to dietary guidance. Meat intake thus parallels a higher intake of ultraprocessed foods, obesity, smoking, and lower physical activity [Alexander et al. 2015; Fogelholm et al. 2015; Grosso et al. 2017; Mihrshahi et al. 2017; Turner & Lloyd 2017; Hur et al. 2018]. Such associations captured by epidemiology then serve as positive feedback to reinforce the dietary recommendations that created them in the first place [Leroy & Hite 2020]. Essentially, the dietary guidelines behave as a self-fulfilling prophecy, reinforcing an animal/plant binary that started out as an ideological construct in the 19th century [see elsewhere].
 
When comparing people with similar socio-demographic characteristics, no differences in mortality are seen between vegetarians and meat eaters [Chang-Claude et al. 2005; Appleby & Key, 2016; Appleby et al. 2016; Mihrshahi et al. 2017]. Also, omission of Seventh-Day Adventist studies, describing a specific subset of communities following healthy lifestyles, from meta-analyses results in the weakening or even disappearance of the beneficial associations between vegetarianism and cardiovascular health [Kwok et al. 2014; Dinu et al. 2017; FCN 2018]. 
 
To further illustrate this point, lifestyle confounding appears when comparing distinct cultural groups. For processed/red meats, associations found in North-America are not necessarily valid elsewhere, where they may be neutral or protective [see below; O'Sullivan et al. 2013; Wang et al. 2016; Grosso et al. 2017; ACC 2018; Iqbal et al. 2021]. Similarly, associations between egg consumption and type 2 diabetes typify US studies but vanish in European (neutral) ans Asian (protective) studies [Djoussé et al. 2016; Tamez et al. 2016; Wallin et al. 2016; Drouin-Chartier et al. 2020].

Cherry picking: dismissal of neutral outcomes 

Developing a case against ASFs, in particular red and processed meats, often leads to a biased selection of evidence. In a critique on an influential meta-analysis favoring vegan and vegetarian diets [i.e., Dinu et al. 2017], it has been shown that such attempts are characterized by selective reporting, exaggerated cause-and-effect statements, and a refusal to refer to studies that find no associations with chronic diseases [Fenton & Gillis 2017]. Also, a critical analysis of studies looking into the link between processed meats and colorectal cancer has shown that '75% of the eligible studies had a moderate to serious risk of missing data, and about half of the studies had issues with bias in the selection of the reported results' [Händel et al. 2021].
 
Many authors were indeed not able to affirm a clear link between veganism, vegetarianism, or meat avoidance and lower mortality or morbidity due to various diseases of modernity, or obtained mixed results for some meat types (mostly processed ones) but not for others [Heilbrun et al. 1989; Thun et al. 1992; Missmer et al. 2002; Flood et al. 2003; Huncharek et al. 2003; Key et al. 2003, 2009; Sato et al. 2006; Lee et al. 2008; Alexander & Cushing 2009; Alexander et al. 2009, 2010a,b,c; 2011; Wallin et al. 2011; Kappeler et al. 2013; Parr et al. 2013; Lee et al. 2013; Kwok et al. 2014; Lippi et al. 2015; Mihrshahi et al. 2017; Hur et al., 2018; Mejborn et al. 2020; Wang et al. 2021; Iqbal et al. 2021]. 
 
As a result, the common narrative that red meat causes colon cancer and heart disease becomes messy. Even if vegetarians may display lower mortality from ischemic heart disease than meat eaters in some large prospective studies, no protective associations with mortality from cerebrovascular disease and various cancers were found in those same datasets (including colon cancer) [Key et al. 1999]. Also, no association was found between red meat and heart disease or diabetes (although processed meats paralleled higher risk) [Micha et al. 2010], while a GRADE analysis of the relationship between vegan diets and cardiovascular disease indicated variable associations and weak to very weak quality of evidence [Kaiser et al., 2021]. One very large study within Harvard’s Pooling Project of Prospect Studies of Diet and Cancer did not find a link between red and processed meat intake and colorectal cancer, but was only abstracted and never published [Cho & Smith-Warner 2004], which has raised suspicion given the dietary beliefs prevailing at Harvard's School of Public Health [Butterworth 2007]. 
 
When arguing for restriction of ASFs based on observational studies, eggs should be exonerated based on mostly neutral findings - with the occasional association being limited mostly to a US context [Rong et al. 2013; Shin et al. 2013; Tran et al. 2014; Alexander et al. 2016; Djoussé et al. 2016; Tamez et al. 2016; Wallin et al. 2016; Dehghan et al. 2020; Xia et al. 2020; Krittanawong et al. 2021]. The case against dairy leads to inconsistent yet mostly neutral results [Jayaraman et al. 2019] and would overall lead to a net protective effect [Scrafford et al. 2020], while even the persistent narrative stating that it would cause prostate cancer is unlikely [Preble et al. 2019].

Cherry picking: dismissal of protective outcomes

Embracing observational data to incriminate omnivore diets would at the same time undermine the premise that vegan diets are the best choice to avoid chronic disease and death. Avoidance of ASFs that are associated with benefits would then have to be considered as suboptimal for some populations, as could be argued for fish and seafood [He et al. 2004; Bouzan et al. 2005; Larsson & Orsini 2011; Chowdhury et al. 2012; Zheng et al. 2012; Jiang et al. 2016; Zhao et al. 2016; Schwingschackl et al. 2017; Qin et al. 2018; Rimm et al. 2018; Black et al. 2019; Kim et al. 2019], white meat [Shi et al. 2015; Xu et al. 2020; Lupoli et al. 2021], eggs [Gopinath et al. 2020; Zhuang et al. 2020; Xu et al. 2020; Krittanawong et al. 2021], and dairy [Shimizu et al. 2003; Aune et al. 2013; Gao et al. 2013; Astrup 2014; Tapsell 2015; Gijsbers et al. 2016; Lovegrove & Hobbs 2016; Wu & Sun 2017; Yoshida et al. 2019; Scrafford et al. 2020; Bhavadharini et al. 2020; Haugsgjerd et al. 2020; Jin et al. 2020; Poppitt 2020; Papadimitriou et al. 2021].

Even the eating of (processed) meat has in some instances been associated with protective outcomes, including a lower morbidity or mortality [Lee et al. 2013; ACC 2018; Yen et al., 2018; Black et al. 2019a, b], increased telomere length, with a potential benefit on life span [Kasielski et al. 2016], and decreased cognitive decline in elderly [Xu et al. 2020]. 
 
Several studies have even linked vegetarian and vegan diets to poorer health and lower quality of life [Sebeková et al. 2001; Krajcovicová-Kudlácková et al. 2002; Ingenbleek & McCully 2012; Burkert et al. 2014; Buscail et al. 2017; Ghoshal & Singh 2017; Iguacel et al. 2019, 2020; Tong et al. 2018; Vanacore et al. 2018; Acer et al. 2019; Borude 2019], including reduced healing of postsurgical scars [Fusano et al. 2020] and psychological well-being [Dobersek et al. 2020]. While meat-containing diets are more frequently linked to ischemic heart attacks, haemorrhagic strokes occur more frequently with vegetarian/vegan diets [Tong et al. 2019; Grüngreiff et al. 2020]. As such studies may be equally criticized for their lack of causal proof, they should at least serve as countermaterial in the overall debate.
 
The Blue-Zone argument
 
A link between 'plant-based' eating in communities with exceptional longevity (so-called 'Blue Zones') is often advanced in support of ASF restriction [Poulain et al. 2004]. Such communities have been identified in Ikaria (Greece), Okinawa (Japan), the Ogliastra Region (Sardinia), Loma Linda (USA), and the Nicoya Peninsula (Costa Rica). The Blue Zones® concept goes back to 2005 [Buettner 2005] and has recently been acquired by the Seventh-Day Adventist Church [Adventist Health 2020], in support of their dietary evangelism and religious calls for meat restriction [see elsewhere]. 
 
The Blue Zones argument may be flawed for various reasons. First, identification of supercentenarians in Blue Zones may suffer from age-reporting errors, skewed interpretations, and registration errors and fraud in remote communities (clerical errors, absence of birth certificates, pension frauds) [Newman 2019, 2020]. 
 
Moreover, the so-called 'traditional' Okinawan diet with allegedly only 9% of protein [cf. Le Couteur et al. 2016] is mainly one of starvation, as it was registered as a post-war dietary snapshot in 1949 by the American administration, after the decimating effects of the war on livestock [Fish 1988]. If anything, Okinawa was not influenced by Buddhism and levels of pork and goat consumption were historically 'exceptional among Japanese food consumption' [Shibata et al. 1992; Poulain & Naito 2005], 'all families raised pigs, and chickens and sometimes other farm animals, such as goats' [Willcox et al. 2014], 'the islands of Okinawa and Jeju are well known for their pork food culture' [Lee & Hyun 2018]. Among other ASF regularly eaten in Okinawa are fish and seafood, especially giant clams [Sho et al. 2001, 2008 ; Claus 2017]. "Animal fats were mainly used for cooking' [Okuyama et al. 1996]
 
The argument also may overlook that not only plants but also dairy and moderate meat intake are independently associated with improved physical function in the Blue zones of Sardinia and Costa Rica [Nieddu et al. 2020] and with longevity in Okinawa [Shibata et al., 1992]. In Costa Rica, a majority of centenarians were found eating dairy and eggs daily, fish and poultry several times per week [Chacon et al. 2017]. In Sardinia, exceptional concentration of centenarians are found in pastoralist population, not among cultivators [Pes et al. 2011]. Studies showed that centenarians in Okinawa ate twice the amount of meat as mainland japanese centenarians [Kagawa 1978 ; Akisaka et al. 1996]. Be that as it may, extracting ASFs from the data sets as an explanatory factor may be problematic in either direction of the health argument. The latter is strongly confounded by other factors since members of Blue Zones lead healthy lifestyles in general and are part of functional social communities.

Translate content