Novel modelling methods produce anomalous findings and fail to adequately control for healthy-user bias.
Sartre (1948) gave the choice between an eclair and a bonbon as an example of a nonexistential decision. Is the choice between a drumstick and a steak an existential decision? To the authors of this paper, it is. Etemadi et al claim to compare white meat (mainly chicken) consumers with red meat consumers by equalising consumption mathematically; "without changing the overall meat intake”, avoiding a more realistic model. How far are their findings reliable?
Healthy user bias is the phenomenon whereby health conscious individuals avoid behaviours advertised in the media as unhealthy, including the behaviour of interest, but also other behaviours which may or may not have been measured or controlled for, and risk-takers or those less able to modify their behaviour for various reasons do the opposite. Red meat consumption has been extensively advertised as unhealthy in the US since at least the 1960’s. The situation is different in Europe and the Far East, where as the authors note there is little or no association between unprocessed red meat and mortality. The potential for healthy user bias can clearly be seen in the baseline data (Table 1), with many more men, a more than doubled rate of smoking, and a mean BMI of 28.3 vs 25.8 in the highest red meat quintile.
Are there anomalous findings in this paper that indicate that residual confounding from the healthy user bias has distorted the results?
The strong association between unprocessed red meat consumption and liver disease mortality, HR 2.14 (1.68,2.72) P<0.0001, stands out. The usual causes of cirrhosis and hepatocellular cancer in the US population are alcoholism, viral hepatitis (chronic HCV and HBV infection), abuse of medicines such as vicodin which combine addictive drugs with the hepatotoxin acetaminophen (as tolerance to the opioid increases so does exposure to the hepatotoxin), and NASH. Although alcohol consumption was controlled for, it is under-reported in health surveys by a factor of 2-3, alcoholism is less likely to be reported, viral hepatitis was not recorded, and exposures to alcohol, drugs, and the risk of viral infection will increase over 16 years in risk-taking individuals.
The authors mention a possible mechanism in that hepatic pathogens can be carried in meat products. Contamination of pork organ meats has been an issue historically in some countries. In 1965 Nanji and French "reported significant correlations between alcohol consumption and cirrhosis mortality and between pork consumption and cirrhosis mortality. They also reported a significant correlation between the product of both alcohol and pork consumption and rates of cirrhosis mortality.”
Bridges confirmed this in 2009, also confirming an earlier finding that “the negative correlation between alcoholic cirrhosis mortality in 2003 and dietary beef implies that dietary beef may be protective factor regarding the pathogenesis of alcoholic cirrhosis."
Nanji and French, with others, most recently Kirpich et al, were consistently able to demonstrate that highly saturated fats such as beef fat, as well as cholesterol, were protective against alcoholic liver disease in animal models. Nanji, French, Kirpich and others also demonstrated that polyunsaturated fats are necessary for, and promote, the development of alcoholic liver disease. This finding is also true for acetaminophen toxicity. Chicken meat is one of the main sources of polyunsaturated fat in the US diet, so, while a pathogenic influence from pork, also a source of polyunsaturated fat, cannot be ruled out, the relationship between the larger beef component of red meat vs chicken consumption in the current study is contrary to known mechanisms of liver disease progression, but consistent with healthy-user bias confounding from viral, drug, and alcohol exposures.
The association of unprocessed red meat with respiratory mortality is also striking, as it is greater than the classical associations of cardiovascular disease and cancer mortality, and no convincing mechanism exists to explain it. However high red meat eaters, who are predominantly male and with lower educational and economic status, are more likely to be industrial workers, or to live or work near busy roads, and thus be exposed to pollutants which affect lung function. Exposure to pollutants is a factor rarely recorded in diet epidemiology, but one that may be highly correlated with outcomes of interest.
The current study did not control for consumption of other nutrients such as sugar and high-GI starches, but sugar and high-GI starch consumption correlates with red meat consumption and may also be associated with the outcomes of interest.
It is possible to correct for healthy-user bias to some extent by excluding all individuals who report unhealthy behaviours. For example, Song et al found that animal protein vs plant protein was significantly associated with mortality in their 2 cohorts, but "These associations were confined to participants with at least 1 unhealthy lifestyle factor based on smoking, heavy alcohol intake, overweight or obesity, and physical inactivity, but not evident among those without any of these risk factors." Similarly, when the EPIC-Oxford study looked at health conscious individuals only, the vegetarians had higher rates of all-cause and cancer mortality, contrary to expectations based on multiple previous studies that did not correct for healthy-user bias.
In populations exposed to consistent health messages in the media, healthy-user bias creates a feedback loop by which public health experts see the wisdom of their recommendations reflected back at them, because they have been influencing the members of their study populations, in the manner of an uncontrolled experiment. Unless healthy-user bias is well controlled for, this affirmation will always be exaggerated, and sometimes completely spurious.
One way to assess the limitations of a prediction is the reductio ad absurdum. Kentucky Fried Chicken supplies no red meat, but only white meat. Furthermore, Kentucky Fried Chicken is cooked in oils high in polyunsaturated fatty acids, which are also associated with reduced mortality, HR 0.81 (95% CI, 0.78-0.84), in 3 US populations, according to the analysis of Wang et al 2016. Ergo, if the findings of these papers can safely be taken at face value, persons who only eat meals from Kentucky Fried Chicken outlets can expect to live longer.
 Nanji AA, French SW. Dietary factors and alcohol cirrhosis. Alcohol. Clin. Exp. Res. 1986;10:271–273.
 Bridges FS. Relationship between Dietary Beef, Fat, and Pork and Alcoholic Cirrhosis. International Journal of Environmental Research and Public Health. 2009;6(9):2417-2425. doi:10.3390/ijerph6092417.
 Kirpich IA, Miller ME, Cave MC, Joshi-Barve S, McClain CJ. Alcoholic Liver Disease: Update on the Role of Dietary Fat. Osna N, Kharbanda K, eds. Biomolecules. 2016;6(1):1. doi:10.3390/biom6010001.
 Hwang J. Diets with corn oil and/or low protein increase acute acetaminophen hepatotoxicity compared to diets with beef tallow in a rat model. Nutrition Research and Practice. 2009;3(2):95-101. doi:10.4162/nrp.2009.3.2.95.
 Song M, Fung TT, Hu FB, Willett WC, Longo VD, Chan AT, Giovannucci EL. Association of Animal and Plant Protein Intake With All-Cause and Cause-Specific Mortality. JAMA Intern Med. 2016;176(10):1453-1463. doi:10.1001/jamainternmed.2016.4182
 Cancer incidence in vegetarians: results from the European Prospective Investigation into Cancer and Nutrition (EPIC-Oxford). Key TJ, Appleby PN, Spencer EA et al. Am J Clin Nutr. 2009 May;89(5):1620S-1626S. doi: 10.3945/ajcn.2009.26736M.
 Wang DD, Li Y, Chiuve SE, Stampfer MJ, Manson JE, Rimm EB, Willett WC, Hu FB. Association of Specific Dietary Fats With Total and Cause-Specific Mortality. JAMA Intern Med. 2016;176(8):1134-1145. doi:10.1001/jamainternmed.2016.2417
Competing interests: No competing interests