London School of Hygiene and Tropical Medicine, Keppel St, London WC1E 7HT, UK.
Alcohol is the epidemiological risk factor with the longest history of systematic study. In the 19th Century many life assurance companies partitioned their business according to whether their policy holders were alcohol abstainers or not. This was done in the belief that abstainers were lower risksand therefore could be offered more favourable premiums or bonuses. This policy decision, like many others since, was taken in the absence of good evidence. It was based on what was regarded by many in the business as the self-evident truth that abstention was physically (and morally) the preferred option. Actuarial analyses that supported this belief started to be published in the late 1890s. The most systematic evidence came a few years later in 1904 when Moore published an analysis of mortality among over sixty thousand policy holders who had taken out life assurance with the United Kingdom Temperance and General Provident Institution over the period 18411901.1 He found the mortality rate among male non-abstainers was unexceptional by the standard used by the life-assurance industry at the time. However, mortality among abstainers was considerably lower, resulting in a 10% greater life expectancy from age 30 years than among non-abstainers. Similar results were reported for women.
Considerable thought went into these analyses, including estimation of the potential biases introduced by people changing from being abstainers to non-abstainers or vice versainformation obtained through the requirement that assured lives had to reaffirm their drinking category on an annual basis. Although as Pearl later commented,2 one might question the validity of such self-reports, this question anticipates the much more recent debate concerning how far those who give up drinking because of ill-health bias estimates of the health risks associated with abstinence.3 He was at pains to point out that while the abstainers and non-abstainers differed in terms of alcohol consumption, they were all nevertheless drawn from the narrow ranks of careful citizens inclined to and able to afford life-assurancethat is they belong to the same social class. Here again Moore's argument prefigures contemporary concerns about confounding by socio-economic circumstances and behaviours.
The issue of the health effects of alcohol in the late 19th and early 20th Century were within the context of broader debates about prohibition. The serious social and medical consequences of excessive drinking, or alcohol abuse, were not in dispute, and had indeed been formally demonstrated in follow-up studies of intemperates as early as 1851.4 What was at issue was whether moderate or proper use of alcohol was harmful to health.5,6 This was one of the central interests of Pearl in his classic study of alcohol and longevity first published in 1923.7 These were based on analyses of family history data collected in Baltimore in the early 1920s in a study whose primary aim was to throw light on genetic factors in tuberculosis. It is notable that the drinking habits of the deceased and living were as reported by the index family selectedwhich to modern epidemiological sensibilities is perhaps a little unsettling. From these data, Pearl concluded that moderate consumption does not sensibly shorten the mean duration of life or increase the rate of mortality, as compared with that enjoyed by total abstainers. (p.278) However, despite his stated conclusions some of his analyses did show abstainers to have slightly higher mortality than moderate drinkershence the interest in Pearl as the originator of the U-shaped mortality association as discussed by Marmot and Klatsky in their commentaries in this issue of the IJE. Whether or not Pearl's work stands up to critical scrutiny is less important than the fact that he along with many others in the first half of the 20th century working on alcohol and health were primarily interested in whether moderate consumption had an adverse effect on mortality.
It is striking how many of the features of these earlier debates find a resonance in our contemporary concerns about the health risks of moderate alcohol consumption. As discussed in the set of papers and commentaries in this issue of the IJE, the debate is now more focused, being principally about whether moderate alcohol consumption has a cardio-protective effect. Looked at dispassionately, this problem has many features that play on the weaknesses of observational epidemiology. The exposure (alcohol consumption) is often poorly reported and underestimated, it is highly confounded with other powerful risk factors (for example smoking) and there are strong a priori grounds for believing that reversecausality could be implicated with people who are sick giving up drinking as a consequence. Despite all these problems, the observational evidence looks surprisingly secure and consistent with physiological studies. It seems that moderate alcohol consumption in countries such as the USA and the UK does impart a cardioprotective effect.
The best evidence for this comes from a recently published meta-analysis by Corrao et al.8 which took great care in only including studies whose quality was above a relatively high threshold. They found that for men, risk of coronary heart disease endpoints decreased from 0 to 25 g ethanol/day, and that there was a cardioprotective effect compared to abstainers up to 87 g/day. For women, there was a decline in risk up to 10 g/day and a cardioprotective effect compared to abstainers up to 31 g/day, thereafter the risk of coronary heart disease increased much more steeply than in men.
The energetic debate over the nature of the dose-response effects of alcohol on coronary heart disease has sometimes taken attention away from the links between alcohol and other conditions. These have been examined in another meta-analysis by Carrao et al. 9 that covered six cancer sites (oral cavity, oesophagus, colorectum, liver, larynx, breast), hypertension, cerebrovascular disease, gastric and duodenal ulcer, liver cirrhosis, pancreatitis and injuries. While they conclude that much more work needs to be done in some cases to establish whether an association exists (as for breast cancer) and for others to determine the precise nature of the dose-response relationship, they show that there is no good evidence that alcohol consumption at any level is protective for these conditions. All of their estimated dose-response curves show increasing risk as alcohol consumption goes up.
The cardioprotective effect of moderate alcohol consumption makes the translation of these epidemiological findings into policy very problematic. In order to determine what the optimal level of alcohol consumption might be in any population it is necessary to take into account the relative importance of different causes of death. In an innovatory and comprehensive report, White et al. have attempted to determine optimal levels of alcohol consumption for men and women at different ages in England and Wales.10 This shows clearly that in England and Wales, it is only among men that a protective effect of alcohol consumption with regard to coronary heart disease outweighs its deleterious effects from other causes. Moreover, this net benefit is only apparent in men over age 55 years and among women over the age of 65 years.
The model developed by White et al. should be applied to other countries. What is abundantly clear, as noted by Bovet and Paccaud in this issue of the IJE, is that the net effects of alcohol consumption on mortality and disease crucially depends upon the relative level of coronary heart disease compared to other alcohol-related conditions. In countries with a relatively low burden of coronary heart disease, the burden of alcohol related disease will be greater than in countries where it is more common. Even with the rapidly rising levels of non-communicable diseases such as coronary heart disease in countries such as India, the net effect of alcohol may still remain more deleterious than in the West. This is because the epidemic of non-communicable diseases is developing in a context where deaths from injuries and violence are at considerably higher levels than seen in Western countries such as the UK. With high levels of mortality from external causes the cardioprotective effects may fail to tip the balance to yield a net benefit of even moderate consumption. This sort of reasoning can be extended further to consider the consequences of the steep decline in coronary heart disease being seen in the UK and other countries at present. Again, this may have important effects upon what is optimal consumption in terms of disease and mortality. Clearly, much more work needs to be done to use the approach developed by White et al. to examine the consequences of such changes.
Finally, there is one important issue with respect to the health effects of alcohol that has not received sufficient attention. It is raised by the extraordinary mortality experience of Russia since the mid-1980s11,12 and some other parts of the former Soviet Union. The introduction of Gorbachev's anti-alcohol campaign in the mid-1980s was associated with a marked improvement in life-expectancy in Russia that was driven largely by mortality reductions in men of working age. The collapse of the Soviet Union in the early 1990s was followed by a catastrophic decline in life expectancy. Between 1990 and 1994 it fell by 6.2 years to 57.6 years in men. Subsequently it rose, only to decline again after the economic crash of August 1998. The causes of death that showed the largest proportional changes were alcohol related, including acute alcohol poisoning. However, coronary heart disease also showed fluctuations that had a similar pattern, while deaths from cancers did not. Data on alcohol consumption are difficult to obtain, but show fluctuations that parallel those for heart disease and other conditions. We have argued that alcohol is indeed one of the main drivers of these fluctuations.13 In particular, we believe that binge drinking, a pattern of consumption common in Russia, may be leading to a huge level of sudden cardiac deaths that are not classic deaths from coronary heart disease. The limited evidence linking pattern of drinking to cardiovascular disease supports this position.14 However, this perspective challenges the new orthodoxy that drinking alcohol exerts a cardioprotective effect. Context, however, may be all. The vast majority of studies included in Corrao's meta-analysis of alcohol and coronary heart disease8 are from the USA. There is now an urgent need for research to be done to clarify how far the very high levels of mortality in Russia and other former Soviet countries can be due to alcohol and in particular binge drinking. This is the really pressing scientific and public health challenge in alcohol research.
References
1 Moore RM. On the comparative mortality among lives of abstainers and non-abstainers from alcoholic beverages. Journal of Institute of Actuaries 1904;38:21376.
2 Pearl R. Alcohol and Longevity. New York: Alfred A. Knopf, 1926.
3 Shaper AG, Wannamethee G, Walker M. Alcohol and mortality in British men: explaining the U-shaped curve. Lancet 1988;2:126773.[ISI][Medline]
4 Neison FGP. On the rate of mortality among persons of intemperate habits. J Stat Soc London 1851;14:20019.
5 Starling EH. The Action of Alcohol on Man. London: Longmans, Green and Co, 1923.
6 Medical Research Council. AlcoholIts action on the human organism. London: HMSO, 1938.
7 Pearl R. Alcohol and mortality. In Starling EH, The Action of Alcohol on Man, London: Longmans, Green and Co, 1923.
8 Corrao G, Rubbiati L, Bagnardi V, Zambon A, Poikolainen K. Alcohol and coronary heart disease: a meta-analysis. Addiction 2000;95:150523.[ISI][Medline]
9 Corrao G, Bagnardi V, Zambon A, Arico S. Exploring the dose-response relationship between alcohol consumption and the risk of several alcohol-related conditions: a meta-analysis. Addiction 1999;94:155173.[ISI][Medline]
10 White IR, Altmann DR, Nanchahal K. Optimal levels of alcohol consumption for men and women at different ages, and the all-cause mortality attributable to drinking. London: Medical Statistics Unit, London School of Hygiene & Tropical Medicine, 2000.
11 Leon DA, Chenet L, Shkolnikov VM et al. Huge variation in Russian mortality rates 198494: artefact, alcohol, or what? Lancet 1997; 350:38388.[ISI][Medline]
12 Shkolnikov V, McKee M, Leon DA. Changes in life expectancy in Russia in the mid-1990s. Lancet 2001;357:91721.[ISI][Medline]
13 McKee M, Shkolnikov V, Leon DA. Alcohol is implicated in the fluctuations in cardiovascular disease in Russia since the 1980s. Ann Epidemiol 2001;11:16.[ISI][Medline]
14 McKee M, Britton A. The positive relationship between alcohol and heart disease in eastern Europe: potential physiological mechanisms. J R Soc Med 1998;91:40207.[Abstract]