The publication by The Lancet recently of an article discussing the risk to health posed by alcohol1 seems to have put paid for all time to the notion that moderate drinking does no harm or is even ‘good for you’. It has thus caused a bit of a stir in some quarters.
The authors found that there was no ‘safe’ limit for consumption of alcohol and that even having one drink2 a day increases the risk of developing one of 23 alcohol-related health problems ranging from injury to liver disease and cancer. While alcohol may have a protective effect in respect of coronary heart disease and diabetes, that benefit is outweighed by the risk of other consequences, and of cancers in particular.
Compared with total abstainers, for drinkers the additional risk of those alcohol-related health events increases progressively with each extra drink per day. Two drinks a day increase the risk by about 10%, and four drinks by about 25%. Nine drinks per day – pretty serious drinking – approximately double the risk. Fifteen drinks triple it – although at that point there is a huge confidence interval at the 95% level.
These findings, of course, are expressed in terms of relative risk – the risk for drinkers compared with non-drinkers as a reference. This is fine in terms of the aims of the investigation, which was to appraise the impact of drinking on public health and to inform healthcare policy, provision and spending. But it is only one side of the risk coin: what about absolute risk? In other words, how many drinkers will suffer the negative impacts of alcohol compared with abstainers?
For a steer on this we have to thank The Lancet’s press office, which asked the authors to crunch a few more numbers. According to the data, in people aged 15-95 years, taking one alcoholic drink a day increases the risk of developing one of the 23 potentially alcohol-related health problems by 0.5% compared with non-drinkers. In terms of absolute risk, for 100,000 people, in any one year an extra four will suffer one of those health problems.
Sir David Spiegelhalter, Winton Professor of the Public Understanding of Risk in the Statistical Laboratory at the University of Cambridge, speaking on the BBC Radio 4 programme More or Less (broadcast on 31 August 2018) put it in terms of gin consumption: if 25,000 people were to drink 16 bottles of gin each over the course of a year – that’s 400,000 bottles of gin in total – then one of them would suffer a serious health event during the year. Put another way, 918 of 100,000 15-95 year-olds consuming one drink a day would suffer a serious health event compared with 914 of the non-drinkers. So an extra risk… but a jolly small one.
In life insurance, of course, we tend to think in terms of events per thousand. So four in 100,000 becomes 0.04 per thousand. If we extrapolate a little using the relative risk curve shown in one of the article’s charts we get roughly:
|Number of drinks/day||Relative risk||Absolute risk per 100,000||Absolute risk per 1,000|
Those per mil values in the far right-hand column represent rates of health events, and not deaths. But before thinking about mortality, a couple caveats are called for. First, these are results from a global study, involving countries ranging from Afghanistan to Zimbabwe and from Timor-Leste to the United States. So risk rates in individual countries will vary considerably. Second, the study does not deal at all with mortality, just morbidity; so any mortality implications discussed here are hypothetical.
OK… Take a thousand nine-a-day drinkers, whose relative risk is doubled (and actually the 95% confidence interval is fairly narrow – roughly 1.8 to 2.2). If two of the eight that develop alcohol-related illnesses were to die in that year (in other words, among that ‘impaired’ group there was a mortality rate of 250 per mil), that would represent a mortality rate of 2 per mil overall – that is, two deaths among the original group of 1,000 drinkers.
But how likely is a 25%/250‰ mortality rate among that impaired group? Suppose one were to equate the risk to a pretty significant treated cancer risk, for which underwriting manuals might recommend a rating of 20 per mil. That would mean 0.16 of the eight likely to die in any one year; that’s 0.16 per mil when applied to the original 1,000 nine-a-dayers. OK, 20 per mil too light? Call it 40 per mil – 0.34 per mil overall. Still not much of an extra risk in insurance terms. To achieve a mortality rate of 1 per mil overall looks like it would require an annual mortality rate among the health-impaired group of about 120 per mil, or 12%.
According to the study, 15 drinks a day increase the relative risk of developing an alcohol-related illness by about three times (but note the earlier comment about how confident we can be about whether that calculated risk level reflects reality – at the upper 95% level the risk is 4.5 times). Do the math. It still doesn’t amount to that much in terms of extra deaths per thousand in life insurance terms, even at mortality rates of up to 25% in the newly health-impaired group.
The mortality rates cited so far have been used to illustrate the potential size of the risk. Quantifying the actual risk in terms of excess death rates (EDRs) or percentage extra mortality is difficult: both are very much age-dependent (remember that the calculations are based on figures for 15 to 95 year-olds – a massive range) and alcohol-related illnesses vary hugely in severity, from mild morbidity and mortality risk in the longer term to possibly immediate fatality. The researchers note that in younger people the consequences of drinking are primarily accident-related, and in older age groups predominantly health issues. Nevertheless, playing around with few numbers makes one think…
A point relevant to quantifying risk for life insurance purposes is that the discussion has compared drinkers with non-drinkers. Generally, normal rates of premium for life insurance include the mortality of moderate drinkers. So in our world, EDRs associated with ‘immoderate’ drinking would be lower than those calculated above (assuming we let ourselves be guided by the study). And it may be that in insured portfolios (and perhaps certain other populations too) non-drinkers have a higher mortality owing to the effect of selection and ‘sick quitters’ – prior drinkers, including alcoholics and other problem drinkers, that have given up owing to illness.
Actually, studies of mortality related to alcohol intake generally have an in-built unreliability factor because of a reliance on individuals disclosing how much they drink. It is difficult to estimate the size of a drink (measures poured at home, beers and wines varying a lot in alcohol content). Individuals may not remember the number of drinks they consume, especially at the higher levels of consumption. Also, it is human nature to under-report, especially if there is high awareness of recommended safe-drinking levels and/or the recipient of the information is a potentially disapproving health professional.
While the Lancet article reports an exceptionally thorough investigation, based on nearly 700 data sources of individual and population-level alcohol consumption, nearly 600 studies on the risk of alcohol use and a methodology that even includes adjustment of alcohol sales estimates to take account of tourist and unrecorded consumption, the statistical outcomes can only be estimates of reality, and even then those estimates of health impact may, for reasons just explained, be on the high side.
In his interview for the BBC, Sir David Spiegelhalter reflected on how concerned people should be about moderate alcohol consumption. His view was that if it gave pleasure then quite possibly the benefits outweigh the statistical risk of morbidity. Bear in mind too that the study seems not to have adjusted for other risk factors such as obesity, smoking, diet and other lifestyle factors. For individuals, the absence of other risk factors may counterbalance the alcohol risk.
The investigation conveys clear messages to those responsible for managing public health provision. Maybe it also provides some useful pointers for underwriters and actuaries, notwithstanding the vast range of participating countries and the question of applicability to local insurance environments.
Much food for thought. And need for careful interpretation. Fifteen drinks a day don’t justify a substantial rating? Well, you know what they say about statistics3. But it’s sobering to think about it all, don’t you agree?
If you have any thoughts on this article, do get in touch.
- Alcohol use and burden for 195 countries and territories, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet 2018 http://dx.doi.org/10.1016/S0140-6736(18)31310-2
- The study defines a ‘drink’ as 10 g alcohol, which equates to:
- A small glass of red wine (100 ml or 3.4 fl oz) at 13% alcohol by volume;
- A can or bottle of beer (375 ml or 12 fl oz) at 3.5% alcohol by volume; or
- A shot of whisky or other spirits (30 ml or 1.0 fl oz) at 40% alcohol by volume.
By comparison a UK ‘unit’ of alcohol is 8 g.
“There are three kinds of lies: lies, damned lies, and statistics.” Attributed by Mark Twain (among others) to the British Prime Minister Benjamin Disraeli (1804-1881).