Tuesday, 13 December 2011

Alcohol-Attributable Fractions

This press release on Saturday reported that there were 1,173,386 hospital admissions related to alcohol in 2010-11, an increase of 11% from the previous year and more than double the number in 2002-3  (Hospital Episode Statistics are calculated from 1st April to 31st March).  Andrew Lansley for the government said:
These figures are disturbing evidence that, despite total consumption of alcohol not increasing recently, we have serious problems with both binge-drinking and long-term excessive alcohol abuse in a minority of people.
These consistent rises show that Labour took their eye off the ball on tackling alcohol abuse during their 13 years in power. Their reckless policies, such as the decision to unleash a 24-hour drinking culture in our country, only made matters worse.
Whereas Diane Abbott for the opposition was of the view that
The alarm bells should be ringing with the publication of these figures. It is clear that this Government is rapidly pushing us towards a binge-drinking crisis.
It is clear that for Andrew Lansley, the be-all and end-all is whether his friends in big business are happy, and, unfortunately, it is costing our NHS and British families an absolute fortune. A recent report predicted that binge-drinking will cost the NHS £3.8 billion by 2015, with 1.5 million A&E admissions a year.
So there you have it.  For the Tories, the problem is one of both binge drinking and long-term alcohol abuse, and it's Labour's fault.  For Labour it's just binge drinking encouraged by big business that's the problem, and the Tories are to blame (Ms Abbott was talking about a different report: I haven't traced it but it's mentioned here)

The press release was reproduced, quotes and all, in most of the papers.  The Guardian fleshed it out a bit.  The Daily Mail took Ms Abbott's word for it that all the admissions were to A&E.  The Times [paywall] got a quote from Mark Bellis, director of the North West Public Health Observatory which compiled the figures "These things are working their way through the system from a massive increase in alcohol consumption over the past two or three decades.  We've probably got more of this to come...Particularly at this time of year, we've got to address our relationship with drunkenness."

There's one dissenting voice, which calls the story a lie and links to this description of how the figures are calculated (the analysis dates from when the 2009-10 figures were published in May this year - the calculation seems to be quicker now).  The statistics are compiled not, as you might think, by asking people admitted to hospital whether they've been drinking (in A&E) or how much they drink (on the heart, liver, and cancer wards), but by applying an "alcohol-attributable fraction" to each admission according to diagnosis, age, and sex.  This methodology is confirmed in a comment by Andy Sutherland of the NHS Information Centre (I'm fairly confident that it's really him, because the press release correction he promises did appear).

The calculation of AAFs specific to England is described in this pdf (the purpose of which is described here). The method for each diagnosis is to identify the best available research on the increased (or decreased) risk associated with various levels of alcohol consumption (by age and sex where the data were available), apply the levels of alcohol consumption determined by the 2005 General Household Survey (by age and sex, adjusted using new estimates of the alcoholic content of drinks), and hence calculate what proportion of hospital admissions in 2005 were related to alcohol consumption.  Ideally alcohol consumption figures integrated over time should be used for diseases which take many years to develop, but the method seems broadly reasonable to me.

The data for all other years since 2002-3 have been calculated using the same AAFs.  Collated data can be downloaded in a spreadsheet here, showing a steady rise in alcohol-related hospital admissions.

Let me say that again: "calculated using the same alcohol-attributable fractions".  So what has happened is that researchers into each diagnosis have analysed data on alcohol consumption for people with and without the disease, and fitted those data to a model in which the diagnosis is due to two perfectly uncorrelated factors, one for alcohol consumption and one for everything else.  Applying this model to data on alcohol consumption in 2005, statisticians have deduced what fraction of 2005 hospital admissions for each diagnosis was due to the alcohol factor - the 2005 AAF.  And then these two perfectly uncorrelated factors have been assumed to be perfectly correlated in every other year, so that the AAF remains constant.  I am shocked that reputable statisticians have put their names to this method.  I can see no good reason not to repeat the 2005 analysis each year (except perhaps that it would take longer to get the numbers out).  Certainly that would give different results, since current alcohol consumption would actually be an input to the analysis.

What are the data on alcohol consumption?  The General Lifestyle Survey reports on weekly alcohol consumption above safe limits:
Following an increase between 1998 and 2000, there has been a decline since 2002 in the proportion of men drinking more than, on average, 21 units a week and in the proportion of women drinking more than 14 units...This trend seems to be continuing under the new methodology; between 2006 and 2009 the proportion of men drinking more than 21 units a week fell from 31 per cent to 26 per cent and the proportion of women drinking more than 14 units a week fell from 20 per cent to 18 per cent. These falls were driven by falls in the younger age groups... 
and on average weekly consumption:
The British Beer and Pub Association (BBPA) makes annual estimates of per capita alcohol consumption using data provided by HM Revenue and Customs. These show a steady increase in consumption from 1998 to 2004, followed by a decline of about 5 per cent to 2006, and then a further decline of about 7 per cent from 2006 to 2009. The decline measured by the GHS is much greater, at about 15 per cent between 2002 and 2006.
(there was a change of methodology in 2006 that makes it difficult to produce a single set of numbers)

It would seem that any increase in hospital admissions must be due either to the after-affects of long-term abuse which may have increased in the years up to 2002 or so before declining thereafter, or to occasional binge drinking not captured by weekly averages.  So I looked at data for three diagnoses: "toxic affect of alcohol", which should be an indication of binge drinking, "alcoholic liver disease", to capture the effects of long-term alcohol abuse, and "atrial fibrillation and flutter" to look at what's happening with a common diagnosis with a small but non-zero AAF.  These are available here, based on the same data as the alcohol-related admissions figures.

I've included series for total admissions and for the alcohol-related admissions data I'm writing about.  All the series are normalized to 2002-3, when the numbers were: total admissions 11,414,074; alcohol-related admissions 510,780; atrial fibrillation and flutter 68,731; alcoholic liver disease 11,582; toxic effect of alcohol 1407.

What's striking is that the alcohol-related admissions numbers have increased far faster than any of the other series (admissions for the toxic effect of alcohol have not increased at all).  How can this be explained?  I looked through the diagnoses to find any that had at least doubled from at least 10,000 since 2002-3: there were 27.  But only one of them had a non-zero AAF: hypertensive renal disease.  I must be looking at the wrong data - AAFs for many diagnoses are higher for younger patients, so there must have been a big increase in these admissions among the relatively young, which don't appear in the totals I looked at.  (The data are there in the spreadsheets, but you get only so much for your money.)

One thing did catch my eye however, which is the increase in admissions for "obesity" from 1,297 to 11,740. This may be associated with increasing availability of bariatric surgery, but it's no secret that there have been big increases in obesity.  Furthermore, obesity is linked to hypertension and diabetes, both of which will increase hospital admissions among the relatively young (not least for hypertensive renal disease).

This is speculative, but my guess is that the alleged rise in alcohol-related hospital admissions is in fact a rise in obesity-related hospital admissions, which are linked to some of the same diagnoses at similar ages.  Perhaps the statisticians behind this weekend's newspaper stories could find time to look into this hypothesis.


  1. For more on this, see here
    and the comments following it.

  2. Thanks for identifying yourself. I'm surprised you didn't notice my paragraph about that piece.

  3. Apologies. Yes, I should have. Was leaving to catch a bus - explanation not excuse.

  4. To my mind this is exactly what I refer to a brilliant article! Do you utilize this site for personal purposes solely or you exploit it as an additional source of income?

  5. It's kind of you to say so. If you want to send me money, please feel free to get in touch using the email address at the top of the page.