After the season of good will, the season of bad omen. More Blu-Tack than tack sharp, Dame Sally Davies, the Chief Medical Officer, stuck at the end of last week to her message that there was no such thing as a safe limit to alcohol consumption, but if you wanted to live dangerously, then she supposed up to 14 units a week was tops. On the Today programme, she was the worthy teacher cajoling the dull child, only to be out-smarted by J Webb, who popped the public health message balloon by pointing out that normal drivers face a similar lifetime risk of death as that implied by the new alcohol limit, yet the Government has yet to advise us that there is no safe level of driving, or that drivers should limit themselves to 14 miles a week. The balloon popped so far above Dame Sally’s head that she missed it. When Jay repeated the point, the response was of the ‘oh no, we don’t need to bother with that sort of nonsense round here’ kind, followed by more chugging rhetoric on the risk of dying from breast cancer.
Given that half the population are all but negligibly likely to die from breast cancer, this was the wrong public health balloon for the Dame to inflate, if she wanted to convey the notion that the new advice was a single guideline that applied to all. If breast cancer was the half-cocked ace in the Dame’s pack, could it be that the other cards were duds? More alarm bells started ringing when any potential benefits from alcohol were given the ‘oh no, we don’t need to bother with that sort of nonsense round here’ treatment. Time, Dr No decided, to have a look at the report itself, all the more so given that British alcohol guidelines have historically been driven more by, ahem, benevolent paternalism – ‘plucked from thin air, dear boy, seemed like a good idea at the time’ – than real evidence. Was there clear new credible scientific evidence, as the health police commissioner claimed, or would there be the usual mish-mash of smoke and mirrors, of selective rhetoric fuelled by hot air what-if models? What were the odds that the usual suspects, the Sheffield lot, had had a hand in the pie?
If you were expecting a dud, the report doesn’t disappoint. Packaged in a plain DoH green cover – is this the insidious hand of the gasper police? Where is the knocked over glass of red wine symbolising the tragic spilling of young blood? – the report also reveals the hand of a ‘Behavioural Expert Group’, the creepy new ‘nudgers’, who aim to influence our behaviour without us even realising we have been ‘nudged’. Indeed, much of the report is devoted not to what information to present, but how to present it, the better to achieve ‘nudge-max’. The ‘information’ itself, surely enough, is largely provided by ScHARR, the Sheffield unit, using SAPM, its Sheffield Alcohol Policy Model, known locally in this parish by another similar but more familiar and indeed more representative acronym. Tucked away towards the back of the report, the list of working party members suggests then whole shebang was most likely one big hail fellow well met back-patting party. At times, the stench of ‘here are the guidelines, now where’s the evidence’ is nose curling. In the corridors of Richmond House, the cry came back, ‘oh no, we don’t need to bother with that sort of nonsense round here’. Paramount instead was the need to have a guideline that anyone, however dense, could understand.
The ScHARR C2H (consumption to harm) report underpinning the Guideline Report tries to get B2B (back to basics), but at it’s heart it remains largely a speculative what-if estimate of actual alcohol related harm. Some parts do draw on recent research to derive estimates of actual harm, but they remain estimates, only as reliable, if that, as the underlying research. Other parts appear to be some sort of mega cross-sectional population level survey that somehow connects – the methodology is not entirely transparent, and in any event cross-sectional population level surveys are notoriously bad at getting at the truth – current reported drinking levels with current morbidity and mortality – a nonsense for long term (chronic) conditions, because today’s chronic alcohol related illnesses and deaths stem from past drinking behaviour – and furthermore current indications are that overall alcohol consumption is in fact declining. Other parts of the report rely on B2S (back to statistics) of a complexity far beyond the comprehension of Dr Average (fractional polynomial regression anyone? No, Dr No doesn’t understand it either), and as Dr No has repeatedly said, if the stats can’t be understood by Dr Average after learning about them at medical school or after half a day’s study, then the research is inherently untrustworthy – because it is black box science, or alchemy – and has no place in clinical practice, let alone national public health guidance.
To be fair, the report does concede, in a roundabout way, that its figures may be (way) out, noting that, when push comes to shove, any guidelines will need to be based on ‘expert judgement’, our old friend, ‘seemed like a good idea at the time’, back from the nineteen eighties. It also freely admits that self reported levels of alcohol consumption are notoriously unreliable, with some surveys ‘reporting’ levels perhaps half that actually consumed – and therein lies a meagre drop of comfort. If the ‘safe’ level is that attached to those who report 14 units a week, then the actual level is perhaps attached to something closer to 28 units – though of course you must never tell the doctor, let alone the Dame, how much you actually consume, lest it mess up the stats.
So much for the figures – lies, damned lies and then statistics, and when it comes to guidelines based on what-if cross sectional population level black box studies of alcohol consumption and harm we are probably drowning in all three. But much more fundamentally – and given the uncertainty – can we really justify guidelines at all?
The Guidelines Report makes much of celebrating transparency and openness, in a spirit of allowing informed choice, and then blows it by saying you can, in the best Ford Motor Company tradition, absolutely drink as much as you like, so long as it is 14 units a week or less. Given the lack of substantive real science (as in proper longitudinal research) behind this guideline, it remains a judgement, albeit expert, but nonetheless inevitably paternalistic (as in we know what level of risk is right for you) and so judgmental. If we really want the public to make informed free choices, perhaps we should cut the crap (the guidelines) altogether, and instead merely report current understandings and indeed uncertainties about alcohol related harm, perhaps benchmarked to other risks, such as that attached to normal driving, sky-diving or whatever. That way the public can make real free informed choices, unfettered by quasi-scientific pronouncements by a largely internally self-appointed and inevitably paternalistic Medico-Political Establishment,
Dr No thinks he can already dimly hear the Establishment reply: ‘oh no, we don’t need to bother with that sort of nonsense round here. What we need is a simple robust guideline going forward that any fool can understand’.
Quasi-Academic Rather Long Footnote after what is already a rather long post: For the academically minded, here are some comments on the research used to ‘inform’ the ScHARR report, and the way the results are presented.
Most partially attributable to alcohol disease risk estimates appear to rely on one meta-analysis (which of course aggregates of many other papers). The meta-analysis may or may not be reliable and/or applicable to the population as a whole. The oesophageal cancer findings appear to be based on a single paper reporting on squamous cell carcinoma, which only accounts for around a quarter of all cases of oesophageal cancer: extrapolating to all cancers is not good science. The breast cancer figures, on the other hand, although still apparently based on one paper, do seem to be more robust, insofar as they are backed up, at least in order of magnitude terms, by other more recent research. Most other partially attributable diseases appear to have undergone a similar one paper/grey box/risk curve result process.
Risk estimates for wholly attributable conditions (those which only ever occur after exposure to alcohol eg alcoholic liver disease) are even more opaque. There appears to have been some sort of black box alchemy used to marry up current estimated reported (with a fair amount of imputation to fill in gaps) population level drinking habits (often severe underestimates of actual consumption) with current mortality (ONS data, from death certificates, probably not too bad for younger people where a cause of death is clear, increasingly much vague as people age) and morbidity (HES, notoriously hopeless). In table 4, the cells on the ‘wholly attributable’ rows in the rightmost column headed ‘Source for risk function’ are intriguingly blank. Nor does the text provide much illumination, noting solely on page 29 that an ‘alternative approach’ (‘hullo?’) whereby risk functions ‘are calculated’ (‘hullo, hullo?’) by age group based on disease burden, consumption levels and group size. No description of the method or calculation is given, meaning there is no way of verifying the reported risk functions.
Any ‘potential’ benefits (harms never get the courtesy of a ‘potential’ label) from alcohol are generally given the ‘oh no, we don’t need to bother too much with that sort of nonsense round here’ treatment, though it is fair to say they do get mentioned, only to be largely dismissed as being either spurious, fanciful and/or trivial.
The report also relies heavily on attributable risk, a notoriously loaded (attribution implies causation) yet slippery concept. The difficulties start with the definition, which, as given in the ScHARR report, is riddled with negatives: (population) attributable (excess) risk (sometimes called fraction – and yes, the proliferation and so confusion of terms does not help either) is the number/proportion of cases that would not occur if the population were not exposed to the risk factor of interest, in this case alcohol. The difficulties (and dangers of over-estimation) increase when multiple categories of exposure (eg 0-10, 11-20 etc units/wk instead of never drinkers vs ever drinkers) are used, with numerous examples in the literature of well meaning researchers ‘accidently’ ramping up the dangers attached to their pet concern. Equation 1 on page 27 is the infamous ‘formula 3’ from Rockhill et al’s useful 1998 paper on the use and misuse of population attributable fractions, an admirable, brief and concise example of a paper by statisticians that can be understood by otherwise well-informed non-statisticians.
Attributable risk is also particularly vulnerable to confounding (where the finger pointing at the moon is mistaken for the moon), especially for common often low level exposures, which of course is exactly what alcohol is (most people drink moderately). Of particular concern here is the complete exclusion of socio-economic status – a huge contributor to both morbidity and mortality – from the analysis altogether (the reason given for this unfortunate exclusion is that the definitions between the four nations of the UK are not comparable).
The headline traffic light tables 2 and 3 in the report, repeated as tables 10 and 11 for good measure, are abundantly clear in their implication (green is good, amber is bad and red is terrible) but are opaque in their actual meaning and derivation. The figures are described as the ‘Absolute lifetime risk of [male/female] alcohol-attributable mortality’ (sounds absolutely worrying, doesn’t it?) but are presented, it seems, as a number (to four decimals places – pretty impressive, eh?) between, possibly, minus one and plus one, with zero being the cross-over between benefit and harm (minus numbers are deaths avoided?). Perhaps minus one is the point at which all deaths are avoided (infinite life, but a miserable one), and plus one is the point at which – well, presumably all deaths are caused by alcohol (a short life but a merry one). Dr No thinks the numbers may be re-jigged percentages (eg 0.0252 in row 2, column 6 in table 2 is another way of saying 2.52% (or 25 in 1000) of all adult male deaths in those who drink 14 units on 2 days a week are ‘attributable’ (note the attribution/causation difficulty mentioned earlier) to alcohol – the reader shouldn’t have to ‘think’, the meaning should be clear – but more worryingly, no method is given for the derivation of these figures, making them unverifiable, and therefore unscientific.