Headlines asserting that hospital admissions attributable to alcohol have doubled since 2004 to over a million a year could soon be just a nasty memory.

New guidelines issued by the Department of Health say that the indicator used to assemble these statistics should be based on just alcohol-related primary diagnoses.

This may seem like a small change, but it has big implications: it will cut numbers dramatically. The claimed figure of over a million admissions a year is based on scrutiny of all the conditions recorded for each patient, not just the one for which the patient is actually admitted. Many seriously-ill patients are admitted with several conditions: a lung cancer patient might, for example, additionally suffer from high blood pressure.  

Although it is the cancer for which the patient is being treated, high blood pressure will also be recorded on his or her notes, and “coded” by clerks. Lung cancer is not an alcohol-related disease, but high blood pressure is. So this admission will be counted as some fraction (around 0.3) of an “alcohol-attributable” admission, even though that was not the actual reason for the patient going into hospital.  

                      The Times, December 10 2011: typical of many such stories

This might not matter much, beyond greatly exaggerating the numbers harmed by alcohol, if it were not for the variation in coding practice over time and between hospitals. The introduction of payment by results by the NHS in England and the publication by health analysts such as Dr Foster of death rates for individual hospitals have encouraged hospitals to increase the number of codes attached to each patient.

If a hospital can show that a patient who died in its care suffered a whole range of conditions on admission, then that death will count for less in the calculation of its Standardised Hospital Mortality Ratio (HSMR) or the Department of Health’s new version of the same thing, Summary Hospital-level Mortality Index (SHMI).

Both of these are based on the ratio of actual to expected deaths. The sicker patients can be shown to be, the higher the number of expected deaths, and the lower the ratio. Hospitals are supposed to improve their HSMRs or SHMIs by reducing the numerator: but many have found it much easier to manipulate the denominator.

The health analysts CHKS have worked out that the average number of codes per patient in the NHS in England rose from just under three in April 2005 to almost 4.5 by December 2010, an increase of 50 per cent. If the number of codes is going up, so will the number of “alcohol-attributable” codes. So a rise of 50 per cent in alcohol-attributable admissions over that period can be accounted for simply by coding creep.

This renders the present method of calculation useless as a way of constructing a time series. Only if coding were consistent would such a series represent reality. So the claims made by the NHS Information Centre that alcohol-related admissions have doubled are worthless.

The effect can be seen by comparing alcohol-related admissions with alcohol-related deaths. According to the NHS IC figures the number of alcohol-related admissions in 2009-10 was 1,057,000, twice what it had been in 2002 (510,800). But over the same period, its figures of alcohol-related deaths have risen more slowly, from 5,842 to 6,584 a year, an increase of 12.7 per cent. (Population growth accounts for about a third of this.)  So while deaths have risen by 12.7 per cent, admissions from the same cause have risen by 107 per cent. That’s not impossible but it is implausible, as I have argued here before.

When patients die, the cause of death is recorded as the primary diagnosis – lung cancer, in my example above. Few hospitals have had the nerve to flatter their HSMRs by coding the cause of death as anything but the primary diagnosis, but even this, incredibly, has happened.

In 2008-09, the Mid-Staffordshire Hospitals NHS Foundation Trust coded only 22.7 per cent of the deaths of patients admitted with fractured neck of femur (broken hip) as the primary diagnosis, while the previous year the proportion had been 92.5 per cent. (Source: written evidence of Professor Sir Brian Jarman to the Mid-Staffs Inquiry.) This made it appear as if the hospital had discovered a magic cure for a common and often fatal condition of the elderly. When I queried this early in 2010, I was told by the hospital that “substantially improved coding procedures” had led to the very low mortality*.

The effect was greatly to improve the trust’s HSMR and lead to Dr Foster declaring Mid-Staffs in its 2009 Good Hospital Guide to be one of the five “most improved hospitals” in the past three years. As we now know, that was a complete misrepresentation of the true position, brought about by the hospital altering its coding practice in a way few would deem to be an improvement.

What would alcohol-related admissions be under the new rule: that is, counting only primary diagnoses? NHS IC figures in Table 4.5 (page 71) Alcohol in England 2011 show 141,700 admissions in 2002-03 with primary diagnoses wholly or partly attributable to alcohol, rising to 194.800 in 2009-10, an increase of 37 per cent. Over the same period, all admissions rose by 27 per cent. That puts the alcohol issue into a more realistic perspective.

If we limit the numbers to admissions wholly attributable to alcohol, the numbers have risen from 45,000 to 68,500, an increase of 52.2 per cent. This is a lot less than 107 per cent, but it’s still pretty substantial and worth looking at in closer detail.

Of these additional 23,500 admissions, 10,800 are due to “acute intoxication”, and show a marked increase around 2004 and 2005 when the four-hour target was introduced into A&E departments. It’s likely, therefore, that some of these admissions are of those still deemed too drunk after four hours in A&E to be shown the door, and who are therefore admitted to “alcohol wards” where a few beds are kept for such people. There is a financial incentive to hospitals to do this, since an admission earns more money than treatment in A&E.

Other increases may be due to better ascertainment or improved services – a very sharp increase of 8,700 admissions a year over the period is shown under the code “withdrawal state” which occurs when alcoholics give up drinking. Or the increase could, as the critics of drinking assert, be due to greater alcohol abuse. It’s difficult to separate out the factors involved.

But as it seems fairly clear that the DH favours an indicator based only on primary diagnoses – as it puts it “to minimise the risk of perverse consequences from any changes in coding practice” - those headlines about a million-plus admissions a year are toast. The North West Public Health Observatory is to consult on the precise methods to be used.

It’ll take some explaining, given the prominence given to the million-plus claims in the past. But better a sinner that repenteth ...

* Patient Coding and the Ratings Game, BMJ, vol 340, pp 950-952, 1 May 2010