The incidence, prevalence, and mortality of many diseases are known to vary by ethnic group.There are well documented inequities in access to preventation, treatment, and palliative health and social care services based on ethnic group. There are, too, reported differences in the quality of services received by different ethnic groups and of outcomes of treatment and care. Many of these inequities are amenable to change. However, in order to address them they must, first of all, be comprehensively defined and documented. Mainstreaming ethnic monitoring/data collection is a vital step in the process. The history of such data collection in the NHS is poor, whichever of the key datasets is examined: hospital episode statistics, general practitioner data, cancer registrations, and disease registers. While steps are now being taken to remedy some of these deficiencies, the continued non-availability of ethnic monitoring data and in some cases of compatible ethnically-coded denominator data remains a problem. In particular the lack of ethnic group in births and deaths data has been the subject of widespread comment by specialists in demography and public health and is probably the single action that could most improve the evidence based for addressing ethnic/racial inequalities in health and health care.

Back to top ^^

Information on ethnicity had been collected in some organisations and systems before April 1991, often without a systematic approach. Coding systems were instituted without a frame of reference, and were therefore open to random self-identification within blank fields or adapted to perceived local need. A particular feature of these coding frameworks was the frequent confusion between the concepts of ethnicity and nationality.

Information on Ethnic Group was first collected routinely in the United Kingdom during the 1980s in the General Household Survey and the Labour Force Survey. Ethnic Group was first universally collected in the United Kingdom as part of the 1991 Census. There were nine Ethnic Groups, chosen after extensive consultation with prospective Census users, with an additional one, Other Asian, added to the output tables without prior testing. These categories were much criticised in following years for their perceived limitations. In particular much was made of their unsuitability to reflect properly the impact of ethnic groups that, while locally relevant to policy and because of their number, were too small to be deemed useful to national level analysis. However, the single most important contribution of the 1991 Census was how it created for the first time a dataset against which data collection could be measured, and means to reflect on how much the locally collected profile matched the local HIulation profile as represented in the Census. A new indicator was then potentially created and so the potential to measure the relationship between the HIulation served and the base.

To enable the National Health Service to introduce Ethnic Group collection where there had been none and enforce uniformity in legacy system, a DSCN (Data Set Change Notice) was issued. This introduced Ethnic Group as a compulsory routine data item in patient records related to Admitted Patient Care, and established a standard system of collection.

Ethnic Group should have been collected at patient's admission; it was self-defined, with each person offered the choice from a list, which could be either the short 10 code or the longer list, sometime supplemented by local codes created using the appropriate one-digit header and an additional digit, usually a letter. This enabled some degree of local flexibility to be retained, while creating a standard output to make the best use of the newly collected item. One of the codes in the table was "Not given", leaving patients who wished to do so the opportunities to withhold the information. In transferring the field to national systems for the creation of reference files such as the Hospital Episode Statistics (HES), used also as the base for international comparisons, any of the additional digits would be dropped, unless it matched the long list.

During the intercensal period, the increased usage being made of ethnic data collection and the interest so created started to impact heavily on the discussion about the appropriateness of the existing coding model, and its impact on services provided both with regard to their appropriateness and their coverage.

Extensive consultations were carried out by the Office for National Statistics in the lead up to the 2001 Census. Many of the discussion centered around four data items:

Items 1. and 2. were dropped from the proposed Census form, the former due to adverse influence on return rate during trials, the latter due to perceived excessive complexity of collection and difficulty to code meaningfully, despite the precedent of collection for Wales.

Item three was introduced for the first time outside of Northern Ireland; item four resulted in a new set of ethnic data collection, defined as Ethnic Category. This new entity, composed of an alphabetical code, has in some cases increased the level of detail collected. By necessity, wishing to maintain a similar number of codes as in the previous system, this resulted in a decrease of specificity in some other categories.

Back to top ^^

Within the NHS, the DSCN (Data Set Change Notice) Ref.: 21/2000 was issued in October 2000 imposing on NHS trust the obligation to implement changes to allow NHS data to be collected in the same format as that of the 2001 census. This document discontinued the collection of Ethnic Group introducing Ethnic Category starting from the first of April 2001.

The change in coding introduced an even higher margin of error in the data field, with many sites finding it difficult to manage the changeover between 1991 and 2001 system and leading to a noticeable increase in the number of invalid codes being recorded. In some cases, instances of both new and old coding frames were found to be in use simultaneously; some sites suddenly started reporting 100% invalid codes.

Additional emphasis on ethnic data collection was recently generated by the introduction of the Race Relation (Amendment) Act 2000 which imposed on public bodies the duty to demonstrate that their activities are not carried out in such a way as to discriminate or cause inequalities in the HIulation they serve.

Despite the obligations on collection of ethnic coding for health data having existed for more than ten years, we are still far from the hoped for completeness and accuracy. This is particularly disappointing, as most other data items appear to be collected with much more success on both counts.

The completeness of ethnic coding has been very patchy from the very beginning, and without systematic use of the data has been slow to improve. Collection of this piece of information has been hampered by several factors, the most commonly cited being the awkwardness of the request as perceived by the person registering the patient's details. Lack of understanding of the purpose of the data collection has been a much more significant underlying reason for poor results. Data usage in a way that makes clear the usefulness of this data item in improving HIulation and individual patients' access and health outcomes could be one way to impress the necessity of collection.

A perverse situation has developed in which the data are not used because of their poor level of completeness, and the data item is not collected as it is not being used and is therefore seen as of poor relevance and possibly even a waste of administrative resources. Initiatives such as the recent introduction of a management target focused on collection and data quality of ethnic coding in London alone piloted by the London Directorate of Health and Social Care is focusing healthcare providers' attention on the issue and is likely to result in more comprehensive and complete data being available.

In this climate, the London Health Observatory (LHO) decided to initiate an Ethnic Health Intelligence Programme, aiming

The brief for the LHO covers the identification and quantification of health inequalities: ethnicity is an important factor in many conditions, from tuberculosis to coronary heart disease, to sickle cell and thalassemia, to diabetes and to some forms of cancer.

Back to top ^^

The analysis involved the use of existing data from the Hospital Episodes Statistics (HES) dataset, covering inpatient and more recently outpatients data and including the ethnicity code.

Two streams of analysis were proposed:

It was felt that attention needed to be addressed to the access to and the outcome of particular types of treatment. Some groups, in particular women and persons belonging to black and minority ethnic groups, have been shown to have differential access rates to particular diagnostic procedures. Difficulties accessing services and making use of them are a major factor contributing to poorer health. The more deprived sectors of the community are often discriminated against either by virtue of difficulties in communication or in understanding different attitudes and approaches to health, or simply by lack of opportunities for access.

Analysis was by necessity limited to disease groups that could be defined by diagnostic procedures or treatment needing to be carried out in hospital, and that could be traced with precision by either diagnosis or operation coding.

The initial areas suggested for analysis were:

All these areas would be examined at the smallest geographical level possible after consideration of confidentiality and event frequency, and aggregated to enable useful comparison. This would allow differential geographical level analysis and comparisons to be carried out, with ward, London borough, primary care trust, Strategic health authority tabulations available in most instances.

A preliminary analysis on the HES dataset was carried out, to assess the overall level of data quality and coding completeness. This resulted in a decision to utilize only the years between 1997/98 and 2000/01, as the previous years' coding was found to be of insufficient quality.

The results were sufficiently different for the separate diagnostic groups to justify by themselves the decision of not utilising previously developed techniques, that would have resulted in the uncoded portion of the record to be apportioned by using an ethnic group breakdown equivalent to that found in the coded records. The premise that the missing coding may be uniform was deemed to be inappropriate in this circumstance, with coding completeness being affected by several factors related to the condition itself, not least the likelihood of the patient being admitted as an emergency and their condition preventing the question about ethnic origin being asked at all.

The choice of tuberculosis as one of the first groups for analysis was made because of the obvious and increased clinical importance of this condition in a cosmopolitan city. At the same time, following the heightened levels of awareness caused both by justified concerns and the increased level of interest in the media, tuberculosis was felt to be an appropriate first choice. The percentage of admissions for tuberculosis showed a proportion of admissions with ethnic code "Not known" of between 31% and 37%, well within the range of the other diagnostic and coding group chosen for analysis (RANGE 25% to 48%.) The proportion of admissions recorded in the "White" category varied between 12% and 24%.For all types of cancer chosen, the proportion of admissions recorded as belonging to the "White" group were highest, ranging from 42% to 69%. The proportion of "Not known" varied from 19% to 37%.

For children, defined as persons aged <16 years of age, the percentage of "Not known" were higher (42% to 47%), reflecting the limitations of the self-definition approach to ethnic coding. This probably reflects ignorance of the rules that allow a parent or guardian to supply the information on ethnic category for children.

For cardiovascular disease the picture was very similar, with "Not known" codes at around 32-35%; the " White" group varied from 40% for hypertensive disease to 59% for dysrhythmia.

Back to top ^^