In this section we therefore consider the use of indices of deprivation as summary measures.
Indices of deprivation and rurality
Although a wide variety of measures of deprivation have been used, there are a few which appear again and again in the health and health service literature (an outline of the main deprivation indices is given in Appendix 1). Conceived and constructed for a particular purpose, many of these are used well beyond what could be construed as their legitimate domain. The emergence of Jarman's UPA score as a generic measure of deprivation is a clear case in point as it was explicitly conceived as a method of comparing the workload of GPs. The near ubiquitous use of Townsend's Index of Material Deprivation raises similar concerns. With a series of pre-calculated indices of deprivation readily available at ward level,
14
it is sometimes questionable whether the choice between them is driven less by any a priori evaluation of their suitability for the task in hand than by the extent to which they support a particular case.
Figure 1: The percentage of English wards classified as ‘rural’ (N = 1690) by deprivation decile for five different measures of deprivation
As illustrated by Figure 1, different indices of deprivation undoubtedly 'reveal' very different levels of deprivation in rural areas. This shows how five key indices of deprivation profile the 1,690 English wards which the ONS classify as either wholly or predominately rural.
15
As might be expected, whatever measure of deprivation is used there is a general tendency for rural wards to fall predominately in the less deprived deciles. Yet significant variation is evident. Jarman's UPA Index, for instance, places very nearly 25% of rural wards into the least deprived decile and only 0.06% into the most deprived decile. (In England as a whole, 10% of wards will fall into each of the ten deprivation deciles.) Indeed, according to Jarman's UPA Index no less than 89.2% of rural wards fall in the five least deprived deciles. The DETR's Index of Multiple Deprivation 2000, on the other hand, places just 7.8% of rural wards into the least deprived decile and only 70.3% of wards in the five least deprived deciles. In brief, rural areas appear far less socially advantaged if the DETR Index of Multiple Deprivation 2000 is used instead of Jarman's UPA or, for that matter, any of the other traditional measures of deprivation. This is not unexpected as the Index of Multiple Deprivation 2000 incorporates a 'Geographical Access to Services' domain in an explicit attempt to capture social isolation as part of its much broader definition of deprivation.
In view of such differences, it is unsurprising that levels and patterns of deprivation 'revealed' in a rural region such as the South West of England are highly dependent on how deprivation is measured. In the following two maps, wards are once again placed into national deprivation deciles according to the Breadline Britain Poverty Index (Figure 2), and the DETR's Index of Multiple Deprivation 2000 (Figure 3). Cornwall and Devon are, in particular, treated profoundly differently by the different measures. Given the lack of a generally accepted definition of deprivation, the adoption of one measure in preference to the others is largely discretionary. In light of this, agencies in the South West could make a stronger political case if they moved away from the more established measures of deprivation towards the DETR's Index of Multiple Deprivation 2000.
Figure 2: Deprivation in South West England: Breadline Britain Index of Poverty
Figure 3: Deprivation in South West England: the DETR’s Index of Multiple Deprivation
Deprivation, health status and
geo
graphical context
Despite the difficulties of establishing a generally accepted definition of deprivation, it is possible to explore empirically whether different indices are more sensitive to how phenomena that are known to be associated with disadvantage are expressed in different types of area. Evidence from a large number of studies has demonstrated the profound social gradients that exist in certain measures of health status such as limiting long-term illness and mortality. Given the availability (e.g. via the Attribution Dataset provided by the NHS Executive to all Health Authorities) of ward-level standardised illness and mortality ratios, it is relatively easy to examine whether different measures of deprivation provide a consistent representation of health status across the urban-rural continuum.
Recent and as yet unpublished work (Barnett et al., forthcoming) has sought to examine how well three commonly used indices of deprivation predict ward-level variations in morbidity
16
and mortality
17
in three ONS-defined geographic contexts, namely: rural areas, the rural fringe and urban areas.
18
The results show that all three indices are better able to predict variations in both morbidity and mortality in urban areas than they are in rural areas.
19
The authors conclude that, in contrast to their established effectiveness in urban areas, standard 'generic' deprivation indices are poor explanatory variables in rural locations. A number of explanations for this are offered, including the possibility that standard deprivation indices are simply not adequately detecting rural deprivation.
Our own elaboration of this analysis, in which we examine the relationship at ward-level between six deprivation indices and the standardised illness ratio for household residents under 75 (SIR<75)
20
in each of the fourteen ONS geodemographic categories, supports this conclusion. As shown in Table 1 below, for all indices of deprivation except the DETR's Index of Multiple Deprivation 2000, relatively strong relationships in a variety of urban contexts contrast markedly with extremely poor (and sometimes even inverse) relationships in rural areas and the rural fringe. A broadly similar pattern, albeit with generally lower correlation coefficients, emerges with respect to standardised mortality ratios (Table 2).
It is possible, as argued by Haynes and Gale (2000), that this urban-rural gradient is in part a statistical artefact produced by an inconsistent scale of analysis. Notwithstanding any such effect, the DETR's Index of Multiple Deprivation 2000 is able to offer both a better overall prediction of standardised rates of morbidity and mortality at ward level (Figures 4 and 5), and is significantly better at discriminating between wards within rural areas and the rural fringe.
It is only in the context of 'deprived city areas' that any of the other measures of deprivation are able to match the ability of the DETR Index to predict variations in morbidity, and only in the context of 'inner city estates' and 'deprived city areas' that other measures match its ability to predict variations in mortality.
The DETR's IMD 2000 also offers a more consistent relationship between deprivation and both morbidity and mortality. Thus, as illustrated in Tables 3 and 4 with respect to the Breadline Britain Poverty Index and Townsend's Index of Material Deprivation, the use of all indices except the DETR's results in a range of very different regression coefficients for the various ONS-defined areas. For instance, using the Breadline Britain Index suggests, improbably, that in 'Rural Areas' both the standardised illness and mortality ratios fall slightly as the level of poverty increases (ß = -0.14 [-0.46 to 0.19] and -0.07 [-0.43 to 0.28] respectively), which contrasts markedly with a regression coefficient of 4.44 [3.47 to 5.41] for the standardised illness ratio in 'Industrial Areas' and 4.38 [3.49 to 5.27] for the standardised mortality ratio in 'Inner City Estates'. This implies that the impact of deprivation on health is profoundly different in different geographic contexts. Using Townsend's Index to measure deprivation results in even greater variability in the relationship with both morbidity and mortality. Not so the DETR's Index of Multiple Deprivation, which results in a range of coefficients from 0.97 [0.86 to 1.08] for 'Rural Areas' to 1.96 [1.78 to 2.13] for 'Prosperous Areas'.
These are unstandardised beta coefficients and thus are not strictly comparable between indices. However, the central point here is that the DETR IMD 2000, whilst confirming previous observations that regression coefficients tend to be lower in rural wards than elsewhere (Haynes and Gale, 1999; Haynes and Gale, 2000), offers a much more consistent and intuitively satisfactory insight into the impact of social deprivation on both morbidity and mortality than do other indices. This lends support to the idea that these other indices incorporate some form of 'urban bias', offering an inadequate insight into the variation of social disadvantage within rural areas, and are thus poor predictors of the relationship between social disadvantage and health status.
The DETR's Index of Multiple Deprivation 2000 is notable because it utilises non-census data and is thus, and can in the future be kept, broadly up-to-date. The other deprivation indices considered are based on 1991 census data, but this difference is unlikely to account for its success (relative to those other indices of deprivation) at predicting variations in mortality and morbidity because standardised mortality ratios, which we have drawn from the December 1999 Attribution Dataset, have been constructed using data for 1989-1993, whilst standardised illness ratios are based on the 1991 census' limiting long term illness question.
What is undoubtedly significant is the fact that the Index of Multiple Deprivation 2000 takes a much broader definition of deprivation than the other indices considered. Townsend's Index, for instance, is based on four variables derived from the 1991 census which address unemployment, household overcrowding, access to a car, and household tenure. The Index of Multiple Deprivation 2000, on the other hand, is based on the premise that deprivation comprises a number of separate dimensions. Within the constraints of the data that were available at ward level, the Index thus attempts to capture these multiple dimensions through the weighted incorporation of six separate 'domain' indices. These relate to Income (25%), Employment (25%), Health Deprivation and Disability (15%), Education, Skills and training (15%), Geographical Access to Services (10%) and Housing (10%). Each domain index is itself constructed using a wide variety of indicator variables. For instance, the Geographical Access to Services domain index is constructed using data relative to access to i) a post office, ii) food shops, iii) to a GP, and iv) to a primary school. The Housing domain index is constructed using data relative to i) homeless households in temporary accommodation, ii) household overcrowding, and iii) poor private sector housing provision.
It might be argued, of course, that part of the success of the DETR's Index in explaining variations in morbidity and mortality lies with the fact that one of its component domains concerns 'Health Deprivation and Disability', and that one of the five variables used to construct this domain index is an age and sex standardised ratio of limiting long term illness (based on 1991 census data). However, using the same methodology (with appropriately changed domain weights) to calculate a modified DETR Index of Multiple Deprivation which excludes the Health Deprivation and Disability component still results in an index which better discriminates between wards than do the traditional indices of deprivation – both overall and, in particular, in rural areas (as shown on Tables 1 and 2).
This modified Index of Multiple Deprivation exhibits a weaker relationship with standardised illness and mortality ratios than does the actual Index, but the effect is slight. This reflects the fact that the DETR's Index uses no less than 32 variables to construct the six domain indices upon which it is based. The exclusion of the Health Deprivation and Disability domain thus has only a marginal effect. Notwithstanding this effect we consider its inclusion entirely legitimate on the grounds that the DETR's Index was originally constructed as a multi-dimensional indicator of the broad condition that is social deprivation.
This broad definition of deprivation adopted by the DETR's Index of Multiple Deprivation probably relates to why it explains 77% of the ward-level variation in standardised illness ratios and 54% of the variation in standardised mortality ratios. These are impressive figures, particularly given that the Index of Multiple Deprivation 2000 was designed as a generic summary measure of deprivation. It was not, in other words, constructed with the explicit purpose of explaining variations in (using variables selected, transformed and weighted to best fit a least squares model of) morbidity and mortality. Its explanatory power (both overall and in specific geodemographic contexts) is evidence of the appropriateness of the definition of deprivation adopted by the DETR and offers clear confirmation of the long recognised link between deprivation and health status.
Conclusion
We conclude that of the measures of social disadvantage explored the Index of Multiple Deprivation 2000 should be utilised. It exhibits the closest overall association with patterns of morbidity and mortality, and has by far the strongest association with those outcomes in rural areas, suggesting that it is able to best express rural disadvantage as it impacts upon health status.
Table 1: Correlations (Pearson) of standard deprivation indices and standardised illness ratios at ward level in different geographic contexts (ONS groups)
Table 2: Correlations (Pearson) of standard deprivation indices and standardised mortality ratios at ward level in different geographic contexts (ONS groups)
Figure 4: Ward level variations of Townsend’s Index of Deprivation against the Standardised Illness Ratio (N = 8,481r
Figure 5: Ward level variations of the DETR’s Index of Multiple Deprivation 2000 against the Standardised Illness Ratio (N = 8,481)
Table 3: Regression co-efficients (with 95% confidence intervals) for ward level relationships between various Indices of Deprivation and the Standardised Illness Ratio
Table 4: Regression co-efficients (with 95% confidence intervals) for ward level relationships between various Indices of Deprivation and the Standardised Mortality Ratio
