urban mortality differences in England and Wales and the effect of deprivation adjustment

Social Science & Medicine 72 (2011) 1685e1694

Contents lists available at ScienceDirect

Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed

Rural/urban mortality differences in England and Wales and the effect of deprivation adjustment Andrea Gartner a, *, Daniel Farewell b, Paul Roach c, Frank Dunstan b a

Public Health Wales Observatory, 14 Cathedral Road, Cardiff CF11 9LJ, UK School of Medicine, Cardiff University, Cardiff, UK c Department of Computing and Mathematical Sciences, University of Glamorgan, Pontypridd, UK b

a r t i c l e i n f o

a b s t r a c t

Article history: Available online 30 March 2011

Perceptions that rural populations are inevitably healthier and live longer than urban populations are increasingly being challenged. But very few publications have investigated the extent to which these putative differences can be explained by variation in area composition. Existing publications have tended to use conventional deprivation measures, often thought to mask rural deprivation by favourable averages. Further, they have typically been based on large and variably-sized geographical units, or confined to studies of a single region or cause of death. This study examines differences in mortality between rural and urban areas in the entire population of England and Wales for 2002e2004. It uses the most up-to-date small geographical units of similar size and homogeneity of population together with the recently-introduced Rural and Urban Area Classification, and adjusts for five different deprivation measures (including modern composite indices). The causes of death investigated were all-cause mortality, cancer, lung cancer, respiratory disease, circulatory disease, suicide and accidents. Particular points of focus for the study were the potential for interaction between deprivation and rurality, and the importance of choice of deprivation measure in quantifying the relationships between mortality, rurality and deprivation. Choice of deprivation measure was not found to alter the substantive conclusions of any analysis, and little evidence for differential effects of deprivation in rural and urban areas was uncovered. Differences between rural and urban areas in all-cause, circulatory disease and cancer mortality could largely be accounted for by adjusting for deprivation. For these causes of death, therefore, rural populations were not found to be inherently healthier than their urban counterparts. However, substantial residual differences between rural and urban areas were found in comparisons of mortality from lung cancer and respiratory disease, mortality being lower in rural areas. Stronger relationships between rurality and mortality were found in ‘village and dispersed’ settlements. Ó 2011 Elsevier Ltd. All rights reserved.

Keywords: Health inequalities Rurality Wales England UK Mortality Deprivation

Introduction Tackling health inequalities features prominently on the policy agendas of all UK nations. Much research has concentrated on health inequalities between areas, often based on gradients of socioeconomic conditions. Relatively few studies have been conducted to examine differences in health experiences between rural and urban areas of the UK, and existing work has largely been confined to regional studies or concentrated on a specific cause of death (Levin & Leyland, 2005; Middleton, Gunnell, Frankel, Whitley, & Dorling, 2003).

* Corresponding author. Tel.: þ44 (0) 29 20827657. E-mail address: [email protected] (A. Gartner). 0277-9536/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.socscimed.2011.03.017

There is a widespread belief that rural areas are inherently healthier places to live than urban areas. Assessing the truth of this is difficult, as people who live in rural areas are different in many ways from those who live in urban areas and these differences may vary between countries. In the United Kingdom, most of the usual measures used to assess the level of deprivation in an area suggest that urban areas are generally more deprived than rural areas, and it is well established that deprivation is a factor that has a strong relationship to health (Smith, Whitley, Dorling, & Gunnell, 2001). The labels ‘urban’ and ‘rural’ are also very broad, and there are substantial differences between areas in both categories. Again within the United Kingdom, rural areas in South East England are very different from those in the Highlands of Scotland; not distinguishing between them might mask real differences. There are also huge differences between countries in the characteristics of

1686

A. Gartner et al. / Social Science & Medicine 72 (2011) 1685e1694

rural areas. Rural areas in Australia and Canada are often very remote and are very different in many ways from those in the United Kingdom. Rural areas in less developed countries are likely to be much poorer than urban areas, in contrast to Western Europe, for example. Several authors have investigated rural/urban differences in health, in a range of countries and with a number of different outcomes; the differences just described mean that results may be expected to vary between countries. A commonly-studied outcome is mortality, either all-cause, due to a specific cause, or premature mortality. Mortality studies are usually based on routine data and use area (rather than individual-specific) measures of deprivation. In Wales, premature mortality was found to be higher in urban areas than in rural areas (Senior, Williams, & Higgs, 2000), but adjusting for area-level deprivation substantially reduced the difference; the effect of the deprivation adjustment varied with the cause of mortality. Mortality was significantly higher in rural areas in Northern Ireland than in urban areas (O’Reilly, O’Reilly, Rosato, & Connolly, 2007), but the difference depended on the cause of death; rates of respiratory disease and lung cancer were substantially higher in urban areas but mortality in young people was worse in rural areas. This last conclusion was reinforced by a study on suicide in South West England (Middleton et al., 2003). They compared trends in suicide rates between urban and rural areas and showed that the trends were least favourable in the most remote areas, similar to findings in other studies in Australia and Norway (Lagace, Desmeules, Pong, & Heng, 2007; Ostry, 2009). International comparisons are not straightforward as the nature of rural areas, and their effect on health, may vary between countries. For example, rural mortality was lower than urban mortality in the Netherlands (Van Hooijdonk, Droomers, Deerenberg, Mackenbach, & Kunst, 2008), though when adjusted for deprivation most of the differences became very small. Adjusted rates were higher in urban areas only for cancer, particularly so for lung cancer. The urban effect appeared much stronger in the United States (House et al., 2000) where the risk of mortality was 60% higher in urban areas, after adjusting for individual characteristics. In China, however, mortality was 30% higher in rural areas (Zimmer, Kaneda, & Spess, 2007). Generalising findings from one country to others is therefore difficult. Studies on different outcomes from mortality are complicated by having different diagnostic criteria and usually much smaller sample sizes, but often have the advantage of incorporating individual measures of socio-economic status. For example, a number of authors have compared mental health between urban and rural areas. A study of common mental disorders in the United Kingdom (Weich, Twigg, & Lewis, 2006) found that they were significantly more common in urban areas; adjustment for factors including socioeconomic status had little effect. Another study in England (Riva, Curtis, Guavin, & Fagg, 2009) also included common mental disorders as an outcome, and found they were more common in cities than in semi-rural areas or villages; adjusting for both individual-level and area-level characteristics removed much, but not all, of the difference and they were still less common in villages. A recent meta-analysis (Peen, Schoevers, Beekman, & Dekker, 2010) found that psychiatric disorders were significantly more common in urban areas and that specific categories of anxiety and mood disorders were also more common in urban areas; the substantial heterogeneity between studies means that results have to be interpreted carefully. Having a measure of individual socioeconomic circumstances is clearly preferable, but many studiesdparticularly those using routine dataduse area-based measures as a proxy. Different deprivation indices are available, but one of the most commonly used in UK academic research is the Townsend index (Townsend,

Phillimore, & Beattie, 1988). The Townsend index is designed to measure material, as opposed to social, deprivation and is based on four area-specific proportions derived from Census data: unemployment, households without a car, households that are not owner occupied and households that are overcrowded. Conventional areabased deprivation measures such as the Townsend index are thought to be less meaningful in rural areas (Farmer, Baird, & Iversen, 2001; Haynes & Gale, 2000; Martin, Brigham, Roderick, Barnett, & Diamond, 2000). Rural deprivation can be hidden by favourable averages, since typically rural areas are more heterogeneous than urban areas (Haynes & Gale, 2000). The Townsend index is also less reliable in rural health research due to the inclusion of car ownership as a component variable; in rural areas, car ownership may be considered essential, regardless of socioeconomic status. Christie and Fone (2003) found that lack of car ownership was a poor proxy for social deprivation in sparsely populated rural areas. In their recent study based in Northern Ireland, O’Reilly et al. (2007) concluded that both car ownership and housing tenure d two components of the Townsend index d were biased measures of material deprivation in the rural/urban mortality context. A more recent composite measure is the Index of Multiple Deprivation 2004 (IMD Noble et al., 2004) for England, which combines data from seven weighted domains (see Table 1). The Welsh Index of Multiple Deprivation 2005 (WIMD, National Assembly for Wales, 2005) is based on different data sources and methodologies, but also combines seven domains (Table 1). The (W)IMD is used extensively in public health departments and other public sector organisations to describe health inequalities and for resource allocation. The (W)IMD is updated every few years, based on recent data, and is now often used in preference to the Townsend index, which relies on decennial Census data. The (W)IMD, with its different domains, also relates to wider aspects of deprivation. Unlike the Townsend index, the (W)IMD does not rely on car ownership to quantify area deprivation but does include health information and, in particular, mortality data. At worst, this may lead to circularity of argument; at best, it introduces the possibility of confounding, with outcome and exposure depending on the same information. For instance, Jordan, Roderick, and Martin (2004) attributed the strong relationship between the IMD 2000 and health in both rural and urban areas to the former’s inclusion of ill health and disability benefits claimants. As a consequence, Adams and White (2006) recommended removal of the health domain when using the (W)IMD in health research, proposing this modified (W)IMD as a third possible measure of area deprivation. For both rural and urban small areas, unemployment has also been suggested as a more reliable and updatable indicator of deprivation than the composite indices (Haynes & Gale, 2000). This is measured in the (W)IMD employment domain, but it includes data on claimants of incapacity benefits in addition to those seeking employment-related benefits, so the possibility of residual

Table 1 Individual IMD 2004/WIMD 2005 deprivation domains and respective weights. IMD domain

Weight

WIMD domain

Weight

Income Employment Health and disability Education, skills and training Barriers to housing and services Living environment Crime

22.5% 22.5% 13.5% 13.5%

Income Employment Health Education, skills and training Housing

25% 25% 15% 15%

Physical environment Geographical access to services

5% 10%

9.3% 9.3% 9.3%

5%

A. Gartner et al. / Social Science & Medicine 72 (2011) 1685e1694

confounding remains. A fifth possible measure of deprivation would therefore rely on the (W)IMD income domain alone; while this (largely, but not entirely) circumvents problems of confounding, it may also fail to capture non-financial aspects of deprivation. Other possible measures of deprivation, pursued no further in this paper, include the Carstairs index (Carstairs & Morris, 1990) or the ‘access to services’ domain of the (W)IMD; both suffer similar drawbacks to those outlined here. Within the wider context of health research, the foregoing discussion raises the question of whether substantive conclusions arising from academic research (and policy decisions derived from these conclusions) would be meaningfully altered by choosing one index of deprivation over a competitor. In the present paper, we propose to compare how conclusions about rural/urban mortality differences vary when adjusting for five of the possible measures discussed. If results are shown to be highly sensitive to choice of deprivation index then, in future health research, great care will be needed in choosing the measure most relevant for the question at hand. If, on the other hand, results are broadly comparable across deprivation indices then, despite possible philosophical objections, this would provide support for seeking and deciding upon a compromise deprivation index that would facilite ease of comparison, updating and interpretation. The challenge of analysing rurality and deprivation is increased further by the possibility that their effects on mortality may be complex, with the potential for differing impacts of rural life on those from different socioeconomic backgrounds. Interactions between rurality and deprivation are another central component of the present paper. In fact, a detailed exploration of rurality and deprivation has benefits going beyond improved understanding of their interplay. As has already been described, many commentators feel that rural disadvantage is not adequately reflected by established measures of deprivation, while others hold that the apparent advantages of rural life are simply a result of better socioeconomic conditions. Flexible modelling of rurality and deprivation allows us to consider these two hypotheses, albeit informally. For all-cause mortality and five specific causes of death, the present paper examines differences between rural and urban areas in England and Wales during the period 2002e2004, adjusting for socioeconomic deprivation. This extensive study formed the basis of an earlier report (Gartner, Farewell, Dunstan, & Gordon, 2008) which presented results using a single deprivation measure. The current paper provides the wider context, investigating differences between rural and urban areas in the relationship between mortality and deprivation, and emphasising possible sensitivity to adjustment by five different deprivation measures. Further, the present paper explores the finer classification provided by a new categorisation of rurality to investigate variation in mortality within urban and rural areas. The rest of the paper is laid out as follows. We first outline sources of data describing deprivation, rurality and mortality, and set out our proposed modelling framework. We then report our findings, followed by a discussion of the main results. We conclude by briefly considering the wider relevance of our study. Data and methods Classification of rural and urban areas and geography To describe rurality, we use the new Rural and Urban Area Classification (RUAC). This was launched in 2004 (Bibby & Shepherd, 2004) and sponsored by a number of Government organisations including the Office for National Statistics (ONS), Department for Environment, Food and Rural Affairs (DEFRA) and the National Assembly for Wales (NAfW). Under the RUAC, settlements with more than 10,000 people are considered urban areas,

1687

with all other settlements deemed to be rural (Bibby & Shepherd, 2004). Within each subgroup there is a further subdivision into areas with sparse and less sparse surroundings, but rural areas are initially divided into ‘town and fringe’ and ‘village and dispersed’. Much of the analysis was undertaken using the rural/urban dichotomy, but an analysis using all six classes was also included. The larger the individual study areas, the greater become both the heterogeneity within and the homogeneity between areas. As has already been mentioned, this can present problems for rural research, since rural deprivation may be quite localised. The geographical unit of study used in the present paper was the Lower Layer Super Output Area (LSOA). These were defined for reporting the results of the 2001 UK Census and on average have approximately 1500 residents with a minimum of 1000 residents. The LSOAs were created taking social homogeneity into account, and were designed to have boundaries that are consistent through time. All 32,482 LSOAs in England and 1896 LSOAs in Wales were included in the study, with populations totalling nearly 53 million individuals. A colour map showing the LSOAs by RUAC classification is shown in Gartner et al. (2008). Deprivation measures As described in the introduction, this paper considers five possible indices of socioeconomic disadvantage. For disclosure reasons, ONS randomly adjust the Census data published for small areas (including LSOAs); however, unadjusted Census data were provided for this work. We were therefore able to calculate the Townsend index for each LSOA from scratch, which provides a more accurate measure. In addition to the Townsend index, we included the (W)IMD; note that, since the IMD (2004) and the WIMD (2005) are not directly comparable, all analyses were undertaken separately for England and Wales. We also make use of the (W)IMD having removed the health domain according to the method published by Adams and White (2006). Finally, we examine deprivation as measured by the individual (W)IMD employment and income component subscales. For analysis, each measure of deprivation was partitioned into quintiles, each of which contained 20% of the LSOAs in their respective country. For rural and urban LSOAs, Fig. 1 shows the distribution of IMD quintiles in England and WIMD quintiles in Wales. Only a small percentage of rural LSOAs fall into the most deprived quintile, though such areas are more prevalent in Wales than in England. Interestingly, the most prevalent quintile in rural areas is the second highest, and this is the least prevalent quintile in urban areas. Clearly then, deprivation as measured by (W)IMD and rurality as measured by RUAC are not independent. It is worth observing, a priori, that the five different deprivation measures actually do differ. Table 2 and Table 3 show kappa values (representing level of agreement and ranging from 1 to 1) between quintiles of the deprivation indices considered in this paper. There is moderately good agreement between most measures, but in some instancesdnotably between the Townsend index and the (W)IMD employment domaindthere may be considerable disagreement about the deprivation quintile in which to place any given LSOA. Since there is variation between different deprivation classifications, it is reasonable to ask whether some measures are better than others at characterising aspects of deprivation important in explaining rural and urban mortality. Mortality data For each LSOA in England and Wales, mortality registration data and population estimates for the years 2002e2004 were provided by the Office of National Statistics. This is based on usual residence

1688

A. Gartner et al. / Social Science & Medicine 72 (2011) 1685e1694

rural

urban

30 25 England

20 15

percentage of LSOAs

10 5 0

30 25 Wales

20 15 10 5 0 least

2

3

4

most

least

2

3

4

most

deprivation quintile Fig. 1. Distribution of IMD 2004/WIMD 2005 quintiles by country and rurality.

in a particular LSOA rather than place of death. The data comprise the entire population of England and Wales and include information on age and sex. With the exception of our analysis of suicides, all ages were included, rather than focussing on ‘premature’ deaths. We included older groups because they have been found to be an important study group with rising inequalities (Levin & Leyland, 2006). For suicides, only individuals aged 15 years or over were included. Mortality data were extracted by ICD-10 code, and the outcome of interest was the number of area-, age- and genderspecific deaths averaged over the period 2002e2004. The causes investigated, their ICD-10 codes and the numbers of deaths in a single year are listed in Table 4. We included all major causes of death, namely cancers, respiratory disease and circulatory disease. Diseases such as lung cancer, perceived to be particularly related to deprivation, were also investigated. Additionally, causes of death

Table 2 Kappa values quantifying agreement between the quintiles of five English deprivation measures.

IMD IMD minus health IMD employment IMD income Townsend

IMD

IMD minus health

IMD employment

IMD Income

Townsend

1 0.881

0.881 1

0.54 0.482

0.617 0.651

0.419 0.467

0.54

0.482

1

0.445

0.258

0.617 0.419

0.651 0.467

0.445 0.258

1 0.539

0.539 1

thought to be higher in rural areas, such as suicides and accidents, were investigated. Table 4 illustrates that there were relatively few deaths from certain causes, particularly suicides and accidents. Analysis All analyses were carried out in SPSS, and were performed separately for England and Wales, and for males and females. Logistic regression was used to consider the dependence of mortality on age, rurality and deprivation. Age was categorised into five-year age bands (0e4, 5e9, 10e14,., 80e84, 85þ) and all regression models incorporated age group and rurality as categorical variables with 85þ and urban categories as reference categories. Models incorporating deprivation quintiles also treated these as categories, rather than as continuous variables, with the most deprived fifth taken as the reference quintile. The results for each regression yielded an estimated odds ratio and a corresponding 95% confidence interval for each factor, including (in particular) the rural/urban dichotomy. Since in any given year death is a relatively rare event in the population, the odds ratio for a particular group can be interpreted approximately as the ratio of their risk of dying to that of the reference group (Kirkwood & Sterne, 2003, pg.212). Demographic adjustment We initially fitted a set of models that allowed us to investigate differences between rural and urban all-cause mortality, having adjusted for age. These analyses are similar to the age-standardised

A. Gartner et al. / Social Science & Medicine 72 (2011) 1685e1694

1689

Table 3 Kappa values quantifying agreement between the quintiles of five Welsh deprivation measures.

WIMD WIMD minus health WIMD employment WIMD income Townsend

WIMD

WIMD minus health

WIMD employment

WIMD income

Townsend

1 0.785 0.546 0.582 0.435

0.785 1 0.527 0.626 0.462

0.546 0.527 1 0.493 0.339

0.582 0.626 0.493 1 0.544

0.435 0.462 0.339 0.544 1

mortality rates for the rural/urban dichotomy at LSOA level that are reported in Gartner et al. (2008). Deprivation adjustment To investigate the extent to which rural/urban differences in allcause mortality could be explained by deprivation, a second set of models added a categorical variable encoding the (W)IMD deprivation quintile. Rurality-deprivation interaction Using the same reference category, a third set of models looked for differences between rural and urban areas in terms of the relationships therein of all-cause mortality and deprivation, as measured by the (W)IMD. Interaction terms allowed each quintile/ rurality combination to have a corresponding odds ratio, admitting complete modelling flexibility in describing the relationships between mortality, deprivation and rurality. Sensitivity to choice of deprivation measure Having investigated the interplay between rurality and deprivation, we then explored in more detail the sensitivity of our findings to the choice of deprivation measure. For each of the five deprivation indices described earlier, a separate model was fitted to examine its relationship with mortality. This exploration focussed on all-cause mortality. Specific causes of death We then proceeded to consider specific causes of death, using the same general approach: logistic regression was used to relate death from a particular cause to demographics, rurality and deprivation. Here again we only report adjustment by the (W)IMD, but details of model fits using other deprivation measures are available from the authors upon request. RUAC classification The final stage in our analysis was to delve more deeply into the classification of rurality, relating all-cause mortality to each of the six categories in the RUAC. This investigation was designed to reveal any specific subcategories within the rural/urban dichotomy that exhibit a particularly strong relationship with mortality. The reference category was chosen to be urban/less sparse, wherein the majority of the population reside. This analysis could unfortunately only be performed for England, with its larger study population. For Wales, there

are too few LSOAs for many of the combinations of deprivation quintile and RUAC class; in some cases, there are none at all. Results Demographic adjustment For England and Wales, the first row of Table 5 shows the results of regressing all-cause mortality on rurality, having adjusted for the age distribution of the population, but without any adjustment for deprivation. This table displays the odds ratios for rural areas, compared to urban ones, together with a corresponding 95% confidence interval. Before any adjustment for deprivation, mortality in England was 15% lower for males (9% for females) in rural areas than in urban areas. The pattern of results was very similar for Wales: mortality was 10% lower for males (8% for females) in rural areas compared to urban areas, before any adjustment for deprivation. Before adjustment, all differences were statistically significant at the usual 5% level. Deprivation adjustment The second row of Table 5 gives odds ratios arising from an analysis of mortality and individual demographics, together with area deprivation as measured by the (W)IMD. In England, rural/ urban differences remaining after adjustment for deprivation were relatively small, and have little substantive importance: 3% lower mortality in males in rural areas, and a (non-significant) 1% lower mortality in females. This suggests that, to a large extent, differences in deprivation distributions were responsible for the rural/ urban differences in all-cause mortality in England reported earlier in the paper. In Wales, the picture was much the same: relatively small (though statistically significant) differences remain after adjustment, with 5% lower mortality estimated for the rural male population, and 4% lower among women. Rurality-deprivation interaction For the investigation of the three-way relationship between mortality, rurality and deprivation, we continue to restrict ourselves to discussion of all-cause mortality and the (W)IMD. To

Table 4 Numbers of deaths and ICD-10 code for each cause of death (single average year). Cause of death

All causes Cancers Lung cancer Respiratory disease Circulatory disease Suicide (15þ) Accidents

ICD-10 code

C00eC97 C33eC34 J00eJ99 I00eI99 X600eX849, Y100eY338, Y340eY349, Y339 excluding ‘inquest adjourned’ V01eX59

England

Wales

Male

Female

Male

Female

234907 66318 16064 30452 89780 3296

260306 60920 10761 36638 98697 1162

15680 4397 1063 2001 6175 247

17467 4024 698 2396 6946 75

5706

4551

400

320

1690

A. Gartner et al. / Social Science & Medicine 72 (2011) 1685e1694

Table 5 Age-adjusted odds ratios for all-cause mortality for rural areas relative to urban areas, and adjustment for deprivation. England male

Unadjusted (W)IMD (W)IMD minus health Townsend Employment Income

England female

Wales male

OR

95% CI

OR

95% CI

OR

95% CI

OR

95% CI

0.85* 0.97* 0.96* 0.98* 0.96* 0.97*

(0.84, (0.96, (0.95, (0.97, (0.95, (0.96,

0.91* 0.99 0.98* 0.98* 0.98* 0.99

(0.90, (0.98, (0.97, (0.97, (0.97, (0.98,

0.90* 0.95* 0.94* 0.96* 0.97 0.97

(0.87, (0.92, (0.91, (0.92, (0.94, (0.93,

0.92* 0.96* 0.95* 0.96* 0.97 0.97*

(0.89, (0.93, (0.91, (0.93, (0.94, (0.93,

0.86) 0.98) 0.97) 0.99) 0.97) 0.98)

illustrate the interplay between mortality and deprivation in rural and urban areas, Fig. 2 plots sex-specific regression coefficients for each quintile of deprivation (as measured by the WIMD in Wales, and the IMD in England). The reference category is urban LSOAs in the most deprived quintile, and all plots show a similar pattern of increasing coefficients (higher mortality) with increasing deprivation, a steeper gradient being seen in the male population. The coefficients for rural areas are almost always lower than the corresponding ones in urban areas, reflecting the lower mortality in rural areas. The difference between the rural and urban coefficients in a given deprivation quintile is also generally much smaller than the difference between quintiles within either the rural or urban population, showing that much more of the area-level variability in mortality is explained by deprivation than rurality.

0.92) 1.00) 0.99) 0.99) 0.99) 1.00)

Wales female

0.93) 0.99) 0.97) 0.99) 1.01) 1.00)

0.95) 1.00) 0.98) 0.99) 1.00) 1.00)

Finally, the differences between urban and rural areas in a deprivation quintile are reasonably constant, suggesting there is little interaction between deprivation and rurality. The association between mortality and deprivation is fairly similar in rural and urban areas, so in subsequent analyses interaction terms are not included. Sensitivity to choice of deprivation measure The bottom four rows of Table 5 show the effect of adjusting the estimated impact of rurality on mortality by the four other deprivation measures. After adjustment for deprivation, mortality in England was between 2% and 4% lower for males (between 1% and 2% for females) in rural areas compared to urban areas, depending

male

female

0.1 0.0 −0.1 England

−0.2 −0.3

parameter estimate

−0.4 −0.5 −0.6 0.1 0.0 −0.1 Wales

−0.2 −0.3 −0.4 −0.5 −0.6 least

2

3

4

most

least

2

3

4

most

deprivation quintile Fig. 2. Plots of regression coefficients for WIMD (males and females) and IMD (males and females). (Urban quintiles are square and connected with a solid line, while rural quintiles are triangular and connected with a dashed line.)

A. Gartner et al. / Social Science & Medicine 72 (2011) 1685e1694

1691

Table 6 Age-adjusted odds ratios for specific causes in England for rural areas relative to urban areas, before and after adjustment for deprivation (IMD 2004). Male

Female

Cause of death

Before adjustment

After adjustment

OR

95% CI

OR

95% CI

OR

95% CI

OR

95% CI

Cancer Lung cancer Respiratory disease Circulatory disease Suicide Accidents

0.88* 0.73* 0.77* 0.88* 0.90* 1.05

(9.86, (0.70, (0.74, (0.87, (0.82, (0.99,

0.96* 0.90* 0.91* 0.99 1.12* 1.23*

(0.94, (0.86, (0.88, (0.97, (1.02, (1.15,

0.91* 0.71* 0.80* 0.95* 0.88 1.04

(0.90, (0.67, (0.78, (0.93, (0.75, (0.97,

0.98* 0.89* 0.91* 1.02* 1.01 1.12*

(0.96, (0.84, (0.88, (1.00, (0.86, (1.04,

0.89) 0.76) 0.79) 0.90) 0.98) 1.12)

on which deprivation measure was used. All differences remained statistically significant, due largely to the large sample sizes, so the presence or absence of a statistically significant difference was not sensitive to the choice of deprivation measure. It could be argued that a doubling of the size of the effect indicates sensitivity to this choice; however, the effects in question are small enough to render these fluctuations of little concern. In Wales, mortality was between 3% and 5% lower for males and females after adjustment for deprivation, depending on which deprivation measure was used. For Wales, statistical significance was found to be dependent on the choice of deprivation measure. In interpreting this apparent difference from England, however, it is important to bear in mind that the population of Wales is an order of magnitude smaller than that of England, and that therefore confidence intervals for the odds ratio are correspondingly wider. When taking this difference in population size into account, it is apparent that in Wales, like in England, deprivation measures are largely interchangeable, at least so far as their relationship to mortality is concerned. Substantively, the inferences drawn from the five adjusted analyses were essentially equivalent, and we conclude that there is little sensitivity to the choice of deprivation measure.

Before adjustment

0.98) 0.94) 0.94) 1.01) 1.24) 1.32)

After adjustment

0.93) 0.74) 0.82) 0.96) 1.02) 1.11)

1.00) 0.93) 0.93) 1.04) 1.19) 1.21)

Unadjusted mortality from lung cancer was 27% lower in rural areas for males, and 29% lower for females (Table 6). After adjustment for deprivation (IMD) these differences were reduced to 10% and 11% respectively. This is still a substantial difference: here the Index of Multiple Deprivation accounts for some, but by no means all, of the difference between rural and urban areas. For cancer and circulatory disease, mortality for males in rural England was 12% lower before adjustment, but after adjustment the difference had reduced to 4% for cancer and 1% for circulatory disease. This pattern was similar to that for all-cause mortality, with deprivation accounting for much of the apparent difference. For accidents, rural/urban differences before adjustment were not statistically significant, though we note again the very small number of such deaths. For males, mortality from accidents was 5% higher in rural areas before adjustment, the equivalent figure being 4% for females. However, after adjustment, mortality was respectively 23% and 12% higher: here adjustment for deprivation causes a widening of the differences. The odds ratio was significantly different from unity after adjustment, and larger in males than in females. The numbers of suicides were also very small, and only the differences for males in England were statistically significant before and after adjustment. Mortality from suicide was 10% lower for males in rural areas (12% for females) before adjustment, but 12% (1%) higher in rural areas after adjustment. Adjustment for deprivation reversed the differences in males, but accounted for the difference in females, the observed difference of 1% not being statistically significant. For most specific causes of death in Wales, statistically significant rural/urban mortality differences were found before adjustment for deprivation, but not after adjustment. As with deaths from all causes, this is, at least in part, a reflection of the smaller sample size and wider confidence intervals in Wales. In common with the English results, the widest differences were observed for lung cancer and respiratory disease. Mortality from lung cancer was 15% lower for males and 22% lower for females in rural areas before adjustment (Table 7). After adjustment for deprivation (WIMD) these differences were reduced to 7% for males and 13% for females. For female suicides and accidents in Wales, mortality was higher in rural areas. Male mortality from accidents was 14% higher in rural

Specific causes of death Six specific causes of death were investigated: all cancers, lung cancer, respiratory disease, circulatory disease, suicide and accidents. Because the choice of deprivation measure made very little difference to the results for specific causes of death, results are shown for only a single deprivation measure, the IMD for England and WIMD for Wales (Tables 6 and 7). For most specific causes in England, statistically significant, but small, differences between rural and urban mortality were found both before and after adjustment for deprivation. For cancer, lung cancer, respiratory and circulatory disease, mortality tended to be lower in rural areas both before and after adjustment, while mortality for accidents was higher in rural areas. The results for suicides were less clear, and differed between males and females. Before adjustment, the widest rural/urban differences in England were observed for lung cancer and respiratory disease.

Table 7 Age-adjusted odds ratios for specific causes in Wales for rural areas relative to urban areas, before and after adjustment for deprivation (WIMD 2005). Male Cause of death

Cancer Lung Cancer Respiratory disease Circulatory disease Suicide Accidents

Female

Before adjustment

After adjustment

Before adjustment

OR

95% CI

OR

95% CI

OR

95% CI

OR

95% CI

0.91* 0.85* 0.87* 0.93* 0.96 1.14

(0.85, (0.75, (0.79, (0.88, (0.74, (0.93,

0.95 0.93 0.96 0.98 1.04 1.27*

(0.89, (0.81, (0.87, (0.92, (0.79, (1.03,

0.94 0.78* 0.87* 0.96 1.16 1.06

(0.89, (0.66, (0.80, (0.91, (0.73, (0.85,

0.98 0.87 0.93 1.00 1.27 1.11

(0.91, (0.74, (0.85, (0.95, (0.78, (0.88,

0.96) 0.96) 0.95) 0.98) 1.25) 1.40)

1.02) 1.06) 1.06) 1.03) 1.36) 1.57)

After adjustment

1.01) 0.92) 0.95) 1.01) 1.85) 1.33)

1.04) 1.03) 1.02) 1.06) 2.07) 1.41)

1692

A. Gartner et al. / Social Science & Medicine 72 (2011) 1685e1694

areas before adjustment, and this difference had widened to 27% after adjustment (and was statistically significant). Unadjusted male suicide rates were slightly lower in rural areas, but this difference could be accounted for by adjusting for deprivation. RUAC classification For the English data, it was possible to investigate variation in all-cause mortality among the six categories detailed by the RUAC (notably, though, even here there were some combinations of deprivation quintile and RUAC category that had very few constituent LSOAs). Relative to the reference category of less sparsely populated urban areas, all other categories had either essentially the same or lower mortality rates, sometimes substantially so. In both male and female subpopulations, the sparse village and dispersed category had an unadjusted odds ratio of 0.81; after adjustment for the IMD, these increased to 0.88 and 0.84 respectively, still representing at least 12% lower mortality in these areas after adjustment for deprivation. The less sparse village and dispersed category also showed lower rural mortality after adjustment for deprivation in males with mortality being 8% lower. Among females mortality was similar to the reference group. However, in the other three RUAC categories (urban sparse, town and fringe sparse, and town and fringe less sparse), initial differences in mortality were essentially explained by adjustment for deprivation (Table 8). It would appear, therefore, that the apparent beneficial effects of rurality on mortality are largely concentrated within the village and dispersed settlement types. Attenuation resulting from combining rural settlement types may partially explain the fact that, in most of our analyses, rurality was not found to have a substantial impact on mortality after adjusting for deprivation. Conversely, for those specific causes of death for which an important difference in rurality was uncovered, it seems plausible that this difference may be even more marked in the village and dispersed subgroup. Discussion Comparison to other studies We note again that many different definitions of rurality, study populations and geographical units have been employed in rural health research. Direct comparison of existing work with results from the present study is therefore difficult; however, the general themes and outcomes are described where possible. Senior et al. (2000) found that, when controlling for the Townsend and Carstairs indices, mortality differences between rural and urban areas of Wales were considerably reduced, a finding that is consistent with the results of the present study. Their findings cannot be compared directly to this study, particularly due to use of different age groups and rurality classifications. The present study also used (among others) the Townsend index as a deprivation measure, and rurality differences in Welsh all-cause,

cancer and circulatory disease mortality were not statistically significant after this adjustment. Likewise, Senior et al. (2000) reported that there were no residual mortality differences for cancers, while for respiratory and circulatory diseases some of the differences could be accounted for through adjusting for employment variables. In the present study, results for cancer and circulatory disease were similar, both showing small residual differences, while for respiratory disease there were substantial remaining differences. While there appear to be some discrepancies between the two studies in the findings on specific causes, the general pattern of reduction of differences through adjustment for deprivation appears similar. The largest rural/urban differences were found for respiratory disease and lung cancer. Deprivation accounted for some of the differences, but important differences remained even after adjustment. O’Reilly et al. (2007) also reported the widest differences for these two specific causes between rural and city areas in Northern Ireland, having adjusted for social class. The studies are not directly comparable, particularly as they refer to ‘rural’ as one of three classes. It is, however, interesting, that the general pattern is similar, even if the differences they reported were not as large as in this study. O’Reilly et al. (2007) speculated that these differences may be related to air quality, or due to insufficient adjustment for deprivation, recognising that the latter is strongly associated with smoking. Law and Morris (1998) also found excess deaths in urban areas compared to rural areas in England and Wales, particularly from lung cancer and chronic bronchitis and emphysema (part of the respiratory disease group), and attributed these mainly to differences in smoking habits. Neither of these studies (Law & Morris, 1998; O’Reilly et al., 2007) provided evidence for a direct link in their studies between rural/urban differences and smoking, but based this on the association between deprivation and smoking. For suicides the pattern was less clear. The number of such deaths was very small and, both before and after adjustment for deprivation, only the rural/urban differences for males in England were statistically significant. In their study based in Scotland, Levin and Leyland (2005) found the highest suicide risk for ‘remote rural’ areas, while the risk was lower in ‘accessible rural’ areas compared to urban areas. Having adjusted for deprivation, the authors reported that the lower risk in ‘accessible rural’ areas compared to urban areas remained. Although the classes are not comparable to the current study, the aggregated rural areas in England, and to a lesser degree in Wales, are dominated by the more numerous ‘rural less sparse’ areas. It is possible that results for the aggregated rural areas in the current study mask higher rates in the less numerous ‘rural sparse’ areas. Future analysis of suicides within rural subclasses might allow more direct comparison of results, but at such a level of detail numbers would inevitably be small. Adjustment for deprivation when calculating the odds ratios for deaths due to accidents in rural versus urban areas produced similar results as for suicide. Adjustment increased the odds ratio in all cases and in England the rates in rural areas became significantly

Table 8 Age-adjusted odds ratios for all-cause mortality in England, after adjustment for IMD and the six RUAC classes. Male Cause of death

Urban sparse Town and fringe sparse Village and dispersed sparse Urban less sparse (ref) Town and fringe less sparse Village and dispersed less sparse

Female

Before adjustment

After adjustment

Before adjustment

After adjustment

OR

95% CI

OR

95% CI

OR

95% CI

OR

95% CI

0.92 0.93 0.81 1.00 0.98 0.84

(0.81, 1.04) (0.86, 1.01) (0.76, 0.86)

0.96 0.99 0.88 1.00 1.00 0.92

(0.85, 1.09) (0.91, 1.07) (0.83, 0.93)

1.01 0.93 0.81 1.00 0.91 0.91

(0.94, 1.09) (0.89, 0.98) (0.78, 0.85)

0.98 0.98 0.84 1.00 1.00 1.00

(0.91, 1.05) (0.93, 1.03) (0.80, 0.88)

(0.94, 1.03) (0.79, 0.89)

(0.96, 1.05) (0.87, 0.98)

(0.90, 0.93) (0.90, 0.92)

(0.98, 1.01) (0.98, 1.01)

A. Gartner et al. / Social Science & Medicine 72 (2011) 1685e1694

higher than those in urban areas. In Wales the point estimates were very close to those for England but the smaller sample size meant the values were not significantly raised above 1. Not adjusting for deprivation would mask what appears to be a real increased risk in rural areas. For both England and Wales, the removal of the health domain from the IMD and WIMD resulted in only a very small difference (0.01) in the rural odds ratio compared to the original IMD/WIMD adjustment. Adams and White (2006) came to a similar conclusion, but recommended removing the health domain as best practice for research aiming to establish how socioeconomic factors relate to health outcomes. Alternatively, it may be advisable to use individual components such as the IMD/WIMD employment or income domains, which may be more easily interpreted and, for Wales, were more strongly related to mortality than the other measures. Limitations and strengths of the study Since our main interest lies in geographical difference in mortality, the present study adopted area-based measures of deprivation. Necessarily, therefore, we cannot draw conclusions about relationships between individual deprivation, rurality and mortality. We emphasise, however, that mortality and demographic aspects of the study were based on individual data, a reasonable design from which to consider the association of area deprivation, rurality and individual mortality. In considering the question of rural health, we are faced with the rather fundamental problem that there is no universally accepted definition of rurality. Thus of the three concepts underlying the investigations in this paperdmortality, rurality and deprivationdonly mortality is unambiguously defined. This study uses the Rural and Urban Area Classification 2004. A number of authors have discussed the shortcomings of previous classifications and the methodological problems of rurality as a unidimensional concept (Higgs, 1999; Martin et al., 2000). It is outside the scope of this study to evaluate the classification, but, by using this latest classification (which has, to date, been little used in health research) and allowing some comparison with other results, the study makes a valuable contribution to the study area. Most of our results focus on the rural/urban dichotomy but we also found differences between some of the rural/urban subclasses. Simply using the rural/urban dichotomy may therefore mask some key distinctions between rural and urban areas, and within rural classes (Higgs, 1999). Our results suggested that all-cause mortality was lower in those areas classified as ‘village and dispersed sparse’, even after adjusting for deprivation. This is similar to the findings of Riva et al. (2009) who compared two types of urban areas and two sets of rural areas with respect to self-reported health, common mental disorders and being overweight or obese. There was some evidence of differences within both urban and rural classifications, with those living in villages generally having better outcomes than residents of semi-rural areas and those living in London having better outcomes than those in other cities. Their results, based on individual data, suggest that there may be interactions between the rural/urban classes and a range of factors including deprivation and that untangling these relationships is a complex task. It must also be noted that some of the classes are relatively rare, so aggregations of the rural classes by settlement type or sparsity may represent a useful compromise. Many authors (Higgs, 1999; Martin et al., 2000) argue that deprivation is underestimated in rural areas, and that some rural areas should be classed as comparatively more deprived. If indeed rural and urban areas are more similar in terms of distribution of deprivation than Fig. 1 suggests, then our adjustment for deprivation may have been larger than necessary. Consequently, it is

1693

possible that the real difference between urban and rural mortality may actually be bigger than our results suggest. Considerable effort has gone into designing this study, taking account of known weaknesses of underlying measures, concepts and classifications. A number of deprivation measures have been employed to avoid reliance on a particular measure and its weaknesses. This study is based on a very large sample: the entire population of England and Wales. This, coupled with the detail of the small area units (of approximately 1500 people), provides a major advantage over previous studies; these tended to use larger area units, or electoral units, such as wards, which vary hugely in size. This is particularly important given the ecological nature of the studied deprivation measures, which become less representative of individuals as areas increase in size. The analysis was performed separately for the two countries and sexes to provide additional comparisons. Also, potential rural/urban differences in the association of deprivation measures and mortality were investigated to adjust rigorously for deprivation. The study also makes a contribution in the comparison of different deprivation measures. The (W)IMD indices have not been used in research of this type before, and the ability to compare how the adjustment by (W)IMD compares with adjustment by the Townsend index is a novel contribution to the field. Detailed exploration of the possible interactions between rurality and deprivation, not only for England but also for Wales, which has extensive rural areas, is also a particular strength of this work. Conclusions The popular notion is that rural populations are inevitably healthier and live longer. However, our large study adds to the growing body of evidence countering this argument: although there were initial differences in mortality between rural and urban areas, the differences in all-cause, circulatory disease and cancer mortality could largely be accounted for by deprivation. This suggests that, for these causes of death, rural populations were not found to be inherently ‘healthier’, but were in fact similar to urban areas. After adjustment for deprivation, there were substantial residual differences for mortality from lung cancer and respiratory disease, where mortality was lower in rural areas. These deaths represent a sizable proportion of deaths in the population, and such large differences are of major public health importance. Further investigation is needed to determine to what extent these differences are related to different smoking habits, air pollution or other factors. It could be argued that for these causes of death the notion of ‘healthier’ rural populations was confirmed. Mortality due to accidents was found to be higher in rural areas after adjustment for deprivation. This may possibly be accounted for by farming and road traffic accidents, thought to be higher in rural areas. Mortality differences from suicide were less clear, but for males in England these were also higher in rural areas after adjustment for deprivation. This suggests that for deaths from accidents and suicide the result is reversed, and urban populations emerge as ’healthier’. This investigation studied small geographical units of similar size and homogeneity of population, a significant improvement on previous studies. However, poor mortality rates may be seen in any area of England and Wales, rural or urban, and in areas classed as least or most deprived. Rural areas, in particular, are reported to be very heterogeneous, with people of diverse backgrounds, income levels and health status living side by side (Haynes & Gale, 2000). Averages, even in such small area units, may not capture all of the intricacies and diversity at the local level. The results of this study should not, therefore, be interpreted as denying the patterns of poor health that may exist in some sections of the rural and urban populations.

1694

A. Gartner et al. / Social Science & Medicine 72 (2011) 1685e1694

This extensive study on the entire population of England and Wales, using the most up-to-date rurality classification, deprivation measures and small area geography, provides a valuable update and significant improvement to this important research area. Given the strength of our findings within specific rural subclasses, the greatest future benefit for research into rural health may now lie in the analysis of inequalities within rural areas, and understanding the specific problems and needs of those populations (Higgs, 1999). In policy development the rural/urban context should certainly be considered in the general drive to reduce health inequalities. Acknowledgements We are grateful to the Office for National Statistics and, in particular, to Myer Glickman and Emma Gordon, for supporting this project and providing data access. References Adams, J., & White, M. (2006). Removing the health domain from the index of multiple deprivation 2004 e effect on measured inequalities in census measure of health. Journal of Public Health, 28(4), 379e383. Bibby, P., & Shepherd, J. (2004). Developing a new classification for urban and rural areas for policy purposes e the methodology. Retrieved from. http://www. statistics.gov.uk/geography/nrudp.asp. Carstairs, V., & Morris, R. (1990). Deprivation and health in Scotland. Health Bulletin, 48, 162. Christie, S., & Fone, D. (2003). Does car ownership reflect socio-economic disadvantage in rural areas? A cross-sectional geographical study in Wales, UK. Public Health, 117(2), 112e116. Farmer, J., Baird, A., & Iversen, L. (2001). Rural deprivation: reflecting reality. The British Journal of General Practice, 51(467), 486. Gartner, A., Farewell, D., Dunstan, F., & Gordon, E. (2008). Differences in mortality between rural and urban areas in England and Wales, 2002e2004. Health Statistics Quarterly/Office for National Statistics, 6. Haynes, R., & Gale, S. (2000). Deprivation and poor health in rural areas: inequalities hidden by averages. Health and Place, 6(4), 275e285. Higgs, G. (1999). Investigating trends in rural health outcomes: a research agenda. Geoforum, 30(3), 203e221. House, J. S., Lepkowski, J. M., Williams, D. R., Mero, R. P., Lantz, P. M., Robert, S. A., et al. (2000). Excess mortality among urban residents: how much, for whom and why? American Journal of Public Health, 90, 1898e1904.

Jordan, H., Roderick, P., & Martin, D. (2004). The Index of Multiple Deprivation 2000 and accessibility effects on health. British Medical Journal, 58(3), 250. Kirkwood, B. R., & Sterne, J. A. (2003). Essential medical statistics. Wiley-Blackwell. Lagace, C., Desmeules, M., Pong, R. W., & Heng, D. (2007). Non-communicable disease and injury-related mortality in rural and urban places of residence: a comparison between Canada and Australia. Canadian Journal of Public Health, 98, S62eS69. Law, M., & Morris, J. (1998). Why is mortality higher in poorer areas and in more northern areas of England and Wales? British Medical Journal, 52(6), 344. Levin, K., & Leyland, A. (2005). Urban/rural inequalities in suicide in Scotland, 1981e1999. Social Science & Medicine, 60(12), 2877e2890, Elsevier. Retrieved from. http://linkinghub.elsevier.com/retrieve/pii/S027795360400557X. Levin, K., & Leyland, A. (2006). A comparison of health inequalities in urban and rural Scotland. Social Science & Medicine, 62, 1457e1464. Martin, D., Brigham, P., Roderick, P., Barnett, S., & Diamond, I. (2000). The (mis) representation of rural deprivation. Environment and Planning A, 32(4), 735e752. Middleton, N., Gunnell, D., Frankel, S., Whitley, E., & Dorling, D. (2003). Urbanerural differences in suicide trends in young adults: England and Wales, 1981e1998. Social Science & Medicine, 57(7), 1183e1194. National Assembly For Wales. (2005). The Welsh index of multiple deprivation. Noble, M., Wright, G., Dibben, C., Smith, G., McLennan, D., Anttila, C., et al. (2004). The English indices of deprivation, 2004. O’Reilly, G., O’Reilly, D., Rosato, M., & Connolly, S. (2007). Urban and rural variations in morbidity and mortality in Northern Ireland. BMC Public Health, 7(1), 123. Ostry, A. S. (2009). The mortality gap between urban and rural Canadians: a gendered perspective. Rural & Remote Health, 9, 1286. Peen, J., Schoevers, R. A., Beekman, A. T., & Dekker, J. (2010). The current status of urbanerural differences in pyschiatric disorders. Acta Psychiatr. Scand., 121, 84e93. Riva, M., Curtis, S., Guavin, L., & Fagg, J. (2009). Unravelling the extent of inequalities in health across urban and rural areas: evidence from a national sample in England. Social Science & Medicine, 68(4), 654e663. Senior, M., Williams, H., & Higgs, G. (2000). Urbanerural mortality differentials: controlling for material deprivation. Social Science & Medicine, 51(2), 289e305. Smith, G. D., Whitley, E., Dorling, D., & Gunnell, D. (2001). Area based measures of social and economic circumstances: cause specific mortality patterns depend on the choice of index. British Medical Journal, 55, 149. Townsend, P., Phillimore, P., & Beattie, A. (1988). Health and deprivation: Inequality and the North. Routledge Kegan & Paul. Van Hooijdonk, C., Droomers, M., Deerenberg, I. M., Mackenbach, J. P., & Kunst, A. E. (2008). Higher mortality in urban neighbourhoods in The Netherlands. J. Epidemiology Comm. Health, 62, 499e505. Weich, S., Twigg, L., & Lewis, G. (2006). Rural/non-rural differences in rates of common mental disorders in Britain. British Journal of Psychiatry, 188, 51e57. Zimmer, Z., Kaneda, T., & Spess, L. (2007). An examination of urban versus rural mortality in China using community and individual data. Journals of Gerontology Series B, 62, S349eS357.