METHODS

Deaths from all causes by five year age groups and sex for local authorities in England and Wales on an annual basis were obtained from the Vital Statistics Tables of the Office for National Statistics (ONS) from 1979 to 1998; mid-year population estimates (also from the ONS) were used to provide denominator populations. A comparable dataset was assembled for Scotland from death records and population estimates from the General Register Office for Scotland. Analysis was restricted to deaths among the population aged less than 65; a summary of these data is provided in table 1. An increase of 4.1% in the population aged less than 65 was accompanied by a 32.8% decline in the number of deaths, meaning that by 1998 there were about 12 fewer deaths per 10 000 population than in 1979. The largest decreases were seen in the older age groups.

The data refer to populations living in local authority districts (England and Wales) or local government districts (Scotland). These are the basic units of the analysis and, although far from homogenous in terms of their populations, they reflect substantial differences in terms of their compositions.²⁶ In 1979 there were 459 such districts. These districts are clustered within the nine regions of England plus Scotland and Wales; however, the intermediate level of county in England and Wales and Health Board in Scotland is also considered. London is considered to be a single county (as well as being a region); it was broken down into five pseudo-counties based on the Nomenclature of Units for Territorial Statistics (NUTS3) areas.²⁷ (As the counties of England and Wales are NUTS3 areas, this creates a standard unit.) These will enable us to determine whether regional variations arise due to the clustering of districts of similar mortality within counties; there were a total of 73 counties in 1979.

The data contain a number of hierarchies. Repeated measures (the number of deaths) were available for each of 20 years in each district. As some districts always tend to have mortality rates that are higher or lower than average, the observations on each district are clearly correlated (just as they would be if repeated measures were made on an individual). Moreover, the districts are nested within counties and the counties within regions. Mortality rates for all of the districts within the same county are likely to be correlated, as are those for all counties within a particular region. Multilevel modelling^28,29 was used to take account of these hierarchies and allow for persistently high (or low) mortality rates in some regions (in contrast with random fluctuations from one year to another). This enabled the partitioning of the variation between districts to that part attributable to each of the district, county and region levels. The multilevel Poisson regression model applied used as an offset the expected number of deaths in each year in each district (based on the 1998 age and sex specific mortality rates for Great Britain) and was therefore equivalent to modelling indirectly standardised rates.³⁰ The software used was MLwiN³¹; further details of the methodology can be found elsewhere.³² Variance components were estimated from the model with the levels of year (observation), district, county and region; however, as there were only 11 regions these were treated as fixed effects as compared with random effects in the models reported (that is, a dummy variable was included for each region).

A trend was fitted to allow for changes over time. The SMRs reported in this paper are those estimated from this trend; using the data for all years in this way has the advantage over the observed SMR for a single year of smoothing out random fluctuations around the trend. The variation between districts within each region is summarised by a standard deviation. As a Poisson multilevel model has been fitted these standard deviations are on the same scale as the log rate ratio. The trend was found to vary significantly between districts, counties, and regions; this meant that the standard deviations changed over time. If the standard deviation is unchanged it implies that the relative dispersion of districts is unchanged (and, with falling mortality rates, this will imply a reduction in the absolute variation).

Local government reorganisation (LGR) between 1995 and 1998 created new unitary authorities; old districts were merged and split as part of this process.²⁷ The death records used in Scotland included both pre-LGR and post-LGR areas of residence; however, the aggregated mortality data in England and Wales were provided using boundaries in existence that year. The impact of these changes, shown in table 1, was a reduction in the number of districts from 459 in 1979 to 432 in 1998. As the trends in the mortality rates in individual areas—and the variance between areas—are of interest, a multiple membership model³³ was used to take account of the geographical changes. In essence, this entails modelling the relative risk of mortality associated with each of the new areas through the weighted sum of the risks for the pre-LGR areas. The weights used were according to the populations in the pre-LGR and post-LGR areas.

RESULTS

Between 1979 and 1998 the mortality rate (standardised to the European standard population) for Great Britain fell from 348 to 219 per 100 000 population aged 0–64 (table 2). Over this time the standard error of the district mortality rates—presented as a measure of variation—decreased from 59 to 48. Although the standard error decreased in absolute terms, the fact that this decrease was smaller than that seen for the mortality rate implies that relative inequalities increased. Also shown in table 2 are the mortality rates and standard errors for each region.

The SMR in 1979 for the whole of Great Britain was 157 (based on 1998 mortality rates); the 19 years therefore saw a 36.4% reduction in age and sex standardised mortality rates. This reduction was not uniform across regions; figure 1 shows yearly SMRs based on national mortality rates in 1998. The reduction in SMR was most marked in Wales—from an above average 165 in 1979 to a below average 96 in 1998, a decrease of 41.9%. The decrease in mortality was relatively slower in Scotland (from 191 to 128, a reduction of 33.1%) and London (150 to 103, 31.4%).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

SMR for country and region, 1979–1998 (SMR for GB 1998 = 100). sco, Scotland; ne, North East; nw, North West; lon, London; y and h, York and Humber; wm, West Midlands; wal, Wales; em, East Midlands; se, South East; sw, South West; eas, East of England.

Figure 1 describes the differences between regions; however, this is only part of the picture. There are also considerable disparities in mortality between districts within the same region. Table 3 shows the districts with the highest and lowest SMRs in 1979 and 1998 in each region. The SMRs given in this table are predicted SMRs, based on long term trends in each district and therefore not subject to year on year fluctuations. In every region there has been a greater reduction in mortality in districts with the lowest mortality rates than among those with the highest rates. The reductions in low mortality areas range from 77% in Scotland (from 137 to 77) to 47% in London; in high mortality areas they range from 37% in Wales to just 5% in Scotland. In general it has not been the same districts that have had the highest or lowest rates in both 1979 and 1998. Also shown in table 3 are the standard deviations for districts within each region; these have increased for all regions by between 16% (Wales) and 90% (East Midlands). These increases suggest a pattern of increasing relative inequalities in all regions.

A variance components model was used to quantify the relative importance of differences between and within regions. The variation between regions accounted for 44% of the total variation in addition to random year on year fluctuations in district mortality rates. There was little variation (17%) between counties within regions, with the remainder (39%) arising because of differences between districts within counties.

Although a total of 56% of the variation in district mortality rates was attributable to differences within regions, this figure varied from one region to another. The relative importance of between regional variation and within regional variation for each region can be quantified in terms of the proportion of mortality that could be avoided if either of these sources of variation were to be eliminated and the other remained unchanged. Firstly, we consider reducing the average mortality rate in each region to that of the East of England—the region with the lowest SMR for most years in figure 1. The relative pattern of inequalities within regions is assumed to be unchanged so that, for example, the relative risk of mortality in Glasgow City remains 2.82 times that in Eastwood in 1998 (as given in table 3). Figure 2 shows the proportion of deaths that would be eliminated in each region in each year in this manner. The excess mortality is calculated as the proportional decrease in observed deaths that would result from having the same long term trend in the risk of mortality as the East of England. Yearly fluctuations in the observed deaths around the long term trend mean that the excess mortality in the East of England will not be exactly 0%. Although some regions such as Wales decreased differences with the East of England, most maintained relative differentials. Excess mortality was about 25% in Scotland, the North East and the North West. There was a persistent increase in excess premature mortality in London, from 14% of all deaths under 65 in 1979 to 19% in 1998.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Percentage of deaths that would be avoided if differences between regions were eliminated. sco, Scotland; ne, North East; nw, North West; lon, London; y and h, York and Humber; wm, West Midlands; wal, Wales; em, East Midlands; se, South East; sw, South West; eas, East of England

Now we consider the effect of reducing the variation within regions. Figure 3 shows the proportion of deaths that would be eliminated in each region if all districts in that region had a mortality rate equivalent to the 10th centile of districts within that region. (It would be possible to choose any other centile, for example, the median or 50th centile. The choice of the 10th centile means that a fairly demanding target is being set.) The chosen centile is obtained from the variance of all districts within the region. From figure 3 we can see that for most regions there was an increase in the proportion of excess deaths over the 20 years; the exceptions to this were a decline in Wales (from 14% in 1979 to 9% in 1998) and little change in London (18% to 19%) and the South West (11% to 13%). Not only did Scotland have a higher percentage of excess deaths than any other region, but this proportion increased substantially from 23% to 33%.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Percentage of deaths that would be avoided if differences within regions were eliminated. sco, Scotland; ne, North East; nw, North West; lon, London; y and h, York and Humber; wm, West Midlands; wal, Wales; em, East Midlands; se, South East; sw, South West; eas, East of England

DISCUSSION

This paper has described changes over time in mortality rates and variation between districts in these rates. There has been no attempt to explain why this variation occurs; rather, the focus has been on the quantification of the variation in terms that are useful when considering inequalities in health. Many recent studies have focused on the relation between contextual (area) characteristics and health, with particular emphasis on social capital,³⁴ socioeconomic deprivation or socioeconomic status,^35–38 and area level income inequality.^39–43 Such factors would provide suitable starting points when trying to explain why these inequalities exist, and the contribution of any of these to inequalities could then be assessed.

The overall pattern of a reduction in SMRs is encouraging, although there must be questions as to why the reductions have varied across regions. It is possible that, at least in part, some of these differences could be explained by compositional differences in terms of factors associated with widening mortality differentials such as social class.⁵ This would not account for the much slower decline in mortality seen in London compared with other regions; by 1998 the relative risk of mortality in London was 15% higher than would have been the case had mortality rates in London declined in line with other regions. Possible causes include variation introduced through differing patterns of selective migration, either from abroad⁴⁴ or within Great Britain.⁴⁵ This slow decline led to an increase in the relative risk of mortality associated with living in London compared with the East of England, the region with the lowest mortality rates. Wales was the only region that closed the gap, with excess mortality falling from 19% to 12%. Inequalities between regions remained unchanged elsewhere.

The reduction in mortality rates was less pronounced in regions with high mortality in 1979; however, this effect was small and made little contribution to mortality differentials. But within regions the gaps between the districts with the lowest and the highest mortality rates increased. As in general it is not the same district that has a high or low mortality rate in a region in 1979 and 1998, the rate at which area mortality rates diverged will be greater than that implied by a simple comparison of the areas in table 3. This increasing polarisation as gaps have opened up between affluent and deprived or healthy and unhealthy areas has translated into increases in the proportion of excess deaths in most regions, with a particularly steep increase seen in Scotland.

Inequalities in health are undesirable and should be reduced when possible.⁴⁶ Many European countries now have specific programmes designed to reduce inequalities in health⁴⁷; it is in recognition of this that the Department of Health recently included the reduction of geographical differences in life expectancy as one of their inequality targets.⁴⁸ The department acknowledges the difficulty in reaching the target—a reduction by at least 10% of the gap in average life expectancy at birth between the 20% of health authorities (in England and Wales) with the lowest life expectancy and the population as a whole—given current trends. The increase in inequalities in mortality between areas is not unique to Great Britain but is an international problem. As the health of certain areas worsens relative to the rest of the country it is important to bear in mind that such inequalities should be tackled nationally and not just at the area or community level.^49,50

Mortality in Scotland is higher than in any region of England or Wales; about a quarter of premature deaths in Scotland would be avoided if Scotland had the same average mortality rate as the East of England. Research based on the early 1990s showed that only 60% of the difference between Scotland and England and Wales could be explained by deprivation—a finding termed the “Scottish effect”.⁵¹ However, a larger and increasing proportion of deaths would be avoided if inequalities between districts within Scotland were eliminated. The failure of current levels of health spending to reduce either source of inequality should be borne in mind in the face of recent calls to abolish⁵² or even maintain⁵³ subsidy through the Barnett formula.