Global and regional trends in COPD mortality, 1990–2010

Introduction

Global patterns of mortality are changing, mainly due to three related transformations: 1) an epidemiological transition driven by change in global risks, with falling death rates in childhood associated with infectious disease; 2) a demographic transition with more people surviving into later life, and 3) a healthcare transition with improved effectiveness of both preventive and curative or palliative treatments. The reassessment of the Global Burden of Disease programme in 2010 identified chronic obstructive pulmonary disease (COPD) as the third most common cause of death worldwide [1], up from fourth in 1990 [2]. Understanding the reasons for this is important in order to identify potential public health interventions. In particular, we need to know whether this change has been caused by an increase in the mortality rate or by an increase in the absolute number of deaths due to increased prevalence, for example associated with ageing in the population. Alternatively, the change may be a relative one, with COPD mortality having declined less than mortality from other causes.

Cigarette smoking is regarded as the most important cause of COPD, with other inhaled hazards from occupational, domestic or outdoor sources thought to play an important role [3]. However, in a study of mortality in local authorities in England and Wales, Barker and Osmond [4] noted that there was a strong association between childhood deaths from pneumonia and bronchitis in the 1920s and deaths from “COPD” in the 1970s, with high death rates from both conditions being strongly associated with poverty. As this association was not noted for lung cancer, this relationship was unlikely to be strongly confounded by cigarette smoking. Subsequently, by following up a cohort of men born in the 1920s, with very complete health records from the first year of life, the authors showed associations between early life factors and death from COPD and adult lung function, again associations that were not explained by smoking habits [5]. More recently, the Burden of Obstructive Lung Disease (BOLD) study has shown a strong ecological association between national levels of poverty and the prevalence of both a low forced vital capacity (FVC) and national COPD mortality rates [6].

Many descriptions of national and regional mortality trends over time have been published previously [7–14]. These have reported variable trends, mostly at a local level, and have rarely analysed the drivers of change. In this study we have examined trends in COPD mortality rates for all 21 Global Burden of Disease regions. We have assessed the impact of population changes on the number of COPD deaths, the associations between changes in age- and sex-specific mortality rates from COPD, and changes in both national income and smoking. By analysing trends within regions over time, we have avoided much of the confounding between national income and other factors that vary across regions, which is inevitable in purely cross-sectional analyses.

Results

Most deaths from COPD occur in low-income regions including East, South and South-East Asia and central Africa, where age-standardised rates are also higher (fig. 1). Globally, the total number of COPD deaths fell from three million in 1990 to 2.8 million in 2010, a fall of 5.5%, whereas the expected number of age- and sex-specific rates were maintained at the 1990 levels, which was very much higher at 5.2 million (table 1). There was a clear fall in age- and sex-specific rates for all groups between 1990 and 2010 (fig. 2a), and COPD mortality fell in all age groups, with the greatest relative declines in younger people (fig. 2b). There was substantial variation in the 2010:1990 mortality rate ratio but clear reductions in all regions, except the North America high-income region where the mortality rate remained stable (fig. 3).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Age- and sex-adjusted chronic obstructive pulmonary disease mortality in 2010.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2

a) Chronic obstructive pulmonary disease (COPD) mortality rates for 1990 and 2010, and b) COPD mortality ratios for 2010:1990 according to sex and age.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3

Mean age- and sex-specific chronic obstructive pulmonary disease (COPD) mortality ratios for 2010:1990 for each of the 21 regions. ^#: high income.

Changes in smoking index varied widely between regions, from a reduction of 15.3% in South Sub-Saharan Africa to an increase of 14.5% in East Asia (table 2). GNI increased in all regions over this period of time, except in West Sub-Saharan Africa. In most regions this increase was substantial and in East Asia it was >700%. Plotting age-standardised COPD mortality rates by GNI showed a negative association. Because this appeared nonlinear we transformed GNI to a log scale.

Multilevel regression analysis, adjusting for age and sex, showed a highly statistically significant association of smoking index and GNI with COPD mortality rate for both years (table 3). Since the model is a log–log regression (exposure and outcome both log transformed), the coefficients for GNI and smoking represent the percentage change in COPD mortality rate given a certain percentage change in GNI or smoking, while keeping all other variables constant. For example, in the 1990 analysis, if we consider the effect of a 50% increase in exposure (1.5 times), the GNI coefficient of −0.47 means a decrease in mortality rate of 17% (1.5⁻⁰·⁴⁷=0.83, i.e. 17% reduction); similarly, the smoking coefficient of 0.43 means a 19% increase in mortality rate. Corresponding figures for 2010 were a 12% (95% CI 3–19%) decrease for a 50% increase in GNI and a 17% (95% CI 12–21%) increase for a 50% increase in smoking index. As expected, for both 1990 and 2010 the model showed COPD mortality rates were higher in older age groups and in men, independent of smoking.

The multilevel regression analysis of the ratio of COPD mortality rates between 2010 and 1990, also analysed as a log–log model, showed a 13% decrease in mortality rate for a 50% increase in GNI from 1990 to 2010 (p=0.02). The effect of change in smoking index (2010:1990 mortality rate ratio) on change in COPD mortality was also statistically significant, although its magnitude was much smaller; a 0.16% (95% CI 0.08–0.17%) increase in COPD mortality rate for a 50% increase in smoking index (table 3). In agreement with figure 2, the multilevel regression model showed large between-region variation in COPD mortality, with an intraclass correlation coefficient of 0.69. When we dropped the four regions with the highest change in GNI from the analysis we obtained the same coefficient for the smoking index and only a small change in the coefficient for GNI, with a 16% (95% CI 4–25%) decrease in COPD mortality rate for a 50% increase in GNI from 1990 to 2010. This remained statistically significant.

Finally, we estimated the percentage change in COPD mortality that was attributable to the change in GNI and smoking exposure. Changes in GNI and smoking exposure accounted for a 26.2% (95% CI 4.7–47.7%) and for 3.9% (95% CI 3.2–4.6%) change in COPD mortality.

Discussion

The relative increased burden from COPD mortality compared with mortality from other causes is largely due to changes in population structure. The apparent paradox that COPD has risen to be the third most common cause of death at a time when age-standardised mortality rates from COPD have been falling is explained by the relatively faster decline in all-cause mortality. Between 1990 and 2010 the total number of COPD deaths fell from 3 to 2.8 million, a fall of 5.5%, while during the same period deaths due to all causes fell by 17%. The increase in COPD mortality due to population ageing explains why the absolute number of COPD deaths has decreased only slightly although age- and sex-specific mortality rates have been decreasing in almost all regions. Much of this decline is associated with changes in GNI, with countries with the greatest improvement in GNI having the greatest relative improvement in COPD mortality rates. Changes over time in the cumulative exposure to cigarette smoking have had relatively little impact.

The ecological design of this study limits conclusions to inferences about populations rather than individuals but, unlike most ecological studies, the longitudinal design and analysis of within-region changes over time largely overcomes the usual problem of confounding with other large between-region differences, which is typical of many ecological studies.

Since the middle of the 19th century, the analysis of cause of death data has provided important insights into disease burden and the causes of ill health [19]. The data are nevertheless limited by incomplete records and inaccuracies in disease coding. Our definition of COPD mortality is deliberately broad, including all chronic lower respiratory tract causes whether bronchitic or obstructive, with the sole exception of asthma. Narrower definitions are more prone to misclassification and to local variation in the practice of both medicine and cause of death coding. With chronic diseases in particular, the assumption of a single underlying cause of death is often unrealistic, and those certifying the cause of death generally have limited information on which to determine the underlying cause of death and limited options on what cause to code it to. Poor practice often exacerbates these limitations. The protocols of the Global Burden of Disease programme have been developed to minimise these disadvantages, and the algorithms used to improve the reliability of these data, taking account of changes in the quality of cause of death coding and changes in the International Classification of Disease over time [15], have been extensively tested by out-of-sample tests of their predictive power.

The smoking index used provides a summary of smoking experience during a lifetime for each age group and has been extensively used, with the original paper being widely cited [16]. However, the estimates of cigarette smoking are indirect and depend on a high and consistent proportion of lung cancer deaths being attributable to smoking. This is an approximation. The population attributable fraction of lung cancer in men associated with smoking has been estimated as 92% in industrialised regions but as only 67% in developing regions [20]. Although the method would be improved by using local estimates of “nonsmoking” lung cancer, these are not universally available thus we have kept to a single estimate. Any bias introduced by this simplification would have only small effects on within-region comparisons between 1990 and 2010. Using this index is better than attempting to reconstruct smoking exposure over a lifetime from alternative sources such as surveys and tobacco sales, providing a relatively consistent source of information on the cumulative lifetime exposure to tobacco.

The downward trend for age- and sex-specific mortality rates from COPD have been demonstrated for the UK. Marks and Burney [7] reviewed mortality rates for respiratory causes of death from the beginning of the 20th century and showed that mortality rates began to decline around the start of the First World War. The effect of the smoking epidemic was to slow, or halt, this decline rather than lead to an absolute increase in rates. Adair et al. [11] and Erbas et al. [12] showed similar results in Australia, with much lower and declining rates in the first half of the 20th century, followed by increasing rates up to 1970–1990 among men and 1990–2000 among women. Pride and Soriano [8] also noted falling rates for bronchitis between 1931 and 1965 among 45–64 year olds in the UK, at a time when death rates from lung cancer were increasing. The strong association between COPD mortality and poverty is not new. In the UK this association was demonstrated at an ecological level by Barker et al. [5] and was shown to be far stronger than the association between poverty and lung cancer. At an individual level, the social class gradient of the mortality rate for COPD has been steeper than those for lung cancer, or even for tuberculosis [21].

The strong association we observed between changes in COPD mortality and GNI and the relatively weak association with changes in smoking index depend on the rate of change in GNI and smoking index, as well as on the strength of their association with COPD mortality. The positive association of COPD mortality with the smoking index in both 2010 and 1990, and the positive association between changes in the smoking index and changes in COPD mortality, confirm the relevance of cigarette smoking to COPD mortality rates. The relatively small impact of smoking changes is largely due to the small change in global smoking rates, with improvements in some regions being offset by major increases in smoking in the populous regions of South, East and South-East Asia. Nevertheless the far greater impact of GNI requires some further explanation.

The BOLD study has shown a strong ecological association between the prevalence of airway obstruction and the prevalence of smoking, but has shown no such association with poverty. However, it did show a strong association between the prevalence of poverty and the prevalence of a low FVC, and the prevalence of a low FVC increases sharply in countries with a per capita GNI <$15 000 per annum [6]. Data from the USA have also shown that survival is much more strongly linked to FVC than to forced expiratory volume in 1 s/FVC, with the latter showing very little association with mortality once smoking had been adjusted for [22]. Although some discount the low FVC in poor countries as being confounded by race, evidence from the USA shows that the outcome from a low FVC is no different in African–Americans and White Americans after adjusting for age, sex and height [23]. These studies suggest that the high mortality from COPD in low-income countries is associated with a low FVC and not obstruction, as has generally been assumed, and raises an important question over the origins of this low FVC. There is good evidence that low ventilatory function in later life is associated with low birthweight [5, 24–27] and possibly with early infections [4, 5, 28], and these conditions are associated with poverty [4, 29, 30]. However, COPD mortality appears to respond immediately to improvements in GNI, and it is therefore unlikely that the changes noted here are solely related to effects from poverty in early life. The mechanism of this more immediate effect and whether it is mediated through general changes in lifestyle, including environmental changes, or through access to healthcare is unclear. It is unlikely that it is due to specific changes in care for COPD, but general access to support could be part of an explanation.

These findings have important implications for both policy and research. The association between COPD mortality and smoking emphasise the need to reduce smoking rates in all regions, particularly those with increasing smoking rates, but our results show that this will not be enough. The strong association with poverty, particularly in the poorest regions of the world, demonstrate that poverty reduction will be critical in reversing the growing importance of COPD mortality. In the meantime, further research is required to understand what mediates the effects of poverty on COPD mortality.

In conclusion, the increase in the global burden of COPD mortality between 1990 and 2010 has been due to changes in the age distribution, while improved economic conditions, particularly in the poorest countries, have been associated with a fall in age- and sex-specific COPD mortality rates. The change in global smoking rates has been relatively small over the past 20 years, with decreases in some areas offset by increases in others. This accounts for the relatively small influence of changes in smoking habits on global COPD mortality changes.