Spatiotemporal Variations in Chronic Obstructive Pulmonary Disease Mortality in China: Multilevel Evidence from 2006 to 2012

Abstract

Mortality of Chronic Obstructive Pulmonary Disease (COPD) is on the decline in China. It is not known if this trend occurs across all areas or whether spatiotemporal variations manifest. We used data from the nationally representative China Mortality Surveillance System to calculate annual COPD mortality counts (2006–2012) stratified by 5-year age groups (aged > 20), gender and time for 161 counties and districts (Disease Surveillance Points, or DSP). These counts were linked to annually adjusted denominator populations. Multilevel negative binomial regression with random intercepts and slopes were used to investigate spatiotemporal variation in COPD mortality adjusting for age, gender and area-level risk factors. COPD mortality rate decreased markedly from 105.1 to 73.7 per 100,000 during 2006 to 2012 and varied over two-fold between DSPs across China. Mortality rates were higher in the west compared with the east (Rate ratio (RR) 2.15, 95% confidence intervals (CI) 1.73, 2.68) and in rural compared with the urban (RR 1.87, 95% CI 1.55, 2.25). Adjustment for age, gender, urban/rural, region, smoking prevalence, indoor air pollution, mean body mass index and socioeconomic circumstances accounted for 67% of the geographical variation. Urban/rural differences in COPD mortality narrowed over time, but the magnitude of the east-west inequality persisted without change. Immediate action via large-scale interventions to enhance the prevention and management of COPD are needed specifically within China's western region in order to tackle this crucial health inequality and leading preventable cause of death.

Keywords:

• fixed effect
• geographic variation
• risk factors
• surveillance

Introduction

Chronic Obstructive Pulmonary Disease (COPD) was the third-leading cause of death worldwide in 2010 (1). In China, where COPD also remains the third-leading cause of death, a recent study (2) has shown a declining trend through time; from 189 deaths per 100,000 in 1990 to approximately 77 per 100,000 two decades later. This overall declining mortality trend may be associated with better access to health services and the introduction of new therapies that have been shown to increase survival among COPD patients in moderate and severe stages of disease progression (1).

Although the prevalence of smoking, as the most important risk factor for COPD, has remained high and ambient air pollution has been getting worse in China, the changes in the distribution of other risk factors, such as indoor air pollution (3, 4), occupational dust and gas exposures(5), socioeconomic circumstances and body mass index (BMI) (6) may also play a role in the change of COPD mortality. Exposure to indoor air pollution by household solid fuel combustion, for example, has decreased in the last 10 years (7) and rising levels of obesity in China may also contribute to the decline, with some research suggesting that higher BMI could be potentially protective against COPD (6).

It is important to appreciate, however, that these changes in risk factors that are often associated with urbanization have not been evenly distributed across China. It is therefore naïve to assume that the aforementioned decline in COPD mortality is being experienced in all areas of China simultaneously, yet the potentially spatiotemporal variation in COPD is an area of research that has received little attention. Some previous work has shown counties and districts in the west of China have higher rates of COPD than their counterparts in the east (7). There are also some small-scale studies set in specific cities indicating that mortality attributable to COPD has decreased in the past 10 to 15 years (9). It is not yet known, however, if the decline in COPD mortality is occurring equally across all regions, or in some areas more so than others. This is important from a policy perspective, as decision-makers are required to allocate scarce health resources efficiently and equitably and to develop tailored intervention strategies across a vast, socioeconomically and topographically diverse country (8). Accordingly, the aims of this study were to investigate recent spatiotemporal variation in COPD mortality and identify potential factors that may explain any potential spatiotemporal variation in COPD mortality.

Methods

Mortality data

COPD mortality counts (International Classification of Diseases 10 (ICD-10): J40-J44) were obtained from the Disease Surveillance Point (DSP) system for 161 counties and districts, cross-classified by 5-year age group (> 20-years), gender and year (2006 to 2012 inclusive). Coverage of the DSP system extends to 73 million people located across all 31 provinces, municipalities and autonomous regions in China. A DSP is a county in rural areas or a district in urban areas, with population ranging from 22 000 to 1.5 million (median 450,000). Counties and districts are the administrative units in China and the areas of each DSP were consistent during the study period. Previous work has demonstrated the representativeness of the DSP system for China's national population following multiple strategies for addressing variation in data quality (9, 10).

Population data

County- and district-level (i.e., DSP) population counts cross-classified by 5-year age group and gender were obtained from the Chinese census in 2000 and 2010. These counts were used as a reference population for calculating age-standardised COPD mortality rates and in statistical models. To account for population change across the study period, we first calculated the annual rate of change in population for each of the 161 DSPs using 2000 and 2010 census data. The total population for each DSP in the years 2006–2012 was then estimated assuming an exponential growth across the time period, in line with the methodology used in the estimation of mortality rates in China in the most recent Global Burden of Disease Study (13).

Area-level risk factor data

Of the 161 DSPs, administrative characteristics were used to distinguish between 97 classified as ‘rural’ (counties) and 64 as “urban” (districts). All DSPs were classified into “east,” “central” and “west,” based on the geographic categorization defined by National Bureau of Statistics (Figure 1). Important risk factors for COPD derived from the literature included tobacco smoking (11), indoor air pollution (3), BMI (6) and socioeconomic circumstances (12). Person-level data from the China Chronic Disease Risk Factor Surveillance (CDRFS) in 2010 (13) was used to estimate the prevalence of risk factors at the DSP level. Detailed design and methods were described elsewhere (13, 14). Briefly, CDRFS is a cross-sectional population-based survey aiming to investigate the prevalence of major chronic diseases and related risk factors in China. The survey included 98,658 adult participants from all 161 DSP sites using a complex, multistage, probability sampling design. Data collection was conducted in examination centers at local health stations or community clinics in the participants’ residential area by trained staff according to a standard protocol. A questionnaire including information on demographic characteristics, medical history, and lifestyle factors was administered by trained interviewers.

Figure 1. Geographic distribution of 161 Disease Surveillance Points (DSP) in China.

The variables used in the current analysis included the prevalence of current or previous cigarette smoking, the level of indoor air pollution and the mean BMI of the adult population. Exposure to indoor air pollution was defined by participants in the CDRFS reporting use of coal, charcoal, crop residues, wood or animal dung as the main source of cooking fuel. BMI was derived from objective measures of height and weight. Each of these variables was classified as tertiles in order to examine potentially curvilinear associations with COPD mortality. The 2010 census was used to obtain a measure of socioeconomic circumstances, using the mean years of education in each DSP. The ethics committee of the Chinese Center for Disease Control and Prevention approved the 2010 CDRFS.

Statistical analysis

To investigate whether the decline in COPD mortality has been consistent across China or area-specific, a multilevel negative binomial regression model was implemented in MLwIN v2.30 (15). This approach of modelling discrete mortality counts offset by denominator populations was preferred over the calculation of standardized mortality ratios as the latter tend to be influenced by small cell sizes. In the case of this study, mortality counts were classified into 29,288 strata by DSP, year, gender and 5-year age groups. Mortality counts for these strata were fitted at level 1 of the multilevel model.

To account for repeated measures of the same DSPs in each of the 7 years of observation, the DSP identifier was fitted at level 2 as a random intercept. Time was fitted as a linear fixed effect, with possible curvilinear trajectories investigated using a square term. Potential spatiotemporal variation was investigated by allowing the time variable to vary for each DSP via the implementation of a random slope (of random coefficient). The collective interpretation of the intercept variance, slope variance and covariance and associated standard errors indicated the extent to which COPD mortality varied spatiotemporally. If the covariance, random intercept and slope variances reached statistical significance, a negative covariance would denote a tendency for DSP-level COPD mortality trajectories to converge through time, whereas a positive covariance would indicate divergence. Furthermore, as gender-differences by age have been shown by previous studies among populations aged 40 years and above (18), a gender*age group interaction term was fitted.

Geographic variation between DSPs at level 2 of the multilevel model was re-expressed in the form of a median rate ratio (MRR)(16) as an aid to interpretation. The MRR is defined as the median value of the rate ratios between two randomly selected areas. Where there is no significant geographical variation in COPD mortality, an MRR would equal 1, whereas an MRR above 1 would denote the presence of geographical variation.

To determine whether geographical inequalities in COPD mortality have changed through time, age-standardized COPD mortality rates were calculated for each year across the study period (2006–2012) for DSPs characterized as urban or rural, and by region (“east,” “central,” or “west”) in order to gauge and understanding of general trends. The urban/rural and region variables were then added to the multilevel model to assess differences across the time period.

To identify potential factors that may explain any potential spatiotemporal variation in COPD mortality, subsequent models also included area-level risk factors (tobacco smoking, indoor air pollution, BMI and socioeconomic circumstances, all averaged over DSP level). The contributions of these risk factor variables were assessed sequentially through observation of any potential attenuation in the fixed effects part of the model and the proportional change in variance (PCV) of the random intercept variance between DSPs (level 2). All fixed effect parameters were exponentiated to RRs and 95% confidence intervals (95%CI).

Results

Table 1 presents the number of COPD deaths and distributions of risk factors. Total number of deaths varied from 54,660 to 45,608 from 2006 to 2012. The number of deaths in rural almost tripled that in^{Table 3} urban areas and there were more deaths in the west than the central and east. The overall prevalence of smoking and indoor air pollution was 35.1% and 44.0% respectively, with rural higher than urban and west higher than central and east. The average BMI was 23.8 and the mean education years was 9.1 with urban higher than rural.

Table 1. Descriptive analysis of death of COPD and distributions of risk factors for COPD by urban/rural and regions.

	Total	Urban	Rural	East	Central	West
Number of deaths
2006	45,608	12,162	33,446	16,180	13,590	15,838
2007	49,373	13,459	35,914	15,286	14,419	19,668
2008	54,660	14,912	39,748	17,169	16,673	20,818
2009	51,094	13,080	38,014	15,159	16,188	19,747
2010	50,390	12,973	37,417	14,565	15,820	20,005
2011	47,700	11,751	35,949	13,400	14,883	19,417
2012	47,460	11,510	35,950	13,400	14,772	19,288
COPD risk factors
Smoking rate (%)	35.1	34.3	35.5	34.9	35.1	36.0
Indoor air pollution (%)	44.0	20.6	58.1	27.9	46.4	57.2
BMI (kg/m^2)	23.8	24.1	23.6	24.0	23.9	23.4
Means years of education	9.1	10.5	8.3	9.5	9.5	8.4

Table 2 shows that age and sex-standardized COPD mortality rates per 100,000 declined across urban areas from approximately 71.9 to 44.7, and in rural areas from 126.6 to 92.7. Similar observations were made from regional age-sex-standardized rates. Comparing the “east” with “west” regions, declines in COPD mortality were observed in both, but the gap remained substantial.

Table 2. Age-sex-standardized mortality rate per 100,000 of COPD in China, 2006–2012

	Age-sex-standardized COPD mortality rate
Year	Total	Urban	Rural	Eastern	Central	Western
2006	105.1	71.9	126.6	80.8	98.5	163.8
2007	99.9	67.3	122.0	72.0	87.5	165.7
2008	98.1	66.1	120.0	70.6	92.8	153.2
2009	89.6	57.6	110.7	62.3	86.5	139.5
2010	81.8	50.1	104.7	55.8	76.0	133.8
2011	77.0	47.5	96.5	53.5	68.0	126.8
2012	73.7	44.7	92.7	51.1	65.0	122.1

As shown in Table 3, a linear decrease in COPD mortality was observed throughout the time period (RR 0.93, 95%CI 0.92, 0.94), (Model 1). A gender*age group interaction term demonstrated such that younger women had lower COPD mortality than men (RR 0.62, 95%CI 0.60, 0.63) and this gap decreased in relative terms as mortality increased by age group. The DSP intercept variance was large, translating to an MRR of 2.40. A small, but statistically significant variance (0.006) in DSP random slopes over time indicated some spatiotemporal variation. Positive covariance (0.034) between intercepts and slopes indicated that while COPD mortality was decreasing through time, the rate of decline in COPD mortality was greater for those DSPs that had lower COPD mortality in 2006. These parameters also indicated that the rate of decline in COPD mortality was least among those DSPs that had higher COPD mortality in 2006; hence the “fanning out” pattern of spatiotemporal variation denoted by the positive covariance.

Table 3. Spatiotemporal variation in COPD mortality in China between 2006 and 2012; multilevel random slope negative binomial regression, adjusted for time, age, gender, urban/rural and region and area-level risk factors.

	Model 1	Model 2	Model 3
Time	0.93 (0.92, 0.94)^*	0.92 (0.91, 0.93)^*	0.92 (0.91, 0.93)^*
Gender (ref: male)	1	1	1
Female	0.62 (0.60, 0.63)^*	0.60 (0.59, 0.62)^*	0.60 (0.58, 0.62)^*
Age (5-year groups)	2.15 (2.14, 2.16)^*	2.18 (2.17, 2.19)^*	2.18 (2.18, 2.19)^*
Female x age	1.02 (1.01, 1.02)^*	1.01 (1.01, 1.02)^*	1.01 (1.01, 1.02)^*
Region (ref: East)		1	1
Central		1.18 (0.95, 1.47)	1.09 (0.89, 1.33)
West		2.15 (1.73, 2.68)^*	1.74 (1.41, 2.14)^*
Urban			1
Rural		1.87 (1.55, 2.25)^*	1.06 (0.83, 1.35)
% smokers (ref: Low)
Moderate			0.92 (0.76, 1.12)
High			1.06 (0.87, 1.29)
% indoor air pollution (ref: Low)
Moderate			1.44 (1.12, 1.86)^*
High			1.57 (1.19, 2.06)^*
Mean body mass index (ref: Low)
Moderate			0.97 (0.79, 1.19)
High			0.79 (0.64, 0.98)^*
Mean years of education			0.84 (0.76, 0.92)^*
Random effects
DSP intercept variance (Standard error)	0.844 (0.094)^*	0.366 (0.041)^*	0.280 (0.032)^*
MRR	2.40	1.78	1.66
PCV		57%	67%
DSP slope variance by year (Standard error)	0.006 (0.001)^*	0.006 (0.001)^*	0.007 (0.001)^*
Covariance between DSP intercepts and slopes	0.034 (0.007)^*	0.016 (0.004)^*	0.014 (0.004)^*

^*p < 0.05, MRR = Median Rate Ratio; PCV = Proportional Change in Variance, compared to model 1; DSPs = Disease Surveillance Points. Model 1: Multilevel random slope negative binomial regression adjusted for time, gender and age; Model 2: Multilevel random slope negative binomial regression adjusted for time, gender, age, region and urban/rural; Model 3: Multilevel random slope negative binomial regression adjusted for time, gender, age, region, urban/rural and area-level risk factors.

Both urban/rural and region variables were added to the multilevel model. The region variable was added first (Model 2), with COPD mortality 2.15 times higher in the west compared with the east (95%CI 1.73, 2.68). This was followed by adjustment for urban/rural, with COPD mortality 1.87 times higher in rural compared with urban areas (95%CI 1.55, 2.25). Overall adjustment for region and urban/rural accounted for 57% of the DSP-level variation in COPD mortality that was estimated in model 1. The contributions of area-level risk factor variables to spatiotemporal variations in COPD mortality are reported in Model 3.

The percentage of current or former cigarette smokers within a DSP was not significantly associated with COPD mortality. Indoor air pollution exposure yielded strong positive association (high versus low RR 1.57, 95%CI 1.19, 2.06) and DSPs with higher average adult BMI tended to have lower COPD mortality rates (high versus low RR 0.79, 95%CI 0.64, 0.98). A 1-year increase in the mean years of education within a DSP was negatively associated with COPD mortality (RR 0.84, 95%CI 0.76, 0.92). Adjusting for these risk factors attenuated the east-west and urban-rural differentials in COPD mortality and explained a further 10% of the DSP variance. Overall, 67% of the variation in COPD mortality between DSPs was explained by the fitted variables.

Discussion

Previous epidemiological studies of COPD mortality in China have focused upon national-level trends in mortality through time (2) with less attention paid to spatiotemporal variation. This study examined whether the decline in COPD mortality has been consistent across China or whether declines have been differential by geographic area. Findings from the present study suggest that there is a greater than two-fold variation in COPD mortality by geographic area, and 67% of the geographic variation in COPD mortality was accounted for by area-level risk factors. Large differences were evident between eastern, central and western regions of China. The magnitude of these geographical inequalities highlight the importance of geographically manifesting factors in understanding COPD burden in China (17). Findings imply that geographically specific interventions are necessary to address these area differences in the social determinants of COPD and related health outcomes in China, focusing in particular within rural areas and the western region in general.

Geographic variations in COPD mortality have been observed in other countries although have tended to report increasing trends of COPD mortality before the year 2000 (18, 19). The recent decline in COPD in China reflects, to some extent, similar trends in the US, Japan, Australia and some European countries (20–24). Although tobacco smoking, as the most important risk factor for COPD, is highly prevalent in China and remains at high levels in the past decade, increased awareness of COPD, improved management of the disease and better healthcare facilities in rural areas as a result of recent healthcare reforms are likely to have contributed to this declining trend. Furthermore, the proportion of COPD deaths attributable to cigarette smoking was 12.1% in men and 5.6% in women in China, suggesting that non-smoking factors play an important role in COPD mortality (11). Some studies have shown that COPD mortality declined faster in men than in women (23). In this study a diverging trend was observed over time between younger men and women, which has not been previously documented. This may be attributable to changes in risk factor distributions among younger persons living in China (25), such as exposure to tobacco smoking and indoor air pollution (3).

The coding of death certificates is an essential issue in interpreting the results of the present study. Differences in COPD mortality in different areas and changes over time may possibly be due to differences in coding practice. However, DSP-based surveillance is a well-established system since the 1980s in China and the attribution of cause and coding were made by trained professional coders in hospitals and county-level staffs from Center for Disease Control and Prevention(9). Additionally, regular centralized training was provided to the coders and strategies are in place to monitor coding accuracy with verifications by city and provincial level coding staff. It is less likely that the coding practice would vary significantly among DSPs.

The “fanning out” pattern of the declining trend of COPD mortality indicated that those DSPs with higher COPD mortality rate, mostly in rural areas and western region, need particular attention from a policy perspective. The increasing public health investments including both surveillance and intervention from the Chinese government should prioritize on those areas with constantly higher mortality rate of chronic diseases such as COPD. The strong association between indoor air pollution and COPD mortality lends further support for interventions aimed at reduced biomass exposures, especially for women and children in western region. Although the government has made substantial efforts to reduce indoor air pollution by improving ventilation and promoting clean fuels, more needs to be done to further reduce the usage of biomass/coal as heating or cooking fuel, which is still commonly seen in many western and rural areas in China (26). Years of education is recognized as an important indicator of socioeconomic circumstances in China (7), and we found a lower COPD mortality rate in areas with higher socioeconomic status, which is consistent with previous studies (27, 28). The observed association between BMI and COPD mortality was also in keeping with results from individual-level prospective studies (29).

The key strength of this study is the nationally representative data on COPD counts across 161 DSPs. To our knowledge, this is the first study investigating the spatiotemporal variation of COPD mortality in a nationally representative dataset. The use of multilevel modelling was another important strength, as it not only accounted for the correlated nature of the time-series analysis through adjusting for repeated measurement of DSPs between 2006 and 2012, but it also afforded the disentangling of contextual (e.g. urban/rural) associations from those attributable to compositional factors (30).

The study, however, is not without limitations. Using mortality counts based on the underlying cause of death might underestimate the true mortality from COPD, because people with COPD often experience multi-morbidity (31). Evidence suggests that among people with COPD as a contributing cause of death, only half have COPD reported as the underlying cause of death (32). Due to the unsatisfactory completeness of data on contributory cause of death in the current DSP system, we could not assess the proportions of COPD as underlying cause of death among those with any mention of COPD as one cause. We believe that similar studies could be done in China with improvement of completeness and accuracy of contributory cause of death in DSP mortality surveillance in the near future.

Second, this is an ecological study of a mortality time series and the risk factor adjustment was limited to population-level point estimates derived from the CDRFS and the 2010 census. Change in COPD mortality coinciding with change in these population-level exposures therefore could not be assessed. This may partly explain why no association between smoking prevalence and COPD mortality was found, along with the temporal incongruity the cumulative effects of the tobacco smoking as the exposure and COPD, which could not be measured in this study (as tobacco smoking information was restricted to a single survey year). Furthermore, risk factors measured by averages over DSP level may not relate to those actually dying of COPD and some of the results may be explained by residual confounding.

Lastly, some important risk factors for COPD including occupational exposures and outdoor air pollution were not available for inclusion in our model. No robust outdoor air pollution data (e.g., particulate matter 2.5) at the DSP scale was available to analyze within this time-series. Previous work has highlighted inconsistent findings between outdoor air pollution and COPD mortality (33). Though it is possible that some of this variation will have accounted for by adjusting for socioeconomic circumstances, future work should incorporate outdoor air pollution data if it were to become available.

Conclusions

In conclusion, this study has demonstrated spatiotemporal variation in COPD mortality between 2006 and 2012. COPD mortality in the west remained far higher than the eastern and central regions across the study period. The narrowing of the urban/rural gap, but the persistence of this east-west regional inequality, despite the general decline in COPD mortality through time, are the key messages for decision makers to note. Immediate action via large-scale interventions to enhance the prevention and management of COPD are needed specifically within China's western region in order to tackle this crucial health inequality and leading preventable cause of death.

Acknowledgments

Peng Yin and Xiaoqi Feng contributed equally to this study.

Declaration of interest

The authors declare no conflict of interest. The authors alone are responsible for the content and writing of the paper.

Funding

This study was funded by China National Science & Technology Pillar Program 2013(2013BAI04B02) from Ministry of Science and Technology, Science & Technology Outstanding Program for Scholars Coming Back from Overseas from Ministry of Human Resource and Social Security of China, and The Australia-China Science and Research Fund (ACSRF17120). TAB's time on the study was supported by the National Heart Foundation of Australia.