The short term burden of ambient fine particulate matter on chronic obstructive pulmonary disease in Ningbo, China

Background Numerous studies have found associations between ambient fine particulate matter (PM2.5) and increased mortality risk. However, little evidence is available on associations between PM2.5 and years of life lost (YLL). We aimed to estimate the YLL due to chronic obstructive pulmonary disease (COPD) mortality related to ambient PM2.5 exposure. Methods A time-series study was conducted based on the data on air pollutants, meteorological conditions and 18,472 registered COPD deaths in Ningbo, China, 2011–2015. The effects of PM2.5 on YLL and daily death of COPD were estimated, after controlling long term trend, meteorological index and other confounders. Results The impact of PM2.5 on YLL due to COPD lasted for 5 days (lag 0–4). Per 10 μg/m3 increase in PM2.5 was associated with 0.91 (95%CI: 0.16, 1.66) years increase in YLL. The excess YLL of COPD mortality were 8206 years, and 0.38 day per person in Ningbo from 2011 to 2015. The exposure-response curve of PM2.5 and YLL due to COPD showed a non-linear pattern, with relatively steep at low levels and flattened out at higher exposures.. Furthermore, the effects were significantly higher in the elderly than those in the younger. Conclusions Our findings explored burden of PM2.5 on YLL due to COPD and highlight the importance and urgency of ambient PM2.5 pollution control and protection of the vulnerable populations. Electronic supplementary material The online version of this article (doi:10.1186/s12940-017-0253-1) contains supplementary material, which is available to authorized users.


Background
Ambient fine particulate matter (PM 2.5 ) is one of the most important environmental issues worldwide, and the adverse impact of PM 2.5 has become a great concern. Many epidemiological studies have provided evidences of associations between PM 2.5 exposure and excess mortality [1][2][3]. The majority of these studies had estimated excess mortality using daily death counts as the dependent variable. However, using this indicator might ignore the differences in ages of deaths [4]. From a public health perspective, deaths occurring at different ages are not equally important.
Years of life lost (YLL) is a measure of disease burden that considers the life expectancy at death and, therefore, assigns higher weights to deaths that occur at younger ages [5]. Besides, it is more accurate than mortality to measure premature death and excess mortality [6]. In this sense, exploring the relationships between PM 2.5 and YLL may provide additional information for policy making and health resource allocation, and has important public health significance.
Chronic obstructive pulmonary disease (COPD), characterized by irreversible airflow obstruction, is a common chronic respiratory disease [7]. Due to the slow progression and chronic nature of the disease, COPD represents a massive and growing disease burden and is an important cause of morbidity and mortality [8]. The Global Burden of Disease study showed that COPD was the third leading cause of deaths in the world and ranked the ninth of global YLL in 2010 [9]. There is compelling evidence that short-term exposure to particulate matter is a risk factor for the development and exacerbation of COPD [10][11][12][13]. However, the availability of studies on the associations between ambient PM 2. 5 and YLL of COPD is rare up to now, and the modifications of socioeconomic factors such as age, gender and marital status are unclear.
With the rapid development in industrialization and urbanization, ambient PM 2.5 pollution has aggravated in China in recent years. Ningbo, located in the Yangtze River Delta in the southern China, is the world's fourthlargest port city. The population in Ningbo was estimated to be 7.83 million in 2015, distributed in an area of approximately 9816 km 2 . The high level of economic development in the city also caused the ambient PM 2.5 pollution to be a subject of highly concern.
In this study, we conducted a time-series study to explore the disease burden of COPD from short term ambient PM 2.5 exposure in Ningbo, China, 2011-2015, using the indicator of YLL..

Data on mortality and YLL
Daily mortality data on deaths of COPD between January 1st 2011 and December 31th 2015 were obtained from Ningbo Municipal Center for Disease Control and Prevention, and the data was restricted to registered residents only. COPD deaths were identified according to the International Classification of Diseases 10th version (ICD-10: J40-44). The dataset comprised date of death, gender, age and marital status. Daily death count was defined as the number of deaths occurring on a single day.
We calculated YLL by matching the patient's age to the life table for each death. The sex-and age-group specific life expectancy of Chinese population was obtained from the World Health Organization (WHO), and life expectancies for 2011-2015 were averaged (Additional file 1: Table S1). Daily YLL were calculated by summing the YLL for all deaths on the same day. Both daily YLL and deaths counts were stratified by age (<75 and ≥75 years), gender (male and female) and marital status (the married and the widowed).

Exposure assessment
Concentrations of PM 2.5 , inhalable particulate matter (PM 10 ), sulfur dioxide (SO 2 ) and nitrogen dioxide (NO 2 ) were monitored continuously at 11 monitoring sites which cover urban and suburban areas of Ningbo. The concentrations of the air pollutants were measured according to the Chinese National Ambient Air Quality Standard [14]. The Environment Monitoring Center of Ningbo collects the pollutants concentrations from all the monitoring sites of the city and records the hourly concentrations, from which the daily average levels for all the individual monitoring sites and the whole city levels were derived. Number of missing is less than 1%, and the missing values were substituted with the mean daily value. Daily meteorological data, including temperature and relative humidity, were obtained from the Ningbo Meteorological Bureau.

Statistical analysis
We used generalized additive model (GAM) to explore the impacts of particulate matter on YLL. Because the dependent variable of daily YLL follows a normally distributed (Additional file 1: Figure S1), the family function for GAM was Gaussian.
The model is as follows: In the model, YLL t is the observed daily YLL at day t; α is the intercept; β is the coefficients of YLL associated with per 1 μg/m 3 increase in particulate matter; Xi is the daily mean concentrations of particulate matter; Cj is the confounding factors including time, daily temperature, daily relative humidity; f j is the smooth functions (natural cubic spline); W t (week) is the dummy variables for day of week on day t. We applied the penalized spline function of time and 7 degrees of freedom per year was used to control for secular trend and seasonality. Degrees of freedom for temperature and relative humidity were set to 3 according to the previous studies [15,16]. Considering the lagging effects of temperature, the 14-day moving average of temperature was used. For relative humidity, the average value of the present day was used in the models.
Then, to investigate the lag pattern of PM 2.5 , potentially delayed and cumulative associations were estimated. We first examined the delayed associations using a single day lag (from lag0 to lag7), as previous studies showed the lag effects of particulate matter were strongest within 7 days [2,17]. Then the cumulative associations were estimated using the moving average over the lag periods from lag01 (moving average concentrations of day0 and day1) to lag07 (moving average concentrations of day0 to day7). Final results were presented as changes in daily YLL with per 10 μg/m 3 increase in PM 2.5 in different lag days. Considering the PM 2.5 effect of 5-day moving average concentrations from day 0 to day 4 (lag04) is strongest, so we use lag04 in our main analysis.
We also calculated the excess YLL as follows: YLL of COPD advanced by PMt ¼ Where PMt is the concentrations of PM 2.5 (lag04) in our main studies at day t, β is the value of 1 μg/m 3 particulate matter increase.
Secondly, we used the penalized spline of PM 2.5 to replace the linear variable of PM 2.5 to plot the doseresponse curve of PM 2.5 and YLL (Fig. 3). The relationship appeared to be nonlinear and the curve became flattening in high level PM 2.5 concentration days, so we detected the turning point. The Akaike Information Criterion (AIC) values of GAM models using 1 unit increment in PM 2.5 within the identified range of threshold were computed between 100 μg/m 3 and 150 μg/m 3 (Additional file 1: Table  S2). The concentration of PM 2.5 with the lowest AIC value was selected as the turning point. Similar method had been used in previous studies [18][19][20]. According to the turning point (128 μg/m 3 ), we created two variables (high concentration level, low concentration level), which were higher or lower than the turning point respectively, to explore the piecewise linear association of PM 2.5 .
In addition, the analyses were stratified by age, gender and marital status. Single pollutant model was used to explore the main association of PM 2.5 with daily YLL, while two pollutants models with SO 2 or NO 2 added was used to examine the stability of these associations. In order to have a comparison, we also examined the effects of PM 10 .
Furthermore, we evaluated the impacts of particulate matter on daily death counts of COPD. The independent variable, lag structure and relevant degrees of freedom in the model were similar to those in YLL model, except time series function with the Poisson link under a GAM framework was used because the dependent variable daily death counts following a Poisson distribution. Results were presented as changes in excess risk (ER) of deaths from COPD per 10 μg/m 3 increase in PM 2.5 and PM 10 in different lag days. The equation of ER was as follows: In the equation, β is the value of daily death counts associated with a unit increase in particulate matter.
An autocorrelation function (ACF) was used in the models to assess whether the residuals were independent over time and the reliability of the models. The results showed no obvious autocorrelations were found (Additional file 1: Figure S2 to Figure S5). Sensitivity analyses were performed on the parameters included in the models to test the robustness of our results. We tested using 4 of degrees of freedom of relative humidity, 4 of degrees of freedom of temperature, and used 6 & 8 of degrees of freedom of time per year in the models.
In this study, the mgcv packages in R software (Version 3.1.2) were used to perform the analyses. P < 0.05(2 side) were considered to be significant. The study was approved by the Institutional Review Board of Ningbo Municipal Center for Disease Control and Prevention (No. IRB 201603).

Results
Mean daily concentrations of PM 2.5 and PM 10 were 49.58 μg/m 3 and 78.59 μg/m 3 , respectively during 2011 to 2015 in Ningbo, China (Table 1). Correlations between air pollutants and meteorological conditions were examined by Spearman correlation function and presented in Table 2. Particulate and gaseous pollutant and meteorological conditions were correlated with each other, with the most significant correlations observed between the four air pollutants.
A total of 18,472 registered COPD deaths were identified for the five study years. Daily death counts and YLL of COPD had a mean of 10 and 80 years, respectively. The average daily YLL of the elderly, male and the married ones were higher than that of the young, female and the widowed ones, respectively (Table 1). Both outcomes showed a seasonal trend, with higher values in November through the next April than other months (Fig. 1).
Estimated changes with 95% confidence intervals in YLL due to COPD associated with 10 μg/m 3 increase in PM 2.5 in different lag days were presented in Fig. 2. In general, the associations with PM 2.5 lasted for five days (lag0 to lag4), and gradually reduced after day 4. From day 5 to day 7, the effects had no significance. In addition, the lag patterns were similar for YLL and ER.
We also analyzed the associations of PM 10 in our study. Per 10 μg/m 3 increase in PM 10 was associated with 0.81 (95%CI: 0.30, 1.33) years increase in YLL and 1.07% (95%CI: 0.51%, 1.63%) increases in ER of COPD death at lag04. Furthermore, the excess YLL of COPD mortality in Ningbo from 2011 to 2015 were estimated, which were 8206 years. Take the population of Ningbo into consideration, the YLL in days per person of the population for a 365 day period of exposure was nearly 0.38 day per person in Ningbo from 2011 to 2015.
In addition, the exposure-response curve of PM 2.5 and YLL due to COPD showed a non-linear pattern (Fig. 3). When stratified by the turning point of PM 2.5 , the associations below the turning point concentration were stronger than those above the turning point concentration (Table 3). Specially, the associations of PM 2.5 -YLL due to COPD was 1.44 (95%CI: 0.57, 2.32) below the turning point concentration, while the value was 0.32 (95%CI:-0.59, 1.23) above the turning point concentration. And the mean age at death for PM 2.5 concentrations above and below the median are 74.50 and 73.87, respectively. Similar pattern were showed for the associations between PM 2.5 and ER of COPD deaths.
In two pollutants models, the estimates for associations with PM 2.5 and PM 10 did not change much when the gaseous pollutants SO 2 or NO 2 were added in the models. For instance, when SO 2 was added in the model, 10 μg/m 3 increasing of PM 2.5 was significantly associated with 1.28 (95%CI: 0.46, 2.11) person year increase in YLL.
The results of modifications of age, gender and marital status were summarized in Table 4. When stratified by age, the associations were stronger in the elderly (≥75 years) than the younger (<75 years). In the elderly, a 10 μg/m 3 increase in PM 2.5 corresponded to 0.98(95%CI: 0.42, 1.54) years increase in YLL. For the younger, no significant effects were found.
For the modification of gender, the associations between female and male were not significant. For different marital status, the associations in the widowed showed an increased trend compared with the married ones, though the effect estimate of the married was well within the widowed ones..
Sensitivity analyses results showed that the estimations were stable based on the variations of the parameters in the models (Table S4).

Discussion
The relationships between ambient PM 2.5 and increased mortality risk were well documented [1][2][3]. Although previous studies indicated that PM 2.5 is a significant environment trigger for acute exacerbation of COPD, and thus leading to increasing symptoms, emergency visits, hospital admissions and mortality [10][11][12][13]. A study conducted in China showed that 83% of the population lived in areas where PM 2.5 concentrations exceeded the Chinese Ambient Air Quality Standard of 35 μg/m 3 , premature mortalities attributed to PM 2.5 nationwide were 0.17 million for COPD [21]. However, the number of death ignores the difference in ages of deaths, and giving equal weights to the deaths occurring at a young age and those occurring at an old age. From a public health perspective, deaths occurring at different ages are not equally important, dying at a young age results in more potential years of life lost (YLL) [4,6]. Nevertheless few studies using YLL as dependent variable to explore the health impact of the current exposure levels of PM 2.5 on COPD deaths in China, possibly due to the lack of individual data needed for calculating YLL.
Our study explored the associations of PM 2.5 exposure with daily YLL due to COPD for the first time. Based on the data of 18,472 COPD deaths over a five year period in Ningbo, China, we found that PM 2.5 exposure were significant associated with increased YLL of COPD. Per 10 μg/m 3 increase in PM 2.5 was associated with 0.91(95% CI: 0.16, 1.66) years increase in YLL of COPD deaths. In the study of He 2016 [5], an increase of 10 μg/m 3 increase in PM 2.5 was related to 2.97(95%CI: -2.01, 7.95) years of all-cause mortality, which using the data of Ningbo from 2009 to 2013. The comparison showed the impact of PM 2.5 on YLL from COPD and all-cause mortality is different, although the impact of PM 2.5 on YLL of all-cause mortality was not significant, there was significant influence on COPD YLL. This indicated more investigations should be carried out to explore the impact of PM 2.5 on YLL from diseases which are leading cause of death and disease burden such as COPD.
In addition, the excess YLL of COPD mortality was also estimated, which were 8206 years. Take the population in Ningbo into consideration, the YLL in days per person of the population for a 365 day period of exposure was nearly 0.38 day per person from 2011 to 2015, which gives a different perspective from excess death number.
Considering the indicator of YLL is a measure of disease burden that considers the life expectancy at death [6], and has been extensively used to identify and prioritize causes of premature death around the world [22], the results of our study provides a complementary indicator to that of excess deaths, and gives additional information for policy making and health resource allocation.
The exposure-response curve of PM 2.5 and YLL due to COPD showed a non-linear pattern in Ningbo, 2011-  [20,23,24]. However, considering there were few observations for PM 2.5 , cautions should be taken when using the estimations in the high concentration level.
In this study, we found that the associations of PM 2.5 on both daily YLL and death counts of COPD were stronger in the elderly (≥75 years) than the younger (<75 years). The difference was plausible from a biological perspective because of the decreased immunologic function and reduced lung function as a natural part of aging, and frailty status which decrease physiologic reserve for the adverse effects of air pollution, etc. [25,26]. In addition, elderly individuals frequently have pre-existing chronic diseases, which may make them more vulnerable to the adverse influence of particulate matter exposure [27,28]. The higher impact in the elderly was consistence with a previous study conduct in Ningbo which explored the air pollution-YLL relationships [5], while was different from the results from Beijing which indicated the influence of particulate matter on daily YLL was higher in the younger people [4]. Considering the annual concentrations of PM 2.5 and PM 10 were 105.1 μg/m 3 and 144.6 μg/ m 3 in the Beijing study, the difference may be due to the different levels of particulate matter in Beijing and  Associations in all concentrations and in low and high concentrations levels were both presented at lag04 (moving average concentrations from day 0 to day4). Single and two pollutants models were used. Data were collected from Ningbo, China, 2011China, -2015 Ningbo, which implied the associations of particulate matter and YLL would be different in various pollution levels. For gender modification, although previous study indicated that female may be more susceptible to particulate matter pollution because female is a risk factor for developing airflow limitation and consequently COPD [29], furthermore, female with COPD would have worse prognosis due to more pronounced dyspnea, lower BMI, and more frequent anxiety [30], we did not found significant differences between gender.
It's the first time to explore the potential effect modification of marital status in the particulate matter exposure and YLL relationships, though non significance was found. Further investigations still need to be carried out in future.
In our study, we also evaluated the particulate matter exposure and ER of death counts in our study. Our results showed a 10 μg/m 3 increase in PM 2.5 and PM 10 were associated with maximum increases in ER of death counts of 1.32% (95%CI: 0.52%, 2.12%) and 1.07% (95%CI: 0.51%, 1.63%) at lag04, respectively. The results were supported by previous studies. For instance, meta-analysis of 31 studies showed that a 10 μg/m 3 increase in PM 10 was associated with increase in COPD mortality with an OR of 1.011 (95%CI: 1.008, 1.014) [31]. A study conducted in Guangzhou also found an increase of 10 μg/m 3 increase in PM 10 was associated with 1.58% (95%CI: 0.12%, 3.06%) increase of COPD mortality [19].
This study has several strengths. First, our study explored the associations between PM 2.5 exposure and YLL due to COPD for the first time, and the excess YLL were also calculated. Using YLL as a key indicator to measure the impact of particulate matter on premature deaths will provide more information for policy making and resources allocation. Secondly, exploring the exposure-response curve of PM 2.5 on YLL due to COPD was the first time. We found the exposure-response curve was nonlinear and the associations were steeper in the low PM 2.5 concentrations and became flattening in the high PM 2.5 concentrations. Thirdly, the modification of socioeconomic factors especially the marital status was first investigated, and provided complementary information for identifying vulnerable subgroups.
However, the data used in the study were only from one city, and cautions should be taken when generalizing the results to other geographic areas. In addition, we used ambient particulate matter from fixed site rather than individual exposure, which may result in measurement error, and the individual risk factors such as smoking, drinking and underlying diseases were unknown and not controlled in the study. Furthermore, as we used time series method to estimate the effects which was considered inappropriate to estimate longterm effects [32], it should be careful when analysing the long-term associations.

Conclusions
In conclusion, our study provided a new insight into the disease burden due to COPD of ambient PM 2.5 exposure. We found that PM 2.5 had significant impacts on YLL due to COPD in Ningbo, China. Furthermore, the exposure-response curve was nonlinear and the associations were steeper at low levels and became flattened out at higher exposures. The elderly were susceptible population. Our findings highlight the importance and urgency of ambient PM 2.5 control and protection of the vulnerable populations.

Additional file
Additional file 1: Table S1. Life expectancy for Chinese population (averaged value for 2011-2015). Table S2. Threshold selection based on the Akaike Information Criterion value. Table S3. Sensitivity analyses of associations of PM 2.5 and PM 10 with YLL and ER of COPD deaths. Figure S1. Distribution of daily deaths and daily YLL of COPD in Ningbo, 2011-2015, China. Figure S2. Auto-correlation function for residuals of PM 2.5 single pollutant models for YLL of COPD at lag04. Figure S3. Auto-correlation function for residuals of PM 2.5 single pollutant models for deaths of COPD at lag04. Figure S4. Auto-correlation function for residuals of PM 10 single pollutant models for YLL of COPD at lag04. Figure S5