Asthma is a major public health issue in the USA, affecting over 23 million persons [
Ambient air pollution has been linked to the development and exacerbation of asthma and its related diseases in Europe, North America, Korea, Japan, and Taiwan ([
Many works have been done in this field and achieved specific findings. Schouten et al. [
However, previous studies have serval limitations: (1) Although elder asthma patients were found to be more fragile to air pollution [
This study aims to investigate the Markov-based acute effects of air pollution on elder asthma hospitalization, which is the key chain of forecasting admission amount, by measuring the acute effects of air pollution on elder asthma admission first. We take sex and seasonality factors into consideration to achieve a systematic research. The Markov model is particularly useful in analyzing risk factors in cohort studies and has been applied successfully to the study of lung cancer, HIV infection [
Our study area covers a region of Sichuan province, China. We obtained inpatient records of asthma hospitalization for adult residents between January 1, 2014, and December 31, 2014 (365 days). Approximately 12 million residents were covered during this period. The main diagnoses of hospital admission were coded according to the International Classification of Diseases, Revision 10 (ICD-10): Asthma (J45.001, J45.005, J45.901, J45.902, and J45.903). The data was also classified by season and sex. Warm season is defined as a period from April to September, and cold season is defined as the rest of time period in a year. Elder person is defined as person older than 65 years.
Daily (24 h) air pollution concentration data including particulate matter less than 2.5 mm in aerodynamic diameter (PM2.5), particulate matter less than 10 mm in aerodynamic diameter (PM10), sulfur dioxide (SO2), and nitrogen dioxide (NO2), from January 1, 2014, to December 31, 2014, were obtained from the website of the Environmental Monitoring Center (EMC) database. To allow adjustment for the effect of weather on hospital admission, meteorological data (daily min temperature) were obtained from the website of Meteorological Bureau.
Daily asthma hospitalization and air pollution levels were linked by date and therefore could be analyzed with a time-series design. Because daily hospital admission for asthma approximately follows a Poisson distribution [
For the purpose of healthcare resource allocation and schedule, it is important to measure the association between air pollution and elder asthma admission amounts and make corresponding forecast. Markov chain is a useful way to describe the asthma admission amount evolution process; it can not only reliably reflect the transition situation but also build a bridge between healthcare management and healthcare resource scheduling optimization. For instance, assuming that Markov transition probability between each admission amount state and future distributions of air pollution condition is given, then, future distributions of the admission amount state are also known. Hence, healthcare resource scheduling according to future distributions of the admission amount state will achieve a better performance. Markov models allow the modelling of patient follow-up as a succession of transitions between states over time. They are quantified as the rate of transition and expressed in number of transitions. The model was considered to be homogeneous; that is, the transition forces are independent of time. To construct the elder asthma admission Markov chain, we use the Lorenz curve and OR analysis to determine admission amount states and severity of air pollution. In economics, the Lorenz curve is a graphical representation of the distribution of income or of wealth, while in this situation, the Lorenz curve is a graphical representation of the distribution of daily admission amount. In statistics, the odds ratio (OR) is one of three main ways to quantify how strongly the presence or absence of property A is associated with the presence or absence of property B in a given population; while in our research, OR is employed to quantify the association between environmental properties and admission amount states. Then, a multistate model (MSM) was used to calculate the Markov transition probabilities. All models were fitted using R software (version 3.3.2, R Foundation for Statistical Computing,
Table
Summary statistics of daily asthma hospital admission, air pollutant concentrations, and weather conditions from January 1, 2014, to December 31, 2014.
|
Mean | SD | Mina | Q1a | Q2a | Q3a | Maxa | IQRa | |
---|---|---|---|---|---|---|---|---|---|
All elder | 1567 | 4.29 | 2.30 | 0 | 4 | 3 | 6 | 13 | 3 |
Sex | |||||||||
Male | 605 | 1.66 | 1.33 | 0 | 1 | 1 | 2 | 7 | 1 |
Female | 962 | 2.64 | 1.76 | 0 | 2 | 13 | 10 | 2 | |
Seasonb | |||||||||
Warm | 725 | 3.96 | 1.97 | 0 | 4 | 3 | 5 | 11 | 2 |
Cold | 842 | 4.63 | 2.55 | 0 | 4 | 36 | 13 | 3 | |
Air pollution concentrations (24 h average) | |||||||||
PM2.5 ( |
— | 72 | 52 | 10 | 38 | 55 | 88 | 396 | 50 |
PM10 ( |
— | 116 | 72 | 20 | 68 | 96 | 147 | 562 | 79 |
SO2 ( |
— | 17 | 10 | 3 | 11 | 15 | 21 | 61 | 10 |
NO2 ( |
— | 52 | 16 | 20 | 41 | 50 | 60 | 109 | 19 |
Meteorological measures | |||||||||
Min temperature | — | 13 | 7 | −2 | 7 | 15 | 20 | 24 | 13 |
PM2.5: particulate matter not greater than 2.5 mm in aerodynamic diameter; PM10: particulate matter not greater than 10 mm in aerodynamic diameter; SO2: sulfur dioxide; NO2: nitrogen dioxide; amin: minimum; Q1: 25th percentile; Q2: 50th percentile; Q3: 75th percentile; max: maximum; IQR: interquartile range (Q3–Q1). bCold season: from October to March; warm season: from April to September.
Generally, PM2.5, PM10, SO2, and NO2 had moderately high correlation coefficients with each other (Table
Pearson correlation coefficients between daily air pollutant concentrations from January 1, 2014, to December 31, 2014.
PM2.5 | PM10 | SO2 | NO2 | |
---|---|---|---|---|
PM2.5 | 1 | |||
PM10 | 0.86 | 1 | ||
SO2 | 0.51 | 0.53 | 1 | |
NO2 | 0.55 | 0.56 | 0.47 | 1 |
Abbreviations are the same as in Table
Percent increase (mean and 95% confidence interval) in daily asthma hospital admission associated with a 10
PM2.5 | PM10 | SO2 | NO2 | |
---|---|---|---|---|
Lag0 | 0.54 (−0.44, 1.52) | 0.24 (−0.47, 0.95) | 6.59 (1.11, 12.36) |
2.4 (−0.87, 5.77) |
Lag0-1 | 0.79 (−0.23, 1.82) | 0.49 (−0.27, 1.25) | 7.27 (1.1, 13.82) |
3.2 (−0.45, 6.98) |
Lag0–2 | 0.82 (−0.24, 1.89)§ | 0.5 (−0.29, 1.3)§ | 6.94 (0.4, 13.91) |
3.26 (−0.66, 7.33)§ |
Lag0–3 | 0.74 (−0.35, 1.84) | 0.42 (−0.4, 1.24) | 5.83 (−0.95, 13.09) | 2.69 (−1.44, 6.99) |
Lag0–4 | 0.65 (−0.47, 1.78) | 0.34 (−0.5, 1.19) | 4.41 (−2.55, 11.87) | 2.01 (−2.28, 6.49) |
Lag0–5 | 0.54 (−0.61, 1.7) | 0.23 (−0.63, 1.1) | 3.12 (−3.97, 10.74) | 1.76 (−2.69, 6.41) |
Abbreviations are the same as in Table
To define different admission amount states, we used the Lorenz curve to analyze the total admission amount. Figure
The Lorenz curve of elder asthma admission.
In the framework of Markov chain, the transition probabilities between different states vary, when decision or situation changes. For instance, when air pollution converts from mild to severe, the transition probability from low-admission amount to high-admission amount increases evidently. In our research, we constructed two transition probability matrices between different states, for mild air pollution and severe air pollution, respectively. Results from Table
Concentration threshold of Chinese Ministry of Environmental Protection for each pollutant.
Pollutant | Concentration threshold | |
---|---|---|
Primary standard | Second standard | |
PM2.5 | 35 |
75 |
PM10 | 50 |
150 |
SO2 | 50 |
150 |
NO2 | 80 |
— |
Abbreviations are the same as in Table
Odds ratio analysis is a vastly used method to select a related index. Table
Odds ratio (mean and 95% confidence interval) between the air pollution index and high elder asthma admission.
Odds ratio | Counts of days exceeding the national standard | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Primary standard | Second standard | ||||||||||||
1 | 2 | 3 | 4 | 5 | 6 | 1 | 2 | 3 | 4 | 5 | 6 | ||
PM2 |
Lag0 | 1.61 (1.31, 1.99) | — | — | — | — | — | 2.04 (1.82, 2.29) | — | — | — | — | — |
Lag0-1 | 1.89 (1.26, 2.84) | 1.9 (1.61, 2.23) | — | — | — | — | 1.84 (1.65, 2.06) | 1.9 (1.67, 2.16) | — | — | — | — | |
Lag0–2 | 0.89 (0.56, 1.4) | 1.75 (1.37, 2.23) | 2.55 (2.18, 2.97) | — | — | — | 1.54 (1.37, 1.71) | 1.9 (1.69, 2.13) | 1.87 (1.61, 2.17) | — | — | — | |
Lag0–3 | 0.41 (0.23, 0.71) | 1.13 (0.85, 1.51) | 1.89 (1.53, 2.32) | 2.7 (2.35, 3.1) | — | — | 1.46 (1.31, 1.63) | 1.69 (1.51, 1.89) | 1.79 (1.57, 2.03) | 2.43 (2.06, 2.85) | — | — | |
Lag0–4 | 0.52 (0.15, 1.84) | 0.89 (0.64, 1.25) | 1.28 (1.02, 1.61) | 2.11 (1.74, 2.55) | 2.83 (2.49, 3.22) | — | 1.27 (1.14, 1.42) | 1.52 (1.36, 1.7) | 1.7 (1.51, 1.92) | 2.05 (1.79, 2.35) | 2.88 (2.4, 3.47) | — | |
Lag0–5 | 0.26 (0.05, 1.38) | 0.72 (0.45, 1.16) | 1.25 (0.94, 1.65) | 1.66 (1.32, 2.07) | 2 (1.69, 2.37) | 3.03 (2.69, 3.43) | 1.2 (1.07, 1.35) | 1.33 (1.19, 1.48) | 1.58 (1.41, 1.77) | 2.04 (1.8, 2.31) | 2.53 (2.17, 2.95) | 3.32§(2.71, 4.06) | |
Lag0 | 3.13 (1.67, 5.86) | — | — | — | — | — | 1.69 (1.48, 1.93) | — | — | — | — | — | |
|
|||||||||||||
PM10 | Lag0-1 | NA | 1.82 (1.39, 2.38) | — | — | — | — | 1.86 (1.65, 2.09) | 1.37 (1.16, 1.63) | — | — | — | — |
Lag0–2 | NA | 3.34§ (1.33, 8.38) | 1.99 (1.59, 2.48) | — | — | — | 1.63 (1.45, 1.82) | 1.69 (1.47, 1.94) | 1.81 (1.48, 2.21) | — | — | — | |
Lag0–3 | NA | 2.68 (0.43, 16.84) | 1.89 (1.26, 2.84) | 2.65 (2.14, 3.3) | — | — | 1.83 (1.64, 2.05) | 1.54 (1.35, 1.74) | 2.01 (1.71, 2.37) | 2.07 (1.63, 2.62) | — | — | |
Lag0–4 | NA | NA | 1.33 (0.67, 2.64) | 1.77 (1.31, 2.39) | 2.33 (1.95, 2.79) | — | 1.69 (1.51, 1.89) | 1.51 (1.34, 1.7) | 2.07 (1.8, 2.39) | 2.14 (1.77, 2.58) | 3.19 (2.43, 4.18) | — | |
Lag0–5 | NA | NA | 1.59 (0.23, 11.13) | 1.63 (0.98, 2.71) | 1.86 (1.42, 2.44) | 2.1 (1.8, 2.46) | 1.61 (1.44, 1.8) | 1.49 (1.33, 1.67) | 2.1 (1.84, 2.39) | 2.16 (1.84, 2.52) | 2.97 (2.39, 3.7) | 3.2 (2.32, 4.42) | |
Lag0 | 3.9 (1.27, 12.01) | — | — | — | — | — | NA | — | — | — | — | — | |
|
|||||||||||||
SO2 | Lag0-1 | 4.02§ (2.26, 7.17) | NA | — | — | — | — | NA | NA | — | — | — | — |
Lag0–2 | 3.27 (2.19, 4.88) | NA | NA | — | — | — | NA | NA | NA | — | — | — | |
Lag0–3 | 2.96 (2.17, 4.05) | NA | NA | NA | — | — | NA | NA | NA | NA | — | — | |
Lag0–4 | 2.81 (2.17, 3.65) | NA | NA | NA | NA | — | NA | NA | NA | NA | NA | — | |
Lag0–5 | 3.13 (2.51, 3.9) | NA | NA | NA | NA | NA | NA | NA | NA | NA | NA | NA | |
Lag0 | 3.2 (2.32, 4.42) | — | — | — | — | — | |||||||
|
|||||||||||||
NO2 | Lag0-1 | 2.97 (2.39, 3.7) | 1.65 (0.73, 3.72) | — | — | — | — | ||||||
Lag0–2 | 2.87 (2.4, 3.42) | 2.82 (1.84, 4.33) | 1.92 (0.54, 6.79) | — | — | — | |||||||
Lag0–3 | 2.44 (2.09, 2.86) | 3.58 (2.63, 4.86) | 2.23 (1.14, 4.38) | NA | — | — | |||||||
Lag0–4 | 2.37 (2.06, 2.73) | 3.51 (2.7, 4.55) | 3.16 (1.8, 5.18) | 0.95 (0.12, 7.63) | NA | — | |||||||
Lag0–5 | 2.04 (1.78, 2.33) | 3.77 (3.02, 4.71) | 3.89§(2.83, 5.34) | 2.61 (1.28, 5.33) | NA | NA |
Abbreviations are the same as in Table
Figure
Markov transition probabilities between the high-admission state (state 2) and the low-admission state (state 1).
Table
Comparison of RR among Shanghai, Milan, and this study.
PM2.5 | PM10 | SO2 | NO2 | ||
---|---|---|---|---|---|
Shanghai (admission) [ |
Lag (increment) | — | Lag0-1 (60 |
Lag0-1 (36 |
Lag0-1 (29 |
RR (95% CI) | — | 1.88 (3.58, 7.35) |
4.79 (1.69, 11.27) |
9.38 (3.24, 15.51) |
|
|
|||||
This study (admission) | Lag (increment) | — | Lag0-1 (60 |
Lag0-1 (36 |
Lag0-1 (29 |
RR (95% CI) | — | 1.44 (−2.8, 5.86) | 25.81 (4.05, 52.12) |
7.11 (−2.5, 17.67) | |
|
|||||
Milan (emergent visit) [ |
Lag (increment) | Lag0–2 (10 |
Lag0–2 (10 |
Lag0–2 (5 |
Lag0–2 (10 |
RR (95% CI) | 3.30 (−4.40, 11.70) | 3.00(−3.60, 10.10) | 9.90(−15.40, 42.80) | 0.80(−4.30, 6.30) | |
|
|||||
This study (admission) | Lag (increment) | Lag0-1 (10 |
Lag0-1 (5 |
Lag0-1 (10 |
Lag0-1 (10 |
RR (95% CI) | 0.79 (−0.23, 1.82) | 0.49 (−0.27, 1.25) | 3.57 (0.55, 6.68) |
3.2 (−0.45, 6.98) |
Abbreviations are the same as in Table
The acute effects of PM10, SO2, and NO2 are all significant in Shanghai [
Table
Percent increase (mean and 95% confidence interval) in asthma hospital admission associated with a 10
Sex | Season | PM2.5 (lag0–2) | PM10 (lag0–2) | SO2 (lag0-1) | NO2 (lag0–2) |
---|---|---|---|---|---|
Both | Both | 0.82 (−0.24, 1.89) | 0.5 (−0.29, 1.3) | 7.27 (1.1, 13.82) |
3.2 (−0.45, 6.98) |
Warm | 4.72 (2.26, 7.23) |
2.48 (0.98, 4.01) |
3.53 (−8.6, 17.26) | −7.73 (−12.98, −2.16) |
|
Cold | 0.94 (0.15, 1.74) |
0.93 (0.31, 1.55) |
9.18 (4.27, 14.32) |
7.39 (4.26, 10.62) |
|
|
|||||
Male | Both | −0.13 (−1.83, 1.59) | −0.12 (−1.39, 1.16) | 2.95 (−6.47, 13.31) | 0.77 (−5.34, 7.28) |
Warm | 10.09 (6, 14.33) |
4.84 (2.41, 7.33) |
12.31 (−7.64, 36.56) | −1.06 (−9.79, 8.51) | |
Cold | −0.02 (−1.26, 1.24) | 0.28 (−0.7, 1.26) | 1.95 (−5.07, 9.5) | 3.28 (−1.4, 8.18) | |
|
|||||
Female | Both | 1.38 (0.03, 2.75) | 0.88 (−0.12, 1.89) | 9.93 (1.95, 18.54) |
4.71 (−0.31, 9.99) |
Warm | 0.92 (−2.3, 4.25) | 0.74 (−1.26, 2.78) | 2.36 (−13.31, 20.85) | −11.39 (−18.04, −4.19) |
|
Cold | 1.5 (0.44, 2.57) |
1.32 (0.49, 2.16) |
13.71 (6.87, 20.99) |
9.31 (5.04, 13.76) |
Abbreviations are the same as in Table
Figure
Markov transition probabilities between the high-admission state (state 2) and the low-admission state (state 1) for the female-cold subgroup.
Corresponding parameters of the female-cold subgroup.
Subgroup | A1 | A2 | A3 | A4 | A5 | A6 |
---|---|---|---|---|---|---|
Female-cold | PM2.5 | Lag0–4 | 35 |
4 days | 2 persons per day | 15.11 (2.72, 83.78) |
A1: pollutant; A2: lag; A3: concentration threshold; A4: counts exceeding the concentration threshold; A5: admission amount threshold; A6: OR.
This study certified that air pollutants have adverse short effects on elder hospital admissions for asthma. Such effects were observed for both the gaseous (SO2 and NO2) and particulate (PM10 and PM2.5) pollutants across all the different sex groups and season groups. PM2.5, PM10, and NO2 showed no significant effects on elders, whereas SO2 was evidently significant from lag0 to lag0–2.
We also made a comparison with the effects in other studies: (1) When compared with Milan (emergent visit), the effects of PM2.5, PM10, and SO2 are stronger than those in this study, which may lie in the fact that emergent visits are more sensitive to air pollution. However, the effect of NO2 in Milan is much weaker than that in this study; this gap needs further study. (2) When compared with Shanghai (admission), for PM10 and NO2, the effects are slightly weaker than those in Shanghai, whereas for SO2, the effects are dramatically stronger than those in Shanghai. These differences may lie when sexual and season factors were not considered.
Sex- and season-specific analysis indicates that for male, the effect was significant only during warm season with PM2.5 and PM10; however, for the female, every effect is significant during cold season.
Precise Markov transition probabilities between high-admission states and low-admission states are obtained by a multistate model. It was also shown that when air pollution gets worse, the transition probabilities from low-admission states and high-admission states to high-admission states increase dramatically. When we focused on the female-cold subgroup, this phenomenon appeared more evidently: the probability increasing due to air pollution worsening of the female-cold subgroup was much dramatic than that of full samples.
When these transition probabilities were combined with the forecast of air pollution, we can obtain the distributions of asthma admission, with reference to asthma healthcare resource demand (such as professional Medicare staff, wards) for a long period. Further, based on these distributions, asthma healthcare resource allocation can be done by the operation research method.
There are three points that should be focused:
Among PM2.5, PM10, SO2, and NO2, only the increment of SO2 was significant with that of elder asthma hospitalization. The effects of PM2.5, PM10, and SO2 in Milan are stronger than those in this study, which may lie in the fact that emergent visits are more sensitive to air pollution. However, the effect of NO2 in Milan is much weaker than that in this study; this gap needs further study. For male, the effect was significant only during warm season with PM2.5 and PM10. However, for the female, every effect is significant during cold season. The strongest effect for each pollutant was in the female-cold subgroup, which indicates that the female-cold subgroup is more fragile to air pollution and that is why we constructed Markov transition probability matrix only for the female-cold subgroup. The difference between full samples and the female-cold subgroup was quite evident: the probability increasing due to air pollution worsening of the female-cold subgroup was much dramatic than that of full samples. That is to say, air pollution matters for the female-cold subgroup more than full samples. This gap may lie when the acute effects between each air pollutant and elder asthma admission vary on sexual and season factors.
In summary, this study mainly achieved three goals: (1) validating air pollution (PM2.5, PM10, NO2, and SO2) has a great impact on elder asthma admission. For different air pollution conditions, the index to forecast high admission differs. (2) Outputting an effective air pollution index was performed to associate it with elder asthma admission. (3) Outputting Markov transition probabilities between high-admission states and low-admission states was performed, which could be used to forecast asthma healthcare resource demand when combined with air pollution forecast and then lead to healthcare resource allocation optimization.
Our study has limitations. First, the study design is ecological in nature, which may limit its ability for causal inference. Second, we simply averaged the monitoring results across various stations as the proxy for population exposure level to air pollution, which may raise a number of issues given that pollutant measurements can differ between monitoring locations and that ambient monitoring results differ from personal exposure level to air pollutants. The resulting measurement error may have substantial implication for interpreting time-series air pollution studies. Finally, this research only focuses on elder asthma admission. In fact, asthma outpatient and emergent patient take a large part of asthma healthcare resource. However, we do not have asthma outpatient and emergent patient.
Our work is a novel and fundamental study in asthma resource management. It not only provides a new prospect in the association between air pollution and asthma admission but also leads to a practical framework to implement asthma intervention and to allocate corresponding resource (such as professional Medicare staff, wards). Future work will be done in the aspect of forecasting admission amount, asthma intervention, and corresponding resource allocation.
Particulate matter not greater than 2.5 mm in aerodynamic diameter
Particulate matter not greater than 10 mm in aerodynamic diameter
Sulfur dioxide
Nitrogen dioxide
Quality-adjusted life years.
The study does not involve human subjects and adheres to all current laws of China.
Informed consent was received from the participants.
The authors declare that they have no competing interests.
Li Luo has made substantial contributions to the conception, design, analysis, and interpretation of the data. Fengyi Zhang has been involved in drafting the manuscript and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Wei Zhang has been involved in drafting the manuscript or revising it critically for important intellectual content. Lin Sun, Chunyang Li, Huang Debin, Han Gao, and Bin Wang have involved in the conception and design of this research. All authors read and approved the final manuscript.
The authors would like to thank the editors and the anonymous reviewers for their insightful and constructive comments and suggestions that have led to this improved version of the paper. The work was supported in part by the National Natural Science Foundation of China (no. 71532007, no. 71131006, and no. 71172197).