Ambient air pollution contributed to pulmonary tuberculosis in China

ABSTRACT Published studies on outdoor air pollution and tuberculosis risk have shown heterogeneous results. Discrepancies in prior studies may be partially explained by the limited geographic scope, diverse exposure times, and heterogeneous statistical methods. Thus, we conducted a multi-province, multi-city time-series study to comprehensively investigate this issue. We selected 67 districts or counties from all geographic regions of China as study sites. We extracted data on newly diagnosed pulmonary tuberculosis (PTB) cases, outdoor air pollutant concentrations, and meteorological factors in 67 sites from January 1, 2014 to December 31, 2019. We utilized a generalized additive model to evaluate the relationship between ambient air pollutants and PTB risk. Between 2014 and 2019, there were 172,160 newly diagnosed PTB cases reported in 67 sites. With every 10-μg/m3 increase in SO2, NO2, PM10, PM2.5, and 1-mg/m3 in CO, the PTB risk increased by 1.97% [lag 0 week, 95% confidence interval (CI): 1.26, 2.68], 1.30% (lag 0 week, 95% CI: 0.43, 2.19), 0.55% (lag 8 weeks, 95% CI: 0.24, 0.85), 0.59% (lag 10 weeks, 95% CI: 0.16, 1.03), and 5.80% (lag 15 weeks, 95% CI: 2.96, 8.72), respectively. Our results indicated that ambient air pollutants were positively correlated with PTB risk, suggesting that decreasing outdoor air pollutant concentrations may help to reduce the burden of tuberculosis in countries with a high burden of tuberculosis and air pollution.


Introduction
Globally, tuberculosis is one of the leading causes of mortality, and pulmonary tuberculosis (PTB) accounts for the vast majority of these deaths.Approximately 10.6 million new tuberculosis cases and 1.6 million tuberculosis-related deaths were documented worldwide in 2021 [1].An estimated quarter of the global population is infected with Mycobacterium tuberculosis (M.tb), namely, latent tuberculosis infection.Risk factors for infection and the development of active tuberculosis include smoking and diabetes [1].Recently, some studies have reported that outdoor air pollutants contribute to tuberculosis risk [2,3].
Ambient air pollution has posed a sizable threat to public health.Millions of deaths worldwide are directly attributed to outdoor air pollution each year [4].A global analysis revealed that PM 10 and PM 2.5 were positively associated with all-cause, cardiovascular, and respiratory mortality [5].Numerous studies have shown that outdoor air pollution probably contributes to various diseases, such as ischemic stroke [6], asthma [7], and childhood pneumonia [8,9].
Exposure to diesel exhaust may reduce the expression of several cytokines, such as tumour necrosis factor-α (TNF-α) and interferon-γ (IFN-γ) in murine lung tissues, which play critical roles in host defense [10,11].Diesel exhaust contains a variety of air pollutants, including SO 2 , NO 2 , particulate matter, and CO.This has led to the hypothesis that outdoor air pollution exposure may increase lung infection acquisition.Although some studies have explored the role of outdoor air pollution in the tuberculosis risk, the results have been heterogeneous.In Chengdu, SO 2 , NO 2 , and PM 10 were positively related to tuberculosis [3].Another study from northern California suggested that PTB was positively associated with NO 2 and CO and negatively associated with PM 10 and O 3 but not significantly associated with SO 2 and PM 2.5 [2].In Seoul, SO 2 was shown to be positively linked to tuberculosis [12].A study from Jinan revealed that PTB was negatively correlated with NO 2 and positively correlated with O 3 [13].
Inconsistent results may be due to the fact that these studies were conducted in a sole region or city, explored diverse exposure times, and applied distinct statistical methods.To address this knowledge gap, we performed a large, multi-province study in 67 study sites throughout China to investigate the impact of outdoor air pollution on the risk of PTB.

Study areas
Currently, China has 34 provincial administrative divisions.We selected Jiangsu Province, Guangdong Province, Qinghai Province, Liaoning Province, and Hubei Province as the study areas for eastern, southern, western, northern, and central China, respectively.According to geographical location, history, culture and nationality, these 34 provincial administrative regions can be divided into seven regions, namely, eastern, southern, southwestern, northwestern, northern, northeastern, and central China.First, we combined these seven regions into five regions, namely, eastern, southern, western (southwestern and northwestern), northern (northern and northeastern), and central China.Each region contained several provincial administrative divisions.Subsequently, we selected two or three provincial administrative divisions from each region as alternative study areas after comprehensively considering the average annual reported PTB incidence (low incidence: < 50/100,000; middle incidence: ≥ 50/ 100,000 and ≤100/100,000; high incidence: > 100/ 100,000) from 2014 to 2019 and the basis of previous cooperation.Then, we contacted the heads of the TB prevention and control institutions in these candidate study areas to ask whether they were interested in cooperating to perform this study.Jiangsu Province (with a low incidence), Guangdong Province (with a moderate incidence), Qinghai Province (with a high incidence), Liaoning Province (with a moderate incidence), and Hubei Province (with a moderate incidence) expressed their willingness to participate in this study.
These five provinces contain 69 municipal administrative divisions.We chose one county-level administrative division in each municipal administrative division.The following criteria were used: (1) availability of data on environmental factors; (2) no change in the administrative division between 2014 and 2019; and (3) the grade of environmental monitoring stations (prioritized order of air pollutant monitoring stations: national control stations > provincial control stations > municipal control stations > county control stations; prioritized order of meteorological factor monitoring stations: national benchmark climate stations > national basic meteorological stations > national general meteorological stations).After excluding two municipal administrative divisions without county-level administrative divisions, a total of 67 county-level administrative divisions were identified as study sites.

Data collection
We collected daily data on newly diagnosed PTB patients between January 1, 2014, and December 31, 2019, at 67 study sites, including sex, age, occupation, and diagnosis date, from the China Tuberculosis Management Information System.Sensitive and identifiable information such as names, phone numbers, and address details of the cases were removed to protect confidentiality and privacy.From this time period, daily data on outdoor air pollutant concentrations and meteorological factors from these sites were extracted from the National Urban Air Quality Real-time Release Platform and the China Meteorological Data Sharing Center, respectively, which were widely utilized in previous studies [14,15].The outdoor air pollutants included SO 2 (μg/m 3 ), NO 2 (μg/m 3 ), PM 10 (μg/ m 3 ), PM 2.5 (μg/m 3 ), and CO (mg/m 3 ).The average meteorological factors included temperature (°C), wind speed (m/s), and relative humidity (%).We calculated the weekly number of PTB cases, the mean concentration of each air pollutant weekly, and the mean value of each meteorological factor weekly.The study period was then divided into 313 weeks.

Statistical analysis
We applied a generalized additive model (GAM) following the quasi-Poisson distribution to evaluate the relationship between outdoor air pollutant concentrations and the PTB risk [16,17].According to previous studies, outdoor air pollutants may have lag effects on tuberculosis risk [3,18].Thus, we determined the maximum lag time up to 25 weeks to capture both short-term (lag 0 week to lag 4 weeks) and relatively long-term (lag 5 weeks to lag 25 weeks) impacts of outdoor air pollutants [17,19].For example, SO 2 at lag 0 week indicated the weekly average concentration of SO 2 at the current week, and SO 2 at lag 1 week indicated the weekly average concentration of SO 2 one week prior.Covariates adjusted in the model included "week" (week = 1, 2, … , 313), the number of cases one week ago, the number of holidays in the week, season, "city" to control city-specific characteristics such as socioeconomic level and population scope, and three weekly average meteorological factors at the same lag week [17,20,21].We created smooth terms for "week" and meteorological factors using the thin plate spline function (TPSF) with maximum degrees of freedom (df) of six and two, respectively, to control their potential nonlinear effects [17,22].
We calculated the influences of ambient air pollutants on the PTB risk at all lag weeks and then identified the lag week with the maximum effect (air pollutant concentration and the PTB risk showed the strongest association at this lag week) for each air pollutant to perform further analyses.The strength of the relationship is presented as percentage changes in the PTB risk and the 95% confidence intervals (CIs) for each 10-μg/m 3 (1-mg/m 3 for CO) increase in air pollutant concentration.We conducted subgroup analyses to estimate the associations between air pollutants and the PTB risk in different populations and seasons.The differences in effects between subgroups were evaluated using the formula:

􏽰
, where b 1 and b 2 are the estimated effects and SE 1 and SE 2 are the standard errors of the estimates.The difference between different sex groups or age groups was considered to be statistically significant if the corresponding value was >1.96, and the difference between different season groups was considered to be statistically significant after the Bonferroni correction if the corresponding value was >2.64 [23].
We further performed a series of sensitivity analyses to assess the stability of the relationships between air pollutants and the PTB risk.First, we constructed dual-pollutant models by additionally adjusting each of the other air pollutants in the model.Notably, due to the high correlation between PM 10 and PM 2.5 (Figure S1), they were not entered into the model simultaneously to address the multicollinearity problem [5,20].Second, we varied the maximum df of the smooth term of "week" between four and eight and the maximum df of the smooth terms of meteorological factors between two and six to investigate the impacts of air pollutants.Third, since monitoring data on ambient air pollutants in some study sites started on December 31, 2014 (week 53), we recalculated the effects at 53-313 weeks.Fourth, we considered "city" as the random effect variable and reevaluated the associations between air pollutant concentrations and the PTB risk using the generalized additive mixed model (GAMM).GAMM is an extension of the GAM that adds a random effect term on the basis of the GAM [24,25].
Moreover, considering our assumption that the links between air pollutants and the PTB risk were linear, we plotted the concentration-response curve between each air pollutant and the PTB risk to examine the reliability of our results.Specifically, we constructed a smooth term for each air pollutant using the TPSF with a maximum df of two to replace the linear term in the model.We applied the piecewise linear function to re-estimate the effect if the curve was nonlinear [17].The optimal cutoff value of the piecewise linear function was determined by the minimum generalized cross-validation score [5].
All analyses were conducted in R software, and the significance level was set as 0.05 (2-tailed).

Outdoor air pollutants and PTB risk
Outdoor air pollutants were significantly associated with the PTB risk at multiple lag weeks, including lag 0 week to lag 6 weeks, lag 8 weeks, and lag 11 weeks to lag 19 weeks for SO 2 ; lag 0 week, lag 15 weeks, lag 19 weeks, lag 21 weeks, lag 24 weeks, and lag 25 weeks for NO 2 ; lag 8 weeks, lag 10 weeks, lag 11 weeks, lag 13 weeks, and lag 15 weeks for PM 10 ; lag 8 weeks and lag 10 weeks for PM 2.5 ; lag 4 weeks, lag 6 weeks, lag 10 weeks to lag 18 weeks, and lag 21 weeks for CO (Figure 1).
The subgroup analyses indicated that the associations between air pollutants and the PTB risk were significant in males, females, people aged <60 years, people aged ≥60 years, autumn, and winter for SO 2 ; in males, females, people <60 years old, and winter for NO 2 ; in males, people aged <60 years, people aged ≥60 years, and winter for PM 10 ; in males, people <60 years old, and winter for PM 2.5 ; and in males, females, people aged <60 years, people aged ≥60 years, summer, and winter for CO (Table 3).Moreover, the differences in the associations between air pollutants and the PTB risk were not statistically  significant between different sex groups or age groups but were statistically significant between spring and winter for NO 2 , PM 10 , and PM 2.5 (Table S2).
The sensitivity analyses showed that the associations between air pollutants and the PTB risk remained significant after adjusting for NO 2 , PM 10 , PM 2.5 , or CO for SO 2 ; after adjusting for PM 10 , PM 2.5 , or CO for NO 2 ; after adjusting for SO 2 , NO 2 , or CO for PM 10 ; after adjusting for SO 2 or NO 2 for PM 2.5 ; and after adjusting for SO 2 , NO 2 , PM 10 , or PM 2.5 for CO (Table 2).In addition, the associations between outdoor air pollutants and the PTB risk were robust when the maximum df of the TPSF varied (Table S3), at 53-313 weeks (Table S4), or based on the GAMM (Table S5).
The concentration-response curves between ambient air pollutants and the PTB risk were almost linear or approximately linear, except for CO (Figure 2).Therefore, we applied the piecewise linear function to re-evaluate its effect.The optimal cutoff value was identified as 2.3 mg/m 3 , corresponding to the minimum generalized cross-validation score (Table S6).For each 1-mg/m 3 increase in CO, the PTB risk increased by 18.57% (95% CI: 10.82, 26.87) when CO was <2.3 mg/m 3 and decreased by 22.51% (95% CI: −50.67, 21.75) when CO was ≥2.3 mg/m 3 (Table S7).

Discussion
In this study from different regions of China, we revealed a positive link between ambient SO 2 , NO 2 , PM 10 , PM 2.5 , and CO and PTB risk.This study goes beyond previous work to provide a rigorous, comprehensive assessment of the interplay between outdoor air pollution and PTB risk at the population level.To our knowledge, this is the largest and first multiprovince, multi-city study to evaluate the impacts of ambient air pollutants on PTB risk.
A time-series analysis from Chengdu suggested that SO 2 had adverse effects on tuberculosis [3].Two spatiotemporal analyses in Seoul and Hubei also revealed a positive link between SO 2 and tuberculosis [12,26].The above studies support our findings.Nevertheless, no significant association was found between SO 2 and tuberculosis in northern California or Lampang [2,27].Northern California is economically developed and had low ambient SO 2 concentrations during the study period (median: 3.43 μg/ m 3 ), while China is a developing country, and the median ambient SO 2 concentration in our study was 13.57μg/m 3 [2].Only a bivariate correlation analysis was conducted for Lampang to evaluate the effect of SO 2 on the tuberculosis risk, but potential confounding factors were not considered [27].These factors may partly explain the inconsistency.
We found that NO 2 was positively associated with PTB at different lag weeks.Previous studies also reported that outdoor NO 2 contributed to tuberculosis [2,3,28].The sensitivity analyses suggested that this positive association was no longer significant when adjusting for SO 2 , possibly because SO 2 and NO 2 often coexist in the outdoor environment due to similar sources, and SO 2 enhances the effect of NO 2 .The effect of NO 2 exposure alone on PTB requires further study.
A study based on seven cities in Korea showed that PM 10 had a lag effect on tuberculosis [29].This finding was consistent with another study from Hong Kong [30].Similarly, we found that PM 10 had a relatively long-term effect on PTB, as the association became significant from lag 8 weeks.Nevertheless, a study from northern California indicated that PM 10 played a protective role in the PTB risk [2].Differences in socioeconomic level, PM 10 composition, and concentration (median: 20.61 vs. 63.43 μg/m 3 ) between the two study areas may be partly responsible for the opposite results.
A study in Beijing and Hong Kong reported that an increase in PM 2.5 resulted in an increased risk of tuberculosis [18].Similarly, we revealed a positive correlation between PM 2.5 and PTB.The concentrationresponse curve between PM 2.5 and PTB was approximately linear.However, the curve was relatively flat Table 3. Subgroup analyses of percentage changes in pulmonary tuberculosis risk and their 95% confidence intervals for each 10μg/m 3 (1-mg/m 3 for CO) increase in ambient air pollutant concentrations*.*: We applied lag 0 week for SO 2 , lag 0 week for NO 2 , lag 8 weeks for PM 10 , lag 10 weeks for PM 2.5 , and lag 15 weeks for CO. † : Adjusted for the week, the number of cases in the previous week, the number of holidays in the week, season, city, average temperature, average wind speed, and average relative humidity at the same lag week.‡ : Adjusted for the week, the number of cases in the previous week, the number of holidays in the week, city, average temperature, average wind speed, and average relative humidity at the same lag week.
when the PM 2.5 concentration was low and became steeper at higher concentrations, suggesting that high concentrations of PM 2.5 had a greater impact on PTB.In addition, the influence of PM 2.5 was slightly stronger than that of PM 10 , probably because, in contrast to PM 10 , PM 2.5 has a smaller particle size and longer suspension time and can not only penetrate deep into the lungs but also enter the circulatory system through the respiratory barrier [31].
Studies from northern California and Hong Kong indicated that CO was positively correlated with tuberculosis [2,30].We also observed a positive correlation between CO and PTB.Moreover, as the concentration-response curve between CO and PTB was nonlinear, we applied a piecewise linear function to re-estimate the effect.The results suggested that CO was positively correlated with PTB when the concentration was <2.3 mg/m 3 and negatively linked to PTB when the concentration was ≥2.3 mg/m 3 , but the latter correlation was not significant.Compared to the piecewise linear function results, although our original results seemed to underestimate the effect of CO, they were still relatively reliable.The Q98 and Q99 of the CO concentration measurements were 2.04 and 2.33 mg/m 3 , respectively.In other words, approximately 99% of the measured concentrations were less than 2.3 mg/m 3 .The protective role of high concentrations of CO needs to be explored in future studies.
Although not statistically significant, we observed several reverse associations between air pollutants and PTB risk at several lag weeks.NO 2 was negatively correlated with the PTB risk at shorter lag weeks, while PM 10 and PM 2.5 were negatively linked to the PTB risk at shorter and longer lag weeks, respectively.The negative correlations at shorter lag weeks may be related to two reasons.First, we utilized the diagnosis date instead of the onset date when counting the number of PTB patients.The onset date is inferred by clinicians based on the patients' recollection, which is often inaccurate.The time from the onset of PTB to its diagnosis is usually several days.Second, PM 10 and PM 2.5 may have relatively long lag effects on the PTB risk, which was also observed in previous studies [18,32].The negative associations at longer lag weeks may be attributed to two factors.One factor is that the lag effects of PM 10 and PM 2.5 on the PTB risk do not persist for long.The other factor is that a longer We applied lag 0 week for SO 2 , lag 0 week for NO 2 , lag 8 weeks for PM 10 , lag 10 weeks for PM 2.5 , and lag 15 weeks for CO.The results were adjusted for the week, the number of cases in the previous week, the number of holidays in the week, season, city, average temperature, average wind speed, and average relative humidity at the same lag week.
lag time results in more unknown confounding factors and a greater likelihood of distortion of the real correlations of PM 10 and PM 2.5 with the PTB risk.Moreover, we found that the relationship between outdoor air pollutants and PTB appeared to vary across populations and seasons, but these differences were statistically significant only between spring and winter for NO 2 , PM 10 , and PM 2.5 , which may be partly attributed to seasonal variations in air pollutant concentrations, sources, and PM compositions.
The mechanism for the role of outdoor air pollutants in PTB risk was beyond the scope of our study.However, several potential biological explanations may be primarily related to immune dysfunction.First, the respiratory tract can produce secretions to wrap M.tb and then clear it to defend against M.tb invasion.Exposure to outdoor air pollutants may weaken the mucociliary clearance of airway secretions, thus raising the possibility of M.tb infection [33].Second, alveolar macrophages (AMs) are the first line of immune defense in the lungs, which can eliminate harmful bacteria.However, PM 10 and PM 2.5 exposure is likely to reduce the expression of phagocytosis-related receptor CD11b on the surface of AMs, impairing the ability of AMs to phagocytose M.tb [34].Third, exposure to outdoor air pollutants decreased the expression of cytokines (TNF-α, IFNγ, etc.) in peripheral blood mononuclear cells (PBMCs), suppressing the role of PBMCs in phagocytosing M.tb and controlling M.tb growth [35][36][37].Fourth, PM 2.5 is likely to cause inflammatory responses and intracellular oxidative stress, leading to the progression of granulomatous lesions in the lungs [31,38].Large studies investigating tuberculosis risk, pollutant exposures, and immunological measures are needed to further address the mechanisms explaining the strong relationship we see in our research.
This study has several limitations.First, we utilized measurements from fixed stations to estimate individual exposure to outdoor air pollutants; however, exposure misclassification is unavoidable.Thus, more precise methods are needed to estimate individual exposure levels to air pollutants in future studies.For example, two studies from Changsha evaluated personal exposure to outdoor air pollutants using the inverse distance weighting method based on data from local air quality monitoring stations and the address of each participant [8,9].Second, other confounders, such as comorbidities (diabetes, etc.) that impact the risk of tuberculosis and the quality of tuberculosis notification, were not included because they were not available in our database.Third, we extracted data on the diagnosis date of PTB patients, an inaccurate proxy of disease onset.This would lead to nondifferential disease misclassification, likely biasing the associations we present here towards the null.

Conclusions
In summary, we revealed that outdoor air pollutants were positively related to PTB risk.As many developing countries suffer from both a high tuberculosis burden and severe outdoor air pollution, reducing outdoor air pollutant concentrations might indirectly benefit tuberculosis control.

Figure 1 .
Figure 1.Percentage changes in pulmonary tuberculosis risk and their 95% confidence intervals for each 10-μg/m 3 (1-mg/m 3 for CO) increase in ambient air pollutant concentrations at all lag weeks.Abbreviation: PTB = pulmonary tuberculosis.The results were adjusted for the week, the number of cases in the previous week, the number of holidays in the week, season, city, average temperature, average wind speed, and average relative humidity at the same lag week.

Figure 2 .
Figure 2. The concentration-response curves between ambient air pollutants and pulmonary tuberculosis risk.The x-axis represents the concentration of air pollutants, while the y-axis represents the contribution of the smooth term to the fitted values.We applied lag 0 week for SO 2 , lag 0 week for NO 2 , lag 8 weeks for PM 10 , lag 10 weeks for PM 2.5 , and lag 15 weeks for CO.The results were adjusted for the week, the number of cases in the previous week, the number of holidays in the week, season, city, average temperature, average wind speed, and average relative humidity at the same lag week.

Table 1 .
Characteristics of pulmonary tuberculosis patients at the 67 study sites.

Table 2 .
Percentage changes in pulmonary tuberculosis risk and their 95% confidence intervals for each 10-μg/m 3 (1-mg/m 3 for CO) increase in ambient air pollutant concentrations*.: We applied lag 0 week for SO 2 , lag 0 week for NO 2 , lag 8 weeks for PM 10 , lag 10 weeks for PM 2.5 , and lag 15 weeks for CO.
*† : Adjusted for the week, the number of cases in the previous week, the number of holidays in the week, season, city, average temperature, average wind speed, and average relative humidity at the same lag week.‡ : Based on Model 1, additionally adjusted for SO 2 at the same lag week.§ : Based on Model 1, additionally adjusted for NO 2 at the same lag week.ll : Based on Model 1, additionally adjusted for PM 10 at the same lag week.**: Based on Model 1, additionally adjusted for PM 2.5 at the same lag week.† † : Based on Model 1, additionally adjusted for CO at the same lag week.