Effect modification of greenness on PM2.5 associated all-cause mortality in a multidrug-resistant tuberculosis cohort

Rationale Evidence for the association between fine particulate matter (PM2.5) and mortality among patients with tuberculosis (TB) is limited. Whether greenness protects air pollution-related mortality among patients with multidrug-resistant tuberculosis (MDR-TB) is completely unknown. Methods 2305 patients reported in Zhejiang and Ningxia were followed up from MDR-TB diagnosis until death, loss to follow-up or end of the study (31 December 2019), with an average follow-up of 1724 days per patient. 16-day averages of contemporaneous Normalised Difference Vegetation Index (NDVI) in the 500 m buffer of patient’s residence, annual average PM2.5 and estimated oxidant capacity Ox were assigned to patients regarding their geocoded home addresses. Cox proportional hazards regression models were used to estimate HRs per 10 μg/m3 exposure to PM2.5 and all-cause mortality among the cohort and individuals across the three tertiles, adjusting for potential covariates. Results HRs of 1.702 (95% CI 1.680 to 1.725) and 1.169 (1.162 to 1.175) were observed for PM2.5 associated with mortality for the full cohort and individuals with the greatest tertile of NDVI. Exposures to PM2.5 were stronger in association with mortality for younger patients (HR 2.434 (2.432 to 2.435)), female (2.209 (1.874 to 2.845)), patients in rural (1.780 (1.731 to 1.829)) and from Ningxia (1.221 (1.078 to 1.385)). Cumulative exposures increased the HRs of PM2.5-related mortality, while greater greenness flattened the risk with HRs reduced in 0.188–0.194 on average. Conclusions Individuals with MDR-TB could benefit from greenness by having attenuated associations between PM2.5 and mortality. Improving greener space and air quality may contribute to lower the risk of mortality from TB/MDR-TB and other diseases.


INTRODUCTION
Earlier studies have shown that particular matter (PM) in alveolar macrophages alters Mycobacterium tuberculosis-induced cytokine production in the lung and systemic components, resulting in relative cellular unresponsiveness. 1 Inhalation exposure to PM impairs important components of the protective human lung and results in suppressing systemic immune response against M. tuberculosis. 2 A previous California cohort found that living near to high traffic volume and density roads was associated with 3%~28% increased risk of death during tuberculosis (TB) treatment, implying adverse impacts of traffic-related air pollution on patients with TB. 3 A recent systematic review suggested that long-term exposure to air pollution could increase infections with M. tuberculosis, development of active TB, and TB-related mortality, while these associations are inconsistent due to the small number of the studies. 4 While a number of studies have examined the associations between air pollution and infections with TB and transmissions, 5 6 few have investigated the impacts of PM 2.5 exposure on mortality after TB infections. Furthermore, fewer have examined the impacts on multidrug-resistant tuberculosis (MDR-TB), a serious form of TB caused by bacteria that do not respond to isoniazid (INH) and rifampicin (RMP), the two most powerful first-line anti-TB

Key messages
What is the key question? ⇒ How greenness protects air pollution-related mortality among patients with multidrugresistant tuberculosis (MDR-TB) is completely unknown.

What is the bottom line?
⇒ Recent experimental and clinical studies suggest that inhalation exposure to PM 2.5 impairs important components of the protective human lung and results in suppressing systemic immune response against Mycobacterium tuberculosis, whereas this impact of air pollution may be overstated if greenness was not considered.

Why read on?
⇒ Consistent with the experimental evidence of oxidative stress and inflammatory effects of PM 2.5 , our study, the largest MDR-TB cohort, shows that exposure to ambient PM 2.5 was significantly associated with risks of mortality among patients with MDR-TB, and highlighted that greenness could benefit patient survival by attenuating the impact of PM 2.5 .
drugs. One previous study in a small Chinese cohort of 752 patients reported the association between air pollution and drugresistant TB development. 7 However, it has been noted that the impact of air pollution may be overstated if greenness, which is generally thought to affect health through multiple mechanisms, such as mitigating exposures to air pollution, cooling weather, and reducing mental and psychological stress, 8 was not taken into account. 9 10 Thus, a gap in evidence exists on current studies of how greenness may modify the impact of air pollution on mortality after getting MDR-TB. We aim to examine the role of residential greenness in modifying the association between chronic PM 2.5 exposure and all-cause mortality in a large Chinese MDR-TB cohort across Zhejiang and Ningxia. These two provinces, representing an affluent southeast and a deprived northwest province, respectively, of China, were involved in the National Health Commission of China-Bill and Melinda Gates Foundation TB Collaboration project, 11 which was initiated in 2009 and employed innovative molecular tests for TB and MDR-TB, and implemented effective follow-up strategies to delivery and manage patients. 12 We present the associations for the full cohort and across tertiles of greenness measured within 500 m around patient's residence during 11 years of follow-up (2009-2019), respectively. Oxidant capacity (O x ) is incorporated in our models as previous studies showed that they were associated with more chronic health outcomes, like TB, than either O 3 or NO 2 alone. 13 14 We hypothesise that living in greener areas attenuated the risk of PM 2.5 -related mortality among patients with MDR-TB by potentially increasing exposure to fresh air, lowering emissions and cooling air.

The study cohort
All MDR-TB cases included in this study were microbiologically confirmed by prefecture-level reference laboratories under rigorous quality control by the Zhejiang Provincial Center for Disease Control and Prevention (CDC) and the Fourth People's Hospital of Ningxia. Patients were required to provide three sputum samples and drug susceptibility testing (DST) at one of the prefecture-level TB-designated hospitals. All the three samples were tested using microscopy, and two were sent for Löwenstein-Jensen solid culture. Culture-positive samples were subject to DST against the first-line anti-TB drug RMP, INH, ethambutol and streptomycin. Both provinces adopted the case diagnosis and detection procedure recommended by the National Health Commission of China-Bill and Melinda Gates Foundation TB Collaboration project. 11 Active MDR-TB cases reported from 1 January first 2009  to 31 December 2017 were eligible for inclusion. We enrolled  all confirmed cases diagnosed between 1 January 2009 and 31  December 2017 in Zhejiang and between 1 January 2015 and 31 December 2017 in Ningxia because earlier cases were not available. We excluded migrant patients to reduce exposure misclassification as they might return to hometowns during TB treatment which were not recorded. We also excluded patients who had no complete record of MDR-TB diagnosis, died, moved out of the provinces before diagnosis, or had no air pollution and greenness data available for their residential addresses.

MEASURES Outcome measures
Patients with MDR-TB were registered and prospectively followed by the Zhejiang Provincial CDC and the Fourth People's Hospital of Ningxia from diagnosis confirmation to death, loss to follow-up or end of the study by 31 December 2019. We verified mortality information through the vital registration database obtained in both provincial CDCs.

Exposure measures
We assigned estimates of air pollution and greenness exposures to patients from date of MDR-TB confirmed to either date of death, loss to follow-up or the end of the study by geocoding patients' home addresses (figure 1). The detailed geocode method was described and verified in our previous studies. 15

Pollutants
High-resolution (1 km) annual PM 2.5 was from the space-time extremely randomised trees model that integrated total column aerosol depth retrievals from the Moderate Resolution Imaging Spectroradiometer (MODIS) of the National Aeronautics and Space Administration (NASA)'s Terra and Aqua satellites. 16 17 The annual PM 2.5 were synthesised from the daily PM 2.5 data of MODIS AOD and a large number of auxiliary data, including meteorological conditions, human distribution and pollution emission, using the method proposed in our previous study. 16 The estimated annual mean PM 2.5 were highly consistent with the ground-based measurements with a high correlation coefficient (R 2 =0.94) in China. 17 Additionally, we calculated the combined oxidant capacity (O x ) of O 3 and NO 2 as a weighted average with weights equivalent to their respective redox potentials (ie, O x =[(1.07×NO 2 ) + (2.075×O 3 )]/3.145). 18 Daily NO 2 and O 3 were obtained from the National Air Quality Network of the Chinese Ministry of Environmental Protection. The annual concentrations of NO 2 were derived from a temporally adjusted national land-use regression model that combined 1500 fixedsite nationwide monitoring data, satellite NO 2 estimates, road length within 10 km, industrial land-use areas within various buffers and meteorological factors. 19 Ozone exposures were estimated based on 8-hour average daily maximum concentrations by integrating monitoring O 3 with the surface temperature, relative humidity, wind speed and direction, land-use areas and various anthropogenic emissions using the Weather Research and Forecasting atmospheric chemistry (WRF-Chem) models. 20 In our model, NO 2 and O 3 were not included to reduce multilinearities with PM 2.5 and O x .

Greenness
The Normalised Difference Vegetation Index (NDVI) has been used in measuring greenness exposure in epidemiological studies. 21 22 NDVI values range from −1 to +1, with −1 for water, 0 for bare soil and +1 for healthy vegetation. 23 Sixteen-day average greenness surface data at 500 m×500 m were derived from MODIS for the study period. The dataset has been used and verified by other studies. 9 21 22 We calculated the cumulative 16-day average NDVI values based on the method introduced in previous study 21 and linked NDVI to patient from diagnosis confirmation to death, loss to follow-up or end of follow-up according to the patient's home address.

Covariates
There were three domains of covariates collected by this study: patient's individual characteristics, treatment and environment: individual characteristics include age at diagnosis, sex, occupation, ethnicity; covariates related to treatment include drugresistant type (ie, MDR, or extensively drug-resistant (XDR) TB), if the patient took MDR-TB treatments or not, and if the case had TB-related treatment before the diagnosis of MDR-TB (suggesting acquired resistance) or the case was initially diagnosed as MDR-TB without any TB treatment before (suggesting primary resistance) 24 25 ; covariates related to environment include province (ie, Zhejiang or Ningxia), place of residence (ie, urban or rural), altitude as a proxy for meteorological factors, like temperature, monthly average Nighttime Light Index (NTL) as a proxy for socioeconomic and urbanisation levels. NTL derived from satellite images represents light intensity level at night with values ranging from 0 to 63, with higher values indicating brighter light and higher urbanisation level. The altitude and monthly NTL were publicly available from the NASA's Socioeconomic Data and Applications Centre with details about NTL data can be found in previous study. 26 The monthly average NTL values were also calculated and linked patients during the follow-up according to their home addresses.

Statistical analysis
The number and proportion of patients with MDR-TB were stratified by individual characteristics, treatment history and environment where they live across tertiles of NDVI. The association between PM 2.5 and mortality was assessed using the Cox proportional hazards regression model with O x adjusted for the full cohort and the three tertiles, separately. HRs were calculated per 10 μg/m 3 and presented with 95% CIs. The three domains of covariates were included in multivariable Cox proportional hazard regression models to adjust for potential confounding. We used the plots of the Martingale-based residuals to check violation of the proportional hazard assumption. 27 Covariates that could not meet the proportional hazards assumption were excluded from the regression. Sensitivity analyses were implemented through a series subgroup tests for the effect modification by age, sex, treatment (treated vs untreated, primary resistance vs acquired resistance), living areas (urban vs rural) and province (Zhejiang vs Ningxia). We fitted separate multivariable Cox models for groups and reported the strata-specific effects at a p value <0.05.
We examined the linearity of exposure-response relationships for the full cohort and the three tertiles of NDVI using the cubic regression spline. 28 Additionally, statistical tests were further conducted to examine if the HRs were significantly different in the three different tertiles and if the risk of PM 2.5 -related mortality was lower for the patients with greater greenness. All the statistical analyses were carried out using R v.3.5.3 with the 'survival' package.

RESULTS
After excluding 443 (15.4%) migrant patients and 10 (0.35%) local patients who did not have complete residential information or no air pollution or NDVI data, we included 2305 MDR-TB cases with 2095 from Zhejiang and 210 from Ningxia in the study. All patients were reported between 2009 and 2017, and followed to 31 December 2019, with an average follow-up of 1724 days per patient. Among the patients, there were 525 deaths during the period of follow-up.

Descriptive statistics
Of 2305 cases, 1674 (72.6%) were male, 62 (2.69%) were ethnic minority, 641 (27.8%) were 60 years and older at diagnosis and 37 (1.61%) were XDR-TB. Prior to MDR-TB diagnosis, 1535 (66.6%)    Table 3 Adjusted HRs (95% CIs) per 10 µg/m 3 increase in PM 2.5 associated with mortality among patients with MDR-TB stratified by treatment and type of drug resistance from single-pollutant and oxidant models in the full cohort (n=2305) and across the three tertiles of greenness (ie, NDVI) within the 500 m buffer around patient's residence    10 we did not find a monotonic trend in the effect of PM 2.5 within the different tertiles of NDVI: the mortality risks were lowered in the greatest tertile of NDVI than those in the other two tertiles in both models for the full cohort and other groups, except for the treated patient group in single-pollutant model. We repeated the oxidant-pollutant adjusted models to examine the association stratified by age (<60 and ≥60 years old), sex, place of residence (urban and rural) and province (Zhejiang and Ningxia). Table 4 shows the stratified analysis results, suggesting that exposure to PM 2.5 were stronger in association with mortality for younger patients (HR 2.434, 95% CI 2.432 to 2.435 (<60 years) vs HR 1.487, 95% CI 1.475 to 1.501 (≥60 years)), female (HR 2.209, 95% CI 1.874 to 2.845 (female) vs HR 1.644, 95% CI 1.611 to 1.678 (male)) and patients from Ningxia province (HR 1.221, 95% CI 1.078 to 1.385 (Ningxia) vs HR 1.059, 95% CI 1.049 to 1.069 (Zhejiang)). For patients living in rural, the HR was 1.780 (95% CI 1.731 to 1.829) slightly higher than those living in urban areas (HR 1.739, 95% CI 1.631 to 1.854). Exposure to PM 2.5 in the lower greenness showed a trend of increasing mortality risk for older (≥60 years old), male and patients living in urban areas, and Zhejiang province. We did not observe a similar increasing trend among patients in Ningxia due to very limited cases in this tertile.

Exposure-response curves
The exposure-response curves shown in figure 3 suggested nonlinearity in the association between PM 2.5 exposure and mortality among patients with MDR-TB for the full cohort (figure 3A) and the three tertiles (figure 3B-D) of NDVI, within the 500 m buffer around patient's residence. Cumulative exposures increased the HRs of PM 2.5 -related mortality, while greater greenness substantially flattened the risk with the HRs reduced in 0.188 (95% CI 0.123 to 0.253) (tertile 2, figure 3C) to 0.194 (95% CI 0.129 to 0.259) (tertile 3, figure 3D) on average, compared with the lowest greenness (tertile 1, figure 3B).

DISCUSSION
To our knowledge, this study represents the largest populationbased study of MDR-TB in China with 11 years of follow-up. Although MDR-TB is treatable and curable with second-line drugs, the disease has high mortality, responsible for an estimate of 2.5 million deaths globally by 2050 if no interventions. 29 Inhaled PM was noted relevant to TB infection and transmission, [30][31][32] while evidence regarding the impacts of PM 2.5 on mortality among patients with TB is still sparse. Furthermore, fewer have examined for MDR-TB and how greenness modifies the risk of PM 2.5 -related mortality. Our study found a 70.2% increase in mortality per 10 μg/m 3 increase in PM 2.5 exposure, after adjusting for O x , in the full MDR-TB cohort. Greater exposure to greenness lowered the impacts of PM 2.5 for untreated and primary resistance patient groups, except those taking MDR-TB treatment. Our data suggested that living in lower greenery neighbourhoods increased risks of mortality associated with PM 2.5 , particularly in elderly (≥60 years), male and patients from urban areas, while it did not show that greenness attenuated the effect of PM 2.5 in the province of Ningxia. These findings are important because they examined the effects of inhaled PM on mortality after infecting with MDR-TB and the effect modification by greenness, which are still unclear to us.
Our study provides reliable exposure measurements as we excluded migrant patients and permanent resident patients without residential information, air pollutant or NDVI data. All 2305 patients included in the analyses have permanent home addresses registered in the TB surveillance systems of the provincial CDC. Our results showed that the associations between PM 2.5 and mortality were consistent for both the full cohort and the three different greenness levels as well.
The adverse effects of ambient PM 2.5 against TB/MDR-TB are biologically plausible. Air pollution has detrimental impact on lung function and immune system through decreasing macrophage functioning, promoting oxidative stress and inflammation, and increasing reactivity and enhanced vulnerability towards pathogens. 33 34 The oxidative stress caused by PM could damage the epithelium of the airways and reduce immune response against M. tuberculosis. 1 35 Inflammation induction and particle deposits in the damaged lungs could cause further reduction of pulmonary function and promote disease progression and deaths, particularly patients with MDR-TB. A Chinese cohort found that exposure to PM 2.5 was associated with a 30% increase in the risks of all-cause mortality and 72% increase in the deaths from TB, respectively. 36 Our risk estimates of exposure to PM 2.5 and all-cause mortality for the full cohort of patients with MDR-TB was higher, suggesting that patient with MDR-TBs might be more vulnerable to ambient PM than those with a regular TB.
Ji et al 10 used a prospective cohort of 12 873 participants from 631 cities and counties in 22 Chinese provinces to measure whether greenness protects against air pollution-related mortality between 2008 and 2014. This large-scale longitudinal study found that increasing greenness exposure could reduce mortality associated with PM 2.5 exposure in the 500 m radium around participant's residence. Although no studies have examined the effect modification by greenness among patients with TB or MDR-TB, Ji et al's estimates of greenness in reducing risks of all-cause mortality relevant to PM 2.5 are consistent with our findings.
Crouse et al 9 conducted a larger cohort study that includes 2.4 million non-immigrant Canadians to examine the role of residential greenness in modifying associations between chronic PM 2.5 exposure and non-accidental mortality during 11 years of follow-up. They found that individuals in deprived neighbourhoods with high greenness benefitted by having more attenuated associations between PM 2.5 and mortality than those living in deprived areas with less greenness. Our study observed greater benefits for patients in urban with higher greenness level than those in urban with less greenness, but not for those in rural areas.
Some previous studies suggested that differences in sex and age regarding the impacts of greenness on PM 2.5 associated with mortality. 10 37 However, the literature has been inconsistent. Our observed more attenuated risks of mortality relevant to PM 2.5 with the greatest greenness among younger (<60 years), untreated and primary resistance patients imply that effect heterogeneity exists in age, sex, treatment and development of drug resistance, although exact mechanisms need to be elucidated.
A strength of this study is its inclusion of all reported patients with MDR-TB in Zhejiang and Ningxia, two provinces implemented rigorous diagnosis methods and 24-month standardised treatment under the China-Bill and Melinda Gates TB project. Additionally, we had detailed demographic data, treatment history and environmental data around patient's residence, which allow us to conduct full adjusted models and a series stratified analysis. Despite these strengths, a few limitations should be noted. Although migrant and local patients who did not have residential information or nor air pollution/NDVI data were excluded, misclassification (overestimation or underestimation) and selection, although low, might still be possible. Although we have adjusted for a set of potential individual and contextual covariates, which minimised but might not have entirely eliminated confounding. For example, some patients with MDR-TB received individual adjusted treatment when standardised MDR-TB regimens were not effective; however, we did not have information regarding change of regimens during treatment. Although we have adjusted for potential regional effects and stratified the analyses by the provinces, the HR of PM 2.5 could not be confirmed for only four cases in the most greenery areas (ie, the third tertile of NDVI) of Ningxia given its relatively small population. Different vegetation types might have different abilities in absorbing air pollutants. 38 The heterogeneity of the greenness modification effects in the two different provinces could be explained if data on vegetation types were available. Lastly, smoking could adversely affect TB treatment 39 and lower survival among patients with MDR-TB, 40 while greenness might attenuate the impact. We could not adjust for smoking as these data were neither available for individuals nor for communities during the follow-up. Future studies should consider adjusting for these potential confounding.
In conclusion, this large MDR-TB cohort study adds evidence of greenness which benefits survival by attenuating the association between PM 2.5 and mortality among patients with MDR-TB. Improving green space and air quality, along with effective treatments, may contribute to the reduction of mortality for MDR-TB.