Article Text
PDF
Abstract
Introduction Long-term exposure to fine particulate matter (≤2.5 µm (PM2.5)) has been associated with pulmonary tuberculosis (TB) notifications or incidence in recent publications. Studies quantifying the relative contribution of long-term PM2.5 on TB notifications have not been documented. We sought to perform a health impact assessment to estimate the PM2.5- attributable TB notifications during 2007–2017 in Ningxia Hui Autonomous Region (NHAR), China.
Methods PM2.5 attributable TB notifications were estimated at township level (n=358), stratified by age group and summed across NHAR. PM2.5-associated TB-notifications were estimated for total and anthropogenic PM2.5 mass and expressed as population attributable fractions (PAFs). The main analysis used effect and uncertainty estimates from our previous study in NHAR, defining a counterfactual of the lowest annual PM2.5 (30 µg/m3) level, above which we assumed excess TB notifications. Sensitivity analyses included counterfactuals based on the 5th (31 µg/m3) and 25th percentiles (38 µg/m3), and substituting effect estimates from a recent meta-analysis. We estimated the influence of PM2.5 concentrations, population growth and baseline TB-notification rates on PM2.5 attributable TB notifications.
Results Over 2007–2017, annual PM2.5 had an estimated average PAF of 31.2% (95% CI 22.4% to 38.7%) of TB notifications while the anthropogenic PAF was 12.2% (95% CI 9.2% to 14.5%). With 31 and 38 µg/m3 as counterfactuals, the PAFs were 29.2% (95% CI 20.9% to 36.3%) and 15.4% (95% CI 10.9% to 19.6%), respectively. PAF estimates under other assumptions ranged between 6.5% (95% CI 2.9% to 9.6%) and 13.7% (95% CI 6.2% to 19.9%) for total PM2.5, and 2.6% (95% CI 1.2% to 3.8%) to 5.8% (95% CI 2.7% to 8.2%) for anthropogenic PM2.5. Relative to 2007, overall changes in PM2.5 attributable TB notifications were due to reduced TB-notification rates (−23.8%), followed by decreasing PM2.5 (−6.2%), and population growth (+4.9%).
Conclusion We have demonstrated how the potential impact of historical or hypothetical air pollution reduction scenarios on TB notifications can be estimated, using public domain, PM2.5 and population data. The method may be transferrable to other settings where comparable TB-notification data are available.
- tuberculosis
- epidemiology
- China
Data availability statement
Data are available in a public, open access repository. Data are available on reasonable request. PM2.5 data are available at: https://sites.wustl.edu/acag/datasets/surface-pm2-5/. Population count data can be found at: https://hub.worldpop.org/geodata/listing?id=30. All other data are available on request from the corresponding author Igor Popovic (i.popovic@uqconnect.edu.au).
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
STRENGTHS AND LIMITATIONS OF THIS STUDY
Long time-series of individual-level tuberculosis (TB) notification data (spanning 11 years and sample size of 32 087 individuals diagnosed with pulmonary TB) used to derive aggregated age-specific TB notification rates.
Use of public domain, global coverage data sets for PM2.5 and population.
Sensitivity of health impact assessment calculations was tested to alternate inputs, including effect estimates and different counterfactual exposure levels.
Causality of the underlying association is not established.
Moderate sensitivity, likely due to uncertainties in concentration-response function, exposure error, counterfactual concentrations and TB notification rates.
Introduction
Ambient fine particulate matter (≤2.5 µm; PM2.5) is a leading environmental risk factor for morbidity and mortality.1 Exposure to ambient PM2.5 was estimated to contribute to over 3 million premature deaths and 100 million disability-adjusted life-years lost globally in 2019.2 Pulmonary Mycobacterium tuberculosis (TB) is also a major contributor to the global burden of disease, with over 10 million incident TB cases and 1.6 million deaths attributed to TB in 2021.3
Low-income to middle-income countries account for the greatest proportion of TB cases (>95%) and some of the highest levels of ambient PM2.5 (population-weighted annual mean PM2.5 from ~60% to ~75 µg/m3).4 5 The association between ambient air pollution and TB disease has attracted a modest but growing body of literature, as we highlighted in a 2019 systematic review,6 various individual studies7–26 and a meta-analysis27 published subsequent to our review. However, most studies have focused on TB-related outcomes among people who already have TB disease (eg, hospitalisations, mortality) while studies of long-term exposure and new TB disease notifications are comparatively scarce, especially in TB-endemic settings.6 20 24 We recently described a small-area analysis of long-term ambient air pollution exposure and TB notifications in Ningxia Hui Autonomous Region (NHAR), a largely rural, TB-endemic area in China.7 We observed consistent effects of annual average PM2.5 exposure on TB-notification rates during 2007–2017.7
Health impact assessments (HIAs) have been widely used to assess the burden of disease attributable to ambient air pollution and the potential health benefits associated with air pollution mitigation. Among other health outcomes, HIAs were used to examine the impact of outdoor air pollutants on global asthma incidence, chronic obstructive pulmonary and cardiovascular disease.28–31 The attractiveness of HIA is its ability to compare the health effects of real or hypothetical pollutant exposure scenarios in a consistent way, using existing effect sizes and uncertainty from both meta-analyses and individual studies. To our knowledge, no studies have systematically used an HIA to estimate long-term outdoor air pollution attributable TB cases, and sensitivity to input parameters in a high TB-burden setting.
We sought to demonstrate the utility of an HIA to estimate TB-notifications attributable to long-term total and anthropogenic PM2.5 mass in NHAR during 2007–2017. We also sought to assess the sensitivity of estimates to changes in the key HIA parameters that could influence the findings, including temporal variation in PM2.5 and other factors affecting TB notification rates.
Ningxia Hui Autonomous Region shows 5 prefecture-level cities and 358 township administrative divisions.39 The population refers to the total population in each prefecture.
Methods
Study location
NHAR (66 400 km2) is a provincial-level autonomous region located in north-western China (figure 1). Inhabited by 7.2 million people, NHAR is divided administratively into five prefecture-level cities, namely Yinchuan (the capital city), Shizuishan, Wuzhong, Guyuan and Zhongwei (figure 1).32 33 Subsequent divisions include districts and counties (n=19) followed by townships (n=358).34 NHAR is primarily an agricultural centre and is ranked 29/31 of the provincial economies in the country in terms of gross domestic product output in 2019.32 35 Compared with other provinces in China, NHAR has had high rates of TB since at least the late 1970s, which has extended into recent decades.36
Study population
TB is a notifiable disease in China. People with TB are reported to local and subsequently provincial and national-level public health authorities. An electronic reporting system was introduced in NHAR in 2005, managed by NHAR’s Provincial Centers for Disease Control and Prevention (CDC).36 People with clinical features suggestive of TB are diagnosed based on clinical assessments (chest radiograph and response to treatment) as well as laboratory confirmation, including sputum smear microscopy and/or sputum culture (adhering to International Standards for Tuberculosis Care).37 On diagnosis, each patient’s age, sex, residential address, occupation, ethnicity, date of symptom onset (self-reported), diagnostic method (clinical or bacteriological) and diagnosis date is recorded in the TB registry. This study focused on the years 2007–2017, as NHAR’s programme did not reach full coverage in 2005 and 2006.
As we have previously documented,7 residential addresses of all TB-notified cases in NHAR (2007–2017) were geocoded using geoChina package in RStudio V.1.1.238 and assigned to one of the 358 non-overlapping township-level divisions in NHAR with ArcGIS Pro (V.2.8.6).7 The aggregated TB-registry data were used as a proxy of annual township TB incidence. Reference population counts (stratified by 5-year age groupings up to 80+) were obtained from the WorldPop Open Population Repository.39 The population estimates (gridded at 100 m) were extracted in ArcGIS Pro and aggregated to township administrative divisions (figure 1).
Exposure assessment
We have described exposure assessment in detail elsewhere.7 Annual average PM2.5 exposure concentrations (for each year during 2007–2017) were estimated (gridded at ~1 km) and averaged over township-level areas (median area=144 km2; IQR=218 km2). These estimates were derived from V.4.CH.03 of the Atmospheric Composition Analysis Group.40 This dataset applies the regional methodology described in van Donkelaar et al,41 with satellite aerosol optical depth retrievals, chemical transport modelling and ground-based observations, but using the updated geophysical values of Hammer et al.42 In these PM2.5 models, ground-based observations over China are directly incorporated from 2014 onward. Due to the paucity of ground-based observations prior to 2014, earlier years are inferred from the interannual changes of the geophysical PM2.5 dataset. In cross-validation, the PM2.5 estimates captured 83% of variability in long-term mean PM2.5 in China (1665 ground monitoring sites) with root-mean-squared error (RMSE) of 6.8 µg/m3. Dust and sea-salt removed PM2.5 mass estimates (ie, to approximate the anthropogenic PM2.5 fraction) were based on applying simulated compositional information to the full composition PM2.5 values, as described by van Donkelaar et al.43
HIA calculations
It is an implied assumption of many HIAs, including ours, that there is a causal relationship between a pollutant and a health outcome, as the interpretation is usually an estimate of excess cases at a given exposure versus a comparator exposure. It is currently unclear if there is a causal relationship between long-term PM2.5 and TB disease incidence. The absence of a sufficient weight of evidence to suggest a causal association, however, does not preclude an HIA, as highlighted by some of the earliest and best-documented examples of the method.44 45
The HIA approach was informed by recent studies by other groups on non-TB health outcomes.28–31 For PM2.5, we observed a single pollutant IRR of 1.35 (95% CI 1.25 to 1.48), per 10 µg/m3 increase in township-level (n=358) average mass on TB notification rate, adjusted for age, sex, ethnicity, education attainment, occupation, remoteness (rural/urban), floor space per person and solid fuel use.7 In this study, we applied a log-linear concentration-response form, which was consistent with the association we observed in the previous study, to estimate the number of TB-notifications attributable to PM2.5, as a proxy for the broader mixture of pollutants, for each township and year during 2007–2017 as follows:
where PM2.5-related notificationst,a is the number of PM2.5 attributable TB notifications in township t for age group a, Pop is the population in township t for age group a, Inc is the baseline TB-notification rate across NHAR in age group a, β is the concentration-response factor of the slope of the relationship between PM2.5 and TB-notification rate described above, and Xt is the annual average PM2.5 concentration in township t.
The lowest observed annual average concentration of PM2.5 in NHAR (30 µg/m3) was set as the counterfactual exposure in the main analysis (ie, we assumed excess notification were attributable above 30 µg/m3). We constrained our estimates to the range from which we derived the concentration response, namely approximately 30 and 70 µg/m3 because it did not require assumptions on the shape of the relationship outside that range. Sensitivity analyses using other counterfactuals are described in the ‘Sensitivity analyses’ section.
Then PM2.5-attibutable notification estimates were calculated by 5-year age groups (up to 80+) per township per year and summed to obtain NHAR annual totals.28 The 0–4 and 5–9 age groups were excluded, given the very low number of TB notifications observed (n=86 over 11 years). Uncertainty in estimates was based on the 95% CIs of the IRR because they are well constrained.29 30 The number of PM2.5-attributable notification of cases was also expressed as population attributable fractions (PAFs, ie, the fraction of the total number of TB notifications in NHAR each year estimated to be attributable to PM2.5 exposure above the counterfactual), and overall rates (per 100 000 population, based on the crude age-specific rate in each age group).
While both natural and anthropogenic PM2.5 sources contribute to the total (natural plus anthropogenic fraction) mass, from a policy perspective, the anthropogenic fraction of PM2.5 can be more directly modified (eg, through changes in emissions or air quality standards). We estimated the anthropogenic fraction of PM2.5 attributable cases by subtracting those due to dust and sea salt from total mass within the concentration-response factor β.29 If the anthropogenic mass in any township was less than the counterfactual, we estimated the anthropogenic part by applying the ratio of natural to total mass to the cases attributable to total mass above the counterfactual and subtracting it from the same (ie, cases attributable above due to total mass) (online supplemental material). We tested for statistically significant linear trends in PAFs over time, for both total and anthropogenic PM2.5, using Sieve-bootstrap Student’s t-test for linear trends in ‘funtimes’ package RStudio V.1.4.1106.46
Supplemental material
Sensitivity analyses
We undertook additional sensitivity analyses to quantify how our estimates changed by varying key input parameters. While the main analysis parameters were selected for the reasons described above, we viewed the main and sensitivity analyses as equivalent and suggest readers do likewise. The sensitivity analyses were (1) substituting the pooled estimate and confidence intervals from a recent meta-analysis by Dimala and Kadia, based on six studies of annual PM2.5 and several proxies of TB incidence rates.27 This meta-analysis spanned a wide range of annual mean PM2.5 concentrations (16–100 µg/m3) than we observed and had a smaller effect size than our main analysis (RR 1.12; 95% CI 1.05 to 1.19 per 10 µg/m3) and (2) using counterfactuals based on the 5th (31 µg/m3) and 25th (38 µg/m3) percentiles of those estimated in NHAR by the exposure models.28 These were more conservative assumptions of the counterfactual (ie, leading to reduction in estimates of attributable notifications) and were empirically informed by the concentrations used to derive our effect estimate.7 28
Drivers of temporal change analysis
Given the 11-year time span, there may be contemporaneous secular or cyclical changes in key HIA parameters (ie, population change, changes postimplementation and uptake of TB-notification system in NHAR due to factors other than PM2.5 and PM2.5 concentrations) on PM2.5 attributable TB notifications. We attempted to disentangle the contribution of each factor using a decomposition analysis, following Anenberg et al 30 and Cohen et al.31 This entailed three sets of ‘rollback’ simulations in which each of the parameters was reverted to the base year (2007). The output for the individual rollback scenarios for each year (2008–2017) was used to estimate the contribution of each parameter to the change in PM2.5-associated TB-notifications between 2007 and all the other years (2008–2017). We elaborate further on this methodology in online supplemental material. We did not examine the interactions between the three parameters, for example, changing PM2.5 on baseline TB-notification rates over time, as they are likely to be negligible relative to the main effect of each.29 30
Patient and public involvement
Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.
Results
PM2.5 exposure concentrations
The median annual mean total PM2.5 during 2007–2017 was 42 µg/m3 (IQR=38–47 µg/m3) (figure 2). Higher concentrations of PM2.5 were observed in northern and western townships (Yinchuan prefecture) and central areas (Wuzhong and Zhongwei prefectures) of NHAR. The anthropogenic PM2.5 fraction had largely similar patterns (figure 2) to total PM2.5.
Annual average total (all composition) and anthropogenic (dust and sea salt removed) PM2.5 concentrations in NHAR (2007–2017). The minimum and maximum of the PM2.5 colour ramp were based on the range of concentrations in both data sets. PM2.5, particulate matter (≤2.5 µm).
Health impact assessment
An estimated 10 043 TB notifications (95% CI 7219 to 12 457), a PAF of 31.2% (95% CI 22.4% to 38.7%) were attributable to PM2.5 in NHAR during 2007–2017 (table 1). The annual proportion of TB-notifications attributable to PM2.5 (total mass) ranged from 42.6% (95% CI 31.2% to 52.0%) in 2013 to 19.0% (95% CI 13.5% to 23.8%) in 2016 and was variable from year to year (online supplemental table 1). There was no statistically significant linear trend over time (p=0.13) (online supplemental table 2).
Total number of TB-notifications attributable to PM2.5 in NHAR from 2007 to 2017
Anthropogenic PM2.5 was linked to an estimated 3933 (95% CI 2950 to 4677) notifications, a PAF of 12.2% (95% CI 9.2% to 14.5%) (table 1). The highest and lowest PAF of notifications associated with anthropogenic PM2.5 only was observed in 2013 (16.6%, 95% CI 12.9% to 19.2%) and 2016 (7.1%, 95% CI 5.3% to 8.6%), respectively (online supplemental table 1). The proportion of TB-notifications attributable to anthropogenic PM2.5 showed a modestly significant decreasing linear trend with a median annual rate of change of −1.2% (p=0.04) (online supplemental table 2).
Figure 3 shows maps of township level PM2.5 attributable TB notifications per 100 000 population for two illustrative years (2007 and 2017) for purpose of visualising indicative spatial variation. In those 2 years, the highest number of PM2.5 attributable TB notifications (between 40 and 50 cases per 100 000 population) was observed in northern and southern areas of Yinchuan and Wuzhong prefectures. While some townships in central and northern NHAR exhibited minimal temporal variability during the study period, reductions in PM2.5-attributable notifications occurred in southern areas of Guyuan and Zhongwei prefectures (figure 3).
Estimates of township-level TB notifications attributable to total PM2.5 mass (per 100 000 population) for illustrative years 2007 and 2017 in NHAR. NHAR, Ningxia Hui Autonomous Region; PM2.5, particulate matter (≤2.5 µm); TB, tuberculosis.
Drivers of change
The temporal variation in PM2.5 attributable TB notifications was influenced to different extents by the three factors investigated (TB notification rate, population growth and PM2.5 concentrations). Of these, the contribution to net change was predominantly driven by TB-notification rates (figure 4). A declining TB-notification rate contributed to a mean reduction of −23.8% in PM2.5 attributable TB notifications between 2007 and 2017, peaking in 2013 (−42.8% contribution). PM2.5 was the second-most dominant factor influencing the change in PM2.5 associated TB-notifications (mean contribution between 2007 and 2017 of −6.2%), with the greatest reduction in 2016 (−21.3%). The influence of decreasing PM2.5 levels on the overall declining net change in PM2.5 associated notifications was also offset by increases in PM2.5 concentrations in years 2009, 2010 and 2013. Following 2013, however, constant reductions in annual average PM2.5 concentrations contributed to a mean decline of −13.6% (range −6.9% to −21.3%) in PM2.5-associated notifications. On the contrary, population growth resulted in a modest but persistent increase (mean contribution of 4.9%) in the number of PM2.5 attributable TB notifications in all years and varied between 1.9% in 2008 and 10.4% in 2017.
Contribution of population growth, tuberculosis (TB)-notification rate and particulate matter (≤2.5 µm) (PM2.5) concentrations to changes in PM2.5 attributable TB notifications in NHAR compared with 2007 (base year) for the other years between 2008 and 2017 following Anenberg et al.30 Changes above 0 are those that led to increased PM2.5-attributable cases and those below 0 are those that reduced PM2.5-attributable cases. The net change represents the sum of each of the factors investigated to the change in PM2.5 associated TB-notifications in NHAR relative to 2007.
Sensitivity analysis
Compared with the main analysis, TB-notifications attributable to PM2.5 were ~6% (9393 attributable TB notifications, 95% CI 6731 to 11 685) and ~51% (4970 attributable TB notifications, 95% CI 3492 to 6298) lower when the 5th (31 µg/m3) and 25th (38 µg/m3) percentiles of annual average PM2.5 exposure in NHAR were used as counterfactual levels (table 1). The corresponding PAFs were 29.2% (95% CI 20.9% to 36.3%) and 15.4% (95% CI 10.9% to 19.6%), respectively (table 1). When we substituted in the PM2.5 effect from the recent meta-analysis,27 the number of PM2.5 attributable TB notifications was lower in both the main analysis (4402 notifications, 95% CI 1994 to 6408) when a counterfactual of 30 µg/m3 was used (table 1 and online supplemental table 3), and also when the 5th percentile (4093 notifications, 95% CI 1850 to 5970) and 25th percentiles (2085 notifications, 95% CI 929 to 3080) were employed (table 1 and online supplemental tables 4 and 5). The corresponding PAFs were 13.7% (95% CI 6.2% to 19.9%) in the main analysis, 12.7% (95% CI 5.8% to 18.6%) for 5th percentile counterfactual and 6.5% (95% CI 2.9% to 9.6%) for the 25th percentile counterfactual (table 1).
Discussion
Health impact assessment
To our knowledge, this is the first study to use an HIA to estimate the number of TB-notifications attributable to long-term ambient PM2.5, comparing total and anthropogenic PM2.5 mass, and assessing temporal change. The overall PAF of total PM2.5 mass (2007–2017) was 31.2%. Anthropogenic PM2.5 was associated with an overall PAF of 12.2%. We have also highlighted that, relative to 2007, the mean annual rate of decline in PM2.5-related TB notifications was −36%, with the greatest reduction (−65%) observed in 2016 in NHAR. We note the estimates were moderately sensitive to alternate effect sizes and counterfactual assumptions, and their combination. However, even the most conservative scenario (total PM2.5 PAF: 6.5%; anthropogenic PM2.5 PAF: 2.6%) suggested non-trivial benefits of PM2.5 reduction.
The burden of TB attributable to known modifiable risk factors, such as diabetes mellitus, tobacco smoking and alcohol consumption, among others, likely outweigh the potential impact of ambient air pollution. However, a reduction in PM2.5, as demonstrated in this study, may still translate to benefits through averted TB disease notifications, given the ubiquitous nature of the exposure.47 48 Here, our main interest was the utility of HIA as a way to compare the impacts of PM2.5 in TB notifications in NHAR, under different scenarios and assumptions, over time. Such information could be used, for example, to show the relative importance of a given PM2.5 reduction or increase scenario, in the context of the various other factors contributing to TB notifications in an area. An important consideration, however, is that the weight of evidence to support or refute causality for specific air pollutant and health outcome takes decades to accumulate.44 45
There has been one study to quantify PM2.5-attributable TB.25 That study used a Bayesian model to assess the effects of several measured pollutants, based on the upper part of the range for PM2.5 (25–35 µg/m3), and concentration-response function from an earlier cohort study.26 While interesting, the Bayesian framework is not directly comparable to ours. Also, we used public domain, global data sets that can be replicated in other locations while Lin et al used data specific to their study location, which may have reduced exposure uncertainty in their study. Other differences relate to the unit of TB-notifications being crude rates for four county-level areas in their study, although it is not stated if this considered the age–sex structure of the population, or if uncertainty of the crude rates, alternate concentration–response relationships and ranges, counterfactuals and temporal drivers were assessed.25
Key drivers of change in PM2.5 attributable TB notifications
The underlying TB-notification rate not related to PM2.5 (changes in access to healthcare, improved TB detection rate and other risk factors) was the dominant driver of temporal variability in PM2.5-associated notifications during the study period. Relative to 2007, TB-notification rate contributed to a persistent decrease in PM2.5-attributable TB notified cases in all years which ranged between −11.7% in 2008 and −42.8% in 2013. Overall, TB-notification rate resulted in a mean annual decline in PM2.5-associated TB notifications of −23.8%, which was substantially greater than the mean contribution of the other HIA parameters (PM2.5 mean contribution: −6.2%; population growth mean contribution: 4.9%).
The overall reduction in TB-notification rate in NHAR may be attributable to several factors, including the establishment of a standardised TB-notification electronic reporting system in addition to locally available treatment and monitoring as well as economic development and various health reforms.49 50 As part of the WHO TB control strategy and in addition to Directly Observed Treatment, Short Course, Ningxia’s Provincial CDC also developed a provincial-level electronic reporting system in 2005.36 Collectively, this WHO initiated programme facilitated standardised case detection and reporting as well as treatment and monitoring of TB notified cases by their local health provider or in the case of Ningxia, village clinics and township health centres. A major benefit of the initiative has been the decentralised administration of treatment and facilitation of represcription of medication for individuals who have been notified by the reporting system and entered the programme. Previously, management of TB treatment was coordinated centrally at the prefecture city level which contributed to delays in diagnosis and treatment, as patients often had to travel long distances and incurred significant costs in the process.51
Since 2000, emerging commercial ventures in manufacturing and agriculture have helped create greater employment opportunities and improve living standards in rural areas, including NHAR.52 China has also implemented various health reforms and funding initiatives including the Cooperative Medical Scheme.53 In 2009, for example, healthcare delivery was optimised through a three-tier rural health service approach consisting of county-level hospitals and more locally, township and village-level healthcare facilities. The three-tiered system has improved access for rural populations, increasing health service utilisation (village clinic visits and township hospital admissions) by 28% from 2006 to 2010.54–56
PM2.5 concentrations were the second-largest contributor to the change in PM2.5 attributable TB notifications. Compared to 2007, in 7 out of 10 years (year 2008, 2011–2012 and 2014–2017), reductions in PM2.5 concentrations contributed to a decline in PM2.5 attributable TB notifications, reaching a peak in 2016 of −21% (figure 4).
It is possible that some of the changes in PM2.5 attributable TB notifications during the study period, particularly from 2013 onwards, occurred due to flow-on effects from China’s Air Pollution Prevention and Control Action Plan (APPCAP), which was started that year and contributed to regional scale reductions in PM2.5 that may have indirectly benefited NHAR and surrounding areas.57–59 APPCAP involved more stringent industrial and vehicle emission standards, modernising outdated industrial infrastructure (upgrading industrial boilers) and promoting cleaner fuels and renewable energies in the residential sector.60 Compared with 2013 levels, these strategies contributed to an estimated reduction of 33.3% (95% CI 16.3% to 50.3%) or 36 µg/m3 (95 CI% 14 to 71) in monitored annual average PM2.5 concentrations by 2017 in areas of implementation (Beijing-Tianjin-Hebei, Yangtze River Delta and Pearl River Delta region). According to Zhang et al, the APPCAP also led to pronounced declines nationally in modelled annual mean PM2.5 of 20 µg/m3 (95% CI 18 to 21) from 2013 to 2017.60
Limitations
Our study has several limitations. First, we assumed, as do other HIAs, a causal association between PM2.5 and TB notifications. As we noted in the methods section, this is not established in the literature at this time, with very few cohort or nested case-control studies of TB disease incidence or proxies drawn from general population-based samples.6 27 Our findings should be considered in that context. Our study is also subject to specific uncertainties relating to the input parameters and assumptions, including, in approximate order of most to least impactful on results, the concentration-response function and counterfactual exposure concentrations, PM2.5 exposure errors, notification rates and population data.28
Second, the log-linear form of the relationship between PM2.5 and TB notifications means the uncertainty in effect estimate and/or exposure errors closer to the assumed counterfactual tend to have greater potential impact.28 We attempted to test sensitivity to alternate inputs, and used the 95% CIs around effect estimates, and different counterfactual concentrations for that reason, but we may still have overestimated or underestimated TB-notifications attributable to PM2.5 within the study area. Third, the error of total PM2.5 concentration estimates is only known at the regional level (ie, East Asia RMSE: 6.8 µg/m3). The uncertainty of the model in NHAR is unclear due to the small number of monitoring sites in Ningxia (n=18) and may, therefore, be greater error than that at regional level. Moreover, it is unknown in our study area how well the anthropogenic and natural components of PM2.5 mass are differentiated, and we did not have any ground observations for PM composition.
Fourth, there are uncertainties concerning the potential time lag between PM2.5 exposure and TB notifications and how this may have impacted our HIA estimates. In our previous study, a 3-year annual average exposure preceding TB notification was estimated, as the risk of developing active TB was found to be highest in the initial 2–3 years following infection and to account for delays in diagnosis from onset of symptoms to presenting to a health facility.36 We also undertook sensitivity analysis substituting exposure in the year of diagnosis. From that, we believe it is unlikely that different exposure lag times would have substantially influenced the observed effect estimates on yearly TB notification rates and our HIA estimates.
Fifth, the underlying TB-registry notification data may not reflect true TB-notification incidence due to the passive case-finding approach of NHAR’s TB programme. Only positive cases who have been admitted to health services were notified to the registry and may have also been underestimated.36 Overall, however, because we had individual-level notifications to derive the aggregated NHAR age-specific rates, this probably captured a reasonable approximation of those rates, and as such introduced less uncertainty than other HIA inputs that were used.
Conclusion
We demonstrated a method to assess the potential impact of annual PM2.5 exposure on TB notifications, in a predominantly rural, high TB burden region in China. A total of 10 043 TB notifications (~17 per 100 000 population per year) were estimated to be attributable to total PM2.5 mass in the 11-year study period (2007–2017). This represented 31.2% of all TB case notifications while estimates under alternative assumptions were between 6.5% and 29.2%. The number of notifications due to the anthropogenic component of PM2.5 was estimated to be 12.2%, which varied between 2.6% and 11.4% under other assumptions. In the period since 2013, reductions in annual average PM2.5 concentrations contributed to a mean decline of −13.6% in PM2.5 associated TB notifications. Our use of public domain, global data may be applicable to other locations, provided valid TB notifications are available, to perform similar estimates relevant to the local context.
Data availability statement
Data are available in a public, open access repository. Data are available on reasonable request. PM2.5 data are available at: https://sites.wustl.edu/acag/datasets/surface-pm2-5/. Population count data can be found at: https://hub.worldpop.org/geodata/listing?id=30. All other data are available on request from the corresponding author Igor Popovic (i.popovic@uqconnect.edu.au).
Ethics statements
Patient consent for publication
Ethics approval
This study involves human participants and was approved by the University of Queensland Ethics Committee on the 22 November 2018 (Ethics Approval Reference # 2018002073). Initial ethics clearance was granted by the Research and Ethics Committee of Ningxia Medical University on 10 March 2016 (Ethics Approval Reference # 2016-117) and renewed on 8 March 2019 (Ethics Renewal Reference # 2019-033). Participants gave informed consent to participate in the study before taking part.
Acknowledgments
The authors also acknowledge the invaluable contribution made by Ningxia’s TB surveillance registry and the Atmospheric Compositional Analysis Group of Washington University in St. Louis which have made the analyses presented in this paper possible.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Contributors IP, RSM, GM and LK conceptualised the study, curated the data, developed methods, edited and critically reviewed the manuscript. IP conducted formal analysis and interpretation of the data and prepared the original draft of the manuscript. YY and SY collected and processed TB notification data. JVB, EG, GF, B-YY, XW and G-HD edited and critically reviewed the manuscript. IP accepts full responsibility for the conduct of the study, had access to the data and controlled the decision to publish.
Funding YY acknowledges funding from the National Science Foundation of China (NSFC) (NSFC Project Grant 81460311).
Disclaimer The funding body had no role in the study or the decision to publish it.
Map disclaimer The inclusion of any map (including the depiction of any boundaries therein), or of any geographic or locational reference, does not imply the expression of any opinion whatsoever on the part of BMJ concerning the legal status of any country, territory, jurisdiction or area or of its authorities. Any such expression remains solely that of the relevant source and is not endorsed by BMJ. Maps are provided without any warranty of any kind, either express or implied.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.