Bijlage 4 Effectenstudie gezondheidsrisico s fijnstof en stikstofdioxide

As the air pollutants under consideration have no known health benefits, this indicated that careful consideration was necessary for any meta-analytic random-effects effect estimate with a two-sided P value of less than 0.10. It is important to realize that the adapted GRADE assessments apply to the whole body of evidence or to some part thereof based on, for example, a selection of studies at low or moderate RoB. No separate GRADE assessments were carried out for the – necessarily smaller number of – studies providing information at the lowest levels of exposure. GRADE assessments for a small number of studies are less robust. Key elements of GRADE such as RoB can be assessed for a smaller number of studies, and this was done where applicable. Step 2. Since the effect estimates examined in the systematic reviews were generally evaluated using linear models and existing evidence generally supports a linear or supralinear, no-threshold relationship for the pollutant–outcome pairs, there must be a procedure to determine the lowest level of observed (measured or modelled) exposure that is sufficiently underpinned with evidence and can, therefore, be used. Pragmatically, the GDG used as a starting point the 5th percentile of the exposure distributions from at least a few studies with the lowest levels of exposure (see below).

62

WHO GLOBAL AIR QUALITY GUIDELINES

The rationale was that below the 5th percentile of an exposure distribution, where data density tends to be sparse, there is typically little confidence in the shape of the CRF. This is evident, for example, from the CIs of splines reported in a number of relevant papers. Confidence depends on the study size. When there are no studies with narrow CIs for effect estimates down to the 5th percentile of the exposure distribution, a higher percentile can be chosen as a starting point. One would hesitate to use the 5th percentile of just one study, but the bodies of evidence considered for AQG level derivation varied considerably in terms of the numbers of studies included in the meta-analyses. In each case, the GDG made a pragmatic choice of studies to include. Step 3. Next, the GDG determined what amount or increase in mortality or other outcome above the lowest level would be considered a relevant increase. This is an a priori decision based on a judgement by the GDG. The GDG decided to use zero as a baseline when reviewing studies (i.e. any increase of the adverse health risk from the lowest long-term concentration – as defined in step 2  would be considered relevant). A zero increase represents a figure that comes closest to the ideal of having an AQG level that is based on health arguments only.

With a positive slope of the CRF at this lowest exposure level, any increase in exposure will result in a non-zero risk increase. See below, however, for a discussion of what zero means in practical terms, and how that differs for the derivation of long- and short-term AQG levels. Step 4. The lowest level – the mean of a number of observed 5th percentile concentrations, as defined in step 2  of measurement is the point above which the GDG assumed (with some confidence) that an increase in risk occurs. Since the GDG decided not to allow any increase in the adverse health risk from the lowest level measured, this is then the starting point for derivation of the AQG level. Step 5. The GDG established an AQG level for all critical health outcomes associated with a specific pollutant following steps 14. Of these, the lowest AQG level is recommended as the WHO AQG level for that pollutant. This will prevent the possibility that, for example, using an AQG level based on all-cause mortality would still allow a substantial amount of asthma to occur. For example, if the AQG level for asthma were lower than the all-cause mortality level, the AQG level based on the asthma outcome would be taken as the WHO AQG level. Step 6.

No separate GRADE assessments were carried out for the relatively few studies reporting the lowest levels of exposure since GRADE was applied to the whole body of evidence and not to single studies.

GUIDELINE DEVELOPMENT PROCESS

63

Nevertheless, a critical discussion was warranted on the merits of studies reporting the lowest exposure levels. This discussion started from an assessment of the RoB, which was conducted at the individual study level. If a study that found a low exposure level was deemed to be at high RoB, then it was excluded from consideration unless the GDG had sound reasons to disagree with this assessment in the relevant systematic review. The GDG also considered whether studies conducted at the lowest exposure levels continued to show increased RRs. Step 7. The systematic reviews concluded their literature searches in early autumn of 2018. Since then, several relevant studies have been published. The GDG considered new evidence up to the meeting in June 2020  after verifying that it met the same standards for inclusion as the studies already included. Step 8. The GDG reconsidered causality of associations for all pollutant–outcome pairs. However, as all pollutant–outcome pairs were considered worthy of further consideration at the start of the process in 2016, such considerations generally did not prevent recommendations of an AQG level whenever the epidemiological evidence was considered to be of moderate or high certainty. Specifically, the GDG referred to the causality assessments shown in Table 2.1, which formed the basis of the current AQG level development process.

The assessment was updated, when necessary, to include newer evaluations published since 2016. These updates are all discussed at the end of section 2.3.3. The steps outlined above produce a rounded integer value as a starting point for AQG level development. This starting point is not equivalent to a threshold of no effect: it is merely a level below which there is less certainty about the existence of an effect. Where there was no threshold, the starting point level was associated with some effect on health. The magnitude of this effect could be estimated from the meta-analytic effect estimate from the systematic review by assuming that, in the absence of a threshold, any level of exposure increases risk. It was useful to do this as a benchmark for comparing the starting points for long-term AQG levels between the pollutants PM2.5, PM10, ozone and nitrogen dioxide. It also provided a benchmark for comparing estimated health effects between long- and short-term AQG levels for the same pollutant.

2.5.2 Formulation of short-term AQG levels There are fundamental differences between AQG levels for short-term and longterm exposures. For long-term exposures, AQG levels are derived based on the lowest long-term exposures that are, with at least moderate certainty, associated with adverse health effects.

64

WHO GLOBAL AIR QUALITY GUIDELINES

Such guidelines are typically expressed as annual averages. Daily and hourly concentrations vary around the annual average, often in a lognormal distribution. If short-term AQG levels are derived based on lowest short-term exposures that are – with at least moderate certainty – associated with adverse health effects, then much lower values are obtained than those determined for long-term AQG levels. The caveat about evidence of less than moderate certainty expressed in section 2.5.1 also applies here. Importantly, the short-term variation in air pollution concentrations is largely driven by meteorology, which cannot be controlled. Short-term guidelines are typically defined as a high percentile of the distribution of daily values, for example the 98th or 99th percentiles equivalent to seven or three days a year exceeding this value (i.e. exceedance days). The rationale for choosing a high percentile and not the maximum is that the maximum of daily values for a given year is a less stable statistic than the high percentiles. For locations in which concentrations are below the annual mean AQG level, days with such high daily mean concentrations will be rare and a large proportion of days will have concentrations below the annual mean AQG level.

Thus, the health burden related to a few days with higher concentrations corresponds to a very small fraction of the total air pollution-related burden. In contrast, the long-term variation in air pollution concentrations is largely driven by spatial variation in air pollution sources and emissions, which can be controlled, although control for some sources such as desert dust, pose unique and much more considerable challenges. Typically, the magnitude of the health effects associated with variations in long-term exposure is larger, per mass unit, than the magnitude of the health effects associated with short-term variations. As a consequence, long-term AQG levels for most health outcomes are more health protective than short-term AQG levels. In such instances, the long-term AQG level is used to derive a short-term AQG level whenever the same health effect is considered (e.g. mortality) for both long- and short-term exposures. According to this line of reasoning, all eight steps outlined for long-term AQG level development remain valid for short-term AQG level development, except for step 3 defining the minimal relevant increase in health outcomes.

2.5.2.1 Procedure In keeping with established practice, as a starting point, short-term AQG levels were considered by the GDG as the 99th percentiles of daily concentrations empirically observed in distributions with a mean equal to the long-term AQG level, for pollutant–outcome pairs for which a long-term AQG level is also being recommended. This is the case for all-cause mortality and PM2.5, PM10, ozone and

GUIDELINE DEVELOPMENT PROCESS

65

nitrogen dioxide. It is also the case for cause-specific mortality and PM2.5 and PM10. The GDG evaluated the percentage of excess daily deaths expected from the meta-analytic linear short-term effect estimate to occur at a day at the 99th percentile of the distribution of daily, 24-hour average concentrations, compared with a day at the annual mean guideline concentration. In the cases of sulfur dioxide and all-cause mortality and hospital admissions and emergency room visits related to asthma and of carbon monoxide and myocardial infarction, no long-term AQG levels were recommended and there are no long-term AQG levels from 2005. The same approach as described at the beginning of step 2 was followed, by evaluating the percentage of excess daily deaths expected from the meta-analytic linear effect estimate to occur at the 99th percentile, relative to a specified and justified low concentration. The rationale for the long-term reference concentrations of sulfur dioxide and carbon monoxide is discussed in Chapter 3. Once the starting point for the short-term AQG level was calculated, it was rounded to the nearest integer value.

The rationale for having short-term AQG levels next to long-term AQG levels for the same pollutant is based on the need to provide air quality managers, health-care providers, vulnerable patients and the general population with tools to communicate health risks and short-term emission controls. The GDG notes that there is substantial evidence that some susceptible groups may be harmed by short-term elevations of some pollutants: those with asthma, coronary heart disease, COPD and other chronic conditions and diseases. Overall, these susceptible groups represent a substantial proportion of the population in many countries. The rationale for having short-term AQG levels in the absence of long-term AQG levels is typically based on documented acute elevation of risk over timescales of minutes to one or a few days. More detailed advice to policy-makers and air quality managers is provided in Chapter 6 of these guidelines. In this protocol a distinction is made between three different scenarios (Table 2.6. Scenario 1. In the first scenario, internally consistent long- and short-term AQG levels is desired, and the argument is that meeting the long-term AQG level protects against serious short-term effects on mortality. This can be shown for PM, nitrogen dioxide and ozone.

66

WHO GLOBAL AIR QUALITY GUIDELINES

Table 2.6. Scenarios in formulation of short-term AQG levels Scenario

Description

Scenario 1

Development of a short-term AQG level for a pollutant for which a long-term AQG level for the same outcome was developed (e.g. all-cause mortality)

Scenario 2

Development of a short-term AQG level for a pollutant for which a long-term AQG level was developed for another outcome (e.g. hospital admissions and emergency room visits related to asthma versus all-cause mortality)

Scenario 3

Development of a short-term AQG level for a pollutant for which no long-term AQG level was developed

First, for PM2.5 and all-cause mortality, in Global update 2005 the annual mean air quality guideline is 10 µg/m and the short-term 99th percentile 24-hour average air quality guideline is 25 µg/m so the ratio between short-term and long-term guideline values was 2.5. At the time, this ratio was not empirically underpinned; the ratio was simply said to be 2.5, with some recognition that it may vary from place to place and from time to time. There is now a very large database – including the 652 cities from the Liu et al. 2019 paper – to document the ratios between higher percentiles of the distributions of 24-hour average concentrations and the annual means. The GDG recommends using the same ratio everywhere for the purpose of deriving a 24-hour average AQG level. The primary motivation is that short-term effect estimates for PM2.5 and all-cause mortality do not significantly vary between different regions of the world. Note that there are differences in effect estimates depending on PM2.5 level, but that is not important when deriving AQG levels for relatively low short-term concentrations; it is important when quantifying the health burdens associated with the higher interim target levels.) The database from the MCC Collaborative Research Network (A.

Gasparrini, London School of Hygiene and Tropical Medicine, unpublished data, 23 June 2020 has descriptive data on long time series of daily average PM2.5 and PM10 concentrations from 384 and 480 cities, respectively. The ratio of the 99th percentile of the daily average concentrations to the multiyear mean is 3.05 for PM2.5, 2.85 for PM10, 2.34 for nitrogen dioxide (398 cities), 2.05 for ozone 244 cities), 3.90 for sulfur dioxide (396 cities) and 2.97 for carbon monoxide 349 cities).

GUIDELINE DEVELOPMENT PROCESS

67

Based on this database, the Network has published a series of articles, which are published as open access (Chen et al., 2021; Liu et al., 2019; Meng et al., 2021; Vicedo-Cabrera et al., 2020. When considering long- and short-term AQG levels for ozone, the GDG realized that long-term AQG levels could be based on the mean peak-season ozone levels, which have a different relationship to the 99th percentile of the daily distributions than the annual means. As an example, the GDG recommended setting the long-term AQG level for PM2.5 at 5 µg/m, and if a ratio of 3 were used to calculate the corresponding 99th percentile of daily means, a 24-hour AQG level of 15 µg/m would be derived. All the recommendations can be found in Chapter 3. Scenario 2. In the second scenario, there may be long-term AQG levels for nitrogen dioxide and ozone based on effect estimates for respiratory mortality, and short-term AQG levels based on effect estimates for all-cause mortality only. For PM2.5 and PM10, there are long- and short-term effect estimates for all-cause mortality as well as a number of cause-specific mortalities. In most cases, these are from the same studies, so there are no serious differences between the 5th percentiles of PM in the lowest-level studies for natural-cause and causespecific mortality.

If there are differences, the expectation is that in the smaller numbers of cause-specific mortality studies the 5th percentiles of the concentration distributions are more likely to be higher than lower, compared with the all-cause mortality studies. The general pattern is that effect estimates for both long- and short-term cause-specific mortality are somewhat bigger than those for all-cause mortality. This is always true for PM2.5; the picture for PM10 is a bit more mixed. Nevertheless, these patterns do not lead to different conclusions for AQG level derivation based on all-cause mortality as compared with AQG levels based on cause-specific mortality. This again assumes that the 5th percentiles from the lowest-level studies are not lower for cause-specific mortality studies than for all-cause mortality studies. There will be short-term AQG levels for ozone, nitrogen dioxide and sulfur dioxide and hospital admissions and emergency room visits related to asthma, and for carbon monoxide and hospital admissions and emergency room visits related to myocardial infarction. The GDG recommends (for ozone and nitrogen dioxide) to start from the long-term AQG level based on mortality, look at the internally consistent short-term AQG level for mortality and then quantify the effect on hospital admissions and emergency room visits related to asthma at that level.

Here, too, data from the Liu et al. 2019 collaboration provide insight into the ratios between 99th percentiles and annual means for ozone as well as for nitrogen dioxide.

68

WHO GLOBAL AIR QUALITY GUIDELINES

A judgement on whether one or the other effect should drive the short-term AQG level is then needed, with potential consequences for consistency between longand short-term AQG levels. As mentioned before, for ozone and nitrogen dioxide, only short-term CRFs for all-cause mortality are available. Therefore, comparisons (as outlined for PM above) are not always possible. Scenario 3. Lastly, in the third scenario, a short-term AQG level needs to be derived for a pollutant for which no long-term AQG level is being developed. A case in point is sulfur dioxide, for which there are systematic reviews of short-term 24-hour associations with all-cause and respiratory mortality and asthma hospital admissions and emergency room visits. For carbon monoxide, there is a 24-hour AQG level based on the systematic review of associations between 24-hour mean carbon monoxide concentrations and hospital admissions and emergency room visits due to myocardial infarction, which can then be compared with the air quality guideline of 7 mg/m for carbon monoxide indoors (WHO, 2014b) and the shorter-term AQG levels as well. To develop an AQG level for a particular pollutant–outcome pair, the GDG examined external evidence for causality of the pollutant–outcome association. Causality judgements were part of the process that produced the PECOS questions for the current process (Table 2.1.

In the case of hospital admissions and emergency room visits related to asthma and myocardial infarction, further adaptations were needed to compare visits/ admissions to deaths. The GDG specified short-term AQG levels for hospital admissions and emergency room visits related to asthma or myocardial infarction based on quantification of the expected increase in such visits/admissions at the proposed short-term AQG level. This recognizes that the health burden related to a few days (three to four per year when using 99th percentiles) with higher concentrations corresponds to a very small fraction of the total air pollutionrelated burden.

2.5.3 Formulation of interim targets and good practice statements 2.5.3.1 Interim targets Interim targets were introduced in Global update 2005 as additional integral elements of guidance, designed to complement the WHO air quality guidelines. Interim targets may be defined as air pollutant concentrations associated with a

GUIDELINE DEVELOPMENT PROCESS

69

specific decrease in health risk that serve as “incremental steps in progressive reduction of air pollution […] intended for use in areas where pollution is high” WHO Regional Office for Europe, 2006. As stated in Global update 2005, “countries may find these interim targets helpful in gauging progress over time in the difficult process of steadily reducing population exposures [to air pollution]”. Moreover, interim targets “aim to promote a shift from high air pollutant concentrations, with acute and serious health consequences, to lower concentrations”WHO Regional Office for Europe, 2006, in line with the AQG levels. Further: [i]f these [interim] targets were to be achieved, one could expect significant reductions in risks for acute and chronic human health effects from air pollution. Progress towards the guideline values should, however, be the ultimate objective of air quality management and health risk reduction in all areas WHO Regional Office for Europe, 2006.

The GDG decided that interim targets, and specifically the 2005 interim targets, should be retained in the updated air quality guidelines for two reasons. ■

■

The first is to promote continuous air quality improvement in places with high levels of ambient air pollution with the goal of achieving AQG levels as expeditiously as possible. Interim targets for reduction of air pollution have been shown to be achievable with abatement measures and have practical value in that several countries have standards equal to some of the interim targets (Kutlar Joss et al., 2017. Importantly, interim targets also have been helpful in achieving AQG levels. The second is to maintain continuity. Policy-makers, nongovernmental organizations and the scientific community in low- and middle-income countries are already familiar with the 2005 interim targets and have employed them since their introduction 15 years ago. Changing the interim targets at this point would be confusing and unnecessary because the interim target levels are still globally relevant, although the 2005 air quality guideline would be added as an interim target in the event that the AQG level is lowered.

Descriptors for each interim target have been provided to inform decision-makers of the implications of achieving the corresponding air pollutant concentrations. These are the risk descriptors calculated using updated CRFs. Lastly, the results of simulating a reduction of the 2016 burden of disease attributable to PM2.5 to the interim target and the new AQG level are provided in

70

WHO GLOBAL AIR QUALITY GUIDELINES

section 3.9, in order to illustrate the mortality and disability adjusted life-year benefits that could be achieved by expeditiously reducing air pollutant levels Evangelopoulos et al., 2020. 2.5.3.2 Good practice statements The WHO handbook for guideline development, 2nd edition WHO, 2014a), provides for the development of good practice statements in certain cases. This occurs when a GDG is confident that a large body of diverse evidence that is difficult to synthesize indicates that the desirable effects of a particular course of action far outweigh its undesirable effects. In other words, when a GDG is confident that implementing a measure would be beneficial with high certainty but when conducting numerous systematic reviews and detailed assessments of evidence would be a poor use of resources (WHO, 2014c). The evidence considered may be of a diverse nature, including linked or indirect evidence, physical and biochemical properties, ethical principles and human rights conventions (WHO, 2019a). Along these lines, the types of evidence that the GDG may consider in the context of air quality guidelines would also include air quality management principles and good practices implemented by reputable institutions.

The option of developing good practice statements was used to provide muchneeded guidance in relation to some specific types of PM identified as critical in the preliminary phase. The GDG chose to closely follow-up major external reviews on BC/EC, UFP and SDS throughout the process. The decision was made to develop good practice statements for these, rather than numerical AQG levels, in the absence of clear quantitative evidence on independent health effects from these pollutants.

GUIDELINE DEVELOPMENT PROCESS

71

3

Recommendations on classical air 3.Recommendations pollutants on classical air pollutants

3.1 Introduction This chapter presents specific recommendations on air quality guideline (AQG levels for the pollutants PM2.5, PM10, ozone, nitrogen dioxide, sulfur dioxide and carbon monoxide, together with the corresponding interim targets. In Chapter 2, a detailed protocol was described that was followed to derive AQG levels for the pollutants and averaging times. Chapter 2 also provide the rationales for including the specific pollutant–outcome associations that formed the basis for the recommendations given in this chapter. The averaging times considered were long term (annual mean or, for ozone, highest six-month average) and short term (24 hours). Long-term effects were considered only for all-cause and cause-specific mortality (PM2.5, PM10, ozone and nitrogen dioxide). For those, any pollutant-attributed increase in long-term mortality was considered harmful. Short-term effects were considered for all non-accidental and cause-specific mortality (PM2.5, PM10, ozone, nitrogen dioxide and sulfur dioxide), for asthma hospital admissions and emergency room visits (ozone, nitrogen dioxide and sulfur dioxide), and for myocardial infarction hospital admissions and emergency room visits (carbon monoxide only).

When both long- and short-term AQG levels were considered for a pollutant–outcome pair, preference was given to the long-term AQG level and the short-term AQG level was aligned using empirical data on frequency distributions of 24-hour concentrations. When only short-term AQG levels were considered, analogy with other pollutant–outcome pairs was used. Information about all the specific pollutant–outcome pairs reviewed can be found in the systematic reviews of evidence available in a special issue of Environment International Whaley et al., 2021.

3.2 PM2.5 3.2.1 General description The general description comes from Global update 2005. PM in urban and non-urban environments is a complex mixture with components having diverse chemical and physical characteristics. Research on PM and the interpretation of research findings on exposure and risk are complicated by this heterogeneity, and the possibility that the potential of particles to cause injury varies with size and other physical characteristics, chemical composition and source(s). Different characteristics of PM may be relevant to different health effects. Newer research findings continue to highlight this complexity and the dynamic nature of airborne PM, as it is formed either primarily or secondarily

74

WHO GLOBAL AIR QUALITY GUIDELINES

and then continues to undergo chemical and physical transformation in the atmosphere. Nonetheless, particles are still generally classified by their aerodynamic properties, because these determine transport and removal processes in the air and deposition sites and clearance pathways within the respiratory tract. The aerodynamic diameter is used as the summary indicator of particle size; the aerodynamic diameter corresponds to the size of a unit-density sphere with the same aerodynamic characteristics as the particle of interest. The differences in aerodynamic properties among particles are exploited by many particle sampling techniques (WHO Regional Office for Europe, 2006.

The focus in recent decades has been on particles with aerodynamic diameters of less than or equal to 2.5 µm (PM2.5) or 10 µm (PM10).

3.2.2 Recommended AQG level for long-term exposure to PM2.5

Based on the methods for deriving an AQG level outlined in the guideline development protocol in Chapter 2, this section provides a recommendation for an annual AQG level for PM2.5 that is based on all non-accidental mortality and cause-specific mortality (Table 3.1. The epidemiological evidence underpinning the AQG level is discussed in a systematic review commissioned by WHO, which is referred to in section 2.4. The review of this pollutant (Chen & Hoek, 2020 was published in Environment International Whaley et al., 2021 as open access. As discussed in section 2.3, there has been no separate, independent assessment of the mechanistic, toxicological and human clinical studies relating ambient particles to human health. The recommendations in this chapter follow the eight steps outlined in the protocol for AQG level development in Chapter 2 (section 2.5. The tables and figures mentioned during the eight steps are listed at the end of the discussion of each recommendation. Step 1. Assess RR estimates and, when available, CRFs The systematic review on PM2.5 and all non-accidental mortality (Chen & Hoek, 2020 reported a meta-analytic effect estimate of RR of 1.08 95% CI 1.061.09 per 10 µg/m³ PM2.5, assuming a linear relationship. The authors found an indication of a supralinear relationship, suggesting a steeper risk increase at lower exposure levels.

RECOMMENDATIONS ON CLASSICAL AIR POLLUTANTS

75

The certainty of the evidence was considered high according to GRADE. CRFs were provided by several studies. These are shown in Fig. 3.1, Fig. 3.2, Fig. 3.3 and Fig. 3.4 (which follow a discussion of the eight steps) for the studies with information at low to very low levels of measured exposure (step 2 Pinault et al., 2016, 2017; Villeneuve et al., 2015; Di et al., 2017a). CRFs were published from four of the six studies with the lowest exposure levels. Two studies did not provide a CRF Weichenthal et al., 2014; Cakmak et al., 2018. For obvious reasons, the uncertainty in the shape of the CRFs is higher in single than in pooled studies, and higher in small than in large studies. Very large studies such as the study by Di et al. 2017a) provide the best evidence for the shape of the CRF at the low end of the exposure range. These shapes generally show linear relationships down to very low concentrations or somewhat steeper curves at low than at higher concentrations. Step 2. Determine the lowest level of exposure measured In 18 of the 25 studies included in the meta-analysis, a 5th percentile of the exposure distribution was reported or could be calculated from the reported mean and standard deviation (Table 3.2. As the concentration distributions are often lognormal, this calculation is not straightforward.

Therefore, preference was given to actual reports of the relevant numbers obtained from the published papers or upon request from the study authors. This is indicated in Table 3.2, Table 3.3, Table 3.4 and Table 3.5. The five lowest levels reported or estimated in these studies were 3.0 µg/m³ Pinault et al., 2016, 3.2 µg/m³ Cakmak et al., 2018, 3.5 µg/m³ Pinault et al., 2017, 4.8 µg/m³ Villeneuve et al., 2015 and 6.7 µg/m³ Weichenthal et al., 2014. Weichenthal et al. 2014 found no effect. The Villeneuve et al. 2015 study provided no evidence of an effect of PM2.5 on all non-accidental mortality below 8 µg/m³. The study by Di et al. 2017a) has the next lowest 5th percentile (7.1 µg/m³) and the study by Hart et al. 2015 the next lowest (7.8 µg/m³). The average PM2.5 level across these five studies with the lowest exposure measurements in the systematic review is 4.2 µg/m³. A sensitivity analysis disregarding the Villeneuve et al. 2015 and Weichenthal et al. 2014 studies produced a mean of 4.9 µg/m³ PM2.5. The sum of weights in the meta-analysis was > 25%, indicating that these studies were influential in the meta-analysis. Step 3. Determine the minimal relevant increase in health outcomes The GDG decided to consider as relevant any increase in risk for an adverse health outcome related to long-term exposure to a pollutant.

76

WHO GLOBAL AIR QUALITY GUIDELINES