Bijlage 4 Effectenstudie gezondheidsrisico s fijnstof en stikstofdioxide

Instead, general risk management principles, based on decades of experience, are summarized in Chapter 6, on implementation of the guidelines. In addition, individual-level interventions, such as the use of personal respiratory protection (e.g. masks, respirators), air purifiers and behavioural measures, are not addressed here but in a report from a separate WHO consultation (WHO, 2020a).

2.3.1 Preliminary consultation Following the conclusions from the Review of evidence on health aspects of air pollution (REVIHAAP project (WHO Regional Office for Europe, 2013a), WHO organized an expert consultation in Bonn in September–October 2015 as a first step for this update of the air quality guidelines. The objective was to gather expert opinion and guidance in order to identify and discuss the latest available evidence on health effects of air pollutants and interventions to reduce exposure to air pollution for the purpose of informing this update of the air quality guidelines. Twenty-eight participants – representing a wide array of expertise and geographical locations – attended the consultation, which included not only a review of the available scientific evidence on a number of ambient air pollutants but also methodological issues and the implications of exposure and intervention studies. Experts recommended that a focus of these guidelines on pollutant-

GUIDELINE DEVELOPMENT PROCESS

31

specific risk assessment was still appropriate and prioritized 32 air pollutants according to four categories to reflect their relative importance in the context of updating the air quality guidelines. Since reviewing the evidence systematically for all air pollutants was infeasible considering the available resources, experts suggested prioritizing the pollutants PM2.5, PM10, ozone, nitrogen dioxide, sulfur dioxide and carbon monoxide for this update. This advice was based on the large body of new health-related evidence that had been published since Global update 2005 WHO Regional Office for Europe, 2016a).

2.3.2 Selection of priority pollutants The final selection of air pollutants took place in September 2016, during the first meeting of the GDG. Prior to the meeting, WHO surveyed GDG members on the final list of air pollutants to be included in this update of the air quality guidelines. The air pollutants identified in the global expert consultation, together with a number of different health outcomes, were included in the survey. In the ensuing discussion, the GDG decided to develop AQG levels for PM10, PM2.5, nitrogen dioxide, ozone, sulfur dioxide and carbon monoxide in relation to health outcomes critical for decision-making and for relevant averaging times. For PM types such as BC/EC and UFP, the GDG agreed that AQG levels for these pollutants could not be formulated owing to the absence of clear quantitative evidence on independent health effects. However, the results of reviews of evidence conducted by other groups would be examined in order to reach a better-informed decision about whether recommendations should be formulated and in which form. Likewise, for SDS, the GDG agreed that any recommendation would likely be qualitative in nature and geared towards guiding countries in moving towards mitigation and adaptation measures.

Following presentations by invited experts, the GDG, at its third meeting, decided to include guidance on BC/EC, UFP and SDS in the form of good practice statements.

2.3.3 Prioritization of health outcomes In order to define the pollutant–outcome pairs that would be systematically reviewed to inform the formulation of AQG levels, the GDG developed a prioritization framework based on the considerations outlined in Box 2.1.

32

WHO GLOBAL AIR QUALITY GUIDELINES

Box 2.1. Health outcome prioritization framework

■

Evidence on causality for a health outcome would be considered first, according to the latest determination (causal or likely causal) from the Health Canada, the International Agency for Research on Cancer, US EPA or other available integrated science assessments. As mandated by the Clean Air Act, the US EPA periodically reviews all scientific evidence about the health effects of so-called criteria pollutants, including all five pollutants considered in this report. These Integrated Science Assessments (ISAs) include a structured analysis of all evidence – including from toxicology – that supports a classification of a specific effect being causal, likely causal, suggestive of a causal relationship, inadequate to infer a causal relationship or not likely to be a causal relationship. For details, see Owens et al. 2017. These classifications have been used in support of identification of the relevant pollutant–outcome associations addressed in this document.

■

Where causality is not determined to be proven or likely (e.g. suggestive causality), the precautionary principle would be used when determining which additional most-severe health outcomes could be included. These outcomes would be based on other considerations such as contribution to burden of disease (e.g. prevalence of disease, disability weight), policy implications and expected increase in exposure to a pollutant in the future.

■

Causality determination would supersede the severity of a health outcome but, in some cases, two (or more) different health outcomes might be systematically evaluated for the same pollutant (e.g. one with a causal or likely causal link to the pollutant, and another health outcome for which the evidence is suggestive only but which is very severe or prevalent in the population). Severity of disease would be informed by considerations proposed by the latest update of the joint European Respiratory Society and American Thoracic Society policy statement on health effects from air pollution (fatality, persistence of effect, susceptible groups and medical/ functional significance, including loss of autonomy and reduced quality of life) (Thurston et al., 2017.

■

Lastly, as health outcomes can be assessed in various ways in studies, the specific health outcome measure(s) would be identified, based on evidence and the expert judgement of the GDG, to be used for quantitative health risk assessment in the guidelines.

GUIDELINE DEVELOPMENT PROCESS

33

By applying the prioritization framework, the GDG identified the following critical health outcomes associated with the selected air pollutants: ■

■

■

■

all-cause (non-accidental) mortality4 (hereafter referred to as all-cause mortality); cause-specific mortality, as per the International Statistical Classification of Diseases and Related Health Problems, 10th edition (ICD10, 2016 version WHO, 2016b): cardiovascular (ICD10 codes I00I99, lung cancer ICD10 codes C30C39 and respiratory (ICD10 codes J00J99; hospital admissions and emergency room visits related to asthma ICD10 code J45; and hospital admissions and emergency room visits related to IHD ICD10 codes I20I25; ultimately restricted to myocardial infarction, ICD10 codes I21I22.5

The pollutant–outcome pairs that were included comprise those for which there is broad scientific consensus regarding the causal nature of the reported relationships; others were chosen based on the strength of the epidemiological evidence. Table 2.1 gives an overview of the different considerations included in the prioritization process. This table was adapted from the document resulting from the population, exposure, comparator, outcome and study design (PECOS process (see section 2.3.4 finalized in November 2016. Since then, it was decided to include one more pollutant–outcome pair: the association between exposure to short-term sulfur dioxide and all-cause and respiratory mortality. This was done to provide continuity with Global update 2005, as explained below. The GDG recognizes that associations for many more pollutant–outcome pairs have been reported and reviewed in the literature. It would be practically impossible, given the resources available for the current guideline update, to include all of these for consideration and review.

 In an epidemiological study of air quality and health, all-cause mortality (ICD10 code A00Z99 refers to all deaths, and non-accidental mortality (ICD10 code A00R99 includes all deaths with the exception of deaths due to accidents, murder, suicide, etc. Although all-cause mortality includes accidental deaths, the proportion of deaths causes by accidents, etc. is typically small (< 10% in comparison with the other causes of death.  The systematic review by Lee et al. 2020 focused on myocardial infarction (ICD10 codes I21I22 as the only IHD outcome because it is not possible to establish the precise time of onset of other IHD outcomes. Further, other conditions within the spectrum of IHD are routinely managed in outpatient settings rather than in the emergency room/ department or hospital wards.

34

WHO GLOBAL AIR QUALITY GUIDELINES

The GDG, however, sees no grounds for assuming that the AQG levels, as derived in Chapter 3, would be very different if more outcomes would have been considered for the pollutants that were included. Obviously, this does not apply to those pollutants that were not considered at all in the current update. A draft version of this document has been reviewed by a large number of experts and stakeholder organizations, and no examples have been provided that would change this assessment. The GDG also emphasizes that there has been no separate, independent assessment of the mechanistic, toxicological and human clinical studies relating air pollution to human health. However, comprehensive evaluations by authoritative bodies such the Committee on the Medical Effects of Air Pollutants (COMEAP in the United Kingdom, Health Canada and the US EPA were taken into account in the development of the AQG levels. Information about all the specific pollutant–outcome pairs reviewed can be found in the systematic reviews of evidence available in a virtual special issue of Environment International entitled Update of the WHO global air quality guidelines: systematic reviews Whaley et al., 2021.

GUIDELINE DEVELOPMENT PROCESS

35

36

Health outcomes used in Global update 2005

Total, cardiopulmonary and lung cancer mortality

Pollutant

PM2.5 and PM10

LONGTERM EXPOSURE

♦

♦

♦

♦

Lung cancer mortality

Respiratory mortality (any, COPD, acute lower respiratory infections)

PM2.5 Causal for cardiovascular and respiratory mortality (US EPA, 2009 ♦ Causal for total and cardiovascular mortality (Health Canada, 2013

♦

PM2.5 Likely causal for respiratory effects (US EPA, 2009 ♦ Likely causal for respiratory effects (Health Canada, 2013

OTHER RELEVANT CAUSAL DETERMINATIONS REFERENCE

PM10 ♦ Health outcome supported by evidence from PM10 and PM2.5

SUPPORTING CONSIDERATIONS

PM ♦ Causal for total mortality in relation to PM Health Canada, 2013 ♦ Group 1ᵇ lung cancer for PM Straif et al., 2013 ♦ Likely causal for lung cancer mortality in relation to PM Health Canada, 2013

♦

CAUSALITY DETERMINATION REFERENCE

Cardiovascular mortality (all, cerebrovascular, IHD

Justification for health outcome selection

All-cause mortality

Health outcomes selected for updating in the 2021 air quality guidelines

Table 2.1. Air pollutants and health outcomes proposed for systematic review in the guideline development processa

WHO GLOBAL AIR QUALITY GUIDELINES

Health outcomes used in Global update 2005

No long-term guideline provided

Pollutant

O3

LONGTERM EXPOSURE

All-cause mortality Respiratory mortality

♦

♦

Health outcomes selected for updating in the 2021 air quality guidelines

GUIDELINE DEVELOPMENT PROCESS

Suggestive causality for respiratory mortality (Health Canada, 2013

Suggestive causality for mortality (US EPA, 2013

Precautionary principle from expected increase of this pollutant due to climate change (policy implications and end-user perspectives)

Severity of health outcome, burden of disease

Likely causal for respiratory effects (US EPA, 2013 Suggestive causality for respiratory effects (Health Canada, 2013

♦

♦

OTHER RELEVANT CAUSAL DETERMINATIONS REFERENCE

♦

♦

SUPPORTING CONSIDERATIONS

♦

♦

CAUSALITY DETERMINATION REFERENCE

Justification for health outcome selection

Table 2.1 contd

37

38

Health outcomes used in Global update 2005

Respiratory effects in children

Pollutant

NO2

LONGTERM EXPOSURE

All-cause mortality Respiratory mortality

♦

♦

Health outcomes selected for updating in the 2021 air quality guidelines

Suggestive causality for total mortality (Health Canada, 2016a))

Suggestive causality for total mortality (US EPA, 2016

The causal determination of US EPA of likely causal for respiratory effects (see other causal determinations below) takes into account respiratory mortality

The causal determination of US EPA for mortality is suggestive, in the light of the limited number of studies properly addressing confounding by other transport-related air pollutants

Recent studies show associations with respiratory mortality, consistent with likely causality for respiratory effects (see other causal determinations below)

Severity of health outcome, burden of disease

Likely causal for respiratory effects (US EPA, 2016 Likely causal for respiratory effects (Health Canada, 2016a)

♦

♦

OTHER RELEVANT CAUSAL DETERMINATIONS REFERENCE

♦

♦

♦

♦

SUPPORTING CONSIDERATIONS

♦

♦

CAUSALITY DETERMINATION REFERENCE

Justification for health outcome selection

Table 2.1 contd

WHO GLOBAL AIR QUALITY GUIDELINES

GUIDELINE DEVELOPMENT PROCESS

COHb levels of below 2% in nonsmokers’ blood (also protective for longterm exposure) (WHO Regional Office for Europe, 2000a, 2010

COHb levels of below 2% in nonsmokers’ blood (also protective for longterm exposure) (WHO Regional Office for Europe, 2000a, 2010

CO

PM2.5 and PM10

Health outcomes used in Global update 2005

Pollutant

SHORTTERM EXPOSURE

Likely causal for cardiovascular effects (Health Canada, 2010

♦

Respiratory mortality

♦

Causal for all-cause, respiratory and cardiovascular mortality Health Canada, 2013

Causal for all-cause, cardiovascular and respiratory mortality US EPA, 2009

PM2.5

Cardiovascular mortality

♦

♦

CAUSALITY DETERMINATION REFERENCE

IHDᶜ appears to be the most consistent outcome from cardiovascular effects associated with short-term CO exposure in epidemiological studies (US EPA, 2010

Consistent with WHO indoor air quality guidelines (WHO Regional Office for Europe, 2010 using COHb levels linked to IHDᶜ-related symptoms (e.g. ST-segment changes, reduced time to exerciseinduced angina)

All-cause mortality

♦

♦

SUPPORTING CONSIDERATIONS

Likely causal for cardiovascular effects (US EPA, 2010

♦

CAUSALITY DETERMINATION REFERENCE

♦

Hospital admissions and emergency room visits related to IHDᶜ

Justification for health outcome selection

♦

♦

Health outcomes selected for updating in the 2021 air quality guidelines

Table 2.1 contd

39

40

(contd)

PM2.5 and PM10

Pollutant

Health outcomes used in Global update 2005

SHORTTERM EXPOSURE

Causal for all-cause mortality (Health Canada, 2013

PM10, supported by evidence from PM2.5

Cardiovascular and respiratory mortality also considered in causal determination of respiratory/cardiovascular effects (US EPA, 2009 (see other relevant causal determinations)

Likely causal for respiratory effects (US EPA, 2009 Causal for cardiovascular effects (US EPA, 2009 Causal for respiratory effects (Health Canada, 2013 Causal for cardiovascular effects (Health Canada, 2013

♦

♦

♦

♦

OTHER RELEVANT CAUSAL DETERMINATIONS PM2.5) REFERENCE

♦

♦

SUPPORTING CONSIDERATIONS

♦

PM (any size fraction)

Health outcomes selected for Justification for health outcome selection updating in the 2021 air quality guidelines

Table 2.1 contd

WHO GLOBAL AIR QUALITY GUIDELINES

Health outcomes used in Global update 2005

Daily mortality

Pollutant

O3

SHORTTERM EXPOSURE

GUIDELINE DEVELOPMENT PROCESS ♦

♦

All-cause mortality

Hospital admissions and emergency room visits related to asthma Likely causal for total mortality (US EPA, 2013 Likely causal for total mortality (Health Canada, 2013

♦

♦

♦

Experimental studies demonstrate decreases in lung function at exposures as low as 6070 ppb O in young healthy adults. Equally strong evidence demonstrates associations of ambient O with asthma hospital admissions and emergency room visits, including for at-risk subpopulations (e.g. children, people with asthma or other pre-existing diseases), who cannot ethically be included in experimental studies

Stronger causal determination for respiratory effects than for mortality outcomes (see other relevant causal determinations below)

SUPPORTING CONSIDERATIONS

Causal for respiratory effects (Health Canada, 2013

♦

Causal for respiratory effects (US EPA, 2013

♦

♦

CAUSALITY DETERMINATION REFERENCE

Health outcomes selected for Justification for health outcome selection updating in the 2021 air quality guidelines

Table 2.1 contd

41

42

(contd)

O3

Pollutant

Health outcomes used in Global update 2005

SHORTTERM EXPOSURE

Mortality also included because of the severity of health outcome, number of exposed individuals and precautionary principle (expected future increase of this pollutant due to climate change, with policy and end-user implications). On the other hand, studies on mortality might target other subgroups of the population such as older people

Likely causal for cardiovascular effects (US EPA, 2013 Suggestive causality for cardiovascular effects (Health Canada, 2013 Likely causal for cardiopulmonary mortality (Health Canada, 2013

♦

♦

♦

OTHER RELEVANT CAUSAL DETERMINATIONS REFERENCE

♦

Health outcomes selected for Justification for health outcome selection updating in the 2021 air quality guidelines

Table 2.1 contd

WHO GLOBAL AIR QUALITY GUIDELINES

Health outcomes used in Global update 2005

Bronchial responsiveness in asthmatics

Pollutant

NO2

SHORTTERM EXPOSURE

GUIDELINE DEVELOPMENT PROCESS ♦

♦

All-cause mortality

Hospital admissions and emergency room visits related to asthma

Health outcomes selected for updating in the 2021 air quality guidelines

Suggestive causality for total mortality (US EPA, 2016

♦

♦

Both Health Canada and US EPA causality assessments for respiratory effects are very recent (2016; limited new evidence might have accumulated since

Stronger causal determination for respiratory effects than for mortality outcomes (see other relevant causal determinations below)

SUPPORTING CONSIDERATIONS

Likely causal for total mortality (Health Canada, 2016a)

♦

Causal for respiratory effects (Health Canada, 2016a)

♦

♦

Causal for respiratory effects (US EPA, 2016

♦

CAUSALITY DETERMINATION REFERENCE

Justification for health outcome selection

Table 2.1 contd

43

44

(contd)

NO2

Pollutant

Health outcomes used in Global update 2005

SHORTTERM EXPOSURE Health outcomes selected for updating in the 2021 air quality guidelines

♦

♦

Mortality is also included because of severity of health outcome and number of exposed individuals. On the other hand, studies on mortality might target other subgroups of the population such as older people

Strongest evidence for relationships of short-term NO2 exposure with respiratory effects is for asthma exacerbations. More uncertainty exists with independent effect of short-term NO2 exposure on non-asthma respiratory effects due to less consistent evidence across scientific disciplines and limited evidence to support biological plausibility. Additionally, studies of short-term NO2 exposure with asthma hospital admissions and emergency room visits include at-risk subpopulations (e.g. children, people with asthma or other pre-existing diseases) who cannot ethically be included in experimental studies

Justification for health outcome selection

Table 2.1 contd

WHO GLOBAL AIR QUALITY GUIDELINES

GUIDELINE DEVELOPMENT PROCESS

All-age mortality and childhood respiratory disease

SO2

All-cause mortalityᵈ Respiratory mortalityᵈ

♦

Hospital admissions and emergency room visits related to asthma

♦

♦

Health outcomes selected for updating in the 2021 air quality guidelines

Causal in adults for respiratory effects (Health Canada, 2016b)

♦

♦

Experimental studies demonstrate lung function decrements and respiratory symptoms at very short-term exposures ♦ (i.e. 510 minutes) to SO2 in adults with asthma. Studies of hospital admissions and emergency room visits related to asthma, including evidence from at-risk subpopulations (e.g. children, people with asthma or other pre-existing diseases) who cannot ethically be included in experimental studies, report positive associations with shortterm SO2 exposures, particularly for children

SUPPORTING CONSIDERATIONS

Causal for respiratory effects (US EPA, 2015

♦

CAUSALITY DETERMINATION REFERENCE

Justification for health outcome selection

COHb: carboxyhaemoglobin; ppb: parts per billion. ᵃ This table was adapted from the document resulting from the PECOS process (see section 2.3.4 finalized in November 2016. ᵇ Group 1 means carcinogenic to humans. ᶜ Ultimately restricted to myocardial infarction (ICD10 codes I21 and I22. ᵈ In the second GDG meeting held in 2018, the GDG agreed to include the pollutant–outcome pairs SO–all-cause mortality and SO–respiratory mortality.

Health outcomes used in Global update 2005

Pollutant

SHORTTERM EXPOSURE

Table 2.1 contd

45

In the second GDG meeting in March 2018, the question was raised of why a systematic review of short-term associations between sulfur dioxide and all-cause mortality would be needed in addition to the reviews on PM, ozone and nitrogen dioxide. It was suggested to also review SO and mortality to ensure continuity with the previous Global update 2005. If a new review is not feasible, the GDG suggested formulating clear justification as to why mortality attributed to SO is not considered. It was subsequently decided to commission a separate review on short-term associations between sulfur dioxide and all-cause as well as respiratory mortality, to be conducted by the team that had already conducted the review on short-term associations of PM, ozone and nitrogen dioxide with mortality. Therefore, the latest US EPA ISA of causality for this association is mentioned below. In 2017 the latest US EPA ISA on sulfur oxides was published (US EPA, 2017. This did not change the assessment noted in Table 2.1 of a causal relationship between short-term sulfur dioxide concentrations and respiratory effects. The association between short-term sulfur dioxide concentrations and total mortality was deemed to be suggestive of a causal relationship. This association was not considered by the GDG in 2016, but was added at a later stage, as previously mentioned.

Therefore, the causality assessment is not reported in Table 2.1. The US EPA published an updated ISA of PM in 2019 US EPA, 2019a). The causality determinations for long- and short-term PM effects on mortality and on respiratory and cardiovascular morbidity were the same as those in the 2009 ISA US EPA, 2009, which was quoted in Table 2.1. The US EPA published an updated ISA of ozone in 2020 US EPA, 2020. The causality determinations for ozone effects on respiratory morbidity were the same as those in the 2009 ISA, which was quoted in Table 2.1. For short-term mortality and for cardiovascular morbidity, the evidence was changed from likely causal to suggestive of a causal relationship, in part because new human exposure studies such as the Multicenter Ozone Study in oldEr Subjects (MOSES study did not clearly demonstrate the cardiovascular effects of ozone (Frampton et al., 2017; Rich et al., 2020. Other reasons quoted were the lack of control for co-pollutants in epidemiological studies and uncertainty about the short-term effects of ozone on cardiovascular emergency room and hospital admissions. Although the US EPA ISA was published in 2020, the literature on long- and short-term effects of ozone has grown since then. Chapter 3 includes five new studies on the long-term effects of ozone on mortality and one very large

46

WHO GLOBAL AIR QUALITY GUIDELINES

worldwide multicity study on the short-term effects of ozone on mortality, which have provided further evidence of the short- and long-term effects on total and respiratory mortality. The reader is referred to section 3.4 for further details. The nitrogen dioxide causality assessments shown in Table 2.1 are based on reviews published in 2016. Since then, COMEAP published a report in 2018 entitled Associations of long-term average concentrations of nitrogen dioxide with mortality, which is somewhat more supportive of a causal role of long-term nitrogen dioxide in increasing all non-accidental mortality, especially respiratory mortality (PHE, 2018. The 2016 EPA ISA classified the relationship between short-term nitrogen dioxide and respiratory effects as causal and the relationship between long-term nitrogen dioxide and respiratory effects as likely causal. A footnote to the causality determination defined the health outcome as “[a] n array of outcomes is evaluated as part of a broad health effects category: physiological measures (e.g. airway responsiveness), clinical outcomes (e.g. hospital admissions), cause-specific mortality”. This suggests the causality determinations also extend to respiratory mortality, although the further detailed assessments in the ISA provide some qualifications for the separate health effects that were evaluated (US EPA, 2016.

A 2018 review by the German Environment Agency (in German, with a summary in English) also supports a role for long-term nitrogen dioxide in causing cardiovascular mortality (Schneider et al., 2018. The GDG notes that one review has specifically investigated how sensitive the associations between long-term nitrogen dioxide concentrations and mortality were to adjustment for different PM metrics (Faustini, Rapp & Forastiere, 2014. Associations with nitrogen dioxide were found to be generally robust. Since 2016, no authoritative reviews have been published on short-term associations between carbon monoxide and hospital admissions for myocardial infarction.

2.3.4 Formulation of review questions As per the WHO procedure of developing guidelines (WHO, 2014a), key questions to guide the review of evidence are best developed using the population, intervention, comparator and outcome format. However, in environmental health risk guidelines such as the WHO air quality guidelines, recommendations are typically given in the form of numerical concentration values to prevent adverse health effects from exposure to environmental pollutants (so-called AQG levels). Typically, the best available evidence from human studies in this field consists mostly, if not exclusively,

GUIDELINE DEVELOPMENT PROCESS

47

of observational studies, as opposed to (randomized) controlled trials. Therefore – and as raised by several expert guideline development methodologists dealing with environmental risk guidelines (Collaboration for Environmental Evidence, 2013  the use of a slightly adapted formulation of the traditional population, intervention, comparator and outcome question was used: a PECOS question. The intervention (I term was replaced by an exposure (E term, reflecting the concentration in ambient air of the particular air pollutant under consideration; also, an S was added to define study designs to be considered in evaluating the evidence, resulting in a PECOS question: population, exposure, comparator, outcome and study design (Table 2.2. The GDG proposed the following PECOS questions (Box 2.2, which were later adapted to the health outcome and specific type of studies relevant for each pollutant and time average (short- or long-term exposure) considered in the updated guidelines.

Box 2.2. Generic PECOS question for long- and short-term exposures

Long-term exposures In any population, including subgroups of susceptible adults and children (P, what is the increase in risk of health outcome x (O per unit increase (C in μg/m of long-term exposure (in the order of months to years) to ambientᵃ concentration of air pollutant y (E, observed in studies relevant for the health outcome and exposure duration of interest (S? In these studies, what is the lowest concentration that produces a measurable increase in risk? Short-term exposures In any population, including subgroups of susceptible adults and children P, what is the increase in risk of health outcome x (O per unit increase (C in μg/m of short-term exposure (in the order of hours to days) to ambienta concentration of air pollutant y (E, observed in studies relevant for the health outcome and exposure duration of interest (S? In these studies, what is the lowest concentration that produces a measurable increase in risk? ᵃ Ambient refers to both outdoor and indoor environments.

48

WHO GLOBAL AIR QUALITY GUIDELINES

These PECOS questions were designed to retrieve the epidemiological evidence necessary to develop updated AQG levels and inform the shape of the CRF for the different pollutant–outcome pairs. For the specific purpose of updating the WHO air quality guidelines, the PECOS terms were defined as follows (Table 2.2.

Table 2.2. Elements of a PECOS question Element

Explanation

Population

The general population, all age groups, from developed and developing countries, living both in urban and in rural areas exposed on a daily basis to the pollutant of interest through ambient air (understood as encompassing exposure in both outdoor and indoor environments), and not exclusively in occupational settings or as a result of indoor exposure alone. Population subgroups that are vulnerable to the effects of air pollution would be included, such as those with specific pre-existing health conditions (e.g. respiratory or cardiovascular diseases), pregnant women, newborns, children or older people. Whenever applicable, the considered health effect of exposure to the pollutant of interest in these vulnerable subgroups of the population would be assessed separately

Exposure

Exposure to air pollutants from any source, measured as long term

Comparator

Exposure to the lowest levels of air pollutants from any source,

Outcome

Health outcome(s) upon which the AQG levels are developed for

Study design

Type of studies evaluated, such as cohort and case–control studies (long term) and time-series, case-crossover and panel studies (short

(months to years) or short term (hours to days)

measured as long- (months to years) or short-term (hours to days)

each air pollutant considered in the guidelines

term)

2.4 Systematic review of the evidence To address the PECOS questions posed by the GDG, a preliminary search of the relevant literature was conducted to identify available systematic reviews and meta-analyses on air quality and health. Based on an overview that assessed the quality of reviews in the field (Sheehan et al., 2016, it was decided that

GUIDELINE DEVELOPMENT PROCESS

49

included peer-reviewed articles that were of sufficient quality and addressed the formulated PECOS would serve as a starting point for most systematic reviews. Missing elements (e.g. specific assessments or syntheses) would be extracted anew and searches updated to the latest possible date within the process. Selected members of the systematic review team, who were mostly identified through the above procedure, reviewed and synthesized all the relevant epidemiological literature in the area of air quality and health, following the principles outlined in the WHO handbook for guideline development, 2nd edition WHO, 2014a) and guidance provided from methodologists and experts in the discipline. The instruments needed to assess the RoB for individual studies included in the reviews and the overall certainty of evidence across studies were adapted to better reflect the particularities of the air quality and health field. PECOS questions were formulated for each of the major pollutant–outcome pairs and relevant averaging times. When the same health outcomes and averaging times were assessed, various air pollutants were grouped under the same systematic review, resulting in six systematic reviews.

All systematic reviews followed a common protocol prepared according to the provisions set out by the WHO handbook for guideline development, 2nd edition WHO, 2014a) and later fine-tuned in relation to the specific exposure–outcome averaging time combinations that each review aimed to address. The protocols for each systematic review are registered on PROSPERO, an international register of systematic reviews, maintained by the University of York (NIHR, 2021. Furthermore, all systematic reviews used in the derivation of AQG levels are publicly available in a special issue of the journal Environment International Whaley et al., 2021, which are also summarized in Annex 3. ■

■

■

50

Long-term exposure to PM and all-cause and cause-specific mortality: a systematic review and meta-analysis (Chen & Hoek, 2020. Long-term exposure to NO2 and O3 and all-cause and respiratory mortality: a systematic review and meta-analysis (Huangfu & Atkinson, 2020. Short-term exposure to particulate matter (PM10 and PM2.5), nitrogen dioxide (NO2), and ozone (O3) and all-cause and cause-specific mortality: systematic review and meta-analysis (Orellano et al., 2020.

WHO GLOBAL AIR QUALITY GUIDELINES

■

■

■

Short-term exposure to sulfur dioxide (SO) and all-cause and respiratory mortality: a systematic review and meta-analysis (Orellano, Reynoso & Quaranta, 2021. Short-term exposure to ozone, nitrogen dioxide, and sulfur dioxide and emergency room visits and hospital admissions due to asthma: a systematic review and meta-analysis (Zheng et al., 2021. Short-term exposure to carbon monoxide and myocardial infarction: a systematic review and meta-analysis (Lee et al., 2020.

The core systematic reviews of adverse health effects were commissioned to address the PECOS questions. To ensure and confirm that no relevant studies in indoor settings had been missed in these reviews (none had been identified in the searches, based on the selected pollutant–outcome pairs), complementary searches were also conducted. In addition, several reviews and analyses were conducted in the context of this update of the guidelines. These included work on the health effects of exposure to particles originating from SDS, the burden of disease attributable to air pollution, the effectiveness of individual-level interventions and the cost–effectiveness of air quality interventions. Relevant review work conducted by other groups was closely monitored (e.g. on the health effects of BC/EC and UFP.6

2.4.1 Identification and retrieval of evidence Based on the PECOS questions, a list of inclusion and exclusion criteria were defined for each systematic review and later fine-tuned by the systematic review team (Table 2.3 and Table 2.4. Because most of the systematic reviews were based on peer-reviewed papers, the original search strategies were revised to reflect any additional eligibility criteria to ensure that all papers addressing the PECOS questions were identified. Specific search strategies using both free text and controlled vocabulary terms were run for each database. More details can be found in the systematic reviews published in Environment International Whaley et al., 2021. All efforts were made to include all relevant papers published, which entailed searching a considerable number of literature sources, the inclusion of papers in languages other than English and the use of time frames spanning from database inception to late 2018.

 With the exception of the review and analysis performed by Evangelopoulos et al. 2020 and the review by Fussell & Kelly (2021, none of the reviews conducted in the context of the update of the guidelines have yet been published.

GUIDELINE DEVELOPMENT PROCESS

51

Table 2.3. Generic eligibility criteria for systematic reviews of long-term exposures PECOS

Population

Inclusion ■

■

Exposure

■

■

Comparator

Outcome

Study design

■

■

■

Exclusion

General human population (including subgroups at risk: children, pregnant women, older people and patients with particular conditions) of all ages, living in developed and developing areas, both urban and rural. No geographical restrictions

■

Exposure to the pollutant of interest predominantly via inhalation through ambient air (this covers exposures in both outdoor and indoor environments) Long-term exposure (in the order of months to years) to ambient air PM2.5, PM10, O3 and NO2 expressed in a concentration unit (µg/m3, ppb)

■

■

For the NO2 systematic review, NOx studies may be included Exposure to lowest levels of the air pollutant of interest in the same or a control population

–

Health outcomes selected in relation to long-term exposure include (ICD10 codes (version 2016 all-cause mortality and cause-specific mortality, including cardiovascular (I00I99, lung cancer C30C39 and respiratory (J00J99

–

Human epidemiological studies such as:

■

- prospective and retrospective studies

■

- cohort studies - case–control and nested case–control studies ■

Exposure to the pollutant of interest in occupational settings or as a result of indoor exposure exclusively

Published (or accepted for publication, i.e. in press) studies in peer-reviewed indexed journals in any language (abstract in English language) and grey literature, where relevant.

■

■

■

Less than 1 year of data available No exclusion criteria applied based on adjustment for co-pollutants

Qualitative studies Studies without individual-level data, that is, fully group-level (ecological) covariates Studies where no original data were analysed Reviews and methodological papers Non-human studies (in vivo, in vitro, other)

ppb: parts per billion.

52

WHO GLOBAL AIR QUALITY GUIDELINES

Table 2.4. Generic eligibility criteria for systematic reviews of short-term exposures PECOS

Inclusion

Population

■

■

Exposure

■

■

Comparator

■

Outcome

■

Study design

■

General human population (including subgroups at risk: children, pregnant women, older people, and patients with particular conditions) of all ages, living in developed and developing areas, both urban and rural. No geographical restrictions

Exclusion ■

Exposure to the pollutant of interest predominantly via inhalation through ambient air (this covers exposures in both outdoor and indoor environments) Short-term exposure (in the order of hours to 7 days) to ambient air PM2.5, PM10, O3, NO2, SO2 and CO, from any source expressed in a concentration unit (µg/m3, ppb)

■

No exclusion criteria were applied based on adjustment for co-pollutants

For NO2 systematic review, NOx studies may be included Exposure to lowest levels of the air pollutant of interest in the same or a control population

–

Health outcomes selected for short-term exposure include (ICD10 codes (version 2016 all-cause mortality and cause-specific mortality, including cardiovascular (I00I99 and respiratory (J00J99, and hospital admissions and emergency room visits related to asthma (J45J46 and myocardial infarction I21I22

–

Human epidemiological studies such as:

■

- time-series studies

■

- case-crossover studies - panel studies ■

Exposure to the pollutant of interest in occupational settings or as a result of indoor exposure exclusively

Published (or accepted for publication, i.e. in press) studies in peer-reviewed indexed journals in any language (abstract in English language) and grey literature, where relevant

■

■

■

Qualitative studies Studies without individuallevel data, that is, fully group-level (ecological) covariates Reviews and methodological papers Non-human studies (in vivo, in vitro, other) Studies with geographical and temporal overlap during meta-analysis

ppb: parts per billion.

GUIDELINE DEVELOPMENT PROCESS

53

For each of the systematic reviews, two reviewers independently screened titles and abstracts of papers identified with the systematic search and identified those that could be excluded based on the eligibility criteria. The full texts of the remaining articles were independently reassessed by two reviewers to ensure that all eligibility criteria were met. Disagreements among reviewers were resolved by discussion or through consultation with a third reviewer. The reasons for excluding articles were recorded. In addition, references of identified relevant articles (and reviews/guidelines, where relevant) were scanned to identify additional papers matching the PECOS question. The resulting list of papers was circulated with the systematic review team and the GDG to identify any potentially relevant missing studies (published or in press). Lastly, papers identified through the peer review process were incorporated as appropriate, either quantitatively or qualitatively, as feasible. Two reviewers extracted all relevant data needed for the process using pre-defined forms. Key data included the elements defined by PECOS and declared conflicts of interest, as well as the data necessary to conduct RoB assessments (e.g. confounding factors) and to derive the AQG levels (i.e.

for onset of the CRF 5th–95th percentiles of population exposure, mean/median, and minimum and maximum pollutant concentrations; for shape of the CRF methods and results of authors’ assessments). Where necessary data were missing, the systematic review team obtained them from the authors of the primary studies or calculated them.

2.4.2 RoB assessment of individual studies To assess RoB for individual studies, a specific instrument was developed by a working group composed of GDG members and methodologists. Based on a review of existing tools, the group agreed to take into account six key domains (confounding, selection bias, exposure assessment, outcome measurement, missing data, selective reporting), each including several subdomains or signalling questions. Judgement options included high, moderate and low RoB. The group also prepared guidance notes to assist the systematic review team in performing the task, including a list of critical and additional potential confounders to consider when making judgements about confounding and key information on the particularities of exposure assessment in the field. To avoid carrying forward the ratings from one domain to the others, the working group considered that an overall judgement of bias at the study level was not appropriate: instead, subgroup analyses were suggested per RoB domain across studies. This approach was considered more suited to identify which particular type of bias had an impact on the pooled effect size, as well as its direction and magnitude (Morgan et al., 2019.

54

WHO GLOBAL AIR QUALITY GUIDELINES

A detailed description of the instrument is available in a dedicated publication WHO Regional Office for Europe, 2020.

2.4.3 Synthesis of evidence Meta-analyses were conducted to obtain summary pooled estimates of the risk for an adverse health outcome per unit increase in exposure to a given air pollutant. When three or more studies were identified for the same pollutant and health outcome, a quantitative synthesis was performed. Otherwise, the effect estimates were described qualitatively. Overall, statistical analyses were performed according to the Guidelines for application of meta-analysis in environmental epidemiology Blair et al., 1995, the Cochrane handbook for systematic reviews of interventions Higgins & Green, 2011 or other authoritative guidance. The approach used was the inverse variance method, assuming a linear concentration–response relationship. When exposure metrics differed among studies, the data were transformed to the same metric, generally the relative risk (RR. Although no dose–response meta-analytic techniques were employed to assess the shape of the CRF, potential deviations from linearity were assessed by other means, for example, by stratifying by mean pollutant concentrations or qualitatively evaluating the determinations and judgements made by study authors.

Because of differences in populations and pollution composition across populations, it was decided that estimates were to be pooled by means of a random-effects meta-analysis (maximum likelihood approach). Several measures of statistical heterogeneity were calculated, including I-squared and tau-squared. If considerable heterogeneity was present, attempts were made to explain the source of heterogeneity by subgroup analysis, meta-regression or sensitivity analysis (only possible if enough studies were available). Other sensitivity analyses included those needed to inform the judgements on RoB, large magnitude of effect size and publication bias within the certainty of evidence approach. Lastly, additional sensitivity analyses were conducted to explore the impact of multipollutant models, conflicts of interest of study authors, population characteristics or lag patterns, where appropriate.

2.4.4 Grading of the certainty of the overall body of evidence Evaluation of the certainty of evidence is foundational for systematic review, with a focus on the validity and precision of effect estimates. In the clinical realm, evidence-informed review has become the starting point for establishing

GUIDELINE DEVELOPMENT PROCESS

55

guidelines for clinical practice, including guidance for therapeutics and diagnostics. Much of the evidence considered in the clinical context comes from randomized controlled trials, where exposures are assigned at random by the investigator to provide some degree of assurance that potential confounders and effect modifiers, both known and unknown, are balanced across treatment groups. In the clinical context, evidence may also come from observational studies, including cohort and case–control studies, case series and other data resources. Given the strength of the randomized controlled trial design for ensuring comparability of treatment and control groups, a hierarchy of evidence sources has been established in which randomized controlled trials (providing the strongest evidence) have the highest ranking and lower rankings are given to other sources. The GRADE approach has been adopted as the basis for evidence review in support of WHO guidelines (Schünemann et al., 2013; WHO, 2014a). GRADE was implemented for the purpose of evaluating evidence in support of formulation of clinical guidelines and, as such, it divided studies into randomized and nonrandomized designs and ranked randomized studies as higher-quality evidence.

The initial certainty level of evidence was determined by the type of study, with randomized controlled trials starting at high certainty and non-randomized studies starting at low certainty. Thereafter, five domains were assessed for downgrading the certainty of the evidence resulting from randomized and non-randomized studies, and three domains were assessed for upgrading the certainty of evidence from non-randomized studies alone (Box 2.3. Consistent with the standard approach, the certainty of the effect estimate was graded as high, moderate, low or very low. The ratings were subsequently used to select the risk functions used to derive AQG levels. With the extension of GRADE to topics for which evidence derives largely or totally from observational studies, there are areas for which evidence from randomized designs is not available and decision-making, of necessity, draws on other evidence. For environmental agents, the evidence foundation is diverse and with very few exceptions does not involve a randomized exposure (e.g. air cleaner with filter versus air cleaner without filter). The human evidence is observational, coming from population-level studies (time-series studies, geospatial analyses, cohort studies, case–control studies and cross-sectional studies).

A further issue that arises with environmental agents is identifying and summarizing the evidence derived from toxicological studies, in vivo animal bioassays and in vitro work addressing mechanisms.

56

WHO GLOBAL AIR QUALITY GUIDELINES

Box 2.3. GRADE domains Domains assessed for downgrading the certainty of evidence by one or two levels ■

Limitations or RoB in all studies that constitute the body of evidence

■

Indirectness of evidence in the studies

■

Inconsistency of results between studies

■

Imprecision of the pooled effect estimate

■

Publication bias detected in the body of evidence.

Domains assessed for upgrading the certainty of evidence by one level ■

■

■

Large magnitude of the pooled effect estimate All plausible confounding shifting the pooled effect estimate towards the null Existence of a concentration−response gradient.

Source: adapted from WHO 2014a).

Recognizing these complexities, different groups have made efforts to adapt GRADE for the assessment of evidence on exposures, but limitations were still under discussion at the time of developing these guidelines (National Research Council, 2014; Morgan et al., 2016; Saracci, 2017; Steenland et al., 2020. In this context, a working group was convened to adjust the standard GRADE approach to the field of air quality and health. The current adaptation was not aimed to assess causality through an examination of all the relevant streams of research Woodruff & Sutton, 2011, but instead aimed to rate how certain one can be that the “true” estimate of the association between an air pollutant and an adverse health effect lies within a particular range (Hultcrantz et al., 2017. The working group decided to start the rating for air pollution observational studies at moderate rather than high certainty evidence because of the risk of unmeasured confounding in observational research. From this level, the certainty of the evidence was then downgraded or upgraded based on several criteria per GRADE domain.

GUIDELINE DEVELOPMENT PROCESS

57

In addition, the working group recognized the need for taking a more nuanced view of the evidence, as well as for incorporating the following additional criteria to complement or replace existing guidance: ■

■

■

■

calculation of an 80% prediction interval, to help assess heterogeneity in conjunction with the 95% confidence interval (CI IntHout et al., 2016; calculation of the sample size needed for a study based on a specific RR and CI, to help guide judgements about imprecision (Rothman & Greenland, 2018; estimation of the extent to which confounding may influence a pooled effect size using the E-value, to facilitate judgements for large magnitude of effect size (Mathur & VanderWeele, 2020; and additional approaches to help assess publication bias, such as a subgroup analysis of multicentre studies compared with single-city studies in case of evidence based on time-series studies, an analysis of differences in effect estimates from earlier versus later studies, and a comparison with published results of attempts to quantify the magnitude of bias.

A detailed description of the adaptation of GRADE is provided in the supplementary materials of the articles published in the special issue of Environment International WHO Global Air Quality Guidelines Working Group on Certainty of Evidence Assessment, 2020.

2.5 From evidence to recommendations The GDG decided that the recommendations (AQG levels) would be primarily based on epidemiological evidence. The GDG discussed how to account for contextual factors in formulating the AQG levels. Given the very large variability in exposures, socioeconomic conditions and other policy considerations across the world, the GDG concluded that retaining and enhancing the widely adopted interim targets from the previous guidelines would be a more useful instrument to assist endusers in implementing the recommendations. Contextual factors should instead be considered during the policy-making process at national, regional or local level, as discussed in Chapters 1 and 6 of this document. The recommendations were based on the certainty of evidence judgements alone, whereby low/very low certainty would prevent the GDG from formulating a recommendation for an AQG level. See, however, the caveats about this in sections 2.5.1 and 2.5.2. Furthermore, two additional elements of guidance are offered in these guidelines. These elements differ from the recommendations in that they are not derived from systematic reviews of evidence of adverse health effects from air pollution.

58

WHO GLOBAL AIR QUALITY GUIDELINES

Instead, they are based on an expert assessment of several types of evidence that included their utility to support end-users in their efforts to improve air quality. These elements of guidance are interim targets and good practice statements. Interim targets are air pollutant levels that are higher than the AQG levels, but which authorities in highly polluted areas can use to develop pollution reduction policies that are achievable within realistic time frames. The interim targets should be regarded as steps towards ultimately achieving AQG levels in the future, rather than as end targets. The number and numerical values of the interim targets are pollutant specific, and they are justified in the relevant sections of Chapter 3. Contextual factors also did not play a direct role in the formulation of this guidance, although some considerations were described in a qualitative manner where relevant (e.g. burden of disease in relation to interim targets, resource considerations in relation to some good practice statements). The following sections provide a detailed description of the approaches used by the GDG to formulate the recommendations and the additional guidance.

2.5.1 Formulation of long-term AQG levels 2.5.1.1 Definition Long-term AQG levels are developed to provide advice to end-users to reduce the adverse effects of long-term exposure to air pollutants and, thereby, reducing associated disease and mortality. Health outcomes in the current process are restricted to all-cause and respiratory mortality (PM2.5, PM10, ozone, nitrogen dioxide). In addition, cardiovascular and lung cancer mortality are considered for some pollutants (PM2.5, PM10). A long-term AQG level is defined as the lowest exposure level of an air pollutant above which the GDG is confident that there is an increase in adverse health effects. Confidence refers primarily to the adapted GRADE qualification confirming that there is high or moderate certainty evidence for an association between a specific pollutant and a specific health outcome. The GRADE certainty rating is based on eight criteria (discussed later in this section). The GDG also took into account additional considerations, including causality determinations. In principle, AQG levels were developed only for pollutant–outcome pairs with at least moderate certainty data. The GDG recognizes that, following the precautionary principle, conditional recommendations could be considered where the certainty of the evidence is less than moderate.

GUIDELINE DEVELOPMENT PROCESS

59

This would be the case, for instance, when exposure is widespread and the effect on population health is severe. However, as will become evident in Chapter 3, there was at least moderate certainty evidence to support long- and short-term AQG levels for all pollutants considered. This approach avoids consideration about what level of exposure should be considered safe, given that the available evidence cannot currently identify levels of exposure that are risk free for any of the pollutant–outcome pairs considered in this document. Moreover, the approach also avoids defining a so-called accepted level of risk, which would violate clean air acts or directives in countries where adverse health effects of air pollution are not accepted.7 It also avoids making inferences for exposure levels below those for which there is solid evidence. The challenge is then to find the lowest level of exposure for which the GDG is still confident that there is at least moderate certainty evidence for adverse health effects. Note that this also requires some consideration of what is an adverse health effect. As a reference for this, the GDG used the latest update of the joint European Respiratory Society and American Thoracic Society policy statement on “what constitutes an adverse health effect of air pollution” Thurston et al., 2017.

The systematic reviews commissioned by WHO formed the starting point for the body of evidence on which an AQG level is based and, therefore, underpin these guidelines. The systematic reviews each provide a summary estimate of the RR derived from the included meta-analyses for each pair of air pollutant and adverse health effect, a 95% CI for this estimate and a GRADE qualification for the certainty of the evidence. In principle, the GDG used this estimate for guideline derivation only if the 95% CI from a random-effects meta-analysis did not include an RR of 1. However, as air pollution has no known health benefits, the GDG decided in specific, well-argued cases to deviate from this principle. 2.5.1.2 Procedure To find the lowest level of long-term exposure for which the GDG would be confident of an adverse health effect, a dedicated working group developed a procedure for each pollutant–outcome pair, based on the following eight steps Table 2.5.

7

60

For example, the US National Ambient Air Quality Standards are based on the Clean Air Act, which stipulates: “National primary ambient air quality standards, prescribed under subsection (a) shall be ambient air quality standards the attainment and maintenance of which in the judgment of the Administrator, based on such criteria and allowing an adequate margin of safety, are requisite to protect the public health” 42 U.S.C. 7409(a)).

WHO GLOBAL AIR QUALITY GUIDELINES

Table 2.5. Eight steps in formulation of long-term AQG levels Step

Description

Step 1

Assess RR estimates and, when available, CRF for each critical health outcome per pollutant as provided by the systematic review. In its first meeting in 2016, based on an initial survey, the GDG decided that the following health outcomes are critical (depending on air pollutant): (i) all-cause mortality (or all, natural-cause mortality, excluding accidental deaths); (ii) respiratory mortality; (iii) cardiovascular mortality, associated with both long- and short-term exposures; (iv) short-term, day-to-day variations in hospital admissions and emergency room visits related to asthma; and (v) myocardial infarction. The GDG recommends AQG levels for all pollutant–outcome pairs identified in 2016 except for those associations not meeting at least moderate levels of certainty. This includes pairs with different likelihoods of causality, according to authoritative reviews by COMEAP, Health Canada, International Agency for Research on Cancer, US EPA and others

Step 2

Determine the lowest level of exposure measured in the studies included in the systematic review or in the subset of studies in the systematic review that estimate risk at this lowest level. For individual studies that used statistical models to evaluate the shape of the CRF, ensure that the lowest level of exposure is associated with a monotonic increase of the CRF curve

Step 3

Determine the minimal relevant increase in health outcomes

Step 4

Determine the starting point for AQG level determination as the long-term concentration of pollutant from which the minimal relevant amount of the health outcome will result

Step 5

Compare the AQG levels for a specific pollutant across critical health outcomes. Take as the final AQG level the lowest AQG level found for any of the critical health outcomes

Step 6

Assess the certainty of the evidence at low levels of exposure. The adapted GRADE assessment is for the entire body of evidence, not the subset of studies conducted at the lowest exposure levels. The evidence provided by these latter studies needs to be discussed, starting from the RoB assessment that was conducted at individual study level

Step 7

Consider new relevant evidence not included in the systematic reviews in a qualitative or, where possible, quantitative manner

Step 8

Reconsider causality of associations between pollutants and outcomes, taking into account whether or not associations have been classified as causal or likely causal in recent reviews by COMEAP, Health Canada, US EPA, WHO or other authoritative bodies

GUIDELINE DEVELOPMENT PROCESS

61

Each of the eight steps is discussed below. Step 1. The GDG used the meta-analytic effect estimate that results from the systematic review and the assessment of the certainty of the evidence that underpins this effect estimate according to GRADE. In principle, effect estimates are only used if the 95% CI does not include an RR of 1 in the random-effects meta-analysis of the relevant body of evidence for a specific exposure–outcome pair. In addition, they are only used when underpinned by moderate to high certainty evidence. This is because there would be little confidence in an AQG level based on a non-significant meta-analytic effect estimate or on an effect estimate for which there is only low-certainty evidence. The GDG recognized that the probability value (P value) generated by a test of statistical significance is a continuous measure and that even a statistically non-significant result may be more consistent with a real increased risk than with the null. Therefore, in cases of borderline significance or where significance is restricted to major subgroups, the GDG decided whether or not to proceed with guideline development, regardless of the overall statistical significance. It was noted that in the meta-analyses of the systematic reviews, statistical significance was based on two-sided tests.