Table 7.

GRADE assessment – NO₂ and all-cause mortality.

Domain	Judgement	Down/Up Grade
Limitations in studies	24 included studies. Risk of bias moderate because although not all studies adjusted for all confounders, exclusion of high risk of bias studies did not reduce the summary RR (Appendix Fig. B5).	No downgrading
Indirectness	All studies included the desired population, exposures and outcomes	No downgrading
Inconsistency	The 80% prediction interval included 1 & > twice CI (Fig. 2). High level of heterogeneity in general population studies. Studies controlling for individual measures of BMI, smoking, SES (Appendix Fig. B3) gave slightly higher, less precise summary RR. Exclusion of patient cohorts (6) did not change summary RR & CI (Appendix Fig. B2).	Downgrade one level
Imprecision	The number of person years in the included studies was greater than 940 000	No downgrading
Publication Bias	According to the funnel plot and Egger’s test (P < 0.1), there were no sign of publication bias/funnel plot asymmetry.	No downgrading
Large Effect Size	Summary RR = 1.02. Precision reduced for cohorts with all individual confounder adjustment but not summary estimate. Insufficient information on unmeasured potential confounders available.	No upgrading
Plausible confounding towards null	Confounding direction unknown but precision may be affected.	No upgrading
Dose-response relation	A linear dose–response relationship was assumed in all studies. 5 studies investigated the shape of the dose response relationship with no evidence to suggest non-linear. 95% CI for linear RR excluded 1.	Upgrade one level
GRADE conclusion	Downgrade one level and upgrade one level	MODERATE CERTAINTY EVIDENCE MEAN RR UNADUSTED FOR CO-POLLUTANTS EQUALS 1.02 PER 10μ/m3