Table 4.

Association between PM_2.5 exposure and stroke incidence in the 11 cohorts under study: results of the sensitivity analyses.

Model	Cohorts (n)	Participants (n)	HR (95% CI)
Main model	11	99,446	1.19 (0.88, 1.62)*
Role of cardiovascular risk factors
Intermediate variables: diabetes and hypertension
Plus diabetes and hypertension	11	99,446	1.15 (0.84, 1.56)*
Physical activity, alcohol consumption, and BMI
Main model, on subset of participants with additional information	8a	76,599	1.32 (0.87, 2.00)*
Plus additional information	8a	76,599	1.30 (0.86, 1.97)*
All cardiovascular risk factors (diabetes, hypertension, physical activity, alcohol, BMI, cholesterol)b
Main model, on subset of participants with additional information	4b	24,948	1.91 (0.96, 3.82)*
Plus additional information	4b	24,948	1.88 (0.99, 3.57)*
Urban/rural
Plus rural indicator	11	99,446	1.18 (0.87, 1.59)
Noise
Main model, on subset of participants with additional information	9c	73,121	1.25 (0.92, 1.71)
Plus noise variable	9c	73,121	1.26 (0.89, 1.78)
Change of address during follow-up
Main model, on cohorts with change of address data	10d	92,216	1.26 (0.93, 1.72)
No change of address during follow-up	10d	62,799	1.19 (0.81, 1.76)
Proportionality-hazards assumption
Variables which don’t meet PH assumption as strata	11	99,446	1.20 (0.89, 1.62)
Exclusion of DCH cohort
10 cohorts (all except DCH)	10	63,753	1.22 (0.86, 1.75)
Performance of the LUR model
LOOCV R2 coefficient > 0.6	6e	32,191	1.75 (1.30, 2.35)
LOOCV R2 coefficient ≤ 0.6	5f	67,255	0.89 (0.70, 1.13)
aAll cohorts except SALT, SIDRIA-Turin and SIDRIA-Rome. bIncludes FINRISK, 60y, HNR, and KORA. cAll cohorts except SDPP and SIDRIA-Rome. dAll cohorts except EPIC-Turin. eIncludes SNAC-K (LOOCV R2 = 0.78), SALT (0.78), 60y (0.78), SDPP (0.78), HNR (0.79), and KORA (0.62). fIncludes FINRISK (LOOCV = R2 0.53), DCH (0.55), EPIC-Turin (0.59), SIDRIA-Turin (0.59), and SIDRIA-Rome (0.60). *pheterogeneity < 0.05, as indicated by Cochran’s Q or I2 > 50%.