Green spaces
The exposure of interest was the amount of greenness in the residential environment. Green space was assessed using the mean Normalized Difference Vegetation Index (NDVI) within 100 m, 300 m, 500 m, and 1000 m of the participant’s residence during pregnancy and the first two years of life in both spring and summer seasons. These area sizes were selected to cover immediate areas of exposure, containing accessible green spaces in accordance with the World Health Organization (WHO) recommendations and within five to ten minute walking distances.20-23 We selected the 300 m buffer for NDVI for the main analyses, in accordance with the WHO and UNICEF recommendations.23,24 The calculation of the NDVI was based on land surface reflectance of visible red (VISR) and near-infrared (NIR) wavelengths,25 applying the following equation (Eq. (1)):
\(NDVI=\ \frac{NIR-VISR}{NIR+VISR}\) (1)
The values range between -1 (water) through zero (rock, sand and snow) to 1, with higher positive values indicating denser green vegetation (i.e., photosynthetically active and healthy vegetation).25 Only cloud-free images from Landsat 5 (spatial resolution: 30 m) during the spring (April–May) and summer (June–August) periods were used from conception (1983-90) to the first two years of life (1984-92) to capture maximum spatial contrasts in greenness (Appendix). In Finland, spring begins in April and summer usually begins in late May in southern Finland (https://en.ilmatieteenlaitos.fi/seasons-in-finland). Negative NDVI values were not included in the calculation. ArcMap 10.5 was used to process satellite images, and QGIS 3.8 was used to extract the average NDVI within each buffer size.

Covariates

Any determinant of allergic rhinitis was considered a potential confounder of the studied associations between green spaces and occurrence of allergic rhinitis.18,26-32 We were able to adjust for the following confounders: age, sex, socioeconomic status measured by parental education level and occupation, maternal smoking during pregnancy and environmental tobacco smoke exposure (Appendix). Data on the covariates was obtained from the baseline questionnaire.

Statistical analysis

The statistical inference was based on estimation of the time of onset of allergic rhinitis in relation to the green space in residential surroundings during pregnancy and the first two years of life. Hazard ratio (HR) was used as the measure of effect. We applied Cox proportional hazards regression models to estimate crude and adjusted hazard ratios for the association between green space, measured as the cumulative exposure to NDVI (during pregnancy and during the first two years of life), and the risk of allergic rhinitis, measured as the time of onset of allergic rhinitis up to six, 12 and 27 years of age. Both crude and adjusted models were fitted, and the effect estimates were expressed for a 1 NDVI-month change in exposure. The final model was adjusted for participants’ sex and age, parental socioeconomic status, maternal smoking, and environmental tobacco smoke exposure, and NDVI during pregnancy (spring and summer) or early-life NDVI exposure (spring and summer). We hypothesized that air pollution (PM2.5, PM10, O3, NO2 and SO2) modifies the association between NDVI and allergic rhinitis through several potential underlying mechanisms (see Discussion).32 The CAUSALMED procedure in SAS was used to calculate the proportions of the mediated effect based on theoretical assumptions proposed by Valeri et al. 33. Based on exploratory analyses, we used the cut-off based on the highest quartile of air pollutant level during pregnancy and early-life to define lower (air pollutants level <4thquartile) or higher air pollution (air pollutant level ≥4th quartile) levels. We stratified the associations between NDVI and allergic rhinitis by air pollution levels. Sensitivity analysis using different buffer sizes (100 m, 500 m, and 1000 m) and quartiles of exposure to NDVI (Table S1) were performed to evaluate the robustness of our findings. Statistical analyses were performed using SAS software version 9.4 TS Level 1M5.