Alan Domínguez, Yu Zhao, Karl Samuelsson, Pol Jimenez-Arenas, Marta Cirach, Gustavo Arevalo, Lluis Barril, Jose Lao, Maria Foraster, Mirea Gascón, Cecilia Persavento, Mark Nieuwenhuijsen, Cathryn Tonne, Teresa Moreno, Xavier Querol, Michael Jerret, Joel Schwartz, Jordi Sunyer, Payam Dadvand, Ioar Rivas, Xavier Basagaña
The findings demonstrate that accounting for dynamic infiltration processes and individual-level behaviors improves the estimation of indoor NO2 exposure.
BACKGROUND: Indoor air pollution is a major contributor to personal exposure to air pollution, yet accurately modeling indoor concentrations remains a major challenge in epidemiological studies. This issue is especially relevant since we spend a great portion of our time indoors.
OBJECTIVE: To develop a two-stage model to estimate indoor nitrogen dioxide (NO2) concentrations.
METHODS: The study was conducted in the Barcelona Life Study Cohort (2018-2021), a cohort of pregnant individuals in Barcelona, Spain. Indoor and outdoor NO2 concentrations were measured using passive NO2 samplers placed in participants' homes at weeks 12 and 32 of pregnancy. A total of 1695 indoor measurements and 1577 outdoor measurements were collected. In the first stage, we modeled the indoor-outdoor (I/O) NO2 ratio as a proxy for infiltration using a linear mixed-effects model with repeated measures using 1528 1-week integrated I/O NO2 measurements. In the second stage, we used a random forest algorithm to estimate weekly indoor NO2 concentrations covering the full pregnancy period, incorporating the predicted I/O ratios, a previously developed hybrid-model outdoor NO2 estimate, meteorological data, and home characteristics.
RESULTS: The I/O NO2 ratio model showed moderate performance, with a cross-validated R2 of 0.27 under leave-one-subject-out (generalized model) validation and 0.70 under leave-one-observation-out validation (cohort performance). The indoor NO2 model achieved a 10-fold cross-validated R2 of 0.53 and RMSE of 1.28 µg/m3. Finally, we estimated indoor NO2 concentrations for each week of pregnancy using a quantile random forest (QRF) algorithm incorporating uncertainty estimation through prediction intervals.
SIGNIFICANCE: The findings demonstrate that accounting for dynamic infiltration processes and individual-level behaviors improves the estimation of indoor NO2 exposure.
IMPACT: Integrating indoor and outdoor NO2 measurements in a two-stage modeling framework and combining it with key indoor determinants improves indoor NO2 estimation.