Association between PM2.5 and respiratory hospitalization in Rio Branco, Brazil: Demonstrating the potential of low-cost air quality sensor for epidemiologic research

Abstract

Background. There is currently a scarcity of air pollution epidemiologic data from low- and middle-income countries (LMICs) due to the lack of air quality monitoring in these countries. Additionally, there is limited capacity to assess the health effects of wildfire smoke events in wildfire-prone regions like Brazil's Amazon Basin. Emerging low-cost air quality sensors may have the potential to address these gaps. Objectives. We investigated the potential of PurpleAir PM2.5 sensors for conducting air pollution epidemiologic research leveraging the United States Environmental Protection Agency's United States-wide correction formula for ambient PM2.5. Methods. We obtained raw (uncorrected) PM2.5 concentration and humidity data from a PurpleAir sensor in Rio Branco, Brazil, between 2018 and 2019. Humidity measurements from the PurpleAir sensor were used to correct the PM2.5 concentrations. We established the relationship between ambient PM2.5 (corrected and uncorrected) and daily all-cause respiratory hospitalization in Rio Branco, Brazil, using generalized additive models (GAM) and distributed lag non-linear models (DLNM). We used linear regression to assess the relationship between daily PM2.5 concentrations and wildfire reports in Rio Branco during the wildfire seasons of 2018 and 2019. Results. We observed increases in daily respiratory hospitalizations of 5.4% (95%CI: 0.8%, 10.1%) for a 2-day lag and 5.8% (1.5%, 10.2%) for 3-day lag, per 10 mu g/m(3) PM2.5 (corrected values). The effect estimates were attenuated when the uncorrected PM2.5 data was used. The number of reported wildfires explained 10% of daily PM2.5 concentrations during the wildfire season. Discussion. Exposure-response relationships estimated using corrected low-cost air quality sensor data were comparable with relationships estimated using a validated air quality modeling approach. This suggests that correcting low-cost PM2.5 sensor data may mitigate bias attenuation in air pollution epidemiologic studies. Low-cost sensor PM2.5 data could also predict the air quality impacts of wildfires in Brazil's Amazon Basin.