Identifying impacts of air pollution on subacute asthma symptoms using digital medication sensors.

Abstract

Objective tracking of asthma medication use and exposure in real-time and space has not been feasible previously. Exposure assessments have typically been tied to residential locations, which ignore exposure within patterns of daily activities. We investigated the associations of exposure to multiple air pollutants, derived from nearest air quality monitors, with space-time asthma rescue inhaler use captured by digital sensors, in Jefferson County, Kentucky. A generalized linear mixed model, capable of accounting for repeated measures, over-dispersion and excessive zeros, was used in our analysis. A secondary analysis was done through the random forest machine learning technique. The 1039 participants enrolled were 63.4% female, 77.3% adult (>18) and 46.8% White. Digital sensors monitored the time and location of over 286980 asthma rescue medication uses and associated air pollution exposures over 193697 patient-days, creating a rich spatiotemporal dataset of over 10905240 data elements. In the generalized linear mixed model, an interquartile range (IQR) increase in pollutant exposure was associated with a mean rescue medication use increase per person per day of 0.201 [95% confidence interval (CI): 0.189-0.214], 0.153 (95% CI: 0.136-0.171), 0.131 (95% CI: 0.115-0.147) and 0.113 (95% CI: 0.097-0.129), for sulphur dioxide (SO2), nitrogen dioxide (NO2), fine particulate matter (PM2.5) and ozone (O3), respectively. Similar effect sizes were identified with the random forest model. Time-lagged exposure effects of 0-3 days were observed. Daily exposure to multiple pollutants was associated with increases in daily asthma rescue medication use for same day and lagged exposures up to 3 days. Associations were consistent when evaluated with the random forest modelling approach.