Using mobile phone big data to estimate the short-term effect of PM2.5 exposures on pharmacy visits

Abstract

OPS 50: Air pollution and health care utilization, Room 117, Floor 1, August 28, 2019, 10:30 AM - 12:00 PM Background: Epidemiological studies have reported associations between PM2.5 and severe health outcomes (e.g., mortality or morbidity), while mild adverse outcomes (e.g., clinical or even subclinical symptoms) have rarely been studied. Big data technology provides a promising way to investigate the short-term effect of PM2.5 exposures on pharmacy visits, a potential proxy of mild adverse outcomes. Methods: We first identified the exact location of 40,875 pharmacies through the pharmacy-related points on interests (P-POIs) in Jiangsu province, China. For each P-POI, normalized daily pharmacy visits numbers from May 28, 2018 to January 17, 2019 (235 days), were extracted from anonymized, geographically and time-referenced Call Detail Records (CDRs) of 136 thousand mobile phone base station macro-cells. Daily pharmacy visits numbers of each P-POIs were matched with the daily PM2.5 concentrations in the closest 10 km grids that were incorporated from PM2.5 stationary monitors using a random forest model. A generalized additive model was applied to estimate the associations of PM2.5 with pharmacy visits for each P-POI. The estimates at the P-POI level were then pooled using a Bayesian hierarchical model at city and province levels. Results and discussion: In general, a significant increased risk of pharmacy visits (0.58%, 95% Confidence Interval: 0.43, 0.72) was associated with a 10 μg/m3 increase in PM2.5 concentrations. The effects were significant in most cities (ranging from 0.54% to 0.83%). Mobile phone big data offers a novel way to quantify air pollution effects on subclinical outcomes, and in regions where administrative health data are lacking.