Crowd-Sourced Traffic Data: Predicting Air Pollution & Ischemic Stroke

摘要

ABSTRACT Air pollution, particularly traffic-related air pollution (TRAP), affects community health including cardiovascular disease and ischemic stroke onset. Stroke is the fifth leading cause of death in the US and the number one cause of long-term adult disability. Stroke has been consistently associated with daily average measurements of TRAP but much less often to traffic itself. Methods are needed to monitor traffic in real time to improve our ability to disentangle the effects of traffic on air pollution and stroke onset at both the neighborhood and citywide scale. Accounting for spatio-temporal variations is important to assessing acute affects since different people can get exposed to TRAP at different times of the day and locations. Key to predicting variations of TRAP is measurement of temporal variations in traffic conditions across the city. Evaluation of data obtained from GPS-enabled mobile phones present in driving vehicles offers the opportunity of inexpensive, real-time traffic monitoring of entire street networks, which cannot be done with traditional traffic monitors. There is great potential of using such crowd-sourced data in air pollution and public health studies. Our overarching hypothe- sis is that the links connecting neighborhood-scale traffic congestion to air pollution and stroke onset can be inexpensively examined by use of crowd-sourced traffic data, e.g., from Google Traffic (GT). In a pilot study, we showed that the five colors that Google uses to indicate congestion of road segments are a measure for vehicle speed based on radar measurements. We also showed that time series of traffic flow (hourly vehicle counts) can be locally inferred from an ordinal Google color code GCC we assigned to the GT colors. Finally we showed that both traffic flow and speed (either derived from a radar-device or GCC) explain levels of black carbon (BC), a tracer for TRAP. BC levels could be predicted from GCC and often readily avail- able annual average daily traffic (AADT) data as well as together with other non-traffic related covariates. Our overall goal is to show that crowd-sourced traffic data can be used to estimate TRAP and that these traffic data estimates can be used directly to investigate associations with health outcomes (without use of air pollution data). Our specific aims are to (1) demonstrate that temporally varying BC levels can be inferred from GT data, and (2) examine the associations between the risk of ischemic stroke onset and traffic conditions in the hours and days preceding each event. We hypothesize that the onset of ischemic stroke will be associated with hourly traffic measurements at the home locations (i.e., within a small buffer zone). The associations we will derive between stroke onset and crowd-sourced traffic data can establish a direct link to the source of TRAP and provide critical information for disease prevention, intervention planning and treatment. Our proposed use of crowd-sourced traffic data can be applied anywhere in the US and to health outcomes other than stroke.