Exposure Misclassification Errors Research Articles

Coupling mobile phone data with machine learning: How misclassification errors in ambient PM2.5 exposure estimates are produced?

BackgroundMost studies relying on time-activity diary or traditional air pollution modelling approach are insufficient to suggest the impacts of ignoring individual mobility and air pollution variations on misclassification errors in exposure estimates. Moreover, very few studies have examined whether such impacts differ across socioeconomic groups. ObjectivesWe aim to examine how ignoring individual mobility and PM2.5 variations produces misclassification errors in ambient PM2.5 exposure estimates. MethodsWe developed a geo-informed backward propagation neural network model to estimate hourly PM2.5 concentrations in terms of remote sensing and geospatial big data. Combining the estimated PM2.5 concentrations and individual trajectories derived from 755,468 mobile phone users on a weekday in Shenzhen, China, we estimated four types of individual total PM2.5 exposures during weekdays at multi-temporal scales. The estimate ignoring individual mobility, PM2.5 variations or both was compared with the hypothetical error-free estimate using paired sample t-test. We then quantified the exposure misclassification error using Pearson correlation analysis. Moreover, we examined whether the misclassification error differs across different socioeconomic groups. Taking findings of ignoring individual mobility as an example, we further investigated whether such findings are robust to the different selections of time. ResultsWe found that the estimate ignoring PM2.5 variations, individual mobility or both was statistically different from the hypothetical error-free estimate. Ignoring both factors produced the largest exposure misclassification error. The misclassification error was larger in the estimate ignoring PM2.5 variations than that ignoring individual mobility. People with high economic status suffered from a larger exposure misclassification error. The findings were robust to the different selections of time. ConclusionsIgnoring individual mobility, PM2.5 variations or both leads to misclassification errors in ambient PM2.5 exposure estimates. A larger misclassification error occurs in the estimate neglecting PM2.5 variations than that ignoring individual mobility, which is seldom reported before.

PM2.5 on the London Underground

IntroductionDespite the London Underground (LU) handling on average 2.8 million passenger journeys per day, the characteristics and potential health effects of the elevated concentrations of metal-rich PM2.5 found in this subway system are not well understood. MethodsSpatial monitoring campaigns were carried out to characterise the health-relevant chemical and physical properties of PM2.5 across the LU network, including diurnal and day-to-day variability and spatial distribution (above ground, depth below ground and subway line). Population-weighted station PM2.5 rankings were produced to understand the relative importance of concentrations at different stations and on different lines. ResultsThe PM2.5 mass in the LU (mean 88 μg m−3, median 28 μg m−3) was greater than at ambient background locations (mean 19 μg m−3, median 14 μg m−3) and roadside environments in central London (mean 22 μg m−3, median 14 μg m−3). Concentrations varied between lines and locations, with the deepest and shallowest submerged lines being the District (median 4 μg m−3) and Victoria (median 361 μg m−3 but up to 885 μg m−3). Broadly in agreement with other subway systems around the world, sampled LU PM2.5 comprised 47% iron oxide, 7% elemental carbon, 11% organic carbon, and 14% metallic and mineral oxides. Although a relationship between line depth and air quality inside the tube trains was evident, there were clear influences relating to the distance from cleaner outside air and the exchange with cabin air when the doors open. The passenger population-weighted exposure analysis demonstrated a method to identify stations that should be prioritised for remediation to improve air quality. ConclusionPM2.5 concentrations in the LU are many times higher than in other London transport Environments. Failure to include this environment in epidemiological studies of the relationship between PM2.5 and health in London is therefore likely to lead to a large exposure misclassification error. Given the significant contribution of underground PM2.5 to daily exposure, and the differences in composition compared to urban PM2.5, there is a clear need for well-designed studies to better understand the health effects of underground exposure.

On the accuracy and potential of Google Maps location history data to characterize individual mobility for air pollution health studies

Appropriately characterizing spatiotemporal individual mobility is important in many research areas, including epidemiological studies focusing on air pollution. However, in many retrospective air pollution health studies, exposure to air pollution is typically estimated at the subjects’ residential addresses. Individual mobility is often neglected due to lack of data, and exposure misclassification errors are expected. In this study, we demonstrate the potential of using location history data collected from smartphones by the Google Maps application for characterizing historical individual mobility and exposure. Here, one subject carried a smartphone installed with Google Maps, and a reference GPS data logger which was configured to record location every 10 s, for a period of one week. The retrieved Google Maps Location History (GMLH) data were then compared with the GPS data to evaluate their effectiveness and accuracy of the GMLH data to capture individual mobility. We also conducted an online survey (n = 284) to assess the availability of GMLH data among smartphone users in the US. We found the GMLH data reasonably captured the spatial movement of the subject during the one-week time period at up to 200 m resolution. We were able to accurately estimate the time the subject spent in different microenvironments, as well as the time the subject spent driving during the week. The estimated time-weighted daily exposures to ambient particulate matter using GMLH and the GPS data logger were also similar (error less than 1.2%). Survey results showed that GMLH data may be available for 61% of the survey sample. Considering the popularity of smartphones and the Google Maps application, detailed historical location data are expected to be available for large portion of the population, and results from this study highlight the potential of these location history data to improve exposure estimation for retrospective epidemiological studies.

A gestational vulnerability window for smoking exposure and the increased risk of preterm birth: how timing and intensity of maternal smoking matter

BackgroundReducing the incidence of preterm birth is a national priority. Maternal cigarette smoking is strongly and consistently associated with preterm birth. The objective of this study was to examine prenatal exposure based on combined measures of timing (by trimester) and intensity level (the number of cigarettes smoked per day) of maternal smoking to identify a pregnancy period with the highest risk of preterm birth.MethodsA sample of 2,485,743 singleton births was drawn from the 2010 National Center of Health Statistics (NCHS) linked birth/infant death file of US residents in 33 states that implemented the revised 2003 birth certificate. Nine mutually exclusive smoking status categories were created to assess prenatal exposure across pregnancy in association with preterm birth. Gestational age was based on the obstetric estimate. Multiple logistic regression analyses were conducted to compare the odds of preterm birth among women who smoked at different intensity levels in the second or third trimester with those who smoked only in the first trimester.ResultsOverall, 7.95% of women had a preterm birth; 8.90% of low intensity (less than a pack/day) smokers in the first trimester only, 12.99% of low and 15.38% of high intensity (pack a day or more) smokers in the first two trimesters, and 10.56% of low and 11.35% of high intensity smokers in all three trimesters delivered preterm. First and second trimester high (aOR 1.85, 95% CI: 1.66, 2.06) and low intensity smokers (aOR 1.51, 95% CI: 1.41, 1.61) had higher odds of preterm birth compared to those who smoked less than a pack a day only in the first trimester, but the odds did not increase for all three trimester smokers relative to the first and second trimester smokers. In sensitivity analysis, adjustment for exposure misclassification error corrected data and testing for effect modification by maternal race/ethnicity found no significant interaction.ConclusionsThis study documented a biologically plausible vulnerability window for smoking exposure and the increased risk of preterm birth. For women who do not modify their smoking behavior preconception, preterm birth risk of smoking remains low until late in the first trimester.

Estimated health benefits of exhaust free transport in the city of Malmö, Southern Sweden

Air pollution is responsible for one in eight premature deaths worldwide, and thereby a major threat to human health. Health impact assessments of hypothetic changes in air pollution concentrations can be used as a mean of assessing the health impacts of policy, plans and projects, and support decision-makers in choices to prevent disease.The aim of this study was to estimate health impacts attributable to a hypothetical decrease in air pollution concentrations in the city of Malmö in Southern Sweden corresponding to a policy on-road transportations without tail-pipe emissions in the municipality. We used air pollution data modelled for each of the 326,092 inhabitants in Malmö by a Gaussian dispersion model combined with an emission database with >40,000 sources. The dispersion model calculates Nitrogen Oxides (NOx) (later transformed into Nitrogen Dioxide (NO2)) and particulate matter with an aerodynamic diameter < 2.5 μg/m3 (PM2.5) with high spatial and temporal resolution (85 m and 1 h, respectively).The average individual reduction was 5.1 (ranging from 0.6 to 11.8) μg/m3 in NO2, which would prevent 55 (2% of all deaths) to 93 (4%) deaths annually, depending on dose-response function used. Furthermore, we estimate that the NO2 reduction would result in 21 (6%) fewer cases of incident asthma in children, 95 (10%) fewer children with bronchitis every year, 30 (1%) fewer hospital admissions for respiratory disease, 87(4%) fewer dementia cases, and 11(11%) fewer cases of preeclampsia every year. The average reduction in PM2.5 of 0.6 (ranging from 0.1 till 1.7) μg/m3 would mean that 2729 (0.3%) work days would not be lost due to sick-days and that there would be 16,472 fewer restricted activity days (0.3%) that year had all on-road transportations been without tail-pipe emissions.Even though the estimates are sensitive to the dose-response functions used and to exposure misclassification errors, even the most conservative estimate of the number of prevented deaths is 7 times larger than the annual traffic fatalities in Malmö, indicating a substantial possibility to reduce the health burden attributed to tail-pipe emissions in the study area.

Using cell phone location to assess misclassification errors in air pollution exposure estimation.

Air pollution epidemiologic and health impact studies often rely on home addresses to estimate individual subject's pollution exposure. In this study, we used detailed cell phone location data, the call detail record (CDR), to account for the impact of spatiotemporal subject mobility on estimates of ambient air pollutant exposure. This approach was applied on a sample with 9886 unique simcard IDs in Shenzhen, China, on one mid-week day in October 2013. Hourly ambient concentrations of six chosen pollutants were simulated by the Community Multi-scale Air Quality model fused with observational data, and matched with detailed location data for these IDs. The results were compared with exposure estimates using home addresses to assess potential exposure misclassification errors. We found the misclassifications errors are likely to be substantial when home location alone is applied. The CDR based approach indicates that the home based approach tends to over-estimate exposures for subjects with higher exposure levels and under-estimate exposures for those with lower exposure levels. Our results show that the cell phone location based approach can be used to assess exposure misclassification error and has the potential for improving exposure estimates in air pollution epidemiology studies.

Satellite-based PM concentrations and their application to COPD in Cleveland, OH.

A hybrid approach is proposed to estimate exposure to fine particulate matter (PM(2.5)) at a given location and time. This approach builds on satellite-based aerosol optical depth (AOD), air pollution data from sparsely distributed Environmental Protection Agency (EPA) sites and local time-space Kriging, an optimal interpolation technique. Given the daily global coverage of AOD data, we can develop daily estimate of air quality at any given location and time. This can assure unprecedented spatial coverage, needed for air quality surveillance and management and epidemiological studies. In this paper, we developed an empirical relationship between the 2 km AOD and PM(2.5) data from EPA sites. Extrapolating this relationship to the study domain resulted in 2.3 million predictions of PM(2.5) between 2000 and 2009 in Cleveland Metropolitan Statistical Area (MSA). We have developed local time-space Kriging to compute exposure at a given location and time using the predicted PM(2.5). Daily estimates of PM(2.5) were developed for Cleveland MSA between 2000 and 2009 at 2.5 km spatial resolution; 1.7 million (∼79.8%) of 2.13 million predictions required for multiyear and geographic domain were robust. In the epidemiological application of the hybrid approach, admissions for an acute exacerbation of chronic obstructive pulmonary disease (AECOPD) was examined with respect to time-space lagged PM(2.5) exposure. Our analysis suggests that the risk of AECOPD increases 2.3% with a unit increase in PM(2.5) exposure within 9 days and 0.05° (∼5 km) distance lags. In the aggregated analysis, the exposed groups (who experienced exposure to PM(2.5) >15.4 μg/m(3)) were 54% more likely to be admitted for AECOPD than the reference group. The hybrid approach offers greater spatiotemporal coverage and reliable characterization of ambient concentration than conventional in situ monitoring-based approaches. Thus, this approach can potentially reduce exposure misclassification errors in the conventional air pollution epidemiology studies.

Gwynneth June Davey

This method was developed to demonstrate the application of probabilistic bias analysis to quantify and adjust for exposure misclassification in a historical cohort mortality study of New Zealand trichlorophenol workers...

Evaluation of land-use regression models used to predict air quality concentrations in an urban area

Cohort studies designed to estimate human health effects of exposures to urban pollutants require accurate determination of ambient concentrations in order to minimize exposure misclassification errors. However, it is often difficult to collect concentration information at each study subject location. In the absence of complete subject-specific measurements, land-use regression (LUR) models have frequently been used for estimating individual levels of exposures to ambient air pollution. The LUR models, however, have several limitations mainly dealing with extensive monitoring data needs and challenges involved in their broader applicability to other locations. In contrast, air quality models can provide high-resolution source–concentration linkages for multiple pollutants, but require detailed emissions and meteorological information. In this study, first we predicted air quality concentrations of PM 2.5, NO x, and benzene in New Haven, CT using hybrid modeling techniques based on CMAQ and AERMOD model results. Next, we used these values as pseudo-observations to develop and evaluate the different LUR models built using alternative numbers of (training) sites (ranging from 25 to 285 locations out of the total 318 receptors). We then evaluated the fitted LUR models using various approaches, including: 1) internal “Leave-One-Out-Cross-Validation” (LOOCV) procedure within the “training” sites selected; and 2) “Hold-Out” evaluation procedure, where we set aside 33–293 tests sites as independent datasets for external model evaluation. LUR models appeared to perform well in the training datasets. However, when these LUR models were tested against independent hold out (test) datasets, their performance diminished considerably. Our results confirm the challenges facing the LUR community in attempting to fit empirical response surfaces to spatially- and temporally-varying pollution levels using LUR techniques that are site dependent. These results also illustrate the potential benefits of enhancing basic LUR models by utilizing air quality modeling tools or concepts in order to improve their reliability or transferability.

Passive sampling to capture spatial variability in PM 10–2.5

This work applied inexpensive passive sampling to measure airborne coarse particles with aerodynamic diameters between 2.5 and 10 μm (PM 10–2.5) over three 7-day periods at 33 sites in a medium-sized Midwest City (Iowa City, IA). The number of sites and their locations were selected using an optimal sampling design that captured 95% of the total variance in PM 10–2.5 as measured with real-time sampling equipment on a mobile sampling platform. Weekly averages of PM 10–2.5 were 15.9 μg m −3 (coefficient of variation between sites, CV=23%), 17.9 μg m −3 (CV=24%), and 6.1 μg m −3 (CV=30%). ANOVA showed that these means were statistically different ( p<0.001), and that the spatial variability plus random error accounted for 29% of the total variability. The maximum coefficient of divergence between sites—a relative measure of uniformity—ranged from 0.21 to 0.36. These values indicate that PM 10–2.5 was heterogeneous even on the fine spatial resolution studied in this work (average distance between sites was 4.4 km). The spatial patterns of PM 10–2.5 measured with the passive samplers closely matched with those of mobile mapping and corresponded with known coarse particle sources in the area. This work demonstrates that passive sampling coupled with effective sampling design may enhance our ability to assess exposure to PM 10–2.5 at a local scale. Compared to exposure estimates made with data from centrally located, filter-based samplers, these highly spatially resolved estimates should reduce exposure misclassification errors in epidemiological studies.

Spatial Variability of PM2.5 in Urban Areas in the United States

Data from the U.S. Environmental Protection Agency's Aerometric Information Retrieval System (now known as the Air Quality System) database for 1999 and 2000 have been used to characterize the spatial variability of concentrations of particulate matter with aerodynamic diameter ≤2.5 μg (PM2.5) in 27 urban areas across the United States. Different measures were used to quantify the degree of uniformity of PM2.5 concentrations in the urban areas characterized. It was observed that PM2.5 concentrations varied to differing degrees in the urban areas examined. Analyses of several urban areas in the Southeast indicated high correlations between site pairs and spatial uniformity in concentration fields. Considerable spatial variation was found in other regions, especially in the West. Even within urban areas in which all site pairs were highly correlated, a variable degree of heterogeneity in PM2.5 concentrations was found. Thus, even though concentrations at pairs of sites were highly correlated, their concentrations were not necessarily the same. These findings indicate that the potential for exposure misclassification errors in time-series epidemiologic studies exists.

The variability of delivered dose of aerosols with the same respirable concentration but different size distributions.

The influences of aerosol size distribution and breath tidal volume on respirable dose estimates were examined for mouth breathing using the ACGIH/ISO/CEN criterion for respirable-equivalent aerosols. Actual tissue doses predicted from a set of pulmonary empirical deposition equations, the Heyder-Rudolf equations, were compared with deposition assumed to occur under the penetration-based respirable dust sampling criterion. Deposition estimate errors ranged from approximately 1/10- to 10-fold, with aerosol mass median aerodynamic equivalent diameter and geometric standard deviation as well as tidal volume each showing a substantial influence under appropriate conditions. These findings demonstrate that reliance on respirable aerosol sampling data obtained with devices performing on a penetration-based sampling criterion may lead to erroneous dose-response relationships in exposure standard development as well as exposure misclassification errors during epidemiological studies. A more reliable dose estimate would be obtained using devices with collection efficiency performance closely matching the alveolar deposition prediction curves of Heyder and Rudolf. We believe that if it is not currently required, the development of a deposition-based aerosol sampling methodology will soon be required for the determination and quantification of inhaled aerosol-induced adverse health effects.