Page 431 - Contributed Paper Session (CPS) - Volume 2
P. 431
CPS1916 Erica P. et al.
in standard analyses. In particular, the measurement of environmental and
pollutant variables is often very likely to be affected by error, due to
instrumental imprecision as well as spatial variability (Goldman et al., 2011).
Some methods have been proposed to correct for ME in environmental and
pollution exposure variables (Dominici et al., 2000; Strand et al., 2006) and
calibration is starting to be performed thanks to the availability of personal
monitoring exposures (PEM), but the problem still needs further thinking and
methodology.
Thanks to their general applicability and lack of strong assumptions,
Bayesian approaches to the problem may provide a general framework to
work on and software like INLA (Rue et al., 2009), STAN (STAN Development
Team, 2017) and JAGS (Plummer, 2003) could promote their use in practical
applications also among scientists without a strong statistical background. The
Bayesian framework provides a very flexible way to account for measurement
error and to model different error types and dependency structures in the
data. Moreover, the possibility to include prior knowledge on the error
components can result in better models and more accurate estimations and
the possibility to model several fixed and random effects, as well as different
link functions, adds flexibility and general applicability to the methods.
Examples of Bayesian approaches used in ME problems can be found in
Wilkinson (2013); Stoklosa et al. (2014); Velasquez-Tibata et al. (2015), Muff et
al. (2015, 2017).
In the present study, we propose to apply these techniques to correct for
measurement error in environmental exposures when considering their
association with high-throughput molecular data. We apply our methods to a
randomized crossover trial, the Oxford Street II Study, whose aim was to
investigate the association between short-term exposure to traffic related air
pollution and perturbation of metabolic pathways.
2. Methodology
The data we use here stem from the Oxford Street Study, a randomized
crossover trial in which 60 volunteers walked for two hours in a traffic-free area
(Hyde Park) and in a busy shopping street (Oxford Street) of London. The
walking experiments were performed on non-rainy weekdays only, from
November to March 2003-2005, to avoid confounding from rain or pollen.
Information on age, sex, BMI, blood pressure, distance walked, diet and
medication use was collected for each participant. For each individual and
each exposure session, three blood samples were collected: two hours before
walking, two hours after walking and 24 hours after walking, on which
untargeted metabolomics analyses were performed.
Finally, real-time measurements of noise, temperature and relative
humidity were obtained at each exposure session. A more detailed description
420 | I S I W S C 2 0 1 9