Discussion of statistical challenges in environmental health with an examination of four cases where these challenges were solved, enabling new insights to be obtained.
Applications of Contemporary Statistical Approaches in Environmental Health April 28 2011
1. Applications of Contemporary
Statistical Approaches in
Environmental Health
B. Rey de Castro, Sc.D.
before the
CDC Emergency Response & Air Toxicants Branch
Atlanta, GA
April 28, 2011
11. MCF-7 Cell Culture Assay
• Dependent data – MCF-7 cell number
• Fixed effect – dose
– 5 estradiol dose levels
– 1 control dose
12. 12-Well Plate
• Random effects, or variance components
– Plate
– Plate Dose interaction
– Well within (Plate Dose)
CD FBS 5% 1E-11M E2
1E-9M E2 1E-12M E2
1E-10M E2 1E-13M E2
13. Generalized Linear Mixed Effects Model
g (Yijkm) ijkm d0 di Pj ( PD)ij W ( PD)k (ij)
where Yijkm = cell number
d0 = mean cell number of no-dose control (intercept; i = 0)
di = fixed effect of ith dose (i = 1, 2, … , 5)
Pj = random effect of jth plate (j = 1, 2, … , 21), P ~ N(0, P2)
(PD)ij = joint random effect of ith dose with jth plate, , PD ~ N(0, PD2)
W(PD)k(ij) = random effect of kth well (k = 1, 2) nested within the ith dose
and jth plate, , W ~ N(0, W2)
In addition, the error term from Y = + is as follows:
m(ijk) = random error of the mth count (m = 1, 2, 3)
14. Findings
• MCF-7 assay data
– Gamma error distribution & reciprocal link
– COV = 3.1 %
– All variance components significant
18. Baltimore Traffic Study
• Observe dynamics of ambient traffic-related
pollutants at a location embedded within an
urban residential neighborhood with high
vehicular volume
19. Baltimore Traffic Study
• 2nd floor row house on commuter street
• Real-time sampling
• Near-simultaneous indoor/outdoor sampling
20. Baltimore Traffic Study
• Black carbon, PM, particle-bound
PAH, CO, O3, NOx, VOCs
• Vehicle counts
• Meteorology
29. Findings
• Neighborhood-level exposure to black carbon
from mobile sources was 65.82 ng/m3 per 100
vehicles
• Background exposure to black carbon without
traffic was estimated to be 899.06 ng/m3
30. Findings
• Winds from the SW-S-SE quarter were
associated with the greatest increases in
black carbon
• Implicates atmospheric processes in
transporting black carbon from
– Baltimore’s central business district
– Interstate highways
– Regional and inter-regional sources
34. Time-Location Data
• Basis for estimating total exposure
– Amount of time in each microenvironment
– Concentration in each microenvironment
• Structured diaries
35. Time-Weighted Exposure
• ith time interval
• jth microenvironment
N M
Exposure twa = ∑ ∑ timeWeight ij × concentration ij
i j
36. Generalized Logit Model
• Regression framework
P
P[Y = j]
log = α j + ∑ β jp X p j = 2, 3, ...,K
P[Y = 1] p=1
37. Time-Weighted Exposure
time ij
timeWeight ij =
time total
subjects ij
=
subjects total
= p ij
P
α j + ∑ β jpX p
ℯ p= 1
P
(j > 1)
= K α k + ∑ β kpX p
1+ ∑ ℯ p= 1
k=2
39. Subjects
• 95 children Sex
• 7 to 11-years-old Age Male Female
7 years 8 9
8 8 12
9 6 9
10 11 13
11 8 11
40. Time-Location
• 12 months
– June 1995 – May 1996
• 4 days
– Thursday - Monday
• 30-minute intervals
– 0600 to 2030
41. Data
• N = 171,000
• 1,800 longitudinal observations/subject
• Missing observations
– Imputation
42. Generalized Logit Model
indoor home
time of day sex
indoor school
Pr indoor other = β 0j + day of week + age
month nonwhite
commuting
lags 1 - 6 televisions
outdoor
50. NATA 2005
• US EPA National-Scale Air Toxics Assessment
• Sources
– Point
– Non-point
– Mobile
• On-road
• Off-road
• Secondary formation and decay
51. NATA 2005
• National Emissions Inventory
• Air monitoring data
• Atmospheric dispersion modeling
• Modeled exposure estimates
– Every United States census tract
• No indoor sources assessed
52. NATA 2005
• Acrolein
– Responsible for 75% respiratory non-cancer health
effects nationwide
53.
54. NHIS 2000 - 2009
• National Health Interview Survey
• Representative
– United States
– Non-institutionalized
– Civilian
• Cross-sectional prevalence
55. NHIS 2000 - 2009
• Adults 18 years-old and over
• Self-reported asthma attack in previous 12
months
56. NATA & NHIS
• NHIS subjects geographically linked to
NATA acrolein exposure estimates
• Census tract
– Survey subject residences
– Area exposure estimates
• Individual-level analysis
57. Preliminary Findings
• At highest quintile of acrolein exposure
– >0.055 g/m3
• pOR 1.11 [1.00:1.23]
• Controlling for
smoking, sex, age, education, race, poverty,
insurance, access to care, urban/rural
residence, survey year