Professor Maria Feychting Institute of Environmental Medicine...Maria Feychting 29 marzo 2008 21...
Transcript of Professor Maria Feychting Institute of Environmental Medicine...Maria Feychting 29 marzo 2008 21...
Exposure indices in ELFepidemiology
Professor Maria Feychting
Institute of Environmental Medicine
29 marzo 2008Maria Feychting 2
Overview of presentation
Sensitivity and specificity of exposure classification
Exposure assessment in occupational studies
Job-titles
”Electrical occupations”
Job-exposure matrices
Exposure assessment in residential studies
Distance or wire codes
Calculated fields
Measurements
Aspects of exposure assessment in overall
assessment of evidence
29 marzo 2008Maria Feychting 3
Sensitivity:
Probability that an exposed subject is classified as
exposed
No. of exposed subjects classified as exposed
Total number of exposed subjects
Specificity:
Probability that an unexposed subject is classified as
unexposed
No. of unexposed subjects classified as unexposed
Total number of unexposed subjects
29 marzo 2008Maria Feychting 4
Misclassification of exposure
Non-differential: independent of the disease
Leads to diluted effect estimates: RR 1
NB: non-differential exposure misclassification has no effect on
relative risk estimates if there is no true association between
exposure and disease risk
Magnitude of bias depends on:
prevalence of the exposure in studied population & sensitivity
and specificity of exposure assessment method
29 marzo 2008Maria Feychting 5
Non-differential exposure misclassification
Example: 2% exposure prevalence, true RR=2
1.0 0.8 0.6 0.4 0.2
1.0 2.00 1.99 1.98 1.98 1.97
0.8 1.09 1.07 1.05 1.02 1.00
0.6 1.05 1.03 1.02 1.00
0.4 1.03 1.02 1.00
0.2 1.02 1.00
Specificity
Sensitivity
Norell SE, 1987
29 marzo 2008Maria Feychting 6
Non-differential exposure misclassification
UnexposedHighexp.
29 marzo 2008Maria Feychting 7
When exposure is rare
High specificity of exposure estimate very important
Specificity = probability that unexposed individual
is classified as unexposed
Even a slight reduction in specificity may dilute risk
estimate considerably
29 marzo 2008Maria Feychting 8
When exposure is rare
Not important to find everyone who is exposed, i.e.
high sensitivity is not needed
Misclassification of exposed subjects as unexposed does
not affect results much
A high sensitvity improves the statistical power of the
study
Affects number of exposed subjects
A high sensitivity is essential for estimation of
exposure prevalence and public health impact
29 marzo 2008Maria Feychting 9
Differential exposure misclassification
i.e. misclassification related to disease
Can affect risk estimate i any direction
Exposure information must be collected in a similar
way for both cases and controls
“Recall bias” may lead to differential exposure
misclassification
Disease influence recall among cases
”Objective” source of information better, i.e. registry
29 marzo 2008Maria Feychting 10
ELF electric and magnetic field exposure
50 or 60 Hz fields (power frequency fields)
Frequencies in occupational settings may vary
Magnetic fields penetrate buildings, trees, humans
Electric fields are shielded by walls, trees, etc.
Epidemiological studies have focused on magnetic
fields – very few have included electric fields
29 marzo 2008Maria Feychting 11
ELF magnetic fields
Generated when electricity is transported or used
Magnetic field directly proportional to the current
Fall of rapidly with distance to the source
Fields are imperceptable – exposure information
cannot easily be obtained through self reports
Need to rely on direct measurements or proxies
29 marzo 2008Maria Feychting 12
ELF magnetic field exposure
Three major groups of exposure sources:
Occupational, residential, electrical appliances
Examples of sources:
Power lines, transformers, substations, various industrial
equipment, public transportation (train, subway), unbalanced
return currents, electric bed heating devices, microwave
ovens, electric razor, etc.
29 marzo 2008Maria Feychting 13
Relevant exposure metric unknown
No known biological mechanism at low levels of
exposure
Time weighted average ?
Peak exposure ?
Intermittent exposure ?
No one is unexposed!
29 marzo 2008Maria Feychting 14
Example: same cumulative dose –
different distribution over time
Duration
Intensity
29 marzo 2008Maria Feychting 15
Time-weighted average (TWA)
Majority of studies have focused on TWA
Some have estimated cumulative exposures
( Thours)
For population exposure: high specificity (91-98%)
when using TWA as an estimate of other metrics
e.g. time above 0.4 T, time above 1.6 T, maximum fields,
sudden changes >0.1 T, intermittence, length of time in a
constant field above 0.2 T
29 marzo 2008Maria Feychting 16
Categorization of exposure
If no prior studies, and no clues from biology –
categorization often based on exposure distribution
among controls, e.g. quartiles
If prior studies available – make sure to present
comparable cut-points
Always present reasons for choice of cut-points
Not valid to search for the cut-points with highest
effect estimates
29 marzo 2008Maria Feychting 17
Residential exposure - high exposures rare
0
10
20
30
40
50
60
70
0.1 0.1-0.19 0.2-0.29 0.3-0.39 0.4-
Magnetic field exposure distribution in the NCI studyLinet et al. 1997
%
T
29 marzo 2008Maria Feychting 18
Residential epxosure - proportion
exposed to >0.4 T
Only 0.4% in the UK Childhood Cancer Study
Similar proportion in Sweden
1% in the US (NCI-study)
3% in a study from Canada (McBride)
29 marzo 2008Maria Feychting 19
Residential magnetic field exposure
assessment
Distance
Wire-codes
Calculated fields
Measurements
Spot measurements
24-48 h measurements
Personal measurements
29 marzo 2008Maria Feychting 20
Distance to power line
Crude estimate of magnetic field level in the home
Correct only very close to high voltage transmission
lines
i.e. less than 40-50 meters from largest transmission lines
(>200 kV)
Few people live close to high voltage transmission
lines
Analysed categories have been within 100 or 200 meters
Low specificity – majority of persons categorized as exposed
will be unexposed
29 marzo 2008Maria Feychting 21
Wire-codes
Originates from Wertheimer & Leeper, 1979
Takes into account distance and type of ELF field
source, e.g. if transmission line, distribution line,
buried cables etc.
Higher specificity than distance alone - but
Does not take variation in power line load into consideration
Measurements have shown that fields are low in most
categories of wire-codes
29 marzo 2008Maria Feychting 22
Wire-codes and magnetic field levels in
studies of EMF and childhood leukemia
Under -
ground
Very
low
Ordinary
low
Ordinary
high
Very high
Savitz Mean
Median
0.05
0.03
0.05
0.03
0.07
0.05
0.12
0.09
0.21
0.22
London Mean
Median
0.05
0.04
0.05
0.04
0.06
0.06
0.07
0.07
0.12
0.11
Linet Mean
Median
0.06 0.05
0.08 0.05
0.12 0.08
0.14 0.10
0.21
0.13
McBride
Mean
0.09
0.08
0.11
0.17
0.26
Green Mean
0.07
(sd 0.06)
0.04
(sd 0.02)
0.14
(sd 0.11)
0.18
(sd 0.15)
0.28 (sd 0.26)
29 marzo 2008Maria Feychting 23
Calculated fields
Takes into account:
Distance to power line
Power line load – historical estimates
Construction of power lineHeight of towers,
Distance between conductors,
Ordering of phases,
Direction of current,
All nearby power lines
Does not take into account:
Other sources of ELF exposure, e.g. unbalancedreturn currents
29 marzo 2008Maria Feychting 24
Calculated fields, cont.
Allows for a quantification of dose-response
relationships
Use of the same unit of measurement (e.g. T)
allows for a better comparability between studies and
between exposure situations
e.g. residential vs occupational
Has primarily improved the specificity of the
exposure assessment
29 marzo 2008Maria Feychting 25
Measurements
+ Captures all sources of exposure
+ Allows for a quantification of dose-response
relationships
- Cannot be made historically
- May vary over time - less reproducible
- Sensitive to time of the day and
- Season of the year
29 marzo 2008Maria Feychting 26
Measurements
Short term spot measurements
May not be representative of long term historical exposure
Highly sensitive to variation of the exposure during the day
and year
24-48 hour measurements
Better representativity than spot-measurements, but may still
not capture historical exposures
Sensitive to exposure variation over different seasons
Personal measurements
For use in prospective cohort studies
Case-control studies: changes in habits after disease
occurrence may influence exposure assessment
29 marzo 2008Maria Feychting 27
Appliances
Magnetic fields falls of rapidly with distance
Can give high localized peak exposures
Electric razors (>0.4 T)
Microwave ovens (0.1-20 T)
Vacuum cleaner (2-20 T)
Small contribution to TWA 24-h exposure
Historically unstable – need to rely on self reports
Recall bias a potential problem
29 marzo 2008Maria Feychting 28
Occupational exposures
Occupational exposures higher than residential
High exposure levels rare – around 4% with >0.5 T
Exposure estimates:
Individual job titles
Grouping of job titles into “electrical occupations”
Job-exposure matrices (JEM)
29 marzo 2008Maria Feychting 29
Job-titles
Underlying exposure information in majority of
occupational studies – also in JEM-based studies
Information from census data or death certificates
One point in time, or several time-points with no information
for intermediate years
Independent of case-control status
Self-reported occupational history
Quality of information may differ between cases and controls
Exposure may vary considerably within the same
occupational title and over time
29 marzo 2008Maria Feychting 30
Electrical occupations
On average higher magnetic field levels than non-
electrical occupations
Electrical & electronic technicians and engineers, repairers of
electronic equipment, telephone & telephone line installers
and repairers, electricians, electric power installers and
repairers, supervisors of electricians and power transmission
installers, power plant operators, motion picture projectionists
Some of the highest exposed occupations are not
included
Welders
Train drivers
Sheet metal workers
29 marzo 2008Maria Feychting 31
Job-exposure matrices (JEM)
Improved exposure assessment – JEMs based on
personal measurements during workday
Aggregated estimates based on several
measurements per occupational title
Summarized into different measures of central tendency
Other metrics
Allows for quantification of dose-response
relationships
Has identified highly exposed occupations other than
”electrical occupations”
29 marzo 2008Maria Feychting 32
Job-exposure matrices
Arithmetic mean – commonly used
Sensitive to outliers
Geometric mean – also frequently used
Not so sensitive to outliers
Median
Not sensitive to outliers
Rate of change
Percent time above a certain exposure level
29 marzo 2008Maria Feychting 33
Weakness of JEMs
Cannot capture historical exposures
Exposure situation has changed in many occupations –
often increased exposures, e.g. librarians, cashiers
Does not take into account individual variations within
occupations
Mainly based on measurements for men – until
recently little was known about female occupations
Greater exposure misclassification for women
Larger number of occupations with no exposure information
for women – reduces statistical power
29 marzo 2008Maria Feychting 34
Further developments of JEMs
Extended from job-title to job-tasks
Takes into account variations within job-titles
Detailed mapping of exposures associated with different job-
tasks
Combines exposure level with estimates of time spent on
each job-task
Increases specificity of exposure estimates
29 marzo 2008Maria Feychting 35
Aspects of exposure assessment in
overall assessment of evidence
First generation of studies used the more crude
exposure assessment methods
Likely to have more extensive exposure misclassification
Later studies have improved exposure assessment
considerably
Likely to have less exposure misclassification
If true association between exposure and disease
–effect estimates should become stronger with
improved exposure assessment
Less dilution of effect estimates from exposure misclassification