Notice: The views expressed here are those of the individual authors and may not necessarily reflect the views and policies of the United States Environmental Protection Agency (EPA). Scientists in EPA have prepared the EPA sections, and those sections

1
Notice The views expressed here are those of the
individual authors and may not necessarily
reflect the views and policies of the United
States Environmental Protection Agency (EPA).
Scientists in EPA have prepared the EPA sections,
and those sections have been reviewed in
accordance with EPAs peer and administrative
review policies and approved for presentation and
publication. The EPA contributed funding to the
construction of this website but is not
responsible for it's contents. Mention of trade
names or commercial products does not constitute
endorsement or recommendation for use.
2
Exploring Connections Between Ecological
Condition and Human Health County-Level
Aggregation for Maryland
John F. Paul National Health and Environmental
Effects Research Laboratory U.S. Environmental
Protection Agency Research Triangle Park, North
Carolina 27711
13 May 2003
3
Acknowledgements Steve Hedtke Michael
McDonald Wayne Munns Hillel Koren Jim
Heltshe Allan Marcus
4
Presentation Outline

• Context
• Definitions
• Working Null Hypothesis
• Methods and datasets
• Results
• Next Steps

5
Context
Public policy continues to evolve in response to
new and complex social, economic, environmental
drivers. Science needs to anticipate,
understand, assess, and reduce risks to human
health and our environment to support mission of
Agency (protect human health and safeguard the
environment)
6
Context (contd)
Holistic approach to environmental problem
solving Future environmental problem solving
requires new and creative approaches -
out-of-the-box thinking and risk taking
7
Long-Term Perspective
Provide the Agency with the information, methods
models needed to make informed selections among
policy options, identify emerging risks,
minimize unintended consequences
8
Hypothetical Example Unintended Consequence of
Improved Ecological Conditions on Human Health
Outcomes
Environmental Issue
Unintended Consequence
Driver
Policy
Increased risk of Lyme disease exposure to
fecal pathogens
surface water quality
increased riparian buffers
environmental protection
9
Definition Eco-Human Connectivity
relationship between ecological systems and human
health
more than exposure-response link condition of
ecological systems with human health not
necessarily cause-effect
10
Starting Point
View humans and ecological systems as one overall
interacting system Some actions can affect both
human health and condition of ecological
systems Web of causation exists
11
Web of Causation
12
Web of Causation
Driver
13
Background for Working Null Hypothesis
Independence two things are independent if the
probability of either is the same whether or not
the other occurs If there are actions that can
affect both human health and condition of
ecological systems, then the appropriate
indicators for human health and condition of
ecological systems are not independent
14
Background for Working Hypothesis (contd)
Conditional probability probability of something
occurring when it is known that something else
has occurred Quantitative measure of eco-human
connectivity is an appropriately determined
conditional probability For example, probability
of occurrence of human health outcome for
specified condition of ecological system Or more
specifically, probability of higher than normal
infant mortality for a county if more than half
of streams in the county are impaired
15
Working Null Hypothesis
P ( Y ) probability of Y (unconditional
probability) P ( Y X ) probability of Y if X
occurs (conditional probability) Eco-human
connectivity null hypothesis P ( Y X ) P ( Y
) or conditional probability equals
unconditional probability
16
Working Null Hypothesis (contd)
If eco-human connectivity exists, then first step
is to disprove the null hypothesis Remainder of
presentation is focused on this
17
Data Bases Used for Analyses
Maryland Biological Stream Survey (1995-97) ?
Lattice sampling of 17 major drainage basins over
three-years with probability design ?
Approximately 300 stream segments sampled each
year ? Use data for benthic IBI from
1995-97 Compressed Mortality Data - CDC Wonder
(1978-1998) ? Under 1 year of age ? All
races ? Both genders ? All causes of
death NOAA Coastal Assessment and Data Synthesis
System (1990) ? Census of agriculture ?
Fertilizer use and sales ? Pesticides ? Land
Use /Land Cover ? Socioeconomics
18
Why Select Infant Mortality and Stream Condition?
Infant mortality has long been considered a
sensitive indicator of the impact of
socioeconomic disparities on the health of
populations (Gortmaker Wise 1997) Condition of
bottom-dwelling communities in streams is
integrative of environmental impacts occurring
over recent time
19
Why Maryland?
Only state that has reported survey data for
stream condition summarized at county level
20
(No Transcript)
21
P ( Y X gt Xo)
22
Application of Conditional Probability Approach
Preliminaries
Ability to identify undesired human health
outcome level exists, e.g., infant mortality
greater than 10 per 1000 Treat county as
fundamental unit equal weighting
Critical concept 50 of stream miles impaired in
county means probability of observing impaired
streams in county is 50
23
Conditional Probability Approach
Given Y undesired human health outcome -
binary (infant mortality in county above 10 per
1000) X ecological condition continuous
(percent of stream miles in county with impaired
benthos) Xo conditional value for ecological
condition Two-step approach to calculate P ( Y
X gt Xo) 1. Identify subset of population with
X gt Xo 2. Determine fraction of subset with
undesired outcome Calculate P ( Y X gt Xo) for
all observed values of X
24
(No Transcript)
25
The basic underlying question Is the association
real or fantasy? Answer by trying to disprove
null hypothesis
26
Test null hypothesis for individual stream
condition value Cant use t-test, two sets tested
are not independent Result using bootstrap
estimation If greater than 78 of stream miles
in county are impaired there is less than 2
chance that value for probability of infant
mortality exceeding 10/1000 could have occurred
randomly Look at entire range of values to
determine when null hypothesis disproved
Bayesian model to account for uncertainty
27
Bayesian Model (mean and 95 posterior intervals)
Probability of infant mortality gt 10/1000
28
Md MBSS Pa, WVA EMAP-SW
WVa
Md
Pa
29
Risk calculations for infant mortality in a
county for fraction of stream miles in county
with impaired benthos (number of counties in
each category)
30
(No Transcript)
31
Where is this all going?
Continue evaluation of eco-human connectivity for
additional data sets Interesting observations can
lead to hypothesis generation Useful for
identification of gaps/needs/unidentified
problems Web of causation - Possible tool to help
target management actions Web of causation -
Conceptual model for problem identification/formul
ation in risk assessment Limitation County is
smallest level of aggregation for mortality data
32
Healthy Ecosystems Healthy People
Unhealthy Ecosystems Unhealthy People
Unhealthy Streams Unhealthy Children
33
THE END