| Summary
During the 1990s, persistent organic pollutants have been
quantified in hundreds of samples collected from polar bears Ursus
maritimus and glaucous gulls Larus hyperboreus in the
Norwegian Arctic. Based on an overall objective to detect temporal
changes in pollutant loads, the primary objective of the present
report is to give advice on alternative strategies for continued
sampling. Special attention was given to statistical power as a
measure of the performance of a monitoring programme. In the polar
bear dataset, the abundant PCB-congener PCB-153 was less variable in
plasma than in blood cells or subcutaneous fat. Therefore, we suggest
that blood plasma may be a preferable sample type for monitoring.
Significant associations were found between concentrations of PCB-153
in polar bear plasma and reproductive status, sampling location,
sampling season, and nutritional condition. A statistical model was
used to adjust for the influence of these confounding variables. Using
the adjusted data, we found that PCB 153 decreased significantly in
polar bear plasma from Svalbard during the 1990s. Furthermore, the
within-year and between-year variation in the polar bear data for
PCB-153 were used to estimate statistical power of various sampling
regimes. Due to random variation between years, we found it unlikely
to make significant findings on temporal trends with less than 7-8
years of sampling. About 15 years of sampling are needed to be >90%
sure to detect a trend with an average change rate of 5% per year. For
the glaucous gull, the different possible sample types were compared
and the merits of repeated sampling on the same individuals were also
discussed. For monitoring of temporal trends in glaucous gulls, we
recommend the use of eggs samples. Both the polar bear and the
glaucous gull results indicate that long-term dedicated effort is
crucial for successful monitoring. "Dedicated effort"
implies that the basic requirement in all monitoring, i.e. to
get a fixed number of similar samples from the same location/colony at
regular time intervals, must not be relaxed in order to accommodate
other research objectives.
Scientific results
Goal 1: Investigate the variability of different
samples from polar bears and glaucous gull (e.g. bloodsamples compared
to fatsamples)
The evaluation of polar bear data was based on PCB congener 153.
PCB 153 was strongly correlated to sum of all PCB congeners, and also
the most abundant congener in the bears. When data from different
tissues were compared, PCB-153 was less variable in plasma than in
blood cells or subcutaneous fat The PCB-153 variance in milk fat was
comparable to the variance in plasma, , but the obvious limitations in
availability of polar bear milk limits the utility of this sample
type.
Different types of sampling procedures from the glaucous gulls were
evaluated. The conclusion was that egg samples were the least
destructive and easiest available samples that at the same times are
least influenced at external factors. It is, however, important that
the eggs are collected from exactly the same place each year. The eggs
don’t catch up the high levels in some males, but this should not be
important when the objective is to monitor changes over time.
Goal 2: Elucidate the consequences of different sampling
procedures (number of samples per year, number of years with sampling,
different tissues sampled from, repeated samples from the same
individuals) for polar bear and glaucous gull, based on existing data.
The polar bear data was used to visualise in which way statistical
power can be expected to increase with number of sampling years.
Different scenarios illustrate the effects of between year variance,
within year variance, different sampling sizes, and the consequences
of biannual sampling. Due to random variation in PCB levels between
years, it is very unlikely to detect any changes in PCB levels the
first 7-8 years of the study, on the condition that the actual change
is about 5% each year. Based on estimates for the random variation
between individuals and the random variation between years, and using
a sample size of 20 each year, about 14 years will pass before it is
possible to be 90% sure to detect a 5% rate of annual change with
statistical significance. The only way to compensate for the random
variation between years is to take samples in a sufficient number of
years. (That is; a large sample number per year does not overcome the
problem of between year random variation.) The incremental value of
each sample decreases rapidly when the sample size passes 20. An
"8 samples per year program" need only two additional
sampling years to attain the statistical power of a "20 samples
per year program". The conclusion is that 10 to 25 samples each
year is the optimal sample size for monitoring studies, depending on
the incremental cost of additional samples and available resources.
It is possible to save resources with biannual sampling, but this
is not recommended. The time needed to attain statistical
"power" is unsatisfactorily long with annual sampling, and
biannual sampling would further increase the time needed to draw
conclusions. In addition, the actual rate of change in pollution
levels could itself change over time. With biannual sampling,
interpretation of the data would more likely be confounded by changes
in rate of change during the sampling period.
A sampling procedure with repeated sampling from the same glaucous
gull, or its egg, was evaluated. This method was not found suitable.
Goal 3: Make a statistical model for
levels of POPs in polar bear with possibility to correct for condition,
gender, and reproductive status. A model could reduce the unexplained
variation in POP levels, and increase the possibility to demonstrate
trends in POP concentrations in time and space.
A regression model was fitted for PCB-153 in polar
bear plasmaHalf of the total variation in PCB-153 could be related to
fat content in plasma, nutritional condition, reproductive status,
eastern longitude of sampling location, and age. After adjustment for
these factors, PCB-153 was 45% lower in the period 1994-97 than the
period 1990-93, significant at p<0.0005. The report further
discusses to what extent it is possible to estimate an average
percentage of yearly change from the data. This is associated with
problems related to large differences in sampling sizes between years.
Random subsamples of the data were selected to simulate the same
sample size each year. In most random subsampling cases, there was a
significant decrease in PCB-153, estimated to about 10% per year.
Goal 4: Develop a general computer program to estimate
statistical power of different sampling procedures based on knowledge
of variance and hypothetical change rates. The costs and performance
of different monitoring programmes could then be estimated when
variation within years is known.
The estimation of statistical power required more advanced
statistical functions than first thought. Visual Basic could not be
used without a supplementary statistical module. The calculations can
bee done on a spreadsheet in the statistical program JMP . This
spreadsheet may be available to NP (with instructions) if required.
The main point is, however, that good estimates of variation within
years and between years are necessary to estimate power. Variation
between years can not be estimated in advance, and it is therefore
necessary to make assumptions. The magnitude of the variation between
years is more important than the number of samples each year. A
program that make precise statements of "power" based on
number of samples and sampling years would be misleading as long as
the important variation between years is poorly known.
Relevance for monitoring
The document gives first hand information on sampling procedures
including numbers of samples for monitoring studies. The document is
based on the last ten years of ecotox studies of glaucous gulls and
polar bears at the Norwegian Polar Institute and the Norwegian
Institute for Nature Research (NINA).
|
