Practical Approaches for Studying Sea Turtle Health and Disease
395
positive result is the proportion of all animals that test positive that really are positive.
In general, the less common the condition, the less predictive value a given test has
and the less confidence can be placed in the result. For example, if the true prevalence
of a given condition is 50%, a test with 95% specificity and 100% sensitivity will
yield 2.5 false positives among 100 animals tested, and the predictive value of the
test will be 95%. If, however, the true prevalence in the population is only 5%, then
4.75 false positives are expected and the predictive value declines to only 51%. That
is, only 51% of the positive test results can be interpreted as being correct.
These statistical artifacts are amplified when a battery of independent tests are
performed. Because each test has its own independent probability of being found
out of range or false positive, the overall probability of finding at least one normal
individual that will have abnormal test results increases with the number of tests
performed. Similarly, when comparing different sample populations to one another
or to a reference distribution, the chances of finding a statistically significant difference increases with the number of independent pair-wise comparisons that are made.
Thus, interpreting the sporadic positive test, out-of-range result, or statistically
significant difference between sample populations becomes somewhat of an intuitive
skill, and is especially difficult when one is surveying an apparently healthy population for conditions that are rare. A strong argument can be made for using the best
tests (high specificity and sensitivity), testing the most closely matched reference
population possible, and employing confirmatory tests when available to help distinguish false positives from true positives (Weisbroth et al., 1998). When a diagnostic test is used to monitor a population for the introduction of a known disease
or infection, or to maintain some level of confidence that the population is free of
a specific disease, it is especially critical to employ confirmatory tests if the surveillance data will be used to support management decisions involving the culling of
positive animals or quarantine of populations.
15.3.1.4 Interpretation of Within-Range and Negative Results
When quantitative test values are compared to an inappropriate reference population,
values that are actually abnormal may be misinterpreted as being within range.
Interpretation of within-range and negative test results also must consider the sensitivity of the test — its ability to identify all the abnormal individuals (true positives)
in the population. Many tests that are used as screening tools are set up to maximize
sensitivity, thereby minimizing false-negative results. Nevertheless, test results that
fall within the normal reference range do not necessarily mean that there is not a
problem. Some tests, such as certain blood biochemistry assays, are relatively insensitive to the underlying disease processes. In many cases, a threshold level of ongoing
tissue damage or loss of function must be reached before abnormalities are detected
on a particular test parameter (Meyer et al., 1992). Because many organ systems
have redundant physiologic capacity, significant pathology and loss of organ function
may go undetected when certain tests are used. For tests that yield categorical results,
there are limits of detection inherent to the method that affect sensitivity. For
example, PCR in theory may be able to detect a single virus genome in a sample,
but in practice, it may require ten or more viral particles to be present (Persing,
1123 book.book Page 395 Monday, November 11, 2002 11:11 AM
395
positive result is the proportion of all animals that test positive that really are positive.
In general, the less common the condition, the less predictive value a given test has
and the less confidence can be placed in the result. For example, if the true prevalence
of a given condition is 50%, a test with 95% specificity and 100% sensitivity will
yield 2.5 false positives among 100 animals tested, and the predictive value of the
test will be 95%. If, however, the true prevalence in the population is only 5%, then
4.75 false positives are expected and the predictive value declines to only 51%. That
is, only 51% of the positive test results can be interpreted as being correct.
These statistical artifacts are amplified when a battery of independent tests are
performed. Because each test has its own independent probability of being found
out of range or false positive, the overall probability of finding at least one normal
individual that will have abnormal test results increases with the number of tests
performed. Similarly, when comparing different sample populations to one another
or to a reference distribution, the chances of finding a statistically significant difference increases with the number of independent pair-wise comparisons that are made.
Thus, interpreting the sporadic positive test, out-of-range result, or statistically
significant difference between sample populations becomes somewhat of an intuitive
skill, and is especially difficult when one is surveying an apparently healthy population for conditions that are rare. A strong argument can be made for using the best
tests (high specificity and sensitivity), testing the most closely matched reference
population possible, and employing confirmatory tests when available to help distinguish false positives from true positives (Weisbroth et al., 1998). When a diagnostic test is used to monitor a population for the introduction of a known disease
or infection, or to maintain some level of confidence that the population is free of
a specific disease, it is especially critical to employ confirmatory tests if the surveillance data will be used to support management decisions involving the culling of
positive animals or quarantine of populations.
15.3.1.4 Interpretation of Within-Range and Negative Results
When quantitative test values are compared to an inappropriate reference population,
values that are actually abnormal may be misinterpreted as being within range.
Interpretation of within-range and negative test results also must consider the sensitivity of the test — its ability to identify all the abnormal individuals (true positives)
in the population. Many tests that are used as screening tools are set up to maximize
sensitivity, thereby minimizing false-negative results. Nevertheless, test results that
fall within the normal reference range do not necessarily mean that there is not a
problem. Some tests, such as certain blood biochemistry assays, are relatively insensitive to the underlying disease processes. In many cases, a threshold level of ongoing
tissue damage or loss of function must be reached before abnormalities are detected
on a particular test parameter (Meyer et al., 1992). Because many organ systems
have redundant physiologic capacity, significant pathology and loss of organ function
may go undetected when certain tests are used. For tests that yield categorical results,
there are limits of detection inherent to the method that affect sensitivity. For
example, PCR in theory may be able to detect a single virus genome in a sample,
but in practice, it may require ten or more viral particles to be present (Persing,
1123 book.book Page 395 Monday, November 11, 2002 11:11 AM