method. The RPDs presented are for TNT or RDX.
the comparison (i.e., data on specific compounds,
The accuracy of colorimetric methods should im-
or total nitroaromatics or nitramines).
prove when compared to total nitroaromatics or
The precision and bias of the screening meth-
nitramines because the methods detect numerous
ods are most appropriately assessed by compari-
related explosives. As the level of nitroaromatics
son to established laboratory methods such as EPA
Method 8330. Methods of comparison that have
other than TNT increases, the accuracy of the
been used include relative percent difference
CRREL and EnSys RISc methods should appear
(RPD), linear regression, correlation, coefficient of
to decrease. When compared to total nitro-
determination (r2), percent false positive and false
aromatics, however, the accuracy should increase.
negative results, analysis of variance, and paired
Thus, to attempt to identify the preferred screen-
t-tests. It should also be remembered that the con-
ing method, it is important to determine specifi-
tribution of analytical error is generally quite small
cally what analytical information is desired from
compared to total error (field error is the major
a screening procedure and the relative concentra-
contributor).
tion of the explosives at a site. Readers should con-
Three studies have been conducted comparing
sult the original studies for more details; however,
the performance of two or more on-site methods
some summary conclusions from the three cited
studies follow.
with Method 8330. The procedures used in the
The EPA (1997) study compared the CRREL,
studies for making the comparisons are given here
EnSys RISc, DTECH, Idetek Quantix, and
and a summary of the results of each study fol-
Ohmicron RaPID Assay methods for TNT and
lows. EPA (1997) calculated RPDs (the difference
concluded that "no single method significantly
between the field and reference method concen-
outperformed other methods" and accuracies for
tration divided by the mean value and expressed
all the on-site methods were comparable. CRREL,
as a percent), established a comparison criterion
EnSys RISc, and Ohmicron were more accurate in
of 50% for RPDs, and determined the frequency
the greater-than-30-mg/kg TNT ranges, and
with which various methods met that criteria
DTECH was more accurate in the less-than-30-
within various sample concentration ranges. EPA
(1997) also calculated regression lines and the r2.
mg/kg range. The same study compared the
CRREL, EnSys RISc, and DTECH methods for
Haas and Simmons (1995) compared on-site meth-
RDX in soil and concluded that they were
ods using the percentage of false positives and
slightly less accurate than the corresponding TNT
false negatives for determining whether samples
methods.
were above or below two proposed remediation
Haas and Simmons (1995) evaluated immu-
criteria for TNT in soil, 48 and 64 mg/kg. They
noassay kits for TNT (DTECH, EnviroGard Tube
also plotted regression data and reported calcu-
lated r2 values. Myers et al. (1994) calculated re-
and Plate, Idetek Quantix, and Ohmicron RaPID
Assay). They concluded that for semiquantitative
gression lines with 99% confidence intervals.
screening, all kits have the potential to accurately
Although no study has compared all the field
screen soil samples for contamination at risk-based
methods under the same conditions, the three
levels (EPA 1993). The study found that, compared
studies evaluated multiple methods under slightly
with HPLC analysis below 1 ppm, several of the
different field conditions (EPA 1997, Haas and
assays had significant bias. Measurements near the
Simmons 1995, Myers et al. 1994). Summary data
detection limit are often problematic; above 1 ppm,
from these studies are provided in Table 6. The
the correlation between the immunoassay kits and
table includes the intercept and slope of regres-
HPLC was generally good.
sion lines for TNT and RDX data for two concen-
Myers et al. (1994) evaluated and compared the
tration ranges, from the detection limit to 100 mg/
kg and from 100 to 1000 mg/kg. Also included
EnSys RISc and DTECH methods for TNT in soil
are the correlation coefficient (r) and the mean RPD
versus EPA Method 8330. The study found that
(absolute value of RPDs). The ideal regression line
EnSys demonstrated a good one-to-one linear cor-
relation with reversed-phase high-performance
would have a slope of 1 and go through the origin
liquid chromatography (RP-HPLC) that can be
(intercept of 0). The correlation coefficient (r)
attributed to the procedure for extraction, i.e., a
shows the degree of association between the on-
large sample size of dried homogenized soil. For
site method and Method 8330 and can range be-
tween 1 and +1. For a perfect positive correla-
the DTECH kit, comparison was more difficult
tion, r = 1. The mean RPD closest to 0 shows the
because of the concentration range-type data and
because one-to-one linear correlation with RP-
greatest agreement with the reference laboratory
21
Previous Page
