ASTM E2677 Limit of Detection Web Portal
Data Input Requirements
All data input must be numerical; two or three significant digits are
usually adequate. Two windows are provided on the
Data Entry Page in which to paste columns of
analyte masses (left window) and detector response data (right window)
from a data storage application such as Microsoft Excel. The data inputted
in each window must be replicated mass-response pairs with the following
requirements:
- at least 10 replications per level
- at least three distinct levels, including one for process blank
response (mass = 0)
Dimensional units for mass levels and responses, while not identified,
must be consistent. The web-tool will report the limit of detection
value in the same dimensional unit as the mass inputs.
Before calculating the limit of detection, the web-tool checks data
quality and will abort for any of the conditions below:
- Less than 10 replications in any level
- Less than three distinct levels
- Absence of a process blank level (mass = 0)
- Number mismatch between mass level and response variables
- Responses in the highest mass level are not significantly different
from those in the process blank level
When data quality passes, the web-tool will perform the calculation and
return the estimate for the limit of detection and the 90% upper
confidence LOD90 limit (a measure of uncertainty). Also, other results
may be returned if those options are preselected. Results may also be
returned with a message that data quality was marginal and offer a
suggestion for improving the quality.
The LOD90 value determined is the best estimate of the minimum mass of a
particular analyte required to elicit a real and reliable response in
the detector tested. This minimum detector response is the Critical Value
(CV, an optional printed output), which should relate to the peak
detection threshold value that can usually be set manually in the trace
detector. The peak detection threshold value is usually set higher than
the CV unless the LOD90 is determined using realistic chemical background.
Caveat: The LOD90 determination assumes normality for the non-zero
measurements but will calculate an LOD90 value with data that may contain
statistical outliers. Large biases in LOD90 estimates are possible depending
on the values of the outliers and the total number of replicates in that
mass level. The user is responsible for checking the quality of the data.
To help with this assessment, the following analyses and plots may be
generated:
- a
Grubbs test for a single outlier is generated for each level
of the mass for the non-zero measurements.
- a scatter plot of the measurements against their associated mass
levels
- a multiplot showing a
normal probability plot of the non-zero
measurements for each distinct level of the mass.
The graphs are generated in Scalable Vector Graphics (SVG) format. SVG
is now natively supported in most major browsers. The one exception is
Internet Explorer (IE). Native SVG support is available for version 9 of
IE. However, an SVG plugin is required if an earlier version is used.
Graphs may be generated in JPEG format if an SVG capable browser is
unavailable.