Long Term-Method Detection Levels (LT-MDL)
for the US Geological Survey, National Water Quality Laboratory

Calculations

The LT-MDL can be calculated in several different ways, depending upon available data. The traditional method for determining method detection limits is through the analaysis of the variability of low-concentration spike samples in a clean matrix. At the National Water Quality Laboratory, an alternative approach uses blind blank samples instead of spiked sample. Blind blanks are submitted weekly throughout the year to those analytical units that report values for blanks (other than less-than's). Blind blanks are the preferential LT-MDL sample medium. Lastly, the laboratory set blank results (internal quality control samples) are used for comparison to other data, and for calculation of the LT-MDL when the data are appropriately collected.

The basic calculation for detection limits assumes a normal distribution of LT-MDL sample results, and assumes that very low concentration spike samples will have an equivalent distribution as the blind blank samples. Therefore, spike sample results are re-centered with zero as the center value, as are blank sample results. The LT-MDL is then calculated by:

LT-MDL = s x t_{(n-1, 1-α = 0.99)}

The laboratory reporting level (LRL) is the less-than value reported when nothing is detected in a sample, or the concentration is <LT-MDL. It is set at two times the LT-MDL to set the chance of a false negative to be less than 1 percent.

LRL = 2 x LT-MDL

Another calculation used to assure the false positive rate is set correctly, is roughly equivalent to the 99th percentile of the blanks (blind or set). Since the definition of the LT-MDL strives to report data that is 99 percent certain to be above the background level, the 99th percentile of the blanks should deliver anideal LT-MDL concentration. For inorganic analyses that provide uncencored blank data, this is almost always the case. When blank detections are infrequent, or are due to random error, this process (99th percentile) may not provide the best estimation of the detection level.

LT-MDL = 99th Percentile of the blank population
(for convenience, set to the 2nd highest ranked blank when n<100).

There can be up to four different LT-MDL values assessed for an analyte when spikes and blind blanks data are available. The highest of the four LT-MDL calculations is selected automatically by our software, but can be overidden at the time of assessment based on extenuating circumstances. The four LT-MDL types are:

1. The calculated LT-MDL based on the blind spike submissions (n=24)
2. The calculated LT-MDL based on the blind blank submissions (n = 50)
3. The 2nd highest ranked ( approximating the 99th percentile) of the blind blank submissions
4. The 99th percentile (actual) of the set blank data (when n>100).

Outliers -

All blind sample results that are returned and are greater than 2xLT-MDL are flagged for re-analysis. The results of the reanalysis will overwrite the existing value if the analyst determines the new value is more appropriate.

The Grubb's Test is used to eliminate outliers at the end of the LT-MDL year. This test assesses suspect values against all of the values, including any outliers. If the calculated Z-value for any value is less than the critical value (look up table), then the P value is greater than 0.05. In other words, there is greater than a 5% chance that this value could be due to chance alone.

Outliers are not included in calculations using the s x t approach. Outliers are included in the 99th percentile calculations.

Exceptions -

In some cases, an LT-MDL determined by this process may be unattainable or impractical in real-life situations. There are several alternatives to this. First, we do not want misrepresentative data to be used to set the LT-MDL. If the analysts determine that something unusual happened to cause the LT-MDL samples to return unusual results, we should not use that LT-MDL.

Rule 1 - Data points should not be manually removed from the data set unless they are known errors (wrong sample or compromised sample). The Grubb's outlier test will remove outliers, reanalysis will fix "errors" where appropriate, but the remainder of the results should be assumed to be part of the normal analytical process.

Rule 2 - If the highest LT-MDL for an analyte contains misrepresentative data that negatively affects the result, choose the next highest LT-MDL. For example if the spike solution degraded over the year, you wouldn't want the variability of that data to represent the variability of the method since it would not affect samples similarly.

Rule 3 - If none of the LT-MDL's for this year are useable, the existing LT-MDL may:

• remain unchanged from the previous year (no update),
• convert to an MRL (minimum reporting level) reporting scheme with an assigned censoring value, or
• convert to an iLTMDL and an IRL (interim reporting level) reporting scheme based on a best-guess value.
  

Frequency -

The LT-MDL year runs from May 1 to April 30. This allows the Branch of Quality Systems time to assess the data before any updated LT-MDLs are entered into the laboratory database on October 1. The updates correspond to the new water year for the USGS projects, and is therefore the most practical time to make any changes.

Updates -

An existing LT-MDL is only updated when the new LT-MDL is outside of the 95th confidence interval of the chi-square distribution. The results based on the 99th percentile calculations are updated when the distribution-free confidence interval for percentiles indicates to do so (starting 10/01/2009).

Rounding -

LT-MDL values are presented by BQS to the least signigicant digit.

REFERENCES:

USGS Series	Open-File Report
Report Number	99-193
Title	New reporting procedures based on long-term method detection levels and some considerations for interpretations of water-quality data provided by the U.S. Geological Survey National Water Quality Laboratory
Edition	-
Language	ENGLISH
Author(s)	Childress, Carolyn J. Oblinger; Foreman, William T.; Connor, Brooke F.; Maloney, Thomas J.
Year	1999
Originating office
USGS Library Call Number	(200) R29o no.99-193
Physical description	iv, 19 p. ill. ;28 cm.
ISBN