Regional Technical Guidance Manual, Risk Assessment

Risk assessments often inappropriately report and handle data near the limits of detection. Common errors include (1) omission of detection limits, (2) failure to define detection limits which are reported, and (3) unjustified treatment of non-detects as zero. This guidance is intended to improve the quality and consistency of handling of data near the detection limit in risk assessments done in Region III.

Statement of the Problem

A. Reporting DLs. The practice of omitting information on DLs from risk assessments is inappropriate, both technically and ethically, because it conceals important uncertainties about potential levels of undetected risk. For example, failure to detect trichloroethene in drinking water at a DL of 50 parts per billion does not establish acceptable levels of health risk; failure to detect TCE at 0.05 ppb does. If risk assessors neglect to consider DLs for analytical data, they may overlook serious health threats. Furthermore, DLs should appear both in data summary tables in the body of the risk assessment, and in tables of raw data in appendices.

In a generic sense, there are two types of analytical lower limits: detection limits (DLs) and quantitation limits (QLs). The DL is the lowest concentration that can reliably be distinguished from zero, but is not quantifiable with acceptable precision. At the DL, the analyte is proven to be present, but its reported concentration is an estimate. The QL is the lowest concentration which can be not only detected, but also quantified with a specified degree of precision. At the QL, the analyte is both proven present and measured reliably.

B. Non-detection v. zero concentration. The routine assump tion that site-related contaminants, if undetected, are absent from samples is often unduly optimistic. Some frequently-encountered carcinogens (e.g., vinyl chloride and tetrachloroethene in drinking water, beryllium in soil) are significant potential health risks at levels below DLs. Risk assessors should use professional judgment, augmented by the decision path described below, to decide if hazardous contaminants should be assumed present at levels below the DL.

Existing Guidance

Section 5.4 of the EPA Risk Assessment Guidance for Superfund (RAGS) IA recommends that all data qualifiers should be reported in the exposure assessment, and that their implications be considered before the data are used for risk assessment. Section 6.5.1 suggests use of models when monitoring data are restricted by the limit of quantitation, and Section 5.3.1 contains guidance for re-analyzing samples and determining which data should be treated qualitatively.

EPA's Guidance for Data Useability in Risk Assessment (DURA) (October, 1990), Section 3.3.4, subdivides generic DLs and QLs, describing six different lower analytical limits. (1) The instrument detection limit (IDL) is three times the standard deviation of seven replicate analyses at the lowest concentration of a laboratory standard that is statistically different from a blank. (2) The method detection limit (MDL) is three times the standard deviation of seven replicate spiked samples handled as environmental samples. (3) The sample quantitation limit (SQL) is the MDL corrected for sample dilution and other sample-specific adjustments. (4) The contract required detection limit (CRDL) is the SQL which CLP laboratories are required to maintain for inorganic analytes. (5) The contract required quantitation limit (CRQL) is the SQL which CLP laboratories must maintain for organic analytes. (6) The limit of quantitation (LOQ) is the level above which analytes may be quantified with a specified precision, often +/- 30%. This precision is usually assumed to occur at ten times the standard deviation measured for the IDL. Section 4.2 of DURA describes a strategy for selecting appropriate analytical methods, which includes consideration of risk at the detection limit.

However, even with an optimum sample and analysis plan, risk assessors must still confront situations where significant risks can occur below the detection limit. Neither RAGS nor DURA presents a procedure for assessing risks from undetected, but potentially present, compounds, nor do they suggest a specific reporting format for detection limits. This Region III guidance document addresses these gaps in national risk assessment guidance. It is intended to augment, not replace, national guidance.

Discussion and Recommendations

A. Reporting DLs. Risk assessments should include analytical limits in all data tables, including summary tables. One of the following should be reported for all undetected analytes, in order of preference: SQL, CRDL (or CRQL), and LOQ (as described in DURA). Each data table in the risk assessment should clearly describe which limits are reported, and define them.

Risk assessments should use the format shown below for all data tables. Undetected analytes should be reported as the DL (i.e., either the SQL, CRDL/CRQL, or LOQ, in that order) with the code "U". Analytes detected above the DL, but below the QL, should be reported as an estimated concentration with the code "J".

Non-detects are reported as the sample quantitation limit, defined as three times the standard deviation of seven replicate spiked samples handled as environmental samples, corrected for sample dilution and other sample-specific adjustments.

B. Non-detection v. zero concentration. Risk assessors have the following methods to choose from, for handling data below the DL:

1. Non-Detects handled as DLs - In this highly conservative approach, all non-detects are assigned the value of the DL, the largest concentration of analyte that could be present but not detected. This method always produces a mean concentration which is biased high, and is not consistent with Region III's policy of using best science in risk assessments.
    
2. Non-Detects reported as zero - This is the best-case approach, in which all undetected chemicals are assumed absent. This method should be used only for specific chemicals which the risk assessor has determined are not likely to be present, using the decision path below.
    
3. Non-Detects reported as half the DL - This approach assumes that on the average all values between the DL and zero could be present, and that the average value of non-detects could be as high as half the detection limit. This method (or method 4, below) should be used for chemicals which the risk assessor has determined may be present below the DL, using the decision path below.
    
4. Statistical estimates of concentrations below the DL - Use of statistical methods to estimate concentrations below the DL is technically superior to method 3 above, but also requires considerably more effort and expertise than the three simpler methods. Also, these statistical methods are effective only for data sets having a high proportion of detects (typically, greater than 50%). Therefore, statistical predictions of concentrations below the DL (as described by Gilbert [1987] and reviewed by Helsel [1990]) are recommended only for compounds which significantly impact the risk assessment and for which data are adequate.

C. Decision Path. Summarizing the discussion above, method 1 (non-detects = DL) consistently overestimates concentrations below the DL, and should not be used. Risk assessors should use the following decision path to select among methods 2 (non-detects = 0), 3 (non-detects = DL/2), and 4 (specialized statistics) to achieve the least biased estimate of reasonable maximum exposure. The choice of method should be based on scientific judgment about whether: (1) the undetected substance poses a significant health risk at the DL, (2) the undetected substance might reasonably be present in that sample, (3) the treatment of non-detects will impact the risk estimates, and (4) the database is sufficient to support statistical analysis. The decision path below, followed by examples of appropriate selections, is recommended:

1. Is the compound present at a hazardous concentration in any site-related sample?  If no, assume non-detects are zero; if yes, continue. (Note that if the compound is not present in any sample at a hazardous level (e.g., 10-6 risk or a hazard quotient of 1), it probably should be dropped from the risk assessment.)

2. Was the sample taken down-gradient of (or, if no gradient exists, adjacent to) a detectable concentration of the chemical? If no, assume non-detects are zero; if yes, continue.

3. Do the chemical's physical-chemical characteristics (e.g., water solubility, octanol-water partitioning, vapor pressure, Henry's law constant, biodegradability, etc.), permit it to reasonably be present in the sample? Are other site-related compounds with similar characteristics present in the sample?  If no (to both questions), assume non-detects are zero; if yes (to either question), continue.

4. Does the assumption that non-detects equal DL/2 signifi cantly impact route-specific quantitative risk estimates? If no, assume non-detects equal DL/2; if yes, consider using statistical methods to estimate concentrations below the DL for that exposure route, assuming data quality permits.

D. Examples.

1. TCE is present in groundwater on site at 500 g/l, a potentially hazardous concentration. Elevated TCE concentrations are measured upgradient of a residential well, but TCE is not detected in the residential well itself. Other site-related chlorinated VOCs are detected in the residential well. The detection limit for TCE was 5 g/l (equivalent to 5 x 10-6 risk under the exposure scenario in the risk assessment).
   • Decision path: Step 1 - continue; step 2 - continue; step 3 - continue; step 4 - assume non-detects are DL/2. If multiple well samples are available, and TCE is detected in some, consider using specialized statistical methods.
      
2. Chromium is present in on-site soils at 10,000 mg/kg, a potentially hazardous concentration under direct contact exposure. Chromium is not detected in an adjacent off-site soil sample, although other site-related metals are. The detection limit for chromium in soil is 0.1 mg/kg, well below a hazardous concentration under the exposure scenario in the risk assessment.
   • Decision path: Step 1 - continue; step 2 - continue; step 3 - continue; step 4 - assume non-detects are DL/2, because using specialized statistics will not appreciably change the risk.
      
3. PCBs are not detected in 20 on-site soil samples. There is no history of PCB disposal at the site, and PCBs were not detected in any other medium.
   • Decision path: Step 1 - assume non-detects are zero.
      
4. Vinyl chloride, a site-related contaminant, is measured in surface water downstream of the site boundary at 10 g/l, a hazardous concentration for a resident receptor. Five hundred meters upstream of the site, vinyl chloride is not detected at a DL of 0.1 g/l.
   • Decision path: Step 1 - continue; step 2 - assume upgradient non-detects equal zero.
      
5. 2,3,7,8-TCDD is detected in an unfiltered monitoring well sample at 5 ng/l, a potentially hazardous concentration. The next downgradient well has no detectable TCDD. Pentachlorophenol, also detected in the first well, is not detected in the second.
   • Decision path: Step 1 - continue; step 2 - continue; step 3 - assume non-detects of both TCDD and PCP equal zero because of low mobility in groundwater.

References

• EPA. 1990. Guidance for Data Useability in Risk Assessment. EPA/540/G-90/008.
• EPA. 1989. Risk Assessment Guidance for Superfund. Volume I, Human Health Evaluation Manual (Part A). EPA/501/1-89/002.
• Gilbert, R.O. 1987. Statistical Methods for Environmental Pollution Monitoring. Van Nostrand Rheinhold Co., New York.
• Helsel, D.R. Less than obvious; statistical treatment of data below the detection limit. Environ. Sci. Technol 24(12);1767-1774.