CONTAMINANT DATA ASSESSMENT AND PROCESSING
Ocean Pollution Research Program (VA CORI)
Internal laboratory protocol document (V. 1 January 18, 2016)
Summary
Due to limitations in chemical analysis procedures, low contaminant concentrations cannot be precisely measured. These concentrations are said to be below the detection limit (DL). Datasets generated by analytical laboratories therefore comprise both quantified results and results below the DL. In order to generate the most reliable workable datasets, substitution of non-detects represents a key exercise. It can be done with two main goals in mind:
- the calculation of total contaminant concentrations by class (e.g. sumPCB or sumPBDE), or;
- the exploration of complex, congener-specific datasets by multivariate statistical analysis (e.g. PCA) wherein DL substitution rules could skew the resulting pattern.
This guidance document describes approaches for these two goals, recognizing that different study designs may require some modulation of data-preparation efforts.
- For the calculation of total contaminant concentrations, all non-detects will be replaced by the DL followed by blank correction.
- For the exploration of congener patterns, only congeners that are detected in more than 70% of the samples will be considered.
STAGE 1
Step 1: Assess data quality
Replicates, matrix spikes, reference samples, surrogate recoveries and/or lab and field blanks should be checked to make sure they meet the data quality objectives (DQO). Below is a table with the criteria used by Capital Regional District (CRD) when dealing with their data. We will follow the ‘pass’ rule defined in Golder (2009).
Table 1: Criteria for distinguishing between marginal and severe DQO failures (Golder, 2009).
Step 2: Non-detect substitution
Substitution of non-detects represents an exercise in support of two potentially divergent end games: i) the calculation of total contaminant concentrations by class (e.g. sumPCB or sumPBDE) using the congeners values, or ii) the exploration of complex, congener-specific datasets by multivariate statistical analysis (e.g. principal component analyses (PCA)) wherein detection limit (DL) substitution rules could skew the resulting pattern. For the latter, a smaller dataset with robust numbers from a precautionary and strictly pre-prepared set of results is necessary.
This guidance document describes approaches for these two end-games, recognizing that different study designs may require some modulation of data-preparation efforts. Users are encouraged to assess the implications of their decision on data preparation (DL substitution) so that they can demonstrate the implications on i) total concentration, and ii) patterns as portrayed through PCA.
For compounds that are non-detects (flagged as ‘U’), Axys reports the DL, make sure to first scan the dataset to identify those samples (highlight the cell in red).
There are various methods available for the replacement of non-detects ranging from a 0 substitution to a 100% DL substitution for all non-detects. However, in many cases, total concentration results may not be affected in any significant way (Table 2).
Table 2: Average difference between total PBDE, PCDE, PCN concentrations in harbour seal blubber after a detection limit substitution and after a 0 substitution.
| Mean difference (%) | |
| PBDEs | 0.8 |
| PCDEs | 2.7 |
| PCNs | 39 |
For preliminary data analyses (notably, the calculation of total contaminant concentrations by class), all non-detects will be replaced by the DL.
Step 3: Blank correction (and lipid correction if needed)
Laboratory blanks are analyzed with each batch of samples and concentrations reported in the blanks should be deducted from contaminant concentrations reported in samples from the same batch of analysis. Contaminant concentrations will also be corrected for field blank, when available.
When appropriate (for example, seal blubber biopsy data), contaminant concentrations should also be lipid corrected.
STAGE 2
Subsequent data exploration for any full or partial dataset is likely to require a revision to the original DL substitution rules applied in Stage 1. This is to ensure maximum comparability among data, where different DLs were generated, or different batches analyzed. This effort is particularly important in the case of multivariate analyses (e.g. PCA), where inappropriate DL substitution rules can lead to skewing of datasets and inaccurate outcomes in graphical representation. For example, DL substitution for congeners that have been largely undetected in a PCA dataset is likely to produce erroneous plots wherein ‘non-detected’ congeners assume equal weight to detected congeners. In order to maximize the strength of the dataset for e.g. a PCA, substitution of DL should only proceed in cases where at least 70% of samples have detectable concentrations.
Step 1: Calculate detection frequency
Detection frequency for a given dataset (i.e the proportion of results that are quantified above DL) should be calculated to support a decision to substitute a non-detect or delete a particular congener. Whenever possible (i.e multiple samples collected at each site), the detection frequency should be calculated on a site-specific basis.
Step 2: Only include congeners with a detection frequency of at least 70%
Only the congeners with a detection frequency of at least 70% will be included. This method has been used for a variety of marine environmental samples (Ross et al., 2004; Cullon et al., 2009; Desforges et al., 2013; Brown et al., 2015).
Decision framework for contaminant data treatment
References
Brown, T. M., Ross, P. S., & Reimer, K. J. (2015). Transplacental transfer of polychlorinated biphenyls, polybrominated diphenylethers, and organochlorine pesticides in ringed seals (Pusa hispida). Archives of environmental contamination and toxicology, 1-8.
Cullon, D. L., Yunker, M. B., Alleyne, C., Dangerfield, N. J., O’Neill, S., Whiticar, M. J., & Ross, P. S. (2009). Persistent organic pollutants in Chinook salmon (Oncorhynchus tshawytscha): implications for resident killer whales of British Columbia and adjacent waters. Environmental Toxicology and Chemistry, 28(1), 148-161.
Desforges, J. P. W., Ross, P. S., & Loseto, L. L. (2013). Metabolic transformation shapes polychlorinated biphenyl and polybrominated diphenyl ether patterns in beluga whales (Delphinapterus leucas). Environmental Toxicology and Chemistry, 32(5), 1132-1142.
Ross, P. S., Jeffries, S. J., Yunker, M. B., Addison, R. F., Ikonomou, M. G., & Calambokidis, J. C. (2004). Harbor seals (Phoca vitulina) in British Columbia, Canada, and Washington State, USA, reveal a combination of local and global polychlorinated biphenyl, dioxin, and furan signals. Environmental Toxicology and Chemistry, 23(1), 157-165.