It has been known that substitution is not a suitable method to address non-detect (ND) data and can result in significant errors. However, reports of reliable statistical methods for handling ND data of polychlorinated dibenzo-p-dioxin/furans (PCDD/Fs) and dioxin-like polychlorinated biphenyls (DL-PCBs) in fish are rare. This work examines US-EPA's recommended statistical methods and traditional substitution-based methods for handling ND data in PCDD/F and DL-PCB datasets obtained from 240 fish samples. The Kaplan-Meier (KM) method was confirmed as a suitable statistical approach for dealing with our fish datasets. The results of the KM and substitution methods were compared; a significant difference was found in TEQs derived from KM and substitution for PCDD/Fs, suggesting that the handling of ND data is a critical issue for PCDD/Fs. For example, the substitution by ½LOD resulted in 20-25% overestimation of PCDD/F TEQs for fish samples with ND data (60-70%). In conclusion, the use of KM method is preferable for average TEQs, and if the use of KM is not applicable, substitution by zero is preferred. Based on this conclusion, the characteristics of fish contamination by PCDD/Fs and DL-PCBs in Korea were investigated.
Keywords: DL-PCBs; Fish; Korea; Non-detect data; PCDD/Fs.
Copyright © 2016 Elsevier Ltd. All rights reserved.