One of the issues in quantitative human health risk assessments is the handling of nondetect data. Typically nondetect data are reported by the analytical laboratory as sample quantitation limits (SQLs) and qualified to indicate that the values were nondetect. Nondetect values, however, could actually fall anywhere within the range of zero to the SQL.
Four substitution methods for handling nondetect data were applied to a database of soil concentrations from a site to evaluate the impact on resulting constituent-specific 95 percent upper confidence limits (UCLs). The four methods utilized were: (1) one-half the SQL; (2) one-fifth the contract required quantitation limit (CRQL) (3) zero; and (4) the use of a random number generator to represent nondetect data with values ranging between zero and the SQL.
At low frequency of detections, the currently recommended USEPA Superfund approach of representing nondetect data with one-half the SQL was the least conservative method. As the frequency of detection increased, the resulting 95 percent UCLs for this method became more conservative (larger).