nondetects

Nondetects refer to measurements in which the substance of interest is present but at concentrations below the method's detection limit. The true value is unknown and is assumed to be less than the detection limit; results are often reported as not detected, below detection limit, or less than LOD. Nondetects are a form of left-censored data, because the data provide a threshold below which values lie but do not specify exact numbers.

In environmental, clinical, and laboratory analytics, nondetects occur when the instrument cannot reliably quantify the compound. Detection limits are determined by instrument sensitivity, sample matrix, and analytical protocol, and are related to the limit of quantification, LOQ. A nondetect does not imply the absence of the substance, only that its concentration is below the method's capability.

Handling nondetects is a methodological challenge. Substitution methods (for example replacing nondetects with LOD/2 or LOD/√2 or zero) are simple but can bias estimated means, variances, and exposure assessments, especially when many nondetects are present. More robust approaches treat nondetects as censored data and use statistical methods such as maximum likelihood estimation for censored data, Tobit regression, regression on order statistics (ROS), or survival-analysis techniques like Kaplan-Meier estimators. Multiple imputation can also be used to reflect uncertainty. In small datasets or when nondetects are rare, simpler methods may be acceptable; otherwise, dedicated censored-data methods are preferred.

Best practices include reporting the detection limit, the number of nondetects, and the method used; employing appropriate censored-data methods; and avoiding overinterpretation of substituted values.