Evaluation of statistical treatment of left-censored contamination data: example on Deoxynivalenol occurrence in pasta and pasta substitute products

The handling of data on food contamination frequently represents a challenge because they are often left censored, i.e., they are composed by both positive and non-detected values; the latter observations are not quantified and provide only the information that they are below a lab-specific threshold value. Besides deterministic approaches, that simplify the treatment by substituting non-detected values by fixed threshold or null values, there has been a growing interest in the application of stochastic approaches to the treatment of not quantified values. In this study, a Multiple Imputation method is applied to analyse contamination data on deoxynivalenol (DON), a mycotoxin that may be present in pasta and pasta substitute products. An application of the proposed method to left-censored DON occurrence data is provided and the results are compared to those attained by using deterministic (substitution methods) methodologies. In this context, the stochastic approach seems to provide a more accurate, unbiased and realistic solution to the problem of left-censored occurrence data, with DON contamination mean values of the food considered equal to 139.4 μg/kg with a 95% confidence interval ranging from 137.0 to 141.9 μg/kg, in which lower and upper bound values of the substitution methods are included (128.1 μg/kg and 160.8 μg/kg respectively). The obtained sample of occurrence values may represent a suitable dataset to be used in the assessment of dietary exposure scenarios of the general population.