Substance concentrations generation

Both chronic and acute dietary exposure assessments rely on assigning substance concentrations to consumed modelled foods. For chronic exposure assessments, this concentration should be the mean concentration of the food and substance, as obtained from the concentration models. For acute, these concentrations are obtained through random sampling, for which there are two distinct approaches: sample-based and substance-based.

Sample-based concentrations generation

In the sample-based approach, the analytical samples from the concentration data form the basis for generating concentrations. For each identified modelled food of a consumption, substance concentrations are generated by drawing a random sample from the set of all samples available for that modelled food. Assuming that for the drawn sample, substance concentration values are known for all substances of interest (i.e., all missing values and censored values are imputed with either a zero concentration or a positive concentration at or below LOR), the substance concentrations for all substances of the assessment group are set to the substance concentrations of the drawn samples. The rationale behind this approach is that it maintains correlations between substance concentrations on the same food.

As mentioned, the sample based approach relies on all samples being analysed for all substances of interest. Often, this is not the case and for a given sample, concentration may missing for one or more substances. Also, this approach requires censored values to be imputed with either positive concentration or a zero concentration.

For imputation of missing values there are two approaches:

  1. Imputation by zero: all missing values are assumed zero.

  2. Imputation using substance-based concentration models: all missing values are imputed by drawing a concentration value from the substance-based concentration models.

For imputation of censored values, two approaches exist:

  1. Replace by zero: Censored values are imputated by a zero concentration value. This is an optimistic approach.

  2. Replace by factor times LOR: Each censored value is replaced by a factor f (e.g., 1 or 1/2) times its LOR.

  3. Replace by factor times LOD LOQ system: Non-detects are replaced by f * LOD; non-quantifications are replaced by LOD + f * (LOQ - LOD) and factor f is e.g., 1 or 1/2.

Substance-based concentrations generation

In the substance-based approach, substance concentrations for a given food are drawn independently per substance from the food/substance concentration models.