Human monitoring analysis calculation
Human monitoring analysis computes substance concentration estimates for a biological matrix (e.g., urine or blood) based on human monitoring data. These estimates are specified at the level of long term average concentrations for individuals in case of chronic assessments, or concentrations for individual-days in case of acute assessments. The concentrations are computed independently for each substance and biological matrix.
The main steps for computing the human monitoring concentration estimates are:
Imputation of censored values.
Imputation of missing values.
Apply conversion of substance concentrations from other biological matrices.
Calculation of individual concentrations (chronic) or individual day concentrations (acute).
Imputation of censored values
Similar to concentrations measurements in food, human monitoring measurements contain measurements below the limit of reporting and similar to concentrations modelling in foods, human monitoring analysis addresses these censored values and replaces them with imputed concentration values. Two approaches are available:
Impute using a non-detects handling method.
Impute using a draw from the left tail of the censored lognormal distribution. See also concentration models and concentration model types.
The available non-detects handling methods for deterministic imputation are:
Replace censored values by zero.
Replace censored values by a factor * LOR, the factor is set between zero and one.
Replace non-detects by a factor * LOD and non-quantifications by LOD + factor * (LOQ - LOD), the factor is set between zero and one.
Note that for the option based on the censored lognormal distribution also a non-detects handling method needs to be specified. Occasionally, fitting the censored lognormal model fails and the deterministic imputation method is used as fallback.
In Figure 80, the human monitoring concentrations data for three bisphenols are imputed with a factor * LOR and the summary statistics are visualized.
Figure 80 Boxplots for imputed concentration data for bisphenols BPA, BPS and BPF. Lower whiskers indicate the p5 and p10 percentiles, upper whiskers the p90 and p95. The edges of the box indicate the p25 and p75 percentiles with the median in the centre of the box.
Imputation of missing values
For missing concentration measurements, three imputation methods are available. The third opion ignores imputation and all missing values remain in the data set.
Replace missing values by zero.
For each substance and combination of biological matrix and sampling type, replace missing values by a random draw from the non-missing concentration values (samples). This is conditional on the specified minimum percentage of non-missing values.
Do not impute missing values
Imputation of missing values by a random draw from the data is conditional on the specified percentage of non-missing values. When the percentage of non-missing values for a specific substance and combination of biological matrix and sampling type in the data is smaller than the specified percentage, imputation is ignored. This is to prevent that imputation takes place using a set of imputation values that is not representative or unrealistically small (e.g. one or a few values). Note that for the second imputation method, more refined methods could be used. E.g., when for a given day multiple samples are available and one is missing, then this sample might be neglected in the computation of an average exposure. Also, when samples have been taken at different time-slots, impute the missing records using samples from the same time-slot.
Apply conversion of substance concentrations from other biological matrices
Conversion of substances measured in biological matrices other than the target matrix is used when the number of substances in the target biological matrix is limited. For instance, in target matrix urine (spot) five substances are measured. For the same individuals also the blood (serum) concentrations are analysed resulting in a additional set of concentrations for five different substances. These five substances from blood (source matrix) are converted to the target matrix urine by checking the option Apply conversion of substance concentrations from other biological matrices and specifying the ‘between matrix conversion’ multiplication factor. Then, the analysis continues with ten substances. Conversion of substance concentrations is performed after imputation of censored and missing values. Selecting specific substances for conversion is possible through the selection options in the primary entity substances module.
Note that after choosing Do not impute missing values or Impute from data, missing value imputation, the data still contains missing values. These values may be imputed from other substance concentrations after applying matrix conversion.
Restricting the monitoring concentrations
After imputation of non-detects and missing values, eventually followed by conversion of substance concentrations from other biological matrices, some individual day monitoring concentration may contain missing values (e.g. after ‘Do not impute missing values or Impute from data, missing value imputation). All individual day samples with missing values for one or more substances are removed from the data set and the analysis continues.
Calculation of acute human monitoring concentrations
For acute assessments, the monitoring concentrations are computed for each substance and biological matrix as average individual-day concentrations. The computation is done after imputation of censored and missing values, eventually followed by a conversion of substance concentrations from other biological matrices. For a given substance and biological matrix, the acute individual-day concentration \(c_{ij}\) for individual \(i\) on day \(j\) is:
where \(n_{\mathtt{samples}}\) is the number of samples available for individual \(i\) on day \(j\), and \(c_{ijk}\) and \(\mathit{sg}_{ijk}\) denote the concentration and specific gravity correction factor, respectively, of the \(k\)-th sample of the individual day \(j\).
Calculation of chronic human monitoring concentrations
For chronic assessments, the monitoring concentrations are computed as the average monitoring concentrations of multiple individual-days for each substance and biological matrix. The computation is done after imputation of censored and missing values, eventually followed by a conversion of substance concentrations from other biological matrices. The chronic concentration \(c_{i}\) for individual \(i\) is computed as:
where \(n_{\mathtt{days}}\) is the number of days that individual \(i\) was monitored, and \(c_{ij}\) denotes the average monitoring concentration of individual \(i\) on day \(j\).