TABLE I1.

Uncertainties identified and their impact on the outcome of the exposure assessment for TBBPA in food.

Description of the uncertainty		Impact on the exposure assessment a
Occurrence data
Analytical measurements	Performance (e.g. specificity for the target compounds) of the analytical method (GC‐ECD, GC–MS, etc.)	1 – Low impact. Methods for TBBPA are well established. Most of the data were analysed by LC–MS and some by GC–MS using isotope labelled standards
	Proportion of left‐censored data and magnitude of difference between risk estimates for LB and UB exposures	1 – Low impact. Although the proportion of left‐censored data is high, the uncertainty is addressed by using the substitution method, and the use of the UB which is an overestimation of the exposure estimates. The impact of this uncertainty was considered to be medium for the exposure assessment, but low for the risk characterisation
	Consideration of recovery (e.g. correction carried out or not)	1 – Low impact. Methods for TBBPA use isotope labelled standards that do not need to be corrected for recovery
Data reporting	Potential errors in reporting the occurrence data (e.g. in the classification of the food category, unit of measurement, parameter, fat vs. whole weight, etc.) – unidentified errors (not apparent from the data provided)	1 – Low impact. Data cleaning procedures that identifies outliers and clarifications received by data providers aim to solve most of the possible reporting errors. Although some errors might remain unidentified their overall impact on the results should be low
	Missing information in reporting the occurrence data (e.g. analytical method)	1 – Low impact. Clarifications received by data providers
	Missing or unclear information about the treatment/processing applied prior to the analysis of the sample that is submitted to EFSA	0 – Negligible impact. Limited information is available on processing, but in general the food is analysed raw. Since sample treatment/processing impact on TBBPA concentration is considered negligible, the impact of this uncertainty is negligible
	Use of food categories at high (often not enough specified) FoodEx/FoodEx2 level	1 – Low impact. A total of 90% of analytical results were reported at Foodex2 Level 3 or lower
Other uncertainties linked to occurrence data, factors and assumptions	Possible effect of food preparation on the chemical concentrations	1 – Low impact. No information available on the impact of food processing on the concentration of TBBPA. Assuming that TBBPA will behave in a similar way to lipophilic POPs, i.e. that TBBPA associates with lipids, and amounts in food may be reduced if fats that come from the food during cooking or processing are discarded; on the other hand if moisture is lost during the heating of foods but no fat is discarded, then the total amount will remain unchanged but the concentrations may increase. Overall, the impact of these changes would be low
	Uncertainty in the reporting of fat content by data providers and need for conversion to whole weight	0 – Negligible impact. When results were expressed on fat content, fat percentage was provided and this allowed the conversion of concentration to whole weight. Some values were clarified with data providers, and data were corrected accordingly
Representativeness of the data	Limited number of analytical results per food categories expected to contribute to the exposure	1 – Low impact. No data for infant formula and/or follow‐on formula were submitted to EFSA. Data from the literature was used: two studies were identified, reporting on 120 samples on infant formula and follow‐on formula collected in two Member States (Spain and France). Considering a common market, these data were considered sufficiently representative and used in the exposure assessment. The impact of this uncertainty was considered to be medium for the exposure assessment, but low for the risk characterisation
		1 – Low impact. Analytical results were only available for a limited number of food categories. Where possible, concentrations were inferred to obtain occurrence data in composite food and derivatives, for food for which occurrence data were available in the raw primary commodity In addition, although very limited number of samples in food categories of non‐animal origin all left‐censored, and these categories were not included in the exposure assessment, TBBPA is not expected to be present at high concentrations in these categories, thus the impact on both the exposure assessment and overall risk assessment is considered low
	Extrapolation of data from one food category to others, and other assumptions	1 – Low impact. Occurrence data have been extended to all similar foods when there was no reason to assume TBBPA was not present in the concerned similar foods Calculations based on recipes were made to obtain TBBPA concentration in composite foods belonging to other FoodEx2 Level 1 categories For some subcategories with limited number of samples the average value at higher levels of the FoodEx2 classification was used as considered more representative/conservative based on expert judgement (e.g. milk vs. cow's milk) When calculating concentrations in derivatives obtained through the process of separation from concentrations in raw primary commodity (e.g. concentration in butter from concentration in milk) the assumption that TBBPA was accumulating in the fat component was made Overall, the impact of this uncertainty was considered to be medium for the exposure assessment, but low for the risk characterisation
	Sampling strategy not fully random	0 – Negligible impact. There were no samples reported as ‘suspect sampling’. Thus it is assumed that sampling was sufficiently randomised
	Uneven distribution of the data per year (e.g. recent years not sufficiently represented)	1 – Low impact. Only the most recent data (from 2011 onwards) were considered in the assessment
	Uneven distribution of the data per country (e.g. large number of MSs not sufficiently represented)	1 – Low impact. Data coming from six Member States plus UK and NO. Most results were provided by France (58%) and Norway (23%). Considering a common market, these data were considered sufficiently representative for the exposure assessment
	Limited number of analytical results per variables that could explain higher/lower levels, such as production method (e.g. wild vs. farmed), processing (e.g. peeled vs. raw), etc	1 – Low impact. The additional information provided is limited but no specific aspects have been identified to have a major impact
Human milk data	Proportion of left‐censored data and inclusion of hydrolysis step to cleave conjugates	1 – Low impact. The study reporting the highest mean concentration in human milk (Martínez et al., 2019) used the MB approach and included a hydrolysis step in the sample preparation to cleave conjugates. Thus the results refer to ‘total TBBPA’ (conjugated + unconjugated). Therefore, the concentration is likely to be an overestimate of the bioavailable fraction of the TBBPA
Consumption data
Data reporting	Unidentified errors in reporting consumption data, e.g. in the classification of the food, portion size, body weight estimation, memory errors, capacity to report details in dietary surveys Different dietary survey methodologies (e.g. dietary record vs. 24‐h re‐call), dietary software, interview options, use of portion‐size measurement Use of national standard recipes and ingredients factors for composite dishes (potentially leading to, e.g. underestimation of minor ingredients, overestimation of standard ingredients) Different sample size and response rate of the dietary surveys Long‐term (chronic) exposure assessed based on few days of consumption per individual information about processing/cooking method not consistently reported in consumption records	1 – Low impact. Uncertainties and limitations related to the use of the EFSA Comprehensive Food Consumption Database have been described by EFSA (EFSA, 2011a). These uncertainties are common to dietary exposure assessments performed using the Comprehensive Database, and have the potential to cause either an over‐ or under‐estimation of the exposure No specific uncertainties affecting the food consumption data used for the TBBPA exposure assessment were identified
Representativeness of the data	Availability of food consumption data for special population groups, including consumers only of specific foods of special interest, or following special diets	1 – Low impact. ‘Consumers only’ of mostly contaminated foods are covered by 95th percentile exposure estimates. Pregnant, lactating women and vegetarian's surveys provided similar exposure results as adults from the general population. No other specific population groups that might have a particular exposure to TBBPA were identified
Exposure estimates
Non‐dietary exposure	Sources of exposure other than dietary – how much important is dietary exposure to the total	1 – Low impact. Exposure from dust can vary greatly between individuals and age classes, and can be significant for some individuals, although not as great as exposure from diet. It can be particularly relevant for young children. Limited data on dermal and inhalation and other potential non‐dietary routes of exposure
Exposure assessment scenario	Consumers loyalty to specific brands or from specific local areas not considered	0 – Negligible impact. Exposure assessment was made for the general population. Brand loyalty and consumer's only scenarios were considered not relevant for the TBBPA exposure assessment. High exposures due to variability across individuals are covered by the 95th percentile of the exposure estimates

Abbreviations: ECD, electron capture detector; FR, France; GC, gas chromatography; LB, lower bound; LC, liquid chromatography; MS, mass spectrometry; MSs, Member States; NO, Norway; POPs, persistent organic pollutants; UB, upper bound; UK, United Kingdom.

^a 0 – Uncertainty with negligible impact; 1 – Uncertainty with low impact; 2 – Uncertainty with medium impact; 3 – Uncertainty with high impact.