What are the challenges in addressing adjustments for data uncertainty? Hildegard Przyrembel, Berlin Federal Institute for Risk Assessment (BfR), Berlin (retired) Scientific Panel for Dietetic Foods, Nutrition and Allergies (NDA) of the European Food Safety Authority (EFSA), Parma
Description of Task: discuss uncertainty of data, not variability Uncertainty = lack of data or knowledge regarding the true value of a quantity - can be reduced only by acquisition of more data; analysis of the impact of sources of uncertainty is descriptive (qualitative) or, more desirable, by mathematical modelling (quantitative) Variability = real life situation due to heterogeneity of a quantity over time, space or members of a population - can be reduced by selection of sample, not by provision of more data; probabilistic assessment methodology is available Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 2
Risk assessment of essential nutrients Adverse health effect A: Deficiency mode (mechanism) of action -response relationship B: Excess mode (mechanism) of action -response relationship Critical multiplied by? Lowest threshold value of intake (= EAR 2SD) NOAEL or LOAEL divided by UF Tolerable upper intake Level (UL) Targeted result: RDA < UL Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 3
The multiple steps (simplified) between intake of a nutrient amount and either physiological or toxic responses depending on External Toxic response Absorbed Concentration in general circulation Concentration in target tissues Clearance Distribution to non-target tissues Local bioactivation Intracellular pathological changes Interaction with intracellular targets modified from IPCS Harmonization Project Document No. 2, 2005 Physiological responses Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 4
Relationship between external (dietary intake) and toxic response for nutrients External Toxic response External Internal Toxic response External Internal Target organ Toxic response External Internal Target organ Target organ metabolism Toxic response External Internal Target organ Target organ metabolism Target organ response Toxic response modified from Renwick et al (2001) Human Ecological Risk Assessment 7: 165-185 Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 5
Relationship between external (dietary intake) and physiological response for nutrients External Physiological response External Internal Physiological response External Internal Target organ Physiological response External Internal Target organ Target organ metabolism Physiological response External Internal Target organ Target organ metabolism Target organ response Physiological response Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 6
Sources of Uncertainty Model: both structure and parameters, i.e. mathematical translation of relationships; algorithms perform sensitivity analysis to identify those parameters with significant impact on model output Input data: quality; variability; measurement and databank errors quantification of effect of missing data is impossible; use of range of fictive data? Scaling algorithm: applicability to model selected problems of scaling propagate errors Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 7
Acceptable Range of Oral Intake, AROI Theoretical -response curves for various effects occurring in a population. The lower end of the response curve for such critical effects related to deficiency (curve 3) and toxicity (curve 6) defines the range of acceptable daily oral intake IPCS, 2002 Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 8
Starting situation is adequate nutrient intake. Intake DECREASES Intake INCREASES Biochemical changes within the homeostatic range no adverse sequelae Biochemical changes outside the homeostatic range no known sequelae Biochemical changes outside the homeostatic range biomarker of potential adverse effects Clinical symptoms minor but reversible Clinical symptoms significant but reversible Clinical signs significant but reversible organ damage Clinical signs irreversible organ damage Intake << requirement Intake >> requirement Increase of effects with increasing (excessive) and decreasing (deficient) intakes of essential nutrients (modified from Renwick et al. (2004) Food Chem Toxicol 42: 1903-1922) Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 9
Combined curve of -incidence relationship for the risks due to deficiency (absence of benefit) and due to toxicity Data requirements: intake that gives a 50% incidence (ED 50) coefficients of variation (CV) of response Assumptions: log-normal distribution CV for benefit 10% (green) CV for benefit 15% (red) CV for toxicity 45% Renwick AG (2006) J Nutr 136: 493S-501S Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 10
Alternative to single-point estimates: benchmark = at which effect selected as significant occurs with predetermined certainty (e.g. 95%) modified from IPCS, 2002 Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 11
Fraction Affected 0.8 0.7 0.6 0.5 0.4 0.3 0.2 A Probit BMD Lower Bound Probit Model with 0.95 Confidence Level Fraction Affected 0.4 0.35 0.3 0.25 0.2 0.15 0.1 B Probit BMD Lower Bound Probit Model with 0.95 Confidence Level 0.1 0.05 0 BMDL BMD 0 0.5 1 1.5 2 2.5 11:44 03/31 2005 0 BMDL BMD 0 0.5 1 1.5 2 2.5 11:50 03/31 2005 C Probit Model with 0.95 Confidence Level 0.3 Probit BMD Lower Bound Fraction Affected 0.25 0.2 0.15 0.1 0.05 0 BMDL BMD 0 0.5 1 1.5 2 2.5 3 11:48 03/31 2005 Fluoride-in-drinking-water benchmark approach applied to occurrence of dental fluorosis in children (data of Dean et al., 1942): a) 5% increase of fluorosis Dean >0.5 : 0.56 mg fluoride/l b) 5% increase of fluorosis Dean >2 : 1.5 mg fluoride/l c) 5% increase of fluorosis Dean >3 : 2.2 mg fluoride/l Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 12
Uncertainties in available data Methodology of balance studies example: calcium Lack of data a) longitudinal example: calcium b) content of diet example: vitamin D, chromium, pantothenic acid c) small number of subjects studied example: pantothenic acid d) absorption efficiency example: pantothenic acid e) storage capacity example: pantothenic acid f) -response relationship example: vitamin K, EPA+DHA Physiological significance example: vitamin K No adverse effect identified example: vitamin B1, vitamin B2, vitamin B12, pantothenic acid, biotin, vitamin K, chromium Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 13
Major sources and types of uncertainties in dietary exposure assessment food consumption (depending on type of survey) body weight (paired values consumption body weight) content in food (analysis, calculation, imputations, recipes) Assessment: qualitative: identify, describe deterministic: different point estimates probabilistic: assessment scenario, distributions, dependencies, quantify resulting variability and uncertainty, sensitivity analysis to examine contribution of each model input on variation and uncertainty in output. Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 14
Uncertainty in relation to food composition databases Number of data and methods of analysis Variability in composition with strain, age, ripeness, growing conditions, storage, processing and preparation Differences in bioavailability of nutrients from different foods Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 15
Uncertainty factors in risk assessment of chemicals for extrapolation animals to humans (interspecies) default value 10 coverage of human variability (interindividual) default value 10 Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 16
Selection of (default) UF in risk assessment of chemicals Animal NOAEL to human: Subchronic to chronic NOAEL: 3 to 10 NOAEL between humans: Human LOAEL to NOAEL: 3 to 10 Deficiencies in the database: up to 2 10 or broken down to nutrient and kinetic and dynamic specific factors 10 or nutrient and kinetic and dynamic specific factors Selection of (default) UF in risk assessment of essential nutrients UL must be > RDA Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 17
Uncertainty factors (UF) selected for endpoint human NOAEL in risk assessment of minerals and vitamins Endpoint Nutrient Population group UF Justification Human NOAEL Vitamin A Vitamin D Vitamin B 6 Phosphorus Fluoride Selenium Copper Iron Manganese Zinc Potassium Fertile women Adults Infants Adults / Infants >8 years 0-6 / >19 years 0-8 years 0-6 months 1.5 1.2 1.8 2.0 2.5 / 3.0 1.0 1.0 / 2.0 1.0 1.0 1.0 1.0 1.0 Variability in sensitivity Insufficient data Insufficient data Insufficient data Sensitive subpopulations IOM 1997 to 2005 Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 18
Uncertainty factors (UF) selected for endpoints human LOAEL and animal NOAEL and LOAEL in risk assessment of minerals and vitamins Endpoint Nutrient Population group UF Justification Human LOAEL Animal NOAEL Animal LOAEL Vitamin A Niacin Folic acid Vitamin C Calcium Magnesium Fluoride Iodine Iron Zinc Sodium Molybdenum Nickel Vitamin E Boron Vanadium Adults Infants 0-8 years >19 years >19 years 5 1.0 1.5 5 1.5 2 1.0 1.0 1.5 1.5 1.5 1.0 30 300 36 30 300 LOAEL LOAEL LOAEL LOAEL Variability in sensitivity Mild reversible effect LOAEL LOAEL LOAEL Animal (!0) variability (3) Animal (10) variability (10) reproduction (3) Animal (3) LOAEL (2) subchronic (2) variability (3) Animal (10) variability (3) Animal (10) LOAEL (3) variability (10) IOM 1997 to 2005 Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 19
Subdivision of the 100-fold Uncertainty Factor into separate factors for species difference (10-fold) and human variability (10-fold) and further subdivision of each into separate default factors for differences in toxicokinetics and toxicodynamics. To be replaced when available by compound-specific adjustment factors. 100-fold UF Interspecies differences 10-fold Interindividual differences 10-fold Toxicodynamic 10 0.4 (2.5) Toxicokinetic 10 0.6 (4.0) Toxicodynamic 10 0.5 (3.2) Toxicokinetic 10 0.5 (3.2) modified according to Renwick (2006) J Nutr 136: 493S-501S Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 20
Toxicokinetics: Toxicodynamics: processes that determine the concentration of the active compound at its site of action (absorption, distribution, metabolism, elimination) processes that relate the concentration in the target organ to an adverse effect Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 21
Importance of knowledge of mode (sometimes even mechanism) of action data both for assessment of toxic risk (but also for assessment of requirement?) Example: Human exposure: Animal toxic effects: Boron Absorption: almost 100% Distribution: Metabolism: Elimination: food, water, detergents, pesticides reproduction development (decreased fetal weight): NOAEL 9.6 mg/kg/day in rats evenly in body fluids and tissues (higher in bones) none clearance rates in rats are 3-4-fold higher than in humans No data on differences in toxicodynamics between rats and humans intraspecies UF=10 No variability assumed in toxicokinetics in humans and default UF for toxicodynamics (3.2) intraspecies UF=3 (FNB, 2001) OR Variability in glomerular filtration rate during pregnancy = 1.8 (toxicokinetic UF) x default toxicodynamic UF (3.2) intraspecies UF=6 (EFSA, 2004) Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 22
Uncertainty Analysis Adequacy of the selected model consistency with scientific theory; logic; inclusion of all necessary variables; predictions should agree with observations; results should have acceptable confidence levels. Uncertainties in model parameters and input data probability theory; Taylor series expansion; Monte Carlo simulation; generalized likelihood uncertainty estimation; Bayesian statistics; sequential partitioning Presentation of results probability distribution of predicted values and ranking of the relative contribution of factors to the total uncertainty of model output (sensitivity analysis). For selection of mathematical methods see e.g. van der Sluijs et al. (2004) RIVM/MNP Guidance for Uncertainty Assessment and Communication. ISBN 90-393-3797-7 For a practical example see e.g. van der Voet & Slob (2007). Integration of probabilistic exposure assessment and probabilistic hazard characterization. Risk Analysis 27: 351-371 Remark: these procedures still require the selection of appropriate exposure data and of relevant endpoints and the availability of trustworthy -response data. Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 23
Conclusions Uncertainty in risk assessment of nutrients and in the definition of requirements of nutrients is unavoidable. It should be characterized with respect to nature and magnitude, and different types of uncertainty should be ranked according to their impact on the result. For nutrients the default uncertainty or adjustment factors conventionally used in risk assessment of chemicals have to be modified ideally by either chemical specific kinetic or process specific dynamic data. However, the different and multiple physiological functions of different nutrients in the human body have to be considered individually and do not permit to set up a system of fixed adjustment factors. Uncertainty due to variability in both kinetics and dynamics can be dealt with by mathematical procedures. Uncertainty, due to gaps in data can be effectively relieved only by acquisition of such data. In the meantime, assumptions about data used in the assessment and the impact of their intentional variation in the calculation need to be identified and communicated both qualitatively and quantitatively. It is, however, uncertain how this communication is understood by the users. Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 24
Thank you for your attention! Hildegard Przyrembel Institute of Medicine Workshops, Washington D.C. 18-20 September 2007 25