National Health Research Institutes 食品風險評估人才訓練課程 食安風險特性化 不確定性及靈敏度分析之實務演練 Yi-Jun Lin Postdoctoral Fellow 2015.10.07
1. What is Risk Characterization ( 風險特性化 )? 2. What are the differences between Qualitative ( 定性 ) and Quantitative ( 定量 )? How to Quantify the Risk? 3. What is Uncertainty ( 不確定性 )? Sources of Uncertainty How to proceed the Uncertainty Analysis? 4. What is Sensitivity Analysis ( 靈敏度分析 )? 5. Case study 2
Risk Assessment of Food (Codex Alimentarius Commission, Working Principles for Risk Analysis for Food Safety for Application by Governments, CAC/GL 62-2007) Hazard Identification Hazard Characterization Exposure Assessment Risk Characterization 3
What is Risk Characterization? 目的 : 描述特定族群對一項已知或對人體健康產生潛在不良影響的發生機率 風險特徵描述包括 : 定性的風險評估 (Qualitative Risk Assessment) 及定量的風險評估 (Quantitative Risk Assessment) 風險評估的不確定性分析 (Uncertainty Analysis) 風險評估的靈敏度分析 (Sensitivity Analysis) 4
What are the differences between Qualitative and Quantitative? Example: Qualitative description Quantitative cancer risk of a chemical 危害物平均暴露量的個體 = 1 10 (Average exposed person) -6 危害物最大暴露量的個體 (Maximally exposed person) = 1 10-6 5
How to Quantify the Risk? 6
非致癌危害商數 Hazard Quotient (HQ) HQ = C IR EF ED 10-3 RfD BW AT nc C: Contaminant concentration in food (μg/g) IR: Ingestion rate (g/day) EF: Exposure frequency (day/year) ED: Exposure duration (year) RfD: Reference dose (mg/kg-day) BW: Body weight (kg) HQ AT nc : Averaging time for noncarcinogens (year) 10-3 : Unit conversion factor >1 <1 Potentially hazardous effects Acceptable risk 7
致癌風險 Target Cancer Risk (TR) TR = C CSF IR EF ED 10-3 BW AT c C: Contaminant concentration in food (μg/g) CSF: Carcinogen slope factor (mg/kg-day) -1 IR: Ingestion rate (g/day) >1 10-6 EF: Exposure frequency (day/year) ED: Exposure duration (year) TR BW: Body weight (kg) AT c : Averaging time for carcinogens (year) <1 10-6 10-3 : Unit conversion factor > 1 cancer cases 1,000,000 Unacceptable cancer risk Acceptable cancer risk 8
What is Uncertainty? Uncertainty stems from lack of knowledge, incomplete information, or incorrect information, either qualitative or quantitative. Types of uncertainty Ambiguity, Measurement, Sampling, Assumption, Extrapolation, Distribution, Others (NRC, Advancing Risk Assessment, 2009) (European Food Safety Authority (EFSA), Public consultation on Draft Guidance document on Uncertainty in Scientific Assessment in 2015) 9
Types and Sources of Uncertainty Example: in exposure assessment (EFSA, Public consultation on Draft Guidance document on Uncertainty in Scientific Assessment in 2015) 舉例 三聚氰胺在奶粉中的最大濃度 受到三聚氰胺汙染的大陸製巧克力的每日最大攝食量 不確定性類型及來源 量測 : 使用準確度未明的方法分析三聚氰胺含量 抽樣 :491 樣本來自 109 家廠商 ( 抽樣代表性 ) 假設 : 假設樣本測量最大濃度值為實際三聚氰胺在奶粉中的最大濃度 抽樣 : 攝食量調查樣本的代表性未知 假設 : 使用攝食量樣本推估值的第 95 百分位作為實際大陸製巧克力的每日最大攝食量 外推 : 將所調查之巧克力攝食量外推為大陸製巧克力的攝食量 10
How to proceed the Uncertainty Analysis? Quantitative Uncertainty Analysis A probabilistic analysis techniques: Monte Carlo (MC) simulation 11
Monte Carlo (MC) simulation 蒙地卡羅模擬法是在二次世界大戰期間用來模擬有關原子 彈發展過程中的複雜問題 ( 以當時最著名的賭場 蒙地卡 羅 命名之 ) (Metropolis, 1987. The beginning of the Monte Carlo method. Los Alamos Science, special issue.) 蒙地卡羅模擬法是一個完全的隨機取樣法 蒙地卡羅模擬法是從參數定義域之機率分布中隨機取樣來 建立風險的機率分佈 12
Probability Probability X: 危害物在食物中濃度 Y: 暴露族群的體重 Z: 食物攝食率 1. Selecting uncertain model parameters X Y Z Risk model = f(x, Y, Z) 2. Determine an appropriate probabilistic distribution 3. A value is randomly sampling from each distribution by MC simulation Run 1: Risk = 0.5 Run 2: Risk = 0.1 Run N: Risk = 0.8 4. Running the model and calculating output values 5. Enough simulations to obtain stable solution Risk 6.Uncertainty in model outcomes 13
What is Sensitivity Analysis? To understand how the parameters of model influence the predicted outcomes (e.g., risk estimates) To identify the most significant parameters 14
Example for Sensitivity Analysis HQ and TR for consumption of As-contaminated tilapia Reference: Lin MP, et al., 2005. A PBTK/TD Modeling-based approach can assess arsenic bioaccumulation in farmed tilapia (Oreochromis Mossambicus) and human health risks. Integrated Environmental Assessment and Management 1: 40 54. Hazard Quotient (HQ) As in tilapia muscle Tilapia ingestion rates Human Body weight Target Cancer Risk (TR) As in tilapia muscle Tilapia ingestion rates Human Body weight 26.1% 25.3% 74.5% 73.9% 0 20 40 60 80 Contribution (%) 15
Case Study Assessing human exposure risk to Zn and Cu through milkfish consumption Reference: Lin MC, 2009. Risk assessment on mixture toxicity of arsenic, zinc and copper intake from consumption of milkfish, Chanos chanos (forsska l), cultured using contaminated groundwater in southwest Taiwan. Bulletin of Environmental Contamination and Toxicology 83: 125 129. 16
Hazard Quotient (HQ) HQ = (C IR EF ED 10-3 ) / (RfD BW AT nc ) Parameters Symbol Estimated value Zn concentration in milkfish C Zn (μg/g) N(37.98, 6.49) Cu concentration in milkfish C Cu (μg/g) N(2.09, 0.40) Milkfish ingestion rate IR (g/day) N(374.07, 134.22) Exposure frequency EF (day/year) 350 Exposure duration ED (year) 30 Body weight for Taiwanese adult BW (Kg) N(60.55, 4.67) Averaging time for noncarcinogens AT nc (day) 10950 Reference dose for Zn RfD Zn (mg/kg-day) 0.3 Reference dose for Cu RfD Cu (mg/kg-day) 0.04 N(a, b) denotes the normal distribution with mean a and SD b 17
To implement MC simulation by Crystal Ball software 18
Download Crystal Ball Free for 30 Days 19
Installing Crystal Ball 20
Open Crystal Ball in Excel 定義參數之機率分布 定義所需要預測的參數 / 選項 清除定義假設與定義預測 預測模擬開始 預測模擬的次數設定 21
Sampling method Monte Carlo (MC) 22
Calculating Food Safety Risk by HQ HQ = (C IR EF ED 10-3 ) / (RfD BW AT nc ) Hazard index (HI) = Total HQ = (HQ Zn + HQ Cu ) 23
參數機率分布設定 : Zn conc. in milkfish 24
參數機率分布設定 : Ingestion rate 25
HQ Forecasts and Simulations 26
HQ Forecasts and Simulations 27
Create Reports Statistics Figures (probabilistic distribution) Percentile 28
HQ Zn HI Zn+Cu HQ Cu 29
Extra Data HQ 95% 信賴區間 30
Sensitivity Analysis 預測 開啟敏感度圖表 Ingestion rate 31
Thank you for your attention