Quantitative Health Risk Analysis Methods. Modeling the Human Health Impacts of Antibiotics Used in Food Animals

Transcription

1 Quantitative Health Risk Analysis Methods Modeling the Human Health Impacts of Antibiotics Used in Food Animals

2 Recent titles in the INTERNATIONAL SERIES IN OPERATIONS RESEARCH & MANAGEMENT SCIENCE Frederick S. Hillier, Series Editor, Stanford University Maros/ COMPUTATIONAL TECHNIQUES OF THE SIMPLEX METHOD Harrison, Lee & Neale/ THE PRACTICE OF SUPPLY CHAIN MANAGEMENT: Where Theory and Application Converge Shanthikumar, Yao & Zijm/ STOCHASTIC MODELING AND OPTIMIZATION OF MANUFACTURING SYSTEMS AND SUPPLY CHAINS Nabrzyski, Schopf & Węglarz/ GRID RESOURCE MANAGEMENT: State of the Art and Future Trends Thissen & Herder / CRITICAL INFRASTRUCTURES: State of the Art in Research and Application Carlsson, Fedrizzi, & Fullér/ FUZZY LOGIC IN MANAGEMENT Soyer, Mazzuchi & Singpurwalla/ MATHEMATICAL RELIABILITY: An Expository Perspective Chakravarty & Eliashberg/ MANAGING BUSINESS INTERFACES: Marketing, Engineering, and Manufacturing Perspectives Talluri & van Ryzin/ THE THEORY AND PRACTICE OF REVENUE MANAGEMENT Kavadias & Loch/PROJECT SELECTION UNDER UNCERTAINTY: Dynamically Allocating Resources to Maximize Value Brandeau, Sainfort & Pierskalla/ OPERATIONS RESEARCH AND HEALTH CARE: A Handbook of Methods and Applications Cooper, Seiford & Zhu/ HANDBOOK OF DATA ENVELOPMENT ANALYSIS: Models and Methods Luenberger/ LINEAR AND NONLINEAR PROGRAMMING, 2 Ed. Sherbrooke/ OPTIMAL INVENTORY MODELING OF SYSTEMS: Multi-Echelon Techniques, Second Edition Chu, Leung, Hui & Cheung/ 4th PARTY CYBER LOGISTICS FOR AIR CARGO Simchi-Levi, Wu & Shen/ HANDBOOK OF QUANTITATIVE SUPPLY CHAIN ANALYSIS: Modeling in the E-Business Era Gass & Assad/ AN ANNOTATED TIMELINE OF OPERATIONS RESEARCH: An Informal History Greenberg/ TUTORIALS ON EMERGING METHODOLOGIES AND APPLICATIONS IN OPERATIONS RESEARCH Weber/ UNCERTAINTY IN THE ELECTRIC POWER INDUSTRY: Methods and Models for Decision Support Figueira, Greco & Ehrgott/ MULTIPLE CRITERIA DECISION ANALYSIS: State of the Art Surveys Reveliotis/ REAL-TIME MANAGEMENT OF RESOURCE ALLOCATIONS SYSTEMS: A Discrete Event Systems Approach Kall & Mayer/ STOCHASTIC LINEAR PROGRAMMING: Models, Theory, and Computation Sethi, Yan & Zhang/ INVENTORY AND SUPPLY CHAIN MANAGEMENT WITH FORECAST UPDATES * A list of the early publications in the series is at the end of the book *

3 Quantitative Health Risk Analysis Methods Modeling the Human Health Impacts of Antibiotics Used in Food Animals by Louis Anthony Cox, Jr. Cox Associates and University of Colorado

4 Louis A. Cox Cox Associates and University of Colorado Denver, CO, USA Library of Congress Control Number: ISBN-13: ISBN-10: ISBN-13: (e-book) ISBN-10: (e-book) Printed on acid-free paper by Springer Science+Business Media, Inc. All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science + Business Media, Inc., 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now know or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks and similar terms, even if the are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed in the United States of America SPIN springeronline.com

5 To Cope and Michael For making good humor, hard work, and complete commitment to scientific truth central in a political world

6 Contents Preface...xiii Acknowledgments... xvii CHAPTER 1 Qualitative and Quantitative Risk Analysis 1. Introduction: A Need for New Methods Challenges and Goals Overview of Contents Background and Motivation Difficult Tradeoffs Recent History The Role of Risk Analysis Qualitative Risk Analysis Some Examples of Qualitative Risk Rating Frameworks The Brenner Scheme for Order-Of-Magnitude Risk Rating US FDA-CVM s Guidance Document # 152 and its limitations Lessons from Previous Approaches A Mathematical Theory of Qualitative Risk Rating What Should Be Done Instead? Discussion and Conclusions APPENDIX Proof of Theorem CHAPTER 2 Risk Analysis: Goals and Methods 1. Introduction to Health Risk Analysis Risk Assessment Quantitative Definition of Risk Risk Management Risk Communication Purposes and Outputs of Risk Analysis Risk Analysis with Uncertain Data Desired Outputs of Risk Analysis Introduction to Risk Assessment A Rapid Risk Rating Technique (RRRT)... 56

7 viii 6. Risk Management Definition of Risk Management Purposes and Outputs of Risk Management Methods for Risk Management Decision-Making Methods of Risk Management to Avoid Validating Risk Management Results Communicating Risk Analysis Results Steps in Risk Communication Purposes of Risk Communication Desired Outputs of Risk Communication Methods for Risk Communication Participatory Risk-Management Decision Processes Summary and Conclusions CHAPTER 3 Hazard Identification 1. Definition of Risk Assessment A Bayesian Network Framework Introduction to Hazard Identification Definition of Hazard Identification Purposes and Goals of Hazard Identification Methods of Hazard Identification Causal Inference From Epidemiological Data Refuting Non-Causal Explanations Seeking Positive Evidence for a Causal Relation Case Study: Did Ending Avoparcin Use Reduce Human Resistance? A Systems Dynamics Approach to Hazard Identification Model Definitions and Equations Model-Based Analysis of Potential Resistance Hazards Hazard Identification Methods to Avoid A Case Study: Presenting Real vs. Hypothesized Hazards Conclusions CHAPTER 4 Exposure Assessment 1. Introduction to Risk Quantification Introduction to Exposure Assessment Definition and Purposes of Exposure Assessment Exposure Assessment Methods: Simulation Practical Simulation-Based Exposure Modeling Simulation Based on Conditioning, Not Imitation

8 Contents ix 4. Attribution-Based Exposure Modeling Attribution through Multiple Stages Systems Dynamics Models of Exposure Model Development and Simplifying Assumptions Estimating Model Parameters from Data Using the Model to Make Predictions Exposure Assessment Methods to Avoid Prevalence-Based Exposure Metrics Holistic Statistical Exposure Modeling Validating and Refining Exposure Models Qualitative Exposure assessment Summary and Conclusions CHAPTER 5 Dose-Response Modeling and Risk Characterization 1. Introduction to Dose-Response Modeling Definitions and Purposes of Dose-Response Modeling Microbial Dose-Response Modeling Empirical Statistical Dose-Response Models Biologically Motivated Statistical Dose-Response Models Multi-Component Dose-Response Models Epidemiological Exposure-Response Models Practical Dose Response Models: Summary Validating Dose-Response Models Showing Uncertainty In Dose-Response Models Health Consequence Modeling Risk Characterization Definition and Purposes of Risk Characterization Desired Outputs of Risk Characterization Methods for Risk Characterization Methods of Simplified Risk Characterization to Avoid Rapid Risk Rating Technique (RRRT) Approach Validation of Risk Characterization Results Uncertainty and Sensitivity Analyses Uncertainty Analysis Sensitivity Analysis Discussion and Summary CHAPTER 6 Human Health Risks from Virginiamycin: A Case Study 1. Introduction Background on QD, VRE, VREF, and VREF A

9 x 2. Risk Assessment Data and Methods Estimating Total Number of VREF Cases Per Quarter Estimating VREF Cases per Quarter in Australia Estimating VREF Cases per Quarter in the United States Estimated Fraction of VRE Cases that are VREF A Estimating the Australian VREF A Fraction of VRE by Upper Bounding Estimating the U.S. VREF A Fraction Fraction of Exogenous (Non-Nosocomial) VREF Cases Fraction of VREF A Cases Attributable to Chickens Fraction of VREF A Cases with QD Resistance QD Resistance Fraction Among VREF A Cases in Australia QD Resistance Fraction Among VREF A Cases in the United States QD Effectiveness Fraction Fraction of VREF A Cases Prescribed QD Over Time Time-Varying QD Resistance Fraction Health Consequences of Treatment Failures Summary Results Results for Australia Results for USA Discussion of Results and Key Drivers Summary and Conclusions APPENDIX: Estimating Human Health Consequences of QD Resistance CHAPTER 7 Dynamic Modeling and Uncertainty Analysis 1. Introduction Motivation: How Does History Constrain Future Risks? A Growing Concern: Managing Potential Future Risks From Qualitative Hazard Identification to Quantitative Risk Assessment Deterministic Modeling and Results VRE Colonization Dynamics Model Bayesian Monte Carlo (BMC) Uncertainty Analysis Selecting Priors for Probabilistic Sensitivity Analysis Data on Historical QD Resistance Rates Bayesian Monte Carlo (BMC) Estimation of Posterior Distributions Stochastic Discrete-Event Simulation (DES) of Time-to-Outbreak Results of Uncertain Dynamic Model Initial Bayesian Analysis and Estimation of Parameters Revised Prescription Rate for QD

10 Contents xi 4.3 Human Health Risks vs. Key Parameter Values Estimated Individual and Population Risks Discussion Summary and Conclusions CHAPTER 8 Potential Human Health Benefits of Animal Antibiotics 1. Introduction A Risk Management Dilemma for Virginiamycin Framework for Assessing Impacts Data For RRRT Risk-Benefit Calculations Calculation of F Calculation of P Results for Virginiamycin impacts Uncertainty and Sensitivity Analyses Extensions to Cattle and Swine Comparison to risks in Australia RRRT for Macrolides and Enrofloxacin Discussion Comparison to Other Risk Assessment Approaches Comparison of Model Predictions to Experience Other Considerations and Extensions Summary and Conclusions APPENDIX: The Log-Exponential Risk Model References Index

11 Preface This book grew out of an effort to salvage a potentially useful idea for greatly simplifying traditional quantitative risk assessments of the human health consequences of using antibiotics in food animals. In 2001, the United States FDA s Center for Veterinary Medicine (CVM) (FDA-CVM, 2001) published a risk assessment model for potential adverse human health consequences of using a certain class of antibiotics, fluoroquinolones, to treat flocks of chickens with fatal respiratory disease caused by infectious bacteria. CVM s concern was that fluoroquinolones are also used in human medicine, raising the possibility that fluoroquinolone-resistant strains of bacteria selected by use of fluoroquinolones in chickens might infect humans and then prove resistant to treatment with human medicines in the same class of antibiotics, such as ciprofloxacin. As a foundation for its risk assessment model, CVM proposed a dramatically simple approach that skipped many of the steps in traditional risk assessment. The basic idea was to assume that human health risks were directly proportional to some suitably defined exposure metric. In symbols: Risk = K Exposure, where Exposure would be defined in terms of a metric such as total production of chicken contaminated with fluoroquinolone-resistant bacteria that might cause human illnesses, and Risk would describe the expected number of cases per year of human illness due to fluoroquinolone-resistant bacterial infections caused by chicken and treated with fluoroquinolones. If it could be made to work correctly, this simple approach would have a lot to recommend it. It obviates the need to develop complex simulation models for exposure or sophisticated dose-response models for human

12 xiv infection and illnesses caused by foodborne bacteria. It roots the estimate of a single parameter, K, firmly in empirical data on concurrent levels of Exposure and Risk. It provides a simple, easily communicated basis for estimating how a reduction in exposure would affect human health risk. And yet, there appeared to be potentially fatal conceptual flaws in the approach. First, it implied a positive value of K whenever Risk and Exposure were both positive and yet, available data showed that handling, cooking, and eating fresh chicken were associated with a reduced level of risk for the bacterial illnesses (resistant or susceptible) of concern. How could such a protective effect be reconciled with the model? (Adding an intercept term to the model was one obvious possibility, but even that would not suffice to explain protective effects for low-level exposures and increased risks for relatively rare, high exposures. Perhaps some nonlinearity was needed, but that would threaten the simplicity of the model.) The approach estimated K from historical levels of Risk and Exposure, but did not appear to give valid predictions of how changes in Exposure would change future levels of Risk the chief concern for practical risk management. And it relied on exposure metrics, such as prevalence of bacterial contamination in chicken carcasses, that could be estimated from historical data, but that are fundamentally inadequate for predicting risks that depend on the quantity of contamination rather than just on its prevalence, i.e., its presence or absence. Most of all, the model assumed a direct proportional causal relation between its selected Risk and Exposure metrics without validation, and then focused on quantifying their ratio, K, without validating that the ratio had causal significance or predictive power in real data sets. It thus skipped the traditional (and rather important) role of hazard identification in human health risk assessment. And yet, the simplicity and intuitive appeal of the approach are strong. Can its technical limitations be remedied and the benefits of simplicity and direct estimation of parameters from data be achieved while allowing valid causal interpretations of the results and useful predictions of human health impacts of changing exposures? This book presents a lengthy, but essentially affirmative, answer. The key steps are to: (a) Allow for other sources of exposure (represented by a Background intercept term to be estimated from data); (b) Acknowledge that changes in animal antibiotic use typically affect multiple types of exposures (e.g., reducing resistant bacteria but perhaps increasing susceptible bacteria in processed food commodities); (c) Focus on causal impacts on risk arising from changes in exposures; and (d) Recognize that the changes in risk predicted by the simple linear model for a given set of changes in exposure can provide valid bounds on the true changes that will occur with a more complex but more realistic model. With these enhancements, the mathematical model becomes slightly more

13 Preface xv complex. For example, the Risk = K Exposure equation must now be generalized to include additional terms, as in: Current Risk = Background + K 1 (exposure to susceptible bacteria) + K 2 (exposure to resistant bacteria) To be interpreted causally, its parameters must be estimated by appropriate statistical methods for causal models, such as structural equations modelling, rather than by straightforward statistical regression. The model s predictions give bounds on true but unknown changes in risk. With these modifications, the goal of obtaining a practical, simple, data-driven approach to estimating probable human health impacts caused by risk management interventions that change exposures to microbial loads can be substantially met. Many other advances are also possible, such as modeling the systems dynamics of bacteria flows over time between resistant and susceptible types and between ill and well subpopulations of humans and animals. These and other extensions of the basic model are explored in the present book. The methods developed here may also prove useful in other areas of health risk analysis where complex simulation models of risk are difficult or impractical and simpler formulae can produce useful bounds on the human health consequences of alternative risk management interventions. The public health implications of this new approach to risk assessment are striking. As demonstrated in this book, considering the effects of animal antibiotics on susceptible bacteria as well as on resistant bacteria makes clear that banning current uses of animal antibiotics is not necessarily an effective way to preserve the efficacy of human antibiotics or to protect human health, as one might at first suppose. To the contrary, withdrawing current animal antibiotics appears to have the potential to create more ill animals per year, more human illnesses per year, and thus greater use of antibiotics in human medicine, where the opportunity to select resistant strains of bacteria may be far more threatening than in animal uses. Applying a systems dynamics perspective to animal antibiotic use and human health suggests that interventions designed to protect human health by reducing or banning animal antibiotic use may unintentionally increase human health risks. This possibility appears sufficiently strong so that it should be carefully evaluated before policies are made that assume that restricting animal antibiotic use will promote human health. Since the FDA-CVM, 2001 risk assessment of fluoroquinolones was completed, the world has moved on. Some, including this author, objected to the use of the Risk = K Exposure framework as giving incorrect predictions, being inconsistent with available epidemiological data, and postulating a causal relation where none was apparent in real data. FDA defended its approach, denied the credibility of these technical objections,

14 xvi and insisted that its risk assessment justified withdrawing approval for fluoroquinolones used in poultry. As this book goes to press, the dispute has proceeded through litigation (with FDA s Administrative Law Judge finding for FDA s CVM in 2004). This litigation clearly demonstrated to both sides and to many stakeholders the need for clearer, sounder, and simpler methods to understand the relevant risks and to improve the basis for risk management decision in this challenging area in the future. The perspective of this book is that risk is fundamentally quantitative, and when an intervention has some helpful and some harmful effects, it is necessary to use quantitative analysis (at least to place bounds on the sizes of different effects) to compare them and to determine the course of action yielding the largest human health benefits. The major goal of this book is to provide and illustrate methods for quantitative risk assessment and for comparing alternative risk management actions, given realistic limitations on scientific knowledge and available data.

15 Acknowledgments First, I thank my colleague Dr. Douglas Popken, who has collaborated with me since 2000 on developing and applying computer simulation models of the human health effects of animal antimicrobials. His tenacity in tracking down real-world data and his creativity and proficiency in simulation modeling led to many of the insights that motivated this book. In addition, Chapters 6 through 8 are based largely on papers that we have written together (Cox and Popken 2004a, 2004b, and 2005). I thank the journals Risk Analysis and Environment International for permission to reuse material from published articles. Chapter 1 includes most of Cox et al., 2005, while Chapter 8 is based largely on material from Cox, The hazard identification framework in Chapter 3 was first sketched in Cox and Ricci, The United States Food and Drug Administration s Center for Veterinary Medicine (CVM) first drew my attention to this area and supported part of my review of their risk assessment approach for fluoroquinolones. Their attempts to develop short-cuts and simplifications in traditional risk assessment have served as an inspiration to me to try to figure out how to do the same thing correctly. Much of the applied research in this book, including development and application of the Rapid Risk Rating Technique (RRRT) approach to quantitative risk assessment, was initially supported by the Animal Health Institute (AHI). Applications to specific animal antibiotics were supported in part by Bayer, Elanco, and Phibro Animal Health. I thank Ken Bafundo, Dennis Copeland, and Tom Shryock for championing the development of better (clearer, more realistic, more valid, more data-driven) approaches to

16 xviii quantitative risk assessment. I gratefully acknowledge many stimulating conversations with them and with Michael Vaughn and Rich Carnevale on details of animal antibiotics and practical quantitative risk assessment methods. I especially thank Stephen Page for providing detailed, thoughtful comments on Chapters 1-3 that substantially improved the text. I have greatly enjoyed discussing the needs and methods of antimicrobial risk assessment practitioners with Stephen and with Scott Hurd, Gay Miller, Greg Paoli, Paolo Ricci and Randy Singer, as well as occasional correspondence and discussions with Gregg Claycamp. I thank Deborah Doherty and her team at Springer along with Doug Wilcox for an outstanding job in preparing the final text and figures. The methods described here are intended to help animal antibiotic manufacturers, as well as veterinarians and regulators involved in antimicrobial risk assessment, to apply sound, practical, methods of quantitative risk assessment that can be carried out using currently available data while acknowledging (and bounding the effects of) current scientific uncertainties.