Suchergebnisse

Intro -- Preface -- Acknowledgments -- Contents -- Part I: Estimating and Simulating Dynamic Health Risks -- Chapter 1: Scientific Method for Health Risk Analysis: The Example of Fine Particulate Matter Air Pollution and COVID-19 Mortality Risk -- Introduction: Scientific Method for Quantitative Risk Assessment -- Scientific Method vs. Weight-of-Evidence Consensus Judgments as Paradigms for Regulatory Risk Analysis -- A Recent Example: PM2.5 and COVID-19 Mortality -- Do Positive Regression Coefficients Provide Evidence of Causation? -- Positive Regression Coefficients Created by Model Specification Error and Other Causes -- Conclusion: Regression Models and Judgment Should Complement Science, not Substitute for it -- Appendix 1: Data -- References -- Chapter 2: Modeling Nonlinear Dose-Response Functions: Regression, Simulation, and Causal Networks -- Introduction -- Why Does Nonlinearity Matter? -- Hazard Identification -- Challenges for Regression-Based Hazard Identification -- Significant Regression Coefficients Arising from Trends and from Omitted Confounders -- Significant Regression Coefficients Arising from Measurement Errors in Confounders -- Significant Regression Coefficients Arising from Model Specification Errors -- Significant Regression Coefficients Arising from Residual Confounding -- Surrogate Variables -- Variable Selection -- Significant Regression Coefficients Arising from Competing Explanations -- Significant Regression Coefficients Arising from Attribution of Joint Effects -- Some Alternatives to Regression for Hazard Identification -- Dose-Response Modeling -- Challenges for Regression-Based Dose-Response Modeling -- Bayesian Networks for Dose-Response Modeling -- Dynamic Simulation for Dose-Response Modeling -- Exposure Assessment -- Risk Characterization, Uncertainty Characterization, and Risk Communication.

AbstractDecision analysis and risk analysis have grown up around a set of organizing questions: what might go wrong, how likely is it to do so, how bad might the consequences be, what should be done to maximize expected utility and minimize expected loss or regret, and how large are the remaining risks? In probabilistic causal models capable of representing unpredictable and novel events, probabilities for what will happen, and even what is possible, cannot necessarily be determined in advance. Standard decision and risk analysis questions become inherently unanswerable ("undecidable") for realistically complex causal systems with "open‐world" uncertainties about what exists, what can happen, what other agents know, and how they will act. Recent artificial intelligence (AI) techniques enable agents (e.g., robots, drone swarms, and automatic controllers) to learn, plan, and act effectively despite open‐world uncertainties in a host of practical applications, from robotics and autonomous vehicles to industrial engineering, transportation and logistics automation, and industrial process control. This article offers an AI/machine learning perspective on recent ideas for making decision and risk analysis (even) more useful. It reviews undecidability results and recent principles and methods for enabling intelligent agents to learn what works and how to complete useful tasks, adjust plans as needed, and achieve multiple goals safely and reasonably efficiently when possible, despite open‐world uncertainties and unpredictable events. In the near future, these principles could contribute to the formulation and effective implementation of more effective plans and policies in business, regulation, and public policy, as well as in engineering, disaster management, and military and civil defense operations. They can extend traditional decision and risk analysis to deal more successfully with open‐world novelty and unpredictable events in large‐scale real‐world planning, policymaking, and risk management.

Dr. Yellman proposes to define frequency as "a time‐rate of events of a specified type over a particular time interval." We review why no definition of frequency, including this one, can satisfy both of two conditions: (1) the definition should agree with the ordinary meaning of frequency, such as that less frequent events are less likely to occur than more frequent events, over any particular time interval for which the frequencies of both are defined; and (2) the definition should be applicable not only to exponentially distributed times between (or until) events, but also to some nonexponential (e.g., uniformly distributed) times. We make the simple point that no definition can satisfy (1) and (2) by showing that any definition that determines which of any two uniformly distributed times has the higher "frequency" (or that determines that they have the same "frequency," if neither is higher) must assign a higher frequency number to the distribution with the lower probability of occurrence over some time intervals. Dr. Yellman's proposed phrase, "time‐rate of events … over a particular time interval" is profoundly ambiguous in such cases, as the instantaneous failure rates vary over an infinitely wide range (e.g., from one to infinity), making it unclear which value is denoted by the phrase "time‐rate of events."

How can we best allocate limited defensive resources to reduce terrorism risks? Dillon et al.'s Antiterrorism Risk‐Based Decision Aid (ARDA) system provides a useful point of departure for addressing this crucial question by exhibiting a real‐world system that calculates risk reduction scores for different portfolios of risk‐reducing countermeasures and using them to rank‐order different possible risk mitigation alternatives for Navy facilities. This comment points out some potential limitations of any scoring system that does not take into account risk externalities, interdependencies among threats, uncertainties that are correlated across targets, and attacker responses to alternative allocations of defensive resources. In at least some simple situations, allocations based on risk reduction scores and comparisons can inadvertently increase risks by providing intelligent attackers with valuable information, or they can fail to reduce risks as effectively as nonscoring, optimization‐based approaches. These limitations of present scoring methods present exciting technical challenges and opportunities for risk analysts to develop improved methods for protecting facilities and infrastructure against terrorist threats.

Hansen et al. (2007) recently assessed the historical performance of the precautionary principle in 88 specific cases, concluding that "applying our definition of a regulatory false positive, we were able to identify only four cases that fit the definition of a false positive." Empirically evaluating how prone the precautionary principle is to classify nonproblems as problems ("false positives") is an excellent idea. Yet, Hansen et al.'s implementation of this idea applies a diverse set of questionable criteria to label many highly uncertain risks as "real" even when no real or potential harm has actually been demonstrated. Examples include treating each of the following as reasons to categorize risks as "real": considering that a company's actions contaminated its own product; lack of a known exposure threshold for health effects; occurrence of a threat; treating deliberately conservative (upper‐bound) regulatory assumptions as if they were true values; treating assumed exposures of children to contaminated soils (by ingestion) as evidence that feared dioxin risks are real; and treating claimed (sometimes ambiguous) epidemiological associations as if they were known to be true causal relations. Such criteria can classify even nonexistent and unknown risks as "real," providing an alternative possible explanation for why the authors failed to find more false positives, even if they exist.

In his comments on our Rapid Risk Rating Technique (RRRT), Dr. Claycamp states that "Cox and Popken were silent on the pivotal expert judgment subsumed in their method: quality weights for illnesses caused by antimicrobial‐resistant and antimicrobial‐sensitive microbes are tacitly assumed to be equal." However, contrary to this comment, we explicitly provide separate quality weights for antimicrobial‐resistant and antimicrobial‐sensitive illnesses (with different values, based on QALYs lost per illness, illustrated in our article). Thus, our model already addresses Dr. Claycamp's central concern. Moreover, since withdrawing macrolides and fluoroquinolones from use as animal antibiotics is projected to cause more of both types of illnesses—antimicrobial‐resistant and antimicrobial‐sensitive—than continuing to use them would, our conclusion that such withdrawals would be poor public health policy is completely robust to changes in the quality weights. Our human health impacts assessment is therefore indeed immune from the "expert judgments in risk management" discussed by Dr. Claycamp.

Suppose that a residential neighborhood may have been contaminated by anearby abandoned hazardous waste site. The suspected contamination consistsof elevated soil concentrations of chemicals that are also found in the absence of site‐related contamination. How should a risk manager decide which residential properties to sample and which ones to clean? This paper introduces an adaptive spatial sampling approach which uses initial observations to guide subsequent search. Unlike some recent model‐based spatial data analysis methods, it does not require any specific statistical model for the spatial distribution of hazards, but instead constructs an increasingly accurate nonparametric approximation to it as sampling proceeds. Possible cost‐effective sampling and cleanup decision rules are described by decision parameters such as the number of randomly selected locations used to initialize the process, the number of highest‐concentration locations searched around, the number of samples taken at each location, a stopping rule, and a remediation action threshold. These decision parameters are optimized by simulating the performance of each decision rule. The simulation is performed using the data collected so far to impute multiple probable values ofunknown soil concentration distributions during each simulation run. This optimized adaptive spatial sampling technique has been applied to real data using error probabilities for wrongly cleaning or wrongly failing to clean each location (compared to the action that would be taken if perfect information were available) as evaluation criteria. It provides a practical approach for quantifying trade‐offs between these different types of errors and expected cost. It also identifies strategies that are undominated with respect to all of these criteria.

Recent reviews of epidemiological evidence on the relation between exposure to diesel exhaust (DE) and lung cancer risk have reached conflicting conclusions, ranging from belief that there is sufficient evidence to conclude that DE is a human lung carcinogen (California EPA, 1994) to conclusions that there is inadequate evidence to support a causal association between DE and human lung cancer (Muscat and Wynder, 1995). Individual studies also conflict, with both increases and decreases in relative risks of lung cancer mortality being cited with 95% statistical confidence. On balance, reports of elevated risk outnumber reports of reduced risk. This paper reexamines the evidence linking DE exposures to lung cancer risk. After briefly reviewing animal data and biological mechanisms, it surveys the relevant epidemiological literature and examines possible explanations for the discrepancies. These explanations emphasize the distinction between statistical associations, which have been found in many studies, and causal associations, which appear not to have been established. Methodological threats to valid causal inference are identified and new approaches for controlling them are proposed using recent techniques from artificial intelligence (AI) and computational statistics. These threats have not been adequately controlled for in previous epidemiological studies. They provide plausible noncausal explanations for the reported increases in relative risks, making it impossible to infer causality between DE exposure and lung cancer risk from these studies. A key contribution is to show how recent techniques developed in the AI‐and‐statistics literature can help clarify the causal interpretation of complex multivariate data sets used in epidemiological risk assessments. Applied to the key study of Garshick et al. (1988), these methods show that DE concentration has no positive causal association with occupational lung cancer mortality risk.