Improving Causal Inferences in Risk Analysis

Abstract

Recent headlines and scientific articles projecting significant human health benefits from changes in exposures too often depend on unvalidated subjective expert judgments and modeling assumptions, especially about the causal interpretation of statistical associations. Some of these assessments are demonstrably biased toward false positives and inflated effects estimates. More objective, data-driven methods of causal analysis are available to risk analysts. These can help to reduce bias and increase the credibility and realism of health effects risk assessments and causal claims. For example, quasi-experimental designs and analysis allow alternative (noncausal) explanations for associations to be tested, and refuted if appropriate. Panel data studies examine empirical relations between changes in hypothesized causes and effects. Intervention and change-point analyses identify effects (e.g., significant changes in health effects time series) and estimate their sizes. Granger causality tests, conditional independence tests, and counterfactual causality models test whether a hypothesized cause helps to predict its presumed effects, and quantify exposure-specific contributions to response rates in differently exposed groups, even in the presence of confounders. Causal graph models let causal mechanistic hypotheses be tested and refined using biomarker data. These methods can potentially revolutionize the study of exposure-induced health effects, helping to overcome pervasive false-positive biases and move the health risk assessment scientific community toward more accurate assessments of the impacts of exposures and interventions on public health.

Has reducing fine particulate matter and ozone caused reduced mortality rates in the United States?

“…The goal of our analysis is to understand the extent to which historical associations between pollutant levels and mortality rates reflect a clear causal relation, rather than merely coincident trends, or the effect of confounders, or modeling choices. Table 2 lists several quantitative methods for causal hypothesis testing, modeling, and analysis that have been extensively developed and applied over the past six decades [5]. Various advantages of these techniques, compared with qualitative causal criteria [31] such as the traditional Hill considerations and other weight-ofevidence and associational methods, are well explained and illustrated [6] in the references for Table 2, along with their limitations [33].…”

“…For at least the past two decades, however, epidemiologists and commentators on scientific methods and results have raised concerns that current practices too often lead to false-positive findings and to mistaken attributions of causality to mere statistical associations [1e4]. Formal training in epidemiology may be a mixed blessing in addressing these concerns, as concepts such as "attributable risk," "population attributable fraction," "burden of disease," "etiologic fraction," and even "probability of causation" are based on relative risks and related measures of statistical association and do not necessarily reveal anything about causation [5,6]. Limitations of human judgment and inference, such as confirmation bias (finding what we expect to find), motivated reasoning (concluding what it pays us to conclude), and overconfidence (mistakenly believing that our own beliefs are more accurate than they really are), do not spare health effects investigators.…”

Towards a Fuzzy Bayesian Network Based Approach for Safety Risk Analysis of Tunnel‐Induced Pipeline Damage

“…As for the test data, we can model the validation process using independent data that are not used in model design. In regard to the test indicators and standards on improving causal inferences in risk analysis, a common approach in epidemiology is to use statistical tests to determine whether there is strong evidence for a nonrandom positive association between input and output, and then to check whether the association can correctly be described by adjectives such as “strong,” “consistent,” “specific,” “temporal,” and “biologically plausible.” Specifically, we previously proposed two indicators to evaluate the rationality and validity of the established BN models: namely, Model Bias and Model Accuracy . The assessment of Model Bias aims to illustrate the correspondence between the central tendency of model output and the actual observation, while the assessment of Model Accuracy aims to illustrate the correspondence between individual pairs of model prediction and observation.…”

Evaluating Weight of Evidence in the Mystery of Balkan Endemic Nephropathy

“…Epidemiologic models (77) -the epidemiologic triad, wheel of causation, web of causation, and causal pie model-were researched and determined if they would provide a more objective guide to causation compared to the BHC. Other causal inference designs, specifically, those listed in Cox, (78) were examined; and those for which a sufficient literature exists regarding BEN and its possible causes were applied to the question of which etiological factor was most likely to be the cause of BEN. Table II summarizes the main suspected risk factors for BEN, and the weight of evidence linking each to the disease according to each of the nine BHC.…”

“…In this specific instance of BEN, such a model would only be useful if it were presupposed that BEN were related to genetic factors, but evidence exists to the contrary. (8,80,81) Finally, causal inference methods outlined by Cox (78) were also considered, and several for which sufficient studies existed on BEN were applied to examining BEN etiology. Cox's article describes formal methods to model and test causal hypotheses, of which three are relevant for this analysis: conditional independence tests, counterfactual and potential outcome models, and negative controls.…”