Although philosophers had discussed the concept of causation since ancient times, the first exposition of causation using scientific principles was proposed by Robert Koch in 1884. Koch developed a series of four postulates to aid microbiologists in determining if exposure to pathogenic bacteria caused a disease. The postulates include the following.
• The microorganisms should be found in all cases of the disease in question and the distribution in the body should be in accordance with the lesions observed.
• The microorganisms should be grown in pure culture in vitro for several generations.
• When the pure culture is inoculated into susceptible animal species, the typical disease must result.
• The microorganism must again be isolated from the lesions of such experimentally produced disease.
In general, Koch's postulates have been satisfied for most known pathogens although some pathogens such as Mycobacterium leprae (leprosy) cannot be grown in culture and others such as Neisseria gonorrhoeae (gonorrhea) have no adequate animal model of infection. Recently, the microbiological community has adopted molecular techniques for identification and experimentation; thus, molecular analogs have been proposed to augment Koch's original postulates.
As long as the focus of environmental health was on infectious disease, Koch's postulates were an adequate basis for determinations of disease causation. In the mid-twentieth century, however, concerns began to arise regarding potential associations between chemical exposure and disease. Rather than taking a strict experimental approach, environmental health scientists began to rely increasingly on epidemiology to make inferences between chemical exposure and disease. For example, the 1964 report of the US Surgeon General on smoking and health concluded that cigarette smoking caused lung cancer. This conclusion was based on historical associations that met tests of strength, consistency, specificity, temporal relationship, and coherence.
In 1965, British epidemiologist Sir Austin Bradford Hill presented a formal system for assisting researchers in determining if an association between chemical exposure and a disease could be considered causal. The factors presented by Hill are often known as the Bradford Hill postulates or Bradford Hill criteria, although it must be noted that Hill never intended these factors to be used in a vacuum or as a checklist in determinations of causation. Hill considered the situation in which observation had revealed an ''association between two variables perfectly clear-cut and beyond what we would care to attribute to the play of chance'' and listed nine factors that should be considered.
Strength of the association. In epidemiology, the strength of the association is usually measured by the relative risk (or odds ratio). A strong association (high numerical value of a relative risk) supports a hypothesis of causality of the association. A weak relative risk fails to support the hypothesis. Hill used well-known examples from historic epidemiology such as Percival Pott's observation of the association between exposure to soot and testicular cancer in chimney sweeps and John Snow's observations about cholera and polluted water to illustrate the concept of strength of the association. Some investigators, and indeed some courts and regulatory bodies, have codified numerical values of relative risk that demark 'strong' or 'weak' in this context; however, none of these limits have been generally accepted. A hypothetical causal relationship is also thought to be more credible if it is precise. Precision in this sense being interpreted as narrow confidence intervals around the relative risk.
Consistency. Consistency normally refers to reproducibil-ity of results. In epidemiology, consistency is demonstrated by repeated findings of associations in different population groups or in different settings. Hill used the example of and association between tobacco smoking and cancer that had been observed in 29 retrospective and seven prospective epidemiologic studies to illustrate consistency.
Specificity. Specificity often applies to the ability of a particular toxicant to cause a response. In epidemiology, the issue is expressed in the elimination of confounders; in medicine the issue is expressed through the concept of differential diagnosis.
Temporality. The exposure must precede the disease. This is a threshold requirement for causation. If a latency period is associated with the development of a particular disease, the duration of the latency should be taken into account.
Biological gradient. The biological gradient often is the dose-response relationship in toxicology. An observed dose-response or concentration-response relationship supports a hypothesis of causation. Locational gradients and variable durations of exposure may also be used in causation analysis.
Plausibility. The assertion should be biologically plausible. Most investigators use information from field observations, laboratory bioassays, chemical measurements, structure-activity relationships, pharmacokinetics, and in vitro testing to reach as judgment regarding causation. When considered as a whole, these various sources of information should be compatible with each other. This factor depends on the state of biological knowledge. For example, a mode of action of a particular toxicant may be unknown at one point in time but discoveries in a related area of research may result in an understanding of the mode of action at a later date. In the first instance, the association may be considered implausible, whereas in the second it may be deemed plausible.
Coherence. Hill believed that an interpretation of causality should not conflict with generally known facts of the natural history or biology of the response.
Experiment. Experimental evidence results when investigators are able to adequately control the variables in a test. With the exception of a randomized clinical trial, human experimental data are generally not available. With species other than humans, experimental data may be the foundation upon which a determination of causation is made.
Analogy. Hill believed that, in some cases, it was possible to draw judgments about causation on the basis of analogy. His examples were birth defects associated with thalidomide and rubella arguing that a similar drug or viral disease could be presumed to have similar effects.
These postulates have been widely applied in human epidemiology and, to a lesser extent, in cognate fields. Although mostly used to analyze observational data, they may also be applied to experimental data. Bradford Hill, in fact, was the originator of the randomized clinical trial in pharmaceutical testing which relies on carefully controlled experimental protocols.
In 1976, A. S. Evans updated Koch's postulates integrating concepts proposed by Hill into a unified concept of disease causation that could be applied to both infectious and noninfectious diseases. Evans postulates included the following.
• The presence of the response should be significantly higher in those exposed than in controls.
• Exposure to the hypothesized cause should be more frequent among those with the response than in controls when all other risk factors are held constant.
• Incidence of the response should be significantly higher in those exposed to the cause than those not so exposed, as shown by prospective studies.
• Temporally, the response should follow exposure to the hypothesized cause.
• A spectrum of host responses should follow exposure to the hypothesized agent along a logical biological gradient.
• A measurable response following exposure to the cause should have a high probability of appearing in those lacking the response prior to exposure.
• Experimental reproduction of the response should occur more frequently in animals or humans exposed to the hypothetical cause compared to an unexposed group.
• Elimination or modification of the hypothetical cause or of the vector carrying it should decrease incidence of the response.
• Prevention or modification of the host's response on exposure to the hypothetical cause should decrease or eliminate the response.
• All of the relationships and findings should be biologically plausible.
In 1991, Mervyn Susser critically analyzed scientific thinking concerning causation and developed a refined group of criteria that he felt were most useful and least tautological:
1. Strength is the size of the estimated risk given the constraints of probability levels, confidence intervals, or other measure of likelihood.
2. Specificity is the precision with which one variable, to the exclusion of others, will predict the occurrence of another:
(a) specificity in the cause implies that a given effect has a unique cause, and
(b) specificity in the effect implies that a given cause has a unique effect.
3. Consistency is the persistence of an association upon repeated testing:
(a) survivability is the number, rigor, and severity of tests of association; and
(b) replicability is the number and diversity of tests of association.
4. Predictive performance is the ability of a causal hypothesis drawn from an observed association to predict an unknown fact that is consequent on the initial association.
5. Coherence is the extent to which a hypothesized causal association is compatible with existing knowledge:
(a) theoretical coherence describes compatibility with existing scientific theory;
(b) factual coherence is compatibility with existing knowledge;
(c) biologic coherence is compatibility with knowledge drawn from a species other than the species from which the causal hypothesis was drawn; and
(d) statistical coherence is compatibility with a conceivable model of the distribution of cause and effect.
The application of sets of criteria or postulates such as those proposed by Bradford Hill, Evans, or Susser has not been without its detractors. Some of the criticism stems from the use of the Bradford Hill postulates as checklists, although this was not intended by its author. Even if the postulates are used correctly, as a broad framework for analysis, many investigators have pointed out deficiencies in the individual postulates or caveats that should be observed in their application. For example, cases where there strong associations but no causation are well known (e.g., the association between water ingestion and mortality). Theoretical objections have also been raised to application of postulates for determination of causation. The school of the twentieth-century philosopher Karl Popper believes that science progresses by rejecting or modifying causal hypotheses rather than demonstrating causation. Because causation is often associated with litigation or regulatory decisions regarding chemical safety, testifying experts or advocates for or against a particular decision have voiced objections to particular postulates particularly when application of the postulate contradicts the authoritative position of the expert. For example, while acknowledging that a strong association is more likely to be causal than a weak one, an authoritative attack on this postulate may note that this does not preclude the possibility of a weak association being causal. Similarly, the authoritative position may argue that biological plausibility may be a mere reflection of the state of biological knowledge and that supporting experimental evidence, while valuable, may be precluded by practical or ethical considerations.
More recently, the concept of evidence-based toxicology has been developed as a comprehensive framework for causation. Drawing on the emerging field of evidence-based medicine, Philip Guzelian and co-workers have attempted to formalize a process for determination of causation that substitutes reliance on a comprehensive scientific database analysis for authoritative statements. In general, Guzelian and co-workers have proposed three main stages of evaluation consisting of a total of 12 individual steps:
1. collecting and evaluating the relevant data,
2. collecting and evaluating the relevant knowledge,
(a) frame the question,
(b) assemble the relevant literature,
(c) assess and critique the literature, and
3. joining data with knowledge to arrive at a conclusion,
(a) general causation (answer to the framed question)
(d) alternative cause
In this system, the first stage is problem specific and corresponds most closely to the problem formulation and exposure assessment stages in risk assessment. Source evaluation includes identification of the chemicals of potential concern and elucidation of exposure-related properties such as bioavailability and persistence. Exposure and dose evaluations are both qualitative and quantitative and include such elements as pathway analysis and dose calculation. Diagnosis presupposes that damage or harm has occurred.
Field biology, laboratory testing, and pathology in addition to other specialties are used to diagnose ecological problems. The second stage corresponds loosely to the toxicological evaluation (i.e., hazard identification and dose-response quantification) stage of a risk assessment. Framing the question involves developing a conceptual model that poses specific questions that may be addressed by recourse to the scientific literature. Guzelian and co-workers list toxicological delimiters that may be used as aids in framing questions (modified slightly here to reflect current risk assessment usage):
1. effect delimiters,
(a) time frame (acute, subchronic, chronic),
(b) site of action (local or systemic),
(c) possibility of toxicological interactions,
(d) certainty of diagnosis,
2. route of exposure delimiters,
(a) source (medium, matrix, single or multiple chemicals),
(b) duration and frequency,
(c) continuous or intermittent,
(d) constant or variable magnitude,
3. dose delimiters,
(c) frequency and duration,
(d) variable or constant,
(e) dose metric (applied, absorbed, target tissue), and
(c) experimental or observational.
The last stage corresponds most closely to the application of classical causation criteria such as those postulated by Bradford Hill and others. Although relatively new and designed for specific application to causation of human disease, evidence-based toxicology bears significant promise for wider application in environmental risk assessment.
Scientific and philosophical debates regarding the nature of causation and applications of criteria to determine causation continue to the present day and are unlikely to be resolved in the immediate future. Rothman and Greenland have recently taken the position that causal inference is most appropriately viewed as an exercise in the measurement of an effect rather than a criteria-driven process for deciding if the effect is present. They present counter-evidence for each one of Bradford Hill's criteria and imply that criteria are only valid to generate hypotheses that may be subsequently tested. Kundi has proposed a dialog approach to evaluate causal inference relying on concepts involving multiple causes and addressing issues of temporality, association, environmental, and population equivalence. Many of these debates are more theoretical and academic rather than pragmatic in nature. Regardless of the ultimate outcome of the debates, causation criteria such as those proposed by Bradford Hill and Susser, especially when used in the context of evidence-based toxicology, are a valuable tool for risk assessors, toxicolo-gists, and epidemiologists to use in evaluating accumulated weight of evidence in complex situations involving multiple stressor stimuli and responses.
Was this article helpful?