International Journal of Epidemiology, Volume 45, Issue 6, .. but a four-way relation between cause, effect, contrast for the cause and contrast. What causes age-related macular degeneration (AMD)? Numerous literature . the effect of age (A) on AMD appears to operate through its relationship with. Epidemiology is the study of the causes of disease. However, a dose response relationship is not necessary to infer causation For example, the the results of which have been published and analyzed in authoritative medical journals. ¡¡.
However, all of these definitions summarized in Table 1 have severe deficits. Not totally unexpected, the definitions found in the literature are insufficient to provide a basis for the notion of disease causation.
As pointed out above for physical phenomena, it is also impossible for disease processes to draw an ontologic demarcation within the indefinite stream of events between causal and noncausal associations. Consider a human being as a complex input—output system that is described by a path through a state space of likely very high dimensionality that may or may not explicitly depend on time.
The task is to solve the equations that relate the input stream, the output stream, and the internal states to each other. If we were in possession of such a tool, we would not need the crutch of a concept of causation. Meanwhile, in a pragmatic sense, it is reasonable to stay with this concept but hold in mind that it is just an economical way to organize the otherwise unfathomable stream of events and to take the necessary steps to counteract or prevent the disease process.
The process of diagnosis itself is one of abstraction and generalization because no two diseased human beings given the same diagnosis have exactly the same features.
In this pragmatic sense, disease cause can be defined as follows: Given two or more populations of subjects that are sufficiently similar for the problem under study, a disease cause is a set of mutually exclusive conditions by which these populations differ that increase the probability of the disease.
Causation in epidemiology: association and causation | Health Knowledge
In some cases, the similarity must be high, such that only homozygous twins can be studied; in other cases, maybe only sex and age must be considered, or the state of immunity. Hence, this temporal relation is a precondition for an agent to be considered a causal factor. This definition is in line with the main designs of epidemiologic studies: It is also in line with the pragmatic definition that assessment of causality affords more than just the observation of an increased incidence or prevalence in some group or the other.
Taking Refuge in Causality It seems that the first time causality entered the discussion on epidemiologic results was during the tobacco controversy in the late s and early s.
In particular, the criticism of Fisher concerning the conclusions drawn from the British Doctors Study by Doll and Bradford Hill initiated a detailed consideration of the concept of causality that led to the famous presidential address by Bradford Hill to the Section of Occupational Medicine of the Royal Society of Medicine in In this talk, Bradford Hill discussed nine issues that should be addressed when deciding whether an observed association is a causal relationship.
The Bradford Hill criteria were established such that, in the case they are met for a specific factor, this would increase our confidence in this factor being causally related to the disease. However, they were not intended to dismiss a factor as potentially causing the disease: First, one has to discriminate between evidence for no causal relationship, and no evidence of a causal relationship Altman and Bland The former expresses an important piece of evidence that may have substantial consequences on steps taken to prevent health hazards, whereas the latter simply expresses lack of knowledge.
It is, however, often misunderstood as an exculpation of the agent in question and is readily misused by interested parties to claim that exposure is not associated with adverse health effects.
Some examples of such statements illustrate the point: There are significant differences between these statements.
Causality and the Interpretation of Epidemiologic Evidence
Hence, it points mainly to the lack of knowledge accumulated so far. The second one goes a step further: It claims that risk assessment based on the weight-of-evidence approach [as applied by the U.
Environmental Protection Agency U. However, there is no category of this type in the weight-of-evidence approaches. Because of the by far higher demands on quality and size of studies set out to dismiss the assumption of carcinogenicity, there is an inherent imbalance of classification concerning carcinogenicity and lack of carcinogenicity.
The first statement goes still further: It claims that an analysis based on the Bradford Hill criteria confirms that there is no causal relationship. All other evidence may reduce the weight in favor of a causal relationship but cannot confirm that there is no causal relationship. Are There Criteria for Causation? It is a complete misinterpretation of the nine issues considered by Bradford Hill that they can be a type of checklist to establish causation.
But it may turn out that they owe their popularity, still persisting after 40 years, exactly to this misconception. Because the definition of a disease cause given above affords the existence of mutually exclusive conditions, in a strict sense, causation can be indicated only by experimental production and control of all relevant conditions.
This, however, leads to ethical problems if the factor is potentially debilitating or lethal. And it is practically impossible if the latency is long, as it is for chronic diseases.
Resorting to animal experimentation can reduce some of these problems but introduces new ones, because inference from results in animals to effects in humans is far from trivial. Hence, we are often left with a number of problems that cannot be optimally solved, and therefore there is no set of criteria that, if fulfilled, would result in attributing a factor as either causally related or not.
This does not mean that we cannot, to the best of our present knowledge, come to a decision concerning the relationship of an agent and a disease. Or, as Bradford Hill said 40 years ago: All scientific work is incomplete—whether it be observational or experimental.
All scientific work is liable to be upset or modified by advancing knowledge.
Causality and the Interpretation of Epidemiologic Evidence
That does not confer upon us a freedom to ignore the knowledge we already have, or to postpone the action that it appears to demand at a given time. A Pragmatic Approach Concerning a particular chemical or physical factor, general medical knowledge may suffice to attribute it as harmful and as causing illness or death but even in extreme cases such derivations may not be altogether valid—e. Everest without respiratory aid. So we are dealing with either less obvious hazards or those that occur only rarely or in a small proportion of the population.
The evidence may stem from all kinds of sources, but often we start only from the pessimistic assumption that an agent either not present in the natural environment or present only at much lower levels may be harmful to health. Or it may be that during routine surveillance, a high prevalence of a rare disease is observed that coincides with a rare environmental condition.
How should we come to a conclusion whether the suspected environmental condition is causing disease? It might be worthwhile to stress that there are cases where we do not need the verdict of causation before we take action e.
Starting from the definition of a disease cause stated above, it is obvious that three main issues need to be addressed to simplify the discussion, let us speak of the set of exclusive conditions as of an agent or determinant A: The philosophical sense, on the other hand, comprises a restrictive set of convictions about how epidemiologists should think about causality.
This is true of the great methodological movements of recent times, such as logical positivism, and it is true of the RPOA too. Epidemiology seeks to be precise and quantitative, but we do not have a precise—let alone quantitative—definition of causation, notwithstanding thousands of years of trying.
Epidemiologists thus find themselves in the awkward position of wanting to say, in precise quantitative terms, things that humankind has so far only been able to say vaguely and qualitatively. One response to this conundrum is to speak only of associations. The alternative to retreating into the associational haven is to take the causal bull by the horns… A proper definition of a causal effect requires well-defined counterfactual outcomes, that is a widely shared consensus about the relevant interventions.
The RPOA does not promote this way of posing and answering causal questions as a universal philosophical analysis, but as a way of thinking about causality that is useful for epidemiologists.
The usefulness comes from the predictive value of causal claims that are relative to specified interventions. Empirical associations uncovered by statistical analysis in observational epidemiology and in the social sciences also allow for prediction.
As we observe associations we can sometimes predict what might happen to a particular individual given certain covariates or given the past. However, the associations that are discovered in such observational research do not in general allow for prediction under contrary to fact scenarios, e. The causal inference literature in statistics, epidemiology, the social sciences etc. So far, the RPOA presents itself as an attractive view.(Epidemiology Course) Introduction to Causality Part 24 out of 26
It identifies an advantage of causal claims over associational ones, namely prediction under hypothetical scenarios; and it advocates restricting our attention to causal claims that clearly specify such hypothetical scenarios. It further restricts the hypothetical scenarios to those we can humanly bring about, again apparently because of a motivation of pragmatism.
The crucial question is then this: What is the point of estimating a causal effect that is not well defined? The resulting relative risk estimate will not be helpful to either scientists, who will be unable to relate it to a mechanism, or policy makers, who will be unable to translate it into effective interventions. In this book, as well as within the causal inference framework that has come to dominate in statistics, epidemiology, and the social sciences, causation is typically conceived of in terms of contrasts in the counterfactual outcomes.
These counterfactual outcomes are themselves typically conceived of as the outcomes under hypothetical interventions; and the hypothetical interventions that give rise to counterfactuals usually consist of some human action; for example, a person takes drug A versus drug B… [p.
This is not only because the latter are randomized although that is of course one reason ; it is also because they are experiments, and thus involve an intervention. An observational study can be conceptualized as a conditionally randomized experiment under the following three conditions: Causal claims allow prediction under hypothetical scenarios, provided the causal claims are well defined.
Causal claims and questions are well defined when interventions are well specified. Epidemiologists should restrict their attention to well-defined causal hypotheses, whose hallmark is well-defined interventions. Except for randomization, observational studies should emulate all aspects of experimental studies because doing so restricts observational studies to investigating well-defined causal hypotheses.
These principles bear little resemblance to the incredibly rich and successful historical practice of epidemiology. Nor are they endorsed by everyone who works on causal inference. First, the RPOA insists that interventions or manipulations must be humanly feasible manipulations, in order to be of interest to epidemiology.
Second, the RPOA denies the meaningfulness and usefulness of causal claims that do not readily yield predictions under hypothetical scenarios. By contrast, other approaches typically seek to offer a framework for accommodating and making sense of such claims. For instance, Judea Pearl, whose work is claimed to be an inspiration for the RPOA, does not subscribe to the idea that observational studies should emulate randomized trials, nor to the idea that non-manipulable factors such as sex and race should not be regarded as causes, as exemplified in the following quotes from his work: Surely we have causation without manipulation.
The moon causes tides, race causes discrimination and sex causes the secretion of certain hormones and not others. Nature is a society of mechanisms that relentlessly sense the values of some variables and determine the value of others; it does not wait for a human manipulator before activating those mechanisms [p. The first is the pragmatic attitude to causality adopted by epidemiologists studying smoking and lung cancer.
The first expression of an explicitly pragmatic approach was perhaps articulated by Lilienfeld in Specificity of the association. There must be a one to one relationship between cause and outcome. Temporal sequence of association. Exposure must precede outcome. Change in disease rates should follow from corresponding changes in exposure dose-response. Presence of a potential biological mechanism. Does the removal of the exposure alter the frequency of the outcome?
According to Rothman , while Hill did not propose these criteria as a checklist for evaluating whether a reported association might be interpreted as causal, they have been widely applied in this way. Rothman contends that the Bradford - Hill criteria fail to deliver on the hope of clearly distinguishing causal from non-causal relations.
For example, the first criterion 'strength of association' does not take into account that not every component cause will have a strong association with the disease that it produces and that strength of association depends on the prevalence of other factors. In terms of the third criterion, 'specificity', which suggests that a relationship is more likely to be causal if the exposure is related to a single outcome, Rothman argues that this criterion is misleading as a cause may have many effects, for example smoking.