Document is current

Article History

Received: 12 May 2021

Revised: 10 September 2021

Accepted: 22 December 2021

First Online: 26 January 2024

CONFLICT OF INTEREST

: The authors of this work declare that they have no conflicts of interest.

: (1) “Here, then, are nine different viewpoints from all of which we should study association before we cry causation. What I do not believe, and this has been suggested, is that we can usefully lay down some hard-and-fast rules of evidence that must be obeyed before we accept cause and effect. None of my nine viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as sine qua non. What they can do, with greater or less strength, is to help us to make up our minds on the fundamental question, is there any other way of explaining the set of facts before us, is there any other answer equally, or more, likely than cause and effect?” [CitationRef removed] (here and below, translation by <i>A.K</i>.).(2) In C.V. Phillips and K.J. Goodman, 2004 [CitationRef removed], entitled “The missed lessons of Sir Austin Bradford Hill,” it is stated that “Hill advised epidemiologists to avoid over-emphasizing statistical significance testing, given the observation that systematic error is often greater than random error.” His suggestions about how to intuitively assess causation are almost completely lost when his address is distilled into a checklist.” Thus, the world’s leading medical statistician of the 20th century [CitationRef removed], A.B. Hill, apparently, was pragmatic about statistics.(3) Some authors (for example, [CitationRef removed, CitationRef removed, CitationRef removed, CitationRef removed]) considered the moment of the absoluteness of “Temporality,” but not in connection with the quotation from A.B. Hill, 1965 [CitationRef removed]. Only in the handbook of epidemiology did A. Aschengrau and G.R. Seage, III, 2014; 2020 [CitationRef removed], pay attention, immediately after quoting from [CitationRef removed], to the absurdity: “We agree with the essence of Hill’s conclusion; however, we think that one part of his statement is incorrect—temporality is a requirement for causality.” We quote it in connection with the rarity of such an obvious conclusion, a single one for many hundreds of sources. Most likely, this is material for psychology.(4) “While data to support some criteria may be incomplete or missing for any given assessment, these criteria offer a useful way to evaluate available information” [CitationRef removed].(5) “The strong consensus among epidemiologists is that conclusions about causation should not be drawn, if at all, until a number of criteria have been considered” [CitationRef removed].(6) “Today, the 1964 Surgeon General’s report and Hill’s subsequent causal criteria are routinely cited as authoritative statements of the proper method for assessing a body of etiological evidence” [CitationRef removed].(7) “It is generally accepted that the Bradford Hill criteria, as formulated in 1965, are still the most relevant criteria to be used in causal inference.” “It is generally accepted that the Bradford Hill criteria, as formulated in 1965, are still the most relevant criteria to be used in causal inference.” “Bradford Hill criteria still remain key components to causal inference” [CitationRef removed].(8) “We believe that Bradford Hill’s guidelines form a useful tool as they stand” [CitationRef removed].(9) “Various modifications of this list have been suggested, and many of the elements remain cornerstones of judgement on whether an exposure really does cause a disease, or whether an intervention is effective in preventing or treating disease” [CitationRef removed].(10) “Another proposed benefit of using Hill’s criteria was that these provided a framework within which allowance could be made for incomplete datasets and missing information” [CitationRef removed].(11) “Guidelines suggested by Bradford Hill are often used to judge the confidence that a risk factor (e.g., radiation exposure) might cause increased incidence in a population of health outcomes (e.g., cancer)” [CitationRef removed].(12) “The Bradford Hill Criteria remain one of the most cited concepts in health research and are still upheld as valid tools for aiding causal inference” [CitationRef removed].(13) “[T]hese have stood the test of time as a valuable way of assessing the strength of causal inference where an association has been demonstrated” [CitationRef removed].(14) “Notwithstanding the renewed interest in other models of causality over the past few years…, Hill’s guidelines remain the cornerstone of causal inference for the practical epidemiologist and health policy expert” [CitationRef removed].(15) “In spite of the rapid development in theoretical epidemiology, they have remained at least as a reference point for causal thinking in review committees and for decision makers” [CitationRef removed].(16) “Hill also stressed that it was meaningless to combine criteria into scores because an association was not more likely to be causal if it met more criteria” [CitationRef removed].(17) “[H]e offered no means of deciding whether these aspects hold when considering a given association, no hierarchy of importance among them, and no method for assessing them to arrive at an inference of causation. Hill included the list of nine viewpoints in four editions of his textbook of medical statistics from 1971 through 1991…, without further elaboration than appeared in the original paper” [CitationRef removed].(18) The criterion “Predictive performance” was proposed by M. Susser as a concession to “Popperian Epidemiology” [CitationRef removed] and was further used by G.A. Fox in the formation of a set of causality guidelines for ecology and eco-epidemiology [CitationRef removed].(19) The fact of the presence of an association is also identified as the criterion by M. Susser [CitationRef removed].(20) “The selection of criteria is more a matter of customary practice than theoretical rigor,” “…is more a matter of personal preference than of careful inquiry” [CitationRef removed].(21) In L. Gordis, 2014 [CitationRef removed], all Hill’s criteria (except “Analogy”) are first named, as we mentioned earlier [CitationRef removed]. But then, in the step-by-step list of evidence-based methodologies, at stage II on the use of guidelines, only seven criteria are listed; not only was “Analogy” absent, so was “Specificity.”(22) Key Event: the observed reproducible stage (which can be measured directly or indirectly, or reasonably deduced based on knowledge of the pathogenesis of the adverse effect), which is a necessary element of “Mode of Action” (MOA) and is crucial for the result. MOA is a biologically plausible series of chemically specific key events that begins with exposure and continues through the interaction of the agent inside the cell, as well as subsequent physiological changes and changes in tissues or organs leading to an adverse effect or outcome [CitationRef removed]. See also, [CitationRef removed].(23) We made an attempt to contact the first author, G. Swaen [CitationRef removed], by e-mail to find out the fate of such an important development, which was not reflected in any way in the subsequent publications of this researcher. The address was official, from the relevant website of the institution, but we did not receive any response.(24) “The challenge of determining causation in public health has always been shaped by the limitations of the available data, the understanding of the underlying biological or sociological processes, and our ability to intervene in the real world. Faced with sometimes limited data and an often poor understanding of a network of connected factors in a complex world, we revert to pragmatism” [CitationRef removed].(25) “If we ask for proof in medicine, or any other empirical science, we may be asking for something that does not exist” [CitationRef removed] (the original source is not available; quoted according to [CitationRef removed, CitationRef removed]).(26) The “counterfactual ideal” justifies an ideal control group when an individual or a group of people who have been exposed are compared with the same individual or with the same group, but without exposure. And it happened <i>at the same time</i>. Such an approach is in fact impossible; therefore, replacement with real comparison groups is used, that is, controls (“contrasts”) imitating the “ideal” [CitationRef removed, CitationRef removed].(27) “All scientific work is incomplete, whether it be observational or experimental. All scientific work is liable to be upset or modified by advancing knowledge. 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” [CitationRef removed].(28) “[N]ow very few philosophers or scientists still think that scientific knowledge is, or can be, proven knowledge. But few realize that with this the whole classical structure of intellectual values falls in ruins and has to be replaced: one cannot simply water down the ideal of proven truth, as some logical empiricists do, to the ideal of ‘probable truth’ or, as some sociologists of knowledge do, to ‘truth by [changing] consensus’” [CitationRef removed].(29) In the work by L. Cox, 2018 [CitationRef removed], causal approaches in epidemiology are expanded to nine, but probabilistic causality is named the first. Thus, (1) Probabilistic causation; (2) Associational causation (higher risks with higher exposure); (3) Attributive causation (R additional cases per person-years per unit of exposure); (4) Counterfactual and potential outcomes causation ( a decrease in exposure per unit of its index causes a decrease in frequency by R person-years); (5) Predictive causation (observation of the effect after counterfactual intervention); (6) Structural causation (the average number of cases per person-years is derived from the magnitude of the exposure); (7) Manipulative causation (decrease in exposure per unit of its index reduces the number of expected cases per R person-years); (8) Explanatory/mechanistic causation (an increase in exposure per unit of its index causes changes propagating through the biological network of causal mechanisms); (9) Without causation.(30) “However, although we can engage in fantasies of deterministic causal attribution, we should recognize that, by ignoring the constraints imposed by how the material world is, we encourage the mythopoetics of personalized medicine. Epidemiologists, surely, should be suspicious of such” [CitationRef removed].(31) Obvious denials of causality criteria in K.J. Rothman’s works up to 2005 [CitationRef removed, CitationRef removed, CitationRef removed, CitationRef removed] were smoothed out in the later books of 2008–2014 [CitationRef removed, CitationRef removed, CitationRef removed]. Continuing to stand on the position of deterministic causality and only the hypothetical–deductive method, K.J. Rothman (alone or with co-authors) at least since 2005 [CitationRef removed, CitationRef removed] pointed out that, although there is no set of causal criteria for philosophical reasons, nevertheless, “lists of causal criteria have become popular, perhaps because they seem to provide a roadmap [for passing] through difficult territory” lists of causal criteria have become popular, possibly because they seem to provide a road map through complicated territory” [CitationRef removed, CitationRef removed, CitationRef removed, CitationRef removed]).(32) “The main causal model used by epidemiologists today is Rothman’s ‘pies’” [CitationRef removed].(33) “[T]he current system of epidemiology is very much determined by the thinking of Rothman” [CitationRef removed].(34) “Rothman’s 1986 book <i>Modern Epidemiology</i> [CitationRef removed] represented a watershed moment in the discipline” [CitationRef removed].(35) As stated in [CitationRef removed], “It is ironic then that the committee [NRC USA] goes on to recommend the GRADE system which is characterized as being ‘closely aligned with the Hill criteria for establishing causality’ ” [CitationRef removed].(36) “Epidemiologists seem to confuse the practical results of epidemiological research at the population level with the metaphysical views about the reality of disease causation at the individual level in their writings on causation” [CitationRef removed].(37) “The scientific method does not operate in isolation, but is conducted by the scientific community, which has specific internal norms to guide the activities of scientists in applying the scientific method. These norms include truthfulness, consistency, coherence, testability, reproducibility, validity, reliability, openness, impartiality, and transparency” [CitationRef removed, CitationRef removed]. These sources have already been cited by us [CitationRef removed].(38) As an example of a conditional, but scientifically based assessment, the UNSCEAR-2012 [CitationRef removed] gives an example of the transgenerational effects of radiation in humans (that is, disorders and pathologies in nonirradiated descendants of irradiated parents). Epidemiology has not provided evidence of the presence of such for more than 70 years of large-scale studies (since 1948 [CitationRef removed]). However, the UNSCEAR concluded about the risks of transgenerational effects of radiation for humans based on animal experiments, for which such effects are clearly detected (see [CitationRef removed, CitationRef removed, CitationRef removed]).(39) “Incorporating nonscientific concerns. This approach may or may not take account of science-based inferences. In this case, decision-makers may take account of norms external to science such as social responsibility, ethics, utility, prudence, precaution, and practicality of application. Such considerations, while important, are outside of the Committee’s remit, and are mentioned here only for the sake of completeness” [CitationRef removed].(40) The unambiguous emphasis in [CitationRef removed, CitationRef removed] on the discrepancies in the conclusions in the studies [CitationRef removed, CitationRef removed] on the effect of alcohol on the incidence of breast cancer does not seem so unambiguous. The use in [CitationRef removed] (1989) of the criteria “Strength of association,” “Consistency of association,” “Specificity,” “Temporality,” and “Coherence with current facts and theoretical knowledge” led to the conclusion that there is no causality. The conclusion of the second study [CitationRef removed] (1990), based on the same criteria, was somewhat different, but not so categorical: “with risk factors for breast cancer, but not for heart disease, may be wise to limit her use of alcohol until more is understood about its effects.” Currently, IARC classifies alcohol as a risk factor for breast cancer [CitationRef removed].(41) “Burch is very critical of the statement that the “causal significance of an association is a matter of judgment which goes beyond any statement of statistical probability.” This is true in science in general. One can place a statement of probability on the findings of a specific experiment, or of a sample survey, but it is impossible to place a probability statement and confidence limits on the truth of a scientific hypothesis. A major reason for this is that a demonstration of causation depends upon a synthesis of sets of data of different types. It would be worthwhile summarizing this logical approach, since it serves as a general background for the more detailed discussion” [CitationRef removed].(42) A method of checking the presence of a confounder to associate an effect with a factor of interest. Another effect or effect is being sought, which may be due to the same confounder, but the association with which is devoid of biological plausibility. If the association is maintained here, then there is the presence of a single confounding factor. So, smoking gives the same association with mortality from both suicides (a biological mechanism can be invented) and homicides (there is no plausibility). In both cases, the influence of behavioral factors is likely [CitationRef removed]. For other examples, see [CitationRef removed].(43) “Scientists these days tend to keep up a polite fiction that all science is equal. Except for the work of the misguided opponent whose arguments we happen to be refuting at the time, we speak as though every scientist’s field and methods of study are as good as every other scientist’s, and perhaps a little better. This keeps us all cordial when it comes to recommending each other for government grants” [CitationRef removed].(44) Alvan R. Feinstein (United States, 1925–2001): “Meta-analysis: statistical alchemy of the 21st century … the idea of getting something for nothing, while simultaneously ignoring established scientific principles” [CitationRef removed]. He was such a critic that even in fact his obituary, written by A. Morabia in 2002, has the title: “The controversial controversy of a passionate controversialist” [CitationRef removed]. A.R. Feinstein died as he lived: of a heart attack during a discussion at one of the symposiums [CitationRef removed]. This author introduced into epidemiology and evidence-based medicine the term “gold standard” for RCT (1982, together with R.I. Horwitz [CitationRef removed]), the name “classical epidemiology” [CitationRef removed, CitationRef removed], etc. [CitationRef removed].(45) “In a controlled trial, as in all experimental work, there is no need in the search for precision to throw common sense out of the window” [CitationRef removed].(46) It is possible, however, to make other selections for every taste: confirming the “standard” position of former times that observational studies overestimate the effects compared to RCT or, less often, that they underestimate them (see sources in systematic reviews [187–189]).(47) “These criteria offered by Hill are saddled with reservations and exceptions” (1982) [CitationRef removed] (cited in [CitationRef removed]) and “As is evident, the standards of epidemiologic evidence offered by Hill are saddled with reservations and exceptions” (2005–2008) [CitationRef removed, CitationRef removed, CitationRef removed, CitationRef removed].(48) “Although checklists may not be appropriate for causal inference, the points laid out by Hill are still important considerations. The criteria may be useful when applied in the context of specific hypotheses” [CitationRef removed].(49) “Assertions that each of Hill’s criteria can be dispensed with (usually advanced authoritatively without supporting evidence) are improvident if the goal is to establish causation” [CitationRef removed].(50) INUS “Insufficient and Non-redundant part of an Unnecessary but Sufficient condition” [CitationRef removed, CitationRef removed]. For example, in a philosophical paper by D.A. Golosnyi, Moscow State University, 2014 [CitationRef removed], where the views of J.L. Mackie are considered, some legal turns are taken (“causing”). In [CitationRef removed], the following construction is used: “it acts as an insufficient, but necessary part of the condition, which in itself is not necessary, but sufficient.” It is not close to the original. In 1965 [CitationRef removed] J.L. Mackie presented as an example of INUS the complex causality of a fire: a short circuit occurred, which is not a necessary condition, since other causes could lead to a fire. Equally, it is not sufficient, because if there were no combustible materials nearby, the short circuit would not cause a fire. The reason was thus complex and consisted of a combination of various factors that made it sufficient. In 1974 [CitationRef removed] J.L. Mackie gave an example of taking poison when a person did not use an antidote and did not have a natural cleansing by vomiting. What is the cause of death? Did a person die from the fact that poison entered the body, or from the fact that he did not take an antidote, or from the fact that his stomach did not work with vomiting? The mere ingestion of poison alone is not sufficient: many took poison without death (because they vomited). But poisoning is part of a combination of conditions that together are sufficient for a lethal outcome. In addition, based precisely on this complex of lethal conditions (not taking into account other complexes sufficient for death from other causes), taking poison alone is not superfluous: deaths <i>from poison</i> do not occur under circumstances when the poison has not been taken. Thus, the intake of poison is an insufficient, but not superfluous part, of an unnecessary, but sufficient condition for death (cited in [CitationRef removed]).(51) “In our ignorance, the best we can do in assessing risk is to classify people according to measured risk indicators and then assign the average risk observed within a class to persons within the class. As knowledge or specification of additional risk indicators expands, the risk estimates assigned to people will depart from average according to the presence or absence of other factors that predict the outcome” [CitationRef removed]. This is the third edition of <i>Modern Epidemiology</i> of 2008. Judging by the quote in [CitationRef removed], practically identical arguments were given in the first edition of 1986 [CitationRef removed].(52) “What is required is much more than the application of a list of criteria. Instead, one must apply thorough criticism, with the goal of obtaining a quantified evaluation of the total error that afflicts the study. This type of assessment is not one that can be done easily by someone who lacks the skills and training of a scientist familiar with the subject matter and the scientific methods that were employed. Neither can it be applied readily by judges in court, nor by scientists who either lack the requisite knowledge or who do not take the time to penetrate the work” [CitationRef removed].(53) According to the “Health belief model,” people, as a rule, do not want to change their behavior if they do not believe in the benefits of such changes [CitationRef removed]. To this it is useful to add a quote from A.B. Hill, 1965 [CitationRef removed]: “we should need very strong evidence before we made people burn a fuel in their homes that they do not like or stop smoking the cigarettes and eating the fats and sugar that they do like”(54) Jeremiah Noah Morris (1910–2009; Scotland) is a well–known epidemiologist in the field of health, who for the first time established the importance of physical activity in preventing cardiovascular disease. Population approach; individual and group risks [CitationRef removed].(55) For P.S. Laplace, the probability is relative: “Probability is relative, in part to ignorance, in part to our knowledge” [CitationRef removed], cited according to [CitationRef removed]. If we could know the world in its entirety and have infinite intelligence, we would presumeably not need probability [CitationRef removed].(56) “The conceptual framework for causes presented here is intended neither as a review nor as an expansion of knowledge, but rather as a viewpoint which bridges the gap between metaphysical notions of cause and basic epidemiologic parameters. The focus, then, is neither metaphysics nor epidemiology, but the gulf between them … may facilitate communication about causes of illness” [CitationRef removed].(57) “[E]pidemiologists seem to confuse the practical results of epidemiological research at the population level with the metaphysical views about the reality of disease causation at the individual level in their writings on causation.” “On the one hand, I answer the question for a useful notion of cause for epidemiological practice by defending the probabilistic account in terms of the average effects as an appropriate one. On the other hand, I answer the question of the metaphysical presuppositions of causal concepts in epidemiology by defending Kenneth Rothman’s view on causation as one providing an insightful answer to that question” [CitationRef removed].(58) “SCC models (the only ones originating in epidemiology) stand apart in requiring specification of mechanisms within the individual units under study. There are rarely data to support such detailed specification, which may explain why SCC models have seen little use beyond teaching examples” [CitationRef removed].(59) “An example is “If sepsis had occurred, the patient would have died” and “If no sepsis had occurred, the patient would not have died.” We need not go into the details of this widely acclaimed theory because its very basis is too speculative and beyond any accessible reality” [CitationRef removed].(60) “It is absolutely impossible to ground them in the experiential world and test them empirically. Accordingly, an advocate of a counterfactual will fail to answer reasonably the basic epistemological question “How do you know that?” Notwithstanding their intriguing metaphysical aura, counterfactuals cannot be relied upon and fruitfully used in empirical–practical domains such as medicine. Diagnostic–therapeutic decision making ought not to rest on untestable etiologic speculation if more reliable knowledge is available. In addition, it has been convincingly demonstrated that counterfactuals lead to inferential absurdities in deductions and should therefore be avoided in scientific reasoning” [CitationRef removed].(61) According to our study (63 cohorts from 18 countries and international groups), the inability to identify an excessive risk of cancer mortality for the majority of workers in the global nuclear industry who have accumulated over the period of employment, average doses rarely exceeding 100 mSv [CitationRef removed], leads to the conclusion that it is difficult to fix stochastic effects for such doses due to uncertainties and confounders [CitationRef removed].(62) It is not easy to find weighty references with the particular “Age–cancer mortality rate” relationship, so we presented all those known.(63) Regarding the question of whether the cancer mortality rate is actually increasing from year to year (rather the opposite according to Rosstat), see our paper [CitationRef removed], note 32.(64) “We use a DAG to visualise complex associations, but when we only have observational data at our disposal, we must find other ways to assess (a) whether a particular association is causal and not due to confounding or other bias, and (b) what the direction of this association is.” “One way to answer this question is by incorporating prior knowledge in Hill’s scheme (or similar criteria) with DAGs to determine which edges can be considered causal” [CitationRef removed].

: This work does not contain any studies involving human and animal subjects.

: There were no timeframes, official requirements, restrictions, or other external objective or subjective confounding factors in the performance of this study.