Vineis and Kriebel refer to a component cause completing the causal pie, and thus “initiating the disease process”. However, matters appear to be more complicated than this. Any complex disease process that leads to say cancer, asthma, or neuro-developmental diseases can involve some, or all of, at least seven stages: preparation; initiation; promotion; retardation; progression, disease onset; and the strengthening/weakening of the severity and /or prevalence of the disease. (“Preparation” includes the genetic, hormonal, immune, age, sex, etc. status of the host)
Each of these stages could be triggered by co-causal factors. Some factors may operate at several stages of the disease process, and via each of several causal chains. Common diseases, such as cancer, could have each of their stages triggered by different, or sometimes overlapping, co-causal factors operating via several causal chains. Identifying co-causal factors in such disease processes (and eventually understanding their different mechanisms of action) will be challenging.
The critical issue of “the timing of the exposure, or dose”, which the authors emphasise, further complicates analysis of disease causation. Timing is not only relevant to the developmental stage reached at the time of exposure/dose (e.g. day x being sensitive to that same exposure whilst, day x plus or minus 1 day may not be so), but also to each of the above stages reached in the disease process. The same exposure may have, or may not have, biologically relevant impact on the disease process, depending on the stage reached in the disease process at the time of exposure. The amount of exposure ( biologically available dose) at the critical times may also determine impact. Too little or too much exposure may have no biological impact compared to the critical level of exposure delivered at the “right” time and at the “right” stage of the disease process.
This view of multi-causality and of multi-stage disease processes also has implications for the conventional methods of identifying attributable fractions of disease, and their later use in estimating “environmental burden of disease”. The authors recommend “considerable humility” and “caution” when interpreting estimated attributable fractions because of the likelihood that the standard methods of estimating “statistical fractions” underestimate the “etiological fraction”, citing Greenland and Robins. However, the example which Vineis and Kriebel describe as “the most important” limitation of standard methods, ie. omitting to include the moving forward in time of the onset of a disease, is likely to be less important than the logical implication of inter-linked causal chains for attributable fractions of disease.
For example in Europe, the “environmental burden of disease” (EBOD) using conventional methods, is estimated to be around 5 %, whilst the oft cited Doll/Peto estimation of the occupational causes of cancer comes out at about the same fraction. Both are likely to be large underestimates. If each of several co-causal and inter-dependent factors in a multi-causal chain are necessary (but not sufficient) links in the chain, each co-causal factor is logically 100 % “responsible” for the disease, as Rothman has pointed out. The policy implications for disease prevention of this very different approach to EBOD are interesting.
Competing interests
none-except I am acknowledged at the end of their article as having provided encouragement and suggestions.
Response to David Gee: "Multi-causality,timing and the disease process"
David Kriebel, University of Massachusetts Lowell
3 August 2006
I thank David Gee for his comments on our paper. In large part, I think they amplify and supplement several of our points in useful ways. I agree with his emphasis on the complexity of disease processes, and with the sad reality that there is so much we do not understand about how environmental chemicals lead to chronic diseases.
I would like to comment more specifically on David Gee's final point. He wrote: "If each of several co-causal and inter-dependent factors in a multi-causal chain are necessary (but not sufficient) links in the chain, each co-causal factor is logically 100 % 'responsible' for the disease, as Rothman has pointed out." One can construct a hypothetical scenario in which there is only one necessary and sufficient causal chain for a disease. In this case, it is quite true that blocking any link in this chain will block the disease, and thus each link is indeed 100% responsible.
While this may be true in principle, the evidence we have to date, even if sadly limited, suggests that most important human diseases have multiple causal chains, not just one. In these cases, blocking a link on one chain will not eliminate the disease, but only the cases of the disease caused by that chain. Thus one would find an attributable fraction for this link of something less than 100%.
As noted by several authors, one would fully expect the sum of all attributable fractions for different causal agents for a disease to be greater than 100%, and perhaps much greater. But I doubt that there will be many examples where a single environmental chemical is 100% responsible for the causation of an important human disease. Would that it were so! To find such an agent would mean tremendous opportunity for preventing disease.
Competing interests
I have no competing interests
Multi causality in expression of environmental effects
Hans Sanderson, The US Soap and Detergent Association
11 August 2006
First of all I wish to thank the authors for an inspiring paper, and secondly the journal for asking me to provide a comment. I believe that many of the authors’ observations are also valid for the study of ecotoxicological and ecosystem effects - in other words Environmental Health, here is a few brief examples.
Toxicological modes of action (MOA): In the field of Ecotoxicology we are facing challenges and potential new paradigms as our science evolves and our understanding of chemicals effects increases. We used to describe toxic effects as a result of one chemical’s toxic mode of action (we typically operate with 1-2 dozen different MOAs). All non specific MOAs are defined as narcosis, which is simply an undefined intrusion of cell membrane integrity. With new technologies, such as toxicogenomics, better understanding of MOA is emerging, first of all documenting that MOAs really are plural (one chemical has multiple modes of action dependent on species, organ, concentration, and environment), the metaphor used by Vineis & Krieble (2006) that causal chains exists. The obvious example is that chemicals in mixtures can alter their effect concentration (synergy, antagonistic, additivity). Finally, the mounting evidence behind the concept of hormesis and non-linear low level dose-response relationships are also suggestions that our implementation of knowledge of MOA in risk assessments might need reconsideration.
Ecosystems: These are by nature integrated and communicating systems. The trophic cascading effects in ecosystems have been documented (e.g. Carpenter, 1985), as well as an appreciation of the importance of indirect effects (e.g. impairing the food source for higher trophic species) are of importance. An important co-variate factor in understanding and extrapolating environmental effects causal chain is climate change – which is often not included in chemical and environmental risk assessments.
Systems Biology: Covers the whole spectra from molecular to ecosystem level and suggests generating methods that allows better understanding and extrapolation of effects – appreciation of causal chains are obviously critically important in this process.
Statistics: Determining causality scientifically is both a quantitative and not least qualitative assessment. Bradford-Hill clearly pointed this out in his criteria more than 40 years ago. Before that Pearson & Nayman pointed this out in 1937 when they formulated their version of statistical null hypothesis testing (which is still widely used), statistical significance (or lack there of) is only a guide for informed decision not a test of truth. Vineis & Krieble (2006) correctly addresses the typical lack of statistical power associated to assessment of complex and variable (= realistic and relevant) situations, and the importance of not interpreting lack of statistical evidence as lack of effect.
In summary, the authors correctly indicate that we need more holistic analysis, which includes more variation and incomplete knowledge to allow us to appreciate the important causal chains and webs resulting in effects at the individual level but also on a population and ecosystem level. Their three recommendations are well taken. Moreover, we need appropriate statistical models and appropriate application of these for elucidating these relationships plus prudent application of qualitative knowledge and societal values to determine and evaluate the ethiologic web and facilitate sustainable decision making.
Competing interests
None declared
Important Limitations to Consider for Causal Models of Neurodevelopmental Disorders
Kathleen Flannery, Saint Anselm College
11 August 2006
Vineis and Kriebel's article identifies four important limitations to consider when conducting and evaluating epidemiological research. With respect to understanding the multiple etiologies of neurodevelopmental disorders these limitations present the following four questions:
First, how many steps and what probability estimates could be identified to create a causal model for neurodevelopmental disorders?
Second, how do we measure acquired susceptibility in children for neurodevelopmental disorders? Next, do our causal models for neurodevelopmental disorders provide us with consistent or inconsistent information when considered at the individual versus population-level? Finally, how would we identify components of sufficient causes relative to predicting neurodevelopmental disorders?
In addition to generating these questions, Vineis and Kriebel's article also provides us with causality models to consider when there are likely to be multiple etiologies for outcomes such as neurodevelopmental disorders. With the rising rate of diagnoses for autism and the well-established finding that males are more likely to be diagnosed with autism compared to females, we may want to consider these four limitations relative to creating and testing our next causal model for this disorder.
Competing interests
No competing interests.
Use of causal models in microbiology and epidemiology
Christian T. K.-H. Stadtländer, University of St. Thomas
22 August 2006
I appreciate the invitation by the editors-in-chief of the journal to provide a comment to the article by Vineis and Kriebel on "Causal models in epidemiology: Past inheritance and genetic future" [1]. The article is undoubtedly an important contribution to the literature on public health in general and to the literature on epidemiology and environmental health in particular. The authors identified limitations in epidemiological research and discussed different approaches to the elucidation of causality in the epidemiology of various diseases that arise from genes, environments, and the interplay between both [1].
I agree with Vineis and Kriebel that a major limitation in modern epidemiology is the identification and quantification of multiple component causes of disease (i.e., multicausality, either additive or multiplicative, in a causal chain). I believe that this limitation plays a role in many (if not in all) studies, including investigations on the etiology of cancer, cardiovascular illnesses, and infectious diseases. Being a microbiologist and epidemiologist, I would like to specifically comment on the use of causal models in microbiology and infectious disease epidemiology.
In microbiology, we typically use "Koch's postulates" as a series of steps or laboratory procedures that need to be followed in order to prove that a specific microorganism is a causal agent of a specific infectious disease [2]. One of the steps includes the establishment of an adequate animal model, which is unfortunately often a difficult task. In the absence of an animal model, causality has sometimes been proven either by accidental laboratory infections in humans (e.g., the accidental swallowing of cholera bacteria [2]) or by inoculation of a culture into human volunteers (e.g., the drinking of a suspension of Helicobacter pylori to prove that this bacterium is the etiologic agent of type-B gastritis [3]). In infectious disease epidemiology, we use a variety of causal models, many of which have been outlined by Vineis and Kriebel [1], including "Hill's nine criteria of disease causation" and "Rothman's causal pie model" [4]. In practice, this means that epidemiologists attempt to identify the microorganism(s) responsible for disease by determining their physical sources and biological relationships, their route of transmission and spread, the genes responsible for virulence, the identification of vaccine-relevant antigens, the determination of drug resistance, and measures to control current and prevent future outbreaks [5,6].
Despite the broad repertoire of causal models available to investigators, microbiologists and infectious disease epidemiologists have still sometimes problems to identify an etiologic agent. In May of 2006, I attended a symposium at the General Meeting of the American Society for Microbiology that was entitled: "Both sides now: Infectious disease outbreaks from the epi and micro perspectives." The presentations clearly demonstrated that both groups of investigators use different causal models, have different resource needs, encounter different challenges, and collect different data. The take-home lesson of this symposium was that microbiologists and epidemiologists should collaborate closer, communicate their causal models, and discuss their results in order to improve outbreak investigations.
Investigators of infectious diseases face numerous obstacles. There is often a lack of knowledge about environmental and genetic factors, an inability to cultivate the microorganism(s), and/or to reproduce disease in experimental models of infection. Furthermore, we do not know enough about the role of the human indigenous microbiota, the effect(s) of the commensal microbiota, the interplay between commensal(s) and pathogen(s), and the general distribution of microorganisms in human hosts over time and space.
It can be expected that future microbiological and epidemiological investigations of infectious diseases will become increasingly more complex, but also more exciting. I believe that the discussion about the use of causal models is important for the elucidation of many illnesses, including infectious diseases, because it helps us to build a much needed bridge between different scientific disciplines, such as microbiology and epidemiology, or infectious diseases and environmental health. This, in turn, will lead to improved public health approaches to disease detection, control, and prevention important in an era of emerging and re-emerging infectious diseases, risk of bioterrorism, and expected pandemics.
References
1. Vineis P, Kriebel D: Causal models in epidemiology: Past inheritance and genetic future.Environ Health 2006, 5:21.
2. Brock TD: Robert Koch: A Life in Medicine and Bacteriology. Washington, DC: American Society for Microbiology Press; 1999.
3. Mobley HLT, Mendz GL, Hazell SL: Helicobacter pylori: Physiology and Genetics. Washington, DC: American Society for Microbiology Press; 2001.
4. Rothman KJ: Epidemiology: An Introduction. New York: Oxford University Press; 2002.
5. Riley LW: Molecular Epidemiology of Infectious Diseases: Principles and Practices. Washington, DC: American Society for Microbiology Press; 2004.
6. Heyman DL: Control of Communicable Diseases Manual. Washington, DC: American Public Health Association; 2004.
Competing interests
The author declares that he has no competing interests.
A Note on Logic & Morality in Models of Causation
Sheldon W. Samuels, Ramazzini Institute for Occupational and Environmental Health Research
23 August 2006
Commentary by Sheldon W. Samuels
Causal Models in Epidemiology
P. Vineis and D. Kriebel
21 July 2006
The Vineis-Kriebel paper contributes to the dialogue on how to gather and analyze environmental epidemiological data. This comment provides a dialectic on issues raised by the authors, and abrades applications in any model of numbers used in logically and morally fallacious methods of risk management that, in teleological fashion, shape the numbers themselves by shaping fallacious methods of risk assessment in which they are generated.
Necessary and sufficient models of causation arose from analysis of relatively simple mechanical systems of motion heuristic for deterministic explorations in Aristotelian-to-Newtonian physical models. In these, a high level of certainty in the assignment of a cause and an effect could be achieved. The models had fruitful use in the more complex biology found in Bernard’s experimental medicine, Cannon’s investigation of homeostatic phenomena, Koch’s postulates in pathology, and in the monadic trait genetics of Mendel. It remains an heuristic canon in species of biomedical inquiry, especially among small populations.
Since Planck and the development of quantum mechanics, especially the advance beyond Hippocratic or simple, repeated observations associating environmental factors and disease to models informed by the current Darwin-Wright conceptual synthesis in the study of natural populations, a second, distinct species of method - stochastic statistical post facto and prospective methods - have been fruitfully introduced in environmental epidemiology. This is illustrated by the work on previously and currently exposed large human populations of Irving Selikoff et al. and in environmental toxicology by the work on prospective lifetime exposure of large rodent populations of Cesare Maltoni et al.
Logical issues of both species of model are grounded in inappropriate distortions of the necessary and sufficient model of causation, generating fallacies of exclusive particularity and simplism.
The fallacy of exclusive particularity occurs in the failure to recognize that a given theory enabling a true answer does not mean that another theory dealing with the same event must enable false answers. Both deterministic and stochastic models of causation investigating the same phenomena using different methods may be useful. Thus, Einstein, a determinist, accepted stochastic explanations as a way of dealing with limitations of available data and finite cognition. Both models may in fact exhibit logically valid and empirically warrantable separatism in another explanation of reality that Bohr called complementarity.
Failed attempts at unification of method, which Rothman appears to present, may result in a confusion of necessary and sufficient conditions, the fallacy of simplism, i.e., when we confuse conditions the absence of which always results in the absence of an effect, with phenomena that may or may not appear to be a condition of appearance or even always associated in time or place. Necessary causes are seldom observed in environmental health. The authors correctly focus on the plethora of sufficient conditions or factors, about which we have some knowledge, heuristically asserting the reality of “multiple causal chains”, reminiscent of Darwin’s tangled patch.
Deterministic and stochastic species of investigation exhibit cosmologic differences that may or may not reflect differences in moral outlook. There is a third set of models characterized by differences in moral outlook that, opportunistically, may or may not reflect differences in cosmology. These are cross breed models of causation used in epidemiological and toxicological experiments exposing small and large human populations conducted or suggested by the pesticide and other industries for product testing and tort or regulatory defense, or conducted or authorized by environmental control agencies through the titration and re-titration of exposure limits and pollutant selection by kind and concentration using risk assessment-management processes tied to cost-benefit and similar economic analyses. Inherently oblivious to the unjust distribution of risk by occupational or socio-economic caste, these practices are found in implementing Europe’s Precautionary Principle. In America, at least during the past 3 1/2 decades, these practices are found in the regulatory analyses of the President’s Office of Management and Budget, and in the facilitation of the marketing of “pollution credits” by the Environmental Protection Agency.
Full discussion of the third set of models of etiology and their moral-cosmologic traits is best left to speculations in the Faustian manner of what environmental health practice might be like if mankind achieves a higher level of civilization. Here we confine our attention to illustrating why the provenance of numbers is vital in judging their value in the rationalization of existing tribal mores.
David Gee’s comment on the Doll/Peto estimation, which was an estimation of the fraction of cancer attributable to occupation, raises a case in point. The estimation was actually a rebuttal to a rough estimate of the range of the portion of cancer attributable to the work environment by competent NIOSH, NCI and NIEHS scientists, calculated to support a carcinogen regulatory policy which was bitterly attacked by industry advocates in Congress. The politicians commissioned the now-defunct Congressional Office of Technology Assessment to undermine the estimate with a "scientific" risk assessment of occupational cancer. Richard Doll, then Warden of an industry-endowed portion of Oxford University's establishment, was selected to do the study. Doll’s estimate, as expected, was a fraction of the government estimate. Selikoff, when asked to intervene in the squabble by labor advocates in Congress, the agencies and labor itself, refused to develop his own number in the face of what he considered to be inadequate data and methods. He believed that we needed to be less concerned about the differences in the numbers than about more serious moral and intellectual mistakes in Doll’s method.
Selikoff called the 'mistakes' in Doll's assessment mistakes of science arising from the external circumstances of Doll’s social perspective. Doll selectively rejected what he considered to be uncertain data, preferring to use uncertain personal factor data that burdened the worker and to exclude uncertain occupational data that burdened the employer. Thus, genetics and diet were arbitrarily given greater weight than occupation. The mistake was in claiming that this judgment was "scientific" or "objective" rather than social or moral and subjective. Doll also arbitrarily excluded occupational data taken from some death certificates in the United States, challenging their certainty to justify their exclusion. Selikoff’s response was to re-examine the records of his asbestos worker studies. The result was a series of four papers, summarized in the last, which was also his last paper. These papers drowned Doll's objections with hard data. Thus, Gee's comment that Doll's number is "likely to be a large underestimate" is off target. Precisely, the number is the shaped product of a shaped method. The shaping, not the number, is worthy of serious discussion in environmental risk management.
Competing interests
none
Asthma: from multi-causality models to social determination?
Maria Regina Cardoso, Faculty of Public Health - University of São Paulo
29 August 2006
Authors: Agnes Soares da Silva & Maria Regina Cardoso
Reading the inspiring article of Vineis and Kriebel, and the comments by David Gee with the respective response of David Kriebel, made us think of going a little further in the already complex model proposed, using asthma as a paradigm for chronic diseases.
If we consider the author’s recommendation of treating causation models separately for individuals and population, our comprehension of the whole process would be segmented because individuals and population would be seen as entities completely independent from each other.
To understand the individual process, we would have to find out the individual inheritance, as well as the different genes involved and the diverse environmental exposures that could lead to the expression of asthma during a life course, and explain why some individuals have or will have asthma and others do not have or will not develop asthma. For the population process, we would have to find out why some populations or social groups have higher prevalences of asthma than others and why they have changed along the time.
Nevertheless, looking at the two processes as independent, we cannot explain the asthma epidemics, therefore we cannot propose any policy action that could eventually intervene in the causation process. The problem is that the high prevalence of atopy among children in industrialized countries and urban populations in non-industrialized countries indicates that atopic genetic traits are extremely common. The atopic phenotype has emerged over the last few decades to become in present time one of the greatest public health concerns, strongly suggesting that ‘environmental’ factors in the broadest sense, thus including socio-economic factors, play a decisive role in this process.
Thus, the complex interactions mentioned by the authors should be thought not only as related to biological and to physical environmental exposures in a linear multi-causality model, but as complex interactions in different and communicating layers of distal and proximal mediators of the disease such as: biological, social, economic, cultural and historical background; psychosocial profile; access to means of communication; social relations; access to social and health services etc. The logical framework in which these relationships occur is given by where they occur in time and geographical location.
Although we know that the context in which asthma occurs determines the individual expression and the prevalence of the disease in different groups, we still do not know how it happens. Following biological processes in individuals, one expects that they should be similar independent of where the individuals are located. However, once we follow environmental processes, the pathways are not necessarily the same.
Accordingly, we should be more revolutionary in our proposition and think about a theory that could account for all this complexity, giving a logical sense to the apparent chaos of adding so many layers and vectors to our limited statistical models.
Competing interests
none
Causal models in epidemiology and the notion of high risk groups
Pietro Comba, Istituto Superiore di Sanità
2 October 2006
The paper by Vineis and Kriebel provides a valuable overview of causal assessment in the health sciences in general, and in epidemiological research in particular. The authors focus on the notion of disentangling the contributions of multiple risk factors in chronic disease, and stress the role of interaction, not simply as a second-order phenomenon to be detected after taking into account main effects, but rather as a fundamental biological phenomenon underlying a wide range of pathogenetic mechanisms.
The prevailing approach adopted by those who have dealt with these issues has been the conduction of large population based case-sontrol studies, that ensure the possibility to investigate the etiological role of various risk factors of the disease of interest. The major limitation of case-control studies is that they may turn out not to be informative if the prevalence of a given exposure in the study population is too low.
A possible improvement in the design of epidemiologic studies is based on sections of the general population defined on a priori knowledge that the prevalence of the exposure of interest be sufficiently elevated. The notion of “high risk groups”, originally developed in the domain of occupational cancer research [1], near toxic waste dumping sites [2] is now applied to the study of a wide range of environmental agents [3]. This development requires an integrated effort of epidemiologists and environmental experts, in order to ensure a valid knowledge of the spatial distribution of pollutant levels both at national level and at small area scale.
This approach may turn out to be beneficial in terms of contrastability between exposure categories. Statistical power may be hampered by obvious constraints on the total number of cases, but this aspect may be counterbalanced by the increase in the prevalence of exposure.
In more general terms, the development of environmental epidemiology studies based on the notion of high risk groups may contribute to the production of knowledge on specific causal webs, and also bring to light exposure circumstances that warrant prompt remedial interventions. The strong connection between epidemiologic research and public health action is thus once more confirmed.
References
1. Terracini B, Segnan N. Identification of high-risk groups. Epidemiol Prev 1997; 0: 17-23 [in italian]
2. Sexton K, Olden K, Johnson B. “Environmental Justice”: the central role of research in establishing a credible scientific foundation for informed decision making. Toxicol Ind Health 1993; 9(5):685-727
3. Fazzo L, Comba P. The role of high risk groups in environmental health research. Ann Ist Super Sanità 2004; 40 (4):417-426 [in italian]
Environmental Health
Multi-causality,timing and the disease process
28 July 2006
Vineis and Kriebel refer to a component cause completing the causal pie, and thus “initiating the disease process”. However, matters appear to be more complicated than this. Any complex disease process that leads to say cancer, asthma, or neuro-developmental diseases can involve some, or all of, at least seven stages: preparation; initiation; promotion; retardation; progression, disease onset; and the strengthening/weakening of the severity and /or prevalence of the disease. (“Preparation” includes the genetic, hormonal, immune, age, sex, etc. status of the host)
Each of these stages could be triggered by co-causal factors. Some factors may operate at several stages of the disease process, and via each of several causal chains. Common diseases, such as cancer, could have each of their stages triggered by different, or sometimes overlapping, co-causal factors operating via several causal chains. Identifying co-causal factors in such disease processes (and eventually understanding their different mechanisms of action) will be challenging.
The critical issue of “the timing of the exposure, or dose”, which the authors emphasise, further complicates analysis of disease causation. Timing is not only relevant to the developmental stage reached at the time of exposure/dose (e.g. day x being sensitive to that same exposure whilst, day x plus or minus 1 day may not be so), but also to each of the above stages reached in the disease process. The same exposure may have, or may not have, biologically relevant impact on the disease process, depending on the stage reached in the disease process at the time of exposure. The amount of exposure ( biologically available dose) at the critical times may also determine impact. Too little or too much exposure may have no biological impact compared to the critical level of exposure delivered at the “right” time and at the “right” stage of the disease process.
This view of multi-causality and of multi-stage disease processes also has implications for the conventional methods of identifying attributable fractions of disease, and their later use in estimating “environmental burden of disease”. The authors recommend “considerable humility” and “caution” when interpreting estimated attributable fractions because of the likelihood that the standard methods of estimating “statistical fractions” underestimate the “etiological fraction”, citing Greenland and Robins. However, the example which Vineis and Kriebel describe as “the most important” limitation of standard methods, ie. omitting to include the moving forward in time of the onset of a disease, is likely to be less important than the logical implication of inter-linked causal chains for attributable fractions of disease.
For example in Europe, the “environmental burden of disease” (EBOD) using conventional methods, is estimated to be around 5 %, whilst the oft cited Doll/Peto estimation of the occupational causes of cancer comes out at about the same fraction. Both are likely to be large underestimates. If each of several co-causal and inter-dependent factors in a multi-causal chain are necessary (but not sufficient) links in the chain, each co-causal factor is logically 100 % “responsible” for the disease, as Rothman has pointed out. The policy implications for disease prevention of this very different approach to EBOD are interesting.
Competing interests
none-except I am acknowledged at the end of their article as having provided encouragement and suggestions.
Response to David Gee: "Multi-causality,timing and the disease process"
3 August 2006
I thank David Gee for his comments on our paper. In large part, I think they amplify and supplement several of our points in useful ways. I agree with his emphasis on the complexity of disease processes, and with the sad reality that there is so much we do not understand about how environmental chemicals lead to chronic diseases.
I would like to comment more specifically on David Gee's final point. He wrote: "If each of several co-causal and inter-dependent factors in a multi-causal chain are necessary (but not sufficient) links in the chain, each co-causal factor is logically 100 % 'responsible' for the disease, as Rothman has pointed out." One can construct a hypothetical scenario in which there is only one necessary and sufficient causal chain for a disease. In this case, it is quite true that blocking any link in this chain will block the disease, and thus each link is indeed 100% responsible.
While this may be true in principle, the evidence we have to date, even if sadly limited, suggests that most important human diseases have multiple causal chains, not just one. In these cases, blocking a link on one chain will not eliminate the disease, but only the cases of the disease caused by that chain. Thus one would find an attributable fraction for this link of something less than 100%.
As noted by several authors, one would fully expect the sum of all attributable fractions for different causal agents for a disease to be greater than 100%, and perhaps much greater. But I doubt that there will be many examples where a single environmental chemical is 100% responsible for the causation of an important human disease. Would that it were so! To find such an agent would mean tremendous opportunity for preventing disease.
Competing interests
I have no competing interests
Multi causality in expression of environmental effects
11 August 2006
First of all I wish to thank the authors for an inspiring paper, and secondly the journal for asking me to provide a comment. I believe that many of the authors’ observations are also valid for the study of ecotoxicological and ecosystem effects - in other words Environmental Health, here is a few brief examples.
Toxicological modes of action (MOA): In the field of Ecotoxicology we are facing challenges and potential new paradigms as our science evolves and our understanding of chemicals effects increases. We used to describe toxic effects as a result of one chemical’s toxic mode of action (we typically operate with 1-2 dozen different MOAs). All non specific MOAs are defined as narcosis, which is simply an undefined intrusion of cell membrane integrity. With new technologies, such as toxicogenomics, better understanding of MOA is emerging, first of all documenting that MOAs really are plural (one chemical has multiple modes of action dependent on species, organ, concentration, and environment), the metaphor used by Vineis & Krieble (2006) that causal chains exists. The obvious example is that chemicals in mixtures can alter their effect concentration (synergy, antagonistic, additivity). Finally, the mounting evidence behind the concept of hormesis and non-linear low level dose-response relationships are also suggestions that our implementation of knowledge of MOA in risk assessments might need reconsideration.
Ecosystems: These are by nature integrated and communicating systems. The trophic cascading effects in ecosystems have been documented (e.g. Carpenter, 1985), as well as an appreciation of the importance of indirect effects (e.g. impairing the food source for higher trophic species) are of importance. An important co-variate factor in understanding and extrapolating environmental effects causal chain is climate change – which is often not included in chemical and environmental risk assessments.
Systems Biology: Covers the whole spectra from molecular to ecosystem level and suggests generating methods that allows better understanding and extrapolation of effects – appreciation of causal chains are obviously critically important in this process.
Statistics: Determining causality scientifically is both a quantitative and not least qualitative assessment. Bradford-Hill clearly pointed this out in his criteria more than 40 years ago. Before that Pearson & Nayman pointed this out in 1937 when they formulated their version of statistical null hypothesis testing (which is still widely used), statistical significance (or lack there of) is only a guide for informed decision not a test of truth. Vineis & Krieble (2006) correctly addresses the typical lack of statistical power associated to assessment of complex and variable (= realistic and relevant) situations, and the importance of not interpreting lack of statistical evidence as lack of effect.
In summary, the authors correctly indicate that we need more holistic analysis, which includes more variation and incomplete knowledge to allow us to appreciate the important causal chains and webs resulting in effects at the individual level but also on a population and ecosystem level. Their three recommendations are well taken. Moreover, we need appropriate statistical models and appropriate application of these for elucidating these relationships plus prudent application of qualitative knowledge and societal values to determine and evaluate the ethiologic web and facilitate sustainable decision making.
Competing interests
None declared
Important Limitations to Consider for Causal Models of Neurodevelopmental Disorders
11 August 2006
Vineis and Kriebel's article identifies four important limitations to consider when conducting and evaluating epidemiological research. With respect to understanding the multiple etiologies of neurodevelopmental disorders these limitations present the following four questions:
First, how many steps and what probability estimates could be identified to create a causal model for neurodevelopmental disorders?
Second, how do we measure acquired susceptibility in children for neurodevelopmental disorders? Next, do our causal models for neurodevelopmental disorders provide us with consistent or inconsistent information when considered at the individual versus population-level? Finally, how would we identify components of sufficient causes relative to predicting neurodevelopmental disorders?
In addition to generating these questions, Vineis and Kriebel's article also provides us with causality models to consider when there are likely to be multiple etiologies for outcomes such as neurodevelopmental disorders. With the rising rate of diagnoses for autism and the well-established finding that males are more likely to be diagnosed with autism compared to females, we may want to consider these four limitations relative to creating and testing our next causal model for this disorder.
Competing interests
No competing interests.
Use of causal models in microbiology and epidemiology
22 August 2006
I appreciate the invitation by the editors-in-chief of the journal to provide a comment to the article by Vineis and Kriebel on "Causal models in epidemiology: Past inheritance and genetic future" [1]. The article is undoubtedly an important contribution to the literature on public health in general and to the literature on epidemiology and environmental health in particular. The authors identified limitations in epidemiological research and discussed different approaches to the elucidation of causality in the epidemiology of various diseases that arise from genes, environments, and the interplay between both [1].
I agree with Vineis and Kriebel that a major limitation in modern epidemiology is the identification and quantification of multiple component causes of disease (i.e., multicausality, either additive or multiplicative, in a causal chain). I believe that this limitation plays a role in many (if not in all) studies, including investigations on the etiology of cancer, cardiovascular illnesses, and infectious diseases. Being a microbiologist and epidemiologist, I would like to specifically comment on the use of causal models in microbiology and infectious disease epidemiology.
In microbiology, we typically use "Koch's postulates" as a series of steps or laboratory procedures that need to be followed in order to prove that a specific microorganism is a causal agent of a specific infectious disease [2]. One of the steps includes the establishment of an adequate animal model, which is unfortunately often a difficult task. In the absence of an animal model, causality has sometimes been proven either by accidental laboratory infections in humans (e.g., the accidental swallowing of cholera bacteria [2]) or by inoculation of a culture into human volunteers (e.g., the drinking of a suspension of Helicobacter pylori to prove that this bacterium is the etiologic agent of type-B gastritis [3]). In infectious disease epidemiology, we use a variety of causal models, many of which have been outlined by Vineis and Kriebel [1], including "Hill's nine criteria of disease causation" and "Rothman's causal pie model" [4]. In practice, this means that epidemiologists attempt to identify the microorganism(s) responsible for disease by determining their physical sources and biological relationships, their route of transmission and spread, the genes responsible for virulence, the identification of vaccine-relevant antigens, the determination of drug resistance, and measures to control current and prevent future outbreaks [5,6].
Despite the broad repertoire of causal models available to investigators, microbiologists and infectious disease epidemiologists have still sometimes problems to identify an etiologic agent. In May of 2006, I attended a symposium at the General Meeting of the American Society for Microbiology that was entitled: "Both sides now: Infectious disease outbreaks from the epi and micro perspectives." The presentations clearly demonstrated that both groups of investigators use different causal models, have different resource needs, encounter different challenges, and collect different data. The take-home lesson of this symposium was that microbiologists and epidemiologists should collaborate closer, communicate their causal models, and discuss their results in order to improve outbreak investigations.
Investigators of infectious diseases face numerous obstacles. There is often a lack of knowledge about environmental and genetic factors, an inability to cultivate the microorganism(s), and/or to reproduce disease in experimental models of infection. Furthermore, we do not know enough about the role of the human indigenous microbiota, the effect(s) of the commensal microbiota, the interplay between commensal(s) and pathogen(s), and the general distribution of microorganisms in human hosts over time and space.
It can be expected that future microbiological and epidemiological investigations of infectious diseases will become increasingly more complex, but also more exciting. I believe that the discussion about the use of causal models is important for the elucidation of many illnesses, including infectious diseases, because it helps us to build a much needed bridge between different scientific disciplines, such as microbiology and epidemiology, or infectious diseases and environmental health. This, in turn, will lead to improved public health approaches to disease detection, control, and prevention important in an era of emerging and re-emerging infectious diseases, risk of bioterrorism, and expected pandemics.
References
1. Vineis P, Kriebel D: Causal models in epidemiology: Past inheritance and genetic future. Environ Health 2006, 5:21.
2. Brock TD: Robert Koch: A Life in Medicine and Bacteriology. Washington, DC: American Society for Microbiology Press; 1999.
3. Mobley HLT, Mendz GL, Hazell SL: Helicobacter pylori: Physiology and Genetics. Washington, DC: American Society for Microbiology Press; 2001.
4. Rothman KJ: Epidemiology: An Introduction. New York: Oxford University Press; 2002.
5. Riley LW: Molecular Epidemiology of Infectious Diseases: Principles and Practices. Washington, DC: American Society for Microbiology Press; 2004.
6. Heyman DL: Control of Communicable Diseases Manual. Washington, DC: American Public Health Association; 2004.
Competing interests
The author declares that he has no competing interests.
A Note on Logic & Morality in Models of Causation
23 August 2006
Commentary by Sheldon W. Samuels
Causal Models in Epidemiology
P. Vineis and D. Kriebel
21 July 2006
The Vineis-Kriebel paper contributes to the dialogue on how to gather and analyze environmental epidemiological data. This comment provides a dialectic on issues raised by the authors, and abrades applications in any model of numbers used in logically and morally fallacious methods of risk management that, in teleological fashion, shape the numbers themselves by shaping fallacious methods of risk assessment in which they are generated.
Necessary and sufficient models of causation arose from analysis of relatively simple mechanical systems of motion heuristic for deterministic explorations in Aristotelian-to-Newtonian physical models. In these, a high level of certainty in the assignment of a cause and an effect could be achieved. The models had fruitful use in the more complex biology found in Bernard’s experimental medicine, Cannon’s investigation of homeostatic phenomena, Koch’s postulates in pathology, and in the monadic trait genetics of Mendel. It remains an heuristic canon in species of biomedical inquiry, especially among small populations.
Since Planck and the development of quantum mechanics, especially the advance beyond Hippocratic or simple, repeated observations associating environmental factors and disease to models informed by the current Darwin-Wright conceptual synthesis in the study of natural populations, a second, distinct species of method - stochastic statistical post facto and prospective methods - have been fruitfully introduced in environmental epidemiology. This is illustrated by the work on previously and currently exposed large human populations of Irving Selikoff et al. and in environmental toxicology by the work on prospective lifetime exposure of large rodent populations of Cesare Maltoni et al.
Logical issues of both species of model are grounded in inappropriate distortions of the necessary and sufficient model of causation, generating fallacies of exclusive particularity and simplism.
The fallacy of exclusive particularity occurs in the failure to recognize that a given theory enabling a true answer does not mean that another theory dealing with the same event must enable false answers. Both deterministic and stochastic models of causation investigating the same phenomena using different methods may be useful. Thus, Einstein, a determinist, accepted stochastic explanations as a way of dealing with limitations of available data and finite cognition. Both models may in fact exhibit logically valid and empirically warrantable separatism in another explanation of reality that Bohr called complementarity.
Failed attempts at unification of method, which Rothman appears to present, may result in a confusion of necessary and sufficient conditions, the fallacy of simplism, i.e., when we confuse conditions the absence of which always results in the absence of an effect, with phenomena that may or may not appear to be a condition of appearance or even always associated in time or place. Necessary causes are seldom observed in environmental health. The authors correctly focus on the plethora of sufficient conditions or factors, about which we have some knowledge, heuristically asserting the reality of “multiple causal chains”, reminiscent of Darwin’s tangled patch.
Deterministic and stochastic species of investigation exhibit cosmologic differences that may or may not reflect differences in moral outlook. There is a third set of models characterized by differences in moral outlook that, opportunistically, may or may not reflect differences in cosmology. These are cross breed models of causation used in epidemiological and toxicological experiments exposing small and large human populations conducted or suggested by the pesticide and other industries for product testing and tort or regulatory defense, or conducted or authorized by environmental control agencies through the titration and re-titration of exposure limits and pollutant selection by kind and concentration using risk assessment-management processes tied to cost-benefit and similar economic analyses. Inherently oblivious to the unjust distribution of risk by occupational or socio-economic caste, these practices are found in implementing Europe’s Precautionary Principle. In America, at least during the past 3 1/2 decades, these practices are found in the regulatory analyses of the President’s Office of Management and Budget, and in the facilitation of the marketing of “pollution credits” by the Environmental Protection Agency.
Full discussion of the third set of models of etiology and their moral-cosmologic traits is best left to speculations in the Faustian manner of what environmental health practice might be like if mankind achieves a higher level of civilization. Here we confine our attention to illustrating why the provenance of numbers is vital in judging their value in the rationalization of existing tribal mores.
David Gee’s comment on the Doll/Peto estimation, which was an estimation of the fraction of cancer attributable to occupation, raises a case in point. The estimation was actually a rebuttal to a rough estimate of the range of the portion of cancer attributable to the work environment by competent NIOSH, NCI and NIEHS scientists, calculated to support a carcinogen regulatory policy which was bitterly attacked by industry advocates in Congress. The politicians commissioned the now-defunct Congressional Office of Technology Assessment to undermine the estimate with a "scientific" risk assessment of occupational cancer. Richard Doll, then Warden of an industry-endowed portion of Oxford University's establishment, was selected to do the study. Doll’s estimate, as expected, was a fraction of the government estimate. Selikoff, when asked to intervene in the squabble by labor advocates in Congress, the agencies and labor itself, refused to develop his own number in the face of what he considered to be inadequate data and methods. He believed that we needed to be less concerned about the differences in the numbers than about more serious moral and intellectual mistakes in Doll’s method.
Selikoff called the 'mistakes' in Doll's assessment mistakes of science arising from the external circumstances of Doll’s social perspective. Doll selectively rejected what he considered to be uncertain data, preferring to use uncertain personal factor data that burdened the worker and to exclude uncertain occupational data that burdened the employer. Thus, genetics and diet were arbitrarily given greater weight than occupation. The mistake was in claiming that this judgment was "scientific" or "objective" rather than social or moral and subjective. Doll also arbitrarily excluded occupational data taken from some death certificates in the United States, challenging their certainty to justify their exclusion. Selikoff’s response was to re-examine the records of his asbestos worker studies. The result was a series of four papers, summarized in the last, which was also his last paper. These papers drowned Doll's objections with hard data. Thus, Gee's comment that Doll's number is "likely to be a large underestimate" is off target. Precisely, the number is the shaped product of a shaped method. The shaping, not the number, is worthy of serious discussion in environmental risk management.
Competing interests
none
Asthma: from multi-causality models to social determination?
29 August 2006
Authors: Agnes Soares da Silva & Maria Regina Cardoso
Reading the inspiring article of Vineis and Kriebel, and the comments by David Gee with the respective response of David Kriebel, made us think of going a little further in the already complex model proposed, using asthma as a paradigm for chronic diseases.
If we consider the author’s recommendation of treating causation models separately for individuals and population, our comprehension of the whole process would be segmented because individuals and population would be seen as entities completely independent from each other.
To understand the individual process, we would have to find out the individual inheritance, as well as the different genes involved and the diverse environmental exposures that could lead to the expression of asthma during a life course, and explain why some individuals have or will have asthma and others do not have or will not develop asthma. For the population process, we would have to find out why some populations or social groups have higher prevalences of asthma than others and why they have changed along the time.
Nevertheless, looking at the two processes as independent, we cannot explain the asthma epidemics, therefore we cannot propose any policy action that could eventually intervene in the causation process. The problem is that the high prevalence of atopy among children in industrialized countries and urban populations in non-industrialized countries indicates that atopic genetic traits are extremely common. The atopic phenotype has emerged over the last few decades to become in present time one of the greatest public health concerns, strongly suggesting that ‘environmental’ factors in the broadest sense, thus including socio-economic factors, play a decisive role in this process.
Thus, the complex interactions mentioned by the authors should be thought not only as related to biological and to physical environmental exposures in a linear multi-causality model, but as complex interactions in different and communicating layers of distal and proximal mediators of the disease such as: biological, social, economic, cultural and historical background; psychosocial profile; access to means of communication; social relations; access to social and health services etc. The logical framework in which these relationships occur is given by where they occur in time and geographical location.
Although we know that the context in which asthma occurs determines the individual expression and the prevalence of the disease in different groups, we still do not know how it happens. Following biological processes in individuals, one expects that they should be similar independent of where the individuals are located. However, once we follow environmental processes, the pathways are not necessarily the same.
Accordingly, we should be more revolutionary in our proposition and think about a theory that could account for all this complexity, giving a logical sense to the apparent chaos of adding so many layers and vectors to our limited statistical models.
Competing interests
none
Causal models in epidemiology and the notion of high risk groups
2 October 2006
The paper by Vineis and Kriebel provides a valuable overview of causal assessment in the health sciences in general, and in epidemiological research in particular. The authors focus on the notion of disentangling the contributions of multiple risk factors in chronic disease, and stress the role of interaction, not simply as a second-order phenomenon to be detected after taking into account main effects, but rather as a fundamental biological phenomenon underlying a wide range of pathogenetic mechanisms.
The prevailing approach adopted by those who have dealt with these issues has been the conduction of large population based case-sontrol studies, that ensure the possibility to investigate the etiological role of various risk factors of the disease of interest. The major limitation of case-control studies is that they may turn out not to be informative if the prevalence of a given exposure in the study population is too low.
A possible improvement in the design of epidemiologic studies is based on sections of the general population defined on a priori knowledge that the prevalence of the exposure of interest be sufficiently elevated. The notion of “high risk groups”, originally developed in the domain of occupational cancer research [1], near toxic waste dumping sites [2] is now applied to the study of a wide range of environmental agents [3]. This development requires an integrated effort of epidemiologists and environmental experts, in order to ensure a valid knowledge of the spatial distribution of pollutant levels both at national level and at small area scale.
This approach may turn out to be beneficial in terms of contrastability between exposure categories. Statistical power may be hampered by obvious constraints on the total number of cases, but this aspect may be counterbalanced by the increase in the prevalence of exposure.
In more general terms, the development of environmental epidemiology studies based on the notion of high risk groups may contribute to the production of knowledge on specific causal webs, and also bring to light exposure circumstances that warrant prompt remedial interventions. The strong connection between epidemiologic research and public health action is thus once more confirmed.
References
1. Terracini B, Segnan N. Identification of high-risk groups. Epidemiol Prev 1997; 0: 17-23 [in italian]
2. Sexton K, Olden K, Johnson B. “Environmental Justice”: the central role of research in establishing a credible scientific foundation for informed decision making. Toxicol Ind Health 1993; 9(5):685-727
3. Fazzo L, Comba P. The role of high risk groups in environmental health research. Ann Ist Super Sanità 2004; 40 (4):417-426 [in italian]
Competing interests
None declared