Hills Criteria of Causation
and disease: association or causation? by Sir Austin Bradford Hill on establishing 2 National Centre for Epidemiology and Population Health, The Australian National . is the only necessary criterion for a causal relationship between. PDF | In , Sir Austin Bradford Hill published nine “viewpoints” to help criteria. Criteria 1: strength of association. Hill's ﬁrst criterion for causation is strength of the associ-. ation. . ated with these two studies underscores how the use of. The correlation is evidence of an association between motor vehicle use and lung Richard Doll and Austin Bradford Hill (shown on the right).
A little exposure should result in a little effect, a large exposure should cause a large effect. Certainly well known to anyone who drinks alcohol; I suppose all homeopaths must be teetotallers. The comparison would be weakened, though not necessarily destroyed, if it depended upon, say, a much heavier death rate in light smokers and a lower rate in heavier smokers.
We should then need to envisage some much more complex relationship to satisfy the cause and effect hypothesis.
Causation in epidemiology: association and causation | Health Knowledge
The clear dose-response curve admits of a simple explanation and obviously puts the case in a clearer light. I suppose that a chiropractor could say your spine is partly unsubluxed as a result of half a spinal manipulation or an acupuncturist saying, your chi is partly unblocked as I used too few needles.
I assume a reader will comment on the validity of this observation. The effect must have biologic plausibility. I would take it a slightly differently: What is biologically plausible depends upon the biological knowledge of the day. I know that there are more things in heaven and earth than are dreamt of in my philosophy.
But you have to prove it to me. In short, the association we observe may be one new to science or medicine and we must not dismiss it too light-heartedly as just too odd. As Sherlock Holmes advised Dr. Sometimes what remains is, however improbable, still nonsense. I know cholera, for example, from the level of the effect of the toxin on cellular receptors to the world wide changes in potable water that lead to the spread of disease and much in between. There is a coherence of understanding of the disease.
The Bradford Hill considerations on causality: a counterfactual perspective
Homeopathy is, above all, totally incoherent. Written inbefore the massive increase in biomedical research funding, experiments were not as vital in understanding diseases and treatments as they are today.
Not that it ever matters to the practitioners. I think about how my practice has changed over the last 25 years: Consider the 44, articles in Pubmed in infectious disease that are published last year. I wonder how much chiropractic articles on or acupuncture articles in or naturopathy 19 articles in or homeopathy articles in practice changed as a result of published studies. It cannot be all that hard to keep up and, so, change accordingly. If one virus, for example, can cause a disease, then it is reasonable to suggest that a second virus could be responsible for a similar disease.
Analogy is not the same as metaphor: He clearly states these are guidelines, and not to be followed blindly. None of my nine viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as a 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? The importance of considering all the data, the preponderance of information, in deciding cause and effect. Hill is also not enthusiastic about statistics, the dreaded p-value: No formal tests of significance can answer those questions.
Such tests can, and should, remind us of the effects that the play of chance can create, and they will instruct us in the likely magnitude of those effects. Perhaps too often generalities were based upon two men and a laboratory dog while the treatment of choice was deducted from a difference between two bedfuls of patients and might easily have no true meaning.
Statistically significant nonsense is still nonsense. The article puts into perspective the ongoing problem of the meta-analysis. I always say that the meta-analysis is good for a general understanding of an intervention but rarely provides definitive answers. As a result, I think meta-analyses are great if they support your prior beliefs and can be safely ignored if they contradict them. Far be it from me to suggest that the Cochrane reviews may be wanting as they are often considered the be all end all of analysis, but their reviews in the few areas I know a little about always leave me unsatisfied.
Dr Hill ends with a discussion on the importance of then using the information of when association merges into causation and to consider at what point we need to act on the information. Stopping a nausea medication because it may cause birth defects has a different impact than stopping the burning of fuels in the home as cause of lung disease.
Does this indicate that motor vehicle use is a cause of lung cancer mortality? Yes, because there was a "strong" correlation, indicating a strong association. No, because the data on the number of motor vehicles being used was probably grossly inaccurate.
No, because the sample size is unknown. No, because these data are from an ecological study, and it is dangerous to conclude that there is a causal relationship from these studies because of the ecological fallacy and the inability to control for many possible confounding factors.
Ina German study reported an association between smoking and lung cancer. Then, in the s and s there was a succession of studies that sought to examine the cause of the epidemic of lung cancer that was claiming more and more lives. Again, the absence of such knowledge would not be indicative of a non-causal explanation.
The difference in Hill's definitions of plausibility and coherence appears to be subtle [[ 7 ], p. Whereas plausibility is worded positively an association that should be in line with substantive knowledgecoherence is verbalised negatively an association that should not conflict with substantive knowledge. Rothman and Greenland [[ 7 ], p. Susser [ 11 ] has tried to retain this consideration by defining different subclasses of coherence depending on where knowledge comes from.
A subtle difference between coherence and plausibility is that plausibility asks: Experiment Causation is more likely if evidence is based on randomised experiments Hill [ 5 ] argued that a causal interpretation of an association from a non-experimental study was supported if a randomised prevention derived from the association confirmed the finding.
For instance, after finding that certain events were related to the number of people smoking, one might forbid smoking to see whether the frequency of the events decrease consecutively. To Rothman and Greenland [[ 7 ], p.
Human experiments were hardly available in epidemiology, and results from animal experiments could not easily be applied to human beings. To Susser [ 11 ], Hill's examples suggested that he meant intervention and active change rather than research design. Both Susser [ 11 ] and Rothman and Greenland [[ 7 ], p. This is motivated by the possibility that a change following a modest intervention could result from the circumstances of a treatment rather than from the treatment itself.
One might add that the Cox and Wermuth consideration requires a modest intervention to be precluded from having a strong influence — an assumption that is certainly context-dependent to a high degree.
In terms of counterfactual causality, the distinction between massive and modest interventions is irrelevant, because a causal effect is only defined for a fixed index and a fixed reference condition.
The Bradford Hill considerations on causality: a counterfactual perspective
Hence, if interpreted in terms of strength of intervention, this is again not a consideration on a specific causal difference, but rather a consideration on a comprehensive causal theory as the one on biological gradient.
Such a theory is required in order to decide what is a modest and what is a strong intervention. If the consideration on experiment is interpreted in terms of avoiding some biases in estimating a specific causal effect by conducting an RCT, it should be generalised as follows: Bias is reduced either by using a study design that avoids major biases or by properly correcting for bias. Clearly, avoiding bias is preferable to correcting for it, but it is often impossible to avoid some biases.
As already mentioned, in RCTs with perfect compliance, confounding cannot occur although confounders might be distributed unequally by chance and there is no measurement error in the exposure. However, bias due to measurement error could still occur in the outcome, and there may be bias due to selection, missing data, etc. Thus, Hill's original formulation [ 5 ] covered only one or two among a variety of possible biases.
Instead, two more general question arise: And, if the optimal study design is not possible, how can bias be accurately corrected for? As in the consideration on strength, this can be summarised by: A causal effect is more likely if, after bias adjustment, the interval estimate excludes the null value, and it is even more likely if the lower boundary is far from the null value.
If adjustment is done properly, systematic error in the corrected interval estimate decreases if the knowledge about biases increases. As a consequence, one can hardly ever demonstrate a causal effect if biases are poorly understood — this is the case even in large samples, because the associated systematic error in the results would remain even while random error decreases. Analogy For analogous exposures and outcomes an effect has already been shown Hill [ 5 ] wrote that it would be sometimes acceptable to "judge by analogy".
He gives the following example: At best, analogy provides a source of more elaborate hypothesis about the associations under study; absence of such analogies only reflects lack of imagination or lack of evidence. The term "analogous" suggests that the entities in external studies are only similar to those in the observed data but not identical. This requires an additional modelling of the counterfactual effects of using analogous but not identical entities in different studies.
This makes the application of the analogy consideration even more uncertain than the application of considerations on plausibility and coherence. Conclusion Hill himself used the terms "viewpoints" and "features to be considered" when evaluating an association.
His aim was to unravel the question: I have argued that the application of seven of the nine considerations consistency, specificity, temporality, biological gradient, plausibility, coherence and analogy involves comprehensive causal theories. Complex causal systems comprise many counterfactuals and assumptions about biases. If complexity becomes very large, the uncertainty regarding whether or not to apply a given consideration can be expected to approach a decision made by coin toss.
Thus, with increasing complexity, the heuristic value of Hill's considerations diminishes. Here, an original argument of Hill [ 5 ] becomes of particularly important: If a causal conclusion needed an action that brought about more harm if wrongly taken than benefit if rightly taken, a correspondingly high amount of evidence would be required.
If the relationship between benefit and harm were converse, less evidence would be necessary.
The major tool to assess the applicability of these considerations is multiple bias modelling. Multiple bias models should be much more frequently used. Moreover, the decision as to whether or not to apply one of these considerations is always implicitly based on one or several multiple bias models. For instance, demanding an association of at least a certain magnitude is logically equivalent to the "true bias model" being part of the set of multiple bias models in which priors on bias parameters would require at least this magnitude of association to be observed.
One may ask the counterfactual question of how epidemiology and medical research would have developped if Hill had been more explicit in recommending when to apply each of his considerations.
I am far from claiming to be able to answer this question, but I consider my speculation being worth mentioning. In their paper entitled "The missed lessons of Sir Austin Bradford Hill" [ 6 ] Phillips and Goodman reviewed malpractices denounced by Hill that were still being made later in practice: Hill's considerations were misused as "causal criteria", and they were taught more often than more sound causal conceptions [ 6 ].
There is no reason to believe that more explicit recommendations on when to apply his considerations would have been better heeded; the cautionary notes that Hill actually made were largely ignored. My own experience is that scientific recommendations are widely followed if they provide easy guidance; recommendations that call for complex action are frequently ignored.
My guess is that this is due to many researchers' desire for simple and globally applicable answers. This desire leads to misinterpretation of scientific texts and to taking individual statements out of their context.
More pessimistically, the question of which guidance is followed depends on which guidelines are in line with the desired answer. Therefore, it seems likely that, even if Hill's paper had not been published, scientists' desire for simple answers would have caused another paper to be written or to be misinterpreted in the same way as happened with Hill's [ 5 ] article.
List of abbreviations MCSA: Monte Carlo sensitivity analyses RCT: Acknowledgements I wish to thank Evelyn Alvarenga for language editing. Before and after Bradford Hill: Some trends in medical statistics. J Roy Stat Soc A. Sir Austin Bradford Hill: A personal view of his contribution to epidemiology. Fisher, Bradford Hill, and randomisation. Fisher and Bradford Hill: The environment and disease: Proceed Roy Soc Medicine — London.
The missed lessons of Sir Austin Bradford Hill. Encyclopedia of Statistics in Behavioral Sciences. The Hill criteria of causation.Causal Relationship - 1. Introduction
Molecular epidemiology of human cancer. Interpreting recent evidence of oral contraceptive studies. Am J Obstet Gynecol. Vascular disorders preceding diagnosis of cancer: What is a cause and how do we know one? A grammar for pragmatic epidemiology. On the origin of Hill's causal criteria. Causal effects in clinical and epidemiological studies via potential outcomes: Concepts and analytical approaches.
Annu Rev Public Health. Causal inference based on counterfactuals. Statistics and causal inference. J Amer Stat Ass. Estimating causal effects of treatments in randomised and nonrandomised studies. The central role of the propensity score in observational studies for causal inference. Rothman KJ, Greenland S, eds. Causality — Models, reasoning and inference. Cambridge University Press; Maldonado G, Greenland S.
Measuring the potency of risk factors for clinical or policy significance.