Causality 1 Akbar soltani Assistant Professor of Medicine
Causality (1) Akbar soltani Assistant Professor of Medicine and Endocrinology Evidence-Based Medicine Working Team (EBMWT) Tehran University of Medical Sciences (TUMS) Shariati Hospital
ASSOCIATION VS CAUSATION To decide whether exposure A causes disease B, we must first find out whether the two variables are associated, i. e. whether one is found more commonly in the presence of the other.
Figure 13 -4. Correlation between dietary fat intake and breast cancer by country. USA Incidence Ratio per 100, 000 Women 250 Switzerland Canada 200 Fed. Repub. Of Germany Italy Israel Sweden France Denmark New Zealand Australia 150 UK Norway Finland Yugoslavia 100 Spain Poland Romania Hong Kong Hungary 50 Japan 0 0 600 Prentice RL, Kakar F, Hursting S, et al: Aspects of the rationale for the Women’s Health Trial. J Natl 800 1000 1200 1400 1600 Per Capita Supply of Fat Calories
Is there a relationship between breast cancer incidence and dietary fat consumption by country? · From the graph , we see that as average dietary fat consumption increases, breast cancer incidence increases. · What is wrong with this data? · The problem is: the ecologic fallacy! Prentice et al. J Natl Cancer Institute 1988 80: 802 -814
The ecologic fallacy · Attributing to members of a group characteristics that they do not possess as individuals · In our example, we only know average values of fat consumption by country - we don’t know if individuals with breast cancer had higher fat intake
Why do an ecologic study? HYPOTHESIS BUILDING! The data is easy to obtain, no follow up or individual contact is needed. An ecologic study can suggest avenues of research that may cast light on an etiologic relationship between exposure and disease HOWEVER An ecologic study does not itself demonstrate that a causal relationship exists
Association vs Cause · To make causal inferences, need study that is ®Internally valid • What factors affect internal validity?
Association vs Cause · To make causal inferences, need study that is ®Internally valid • Findings not likely due to chance • Bias controlled • Findings not due to confounding ®Of appropriate design • What designs does this include?
Association vs Cause · To make causal inferences, need study that is ®Internally valid • Findings not likely due to chance • Bias controlled • Findings not due to confounding ®Analytic or intervention design • Cohort, case-control, controlled trial • Cross-sectional? – Usually not
Association vs Cause · Given study that is ®Internally valid ®Analytic or intervention design · How would you know that an association is causal? ®i. e. , that factor (a) causes outcome (b)?
Case-control and cohort studies · Unlike ecologic studies, in case-control and cohort studies we have information on both exposure and outcome for individual subjects · Case-control and cohort studies have a time element - in each, we are assuming that exposure occurred before disease
Assessing Causality • When statistical associations emerge from clinical research, the next step is to judge what type of association exists. Statistical associations do not necessarily imply causal associations. • Classifications: spurious, indirect, and causal. • Spurious associations are the result of selection bias, information bias, and chance. • By contrast, indirect associations (which stem from confounding) are real but not causal. • Causal
Interpreting Associations- Causal and Non-Causal (due to confounding) Coffee Consumption Causal Coffee Consumption Real Association Smoking Spurious Association Real Association Pancreatic Cancer
Real or Spurious Associations · Distinction is important ®If relation is causal Prevent or lower risk of disease if lower or eliminate exposure ®If relation is due to confounding Changes in exposure will have no effect on the risk of disease ®Intervention based on which study?
Web of causation for myocardial infarction.
Sufficient Cause Cluster: · A sufficient cause cluster which means a complete causal mechanism, can be defined as a set of minimal conditions and events that inevitably produce effect. · Minimal implies that all of the conditions and events are necessary.
Necessary Cause · A necessary cause can be defined as a conditions and events that without which the effect does not occur.
An effect with one sufficient cause cluster with two component cause A B · A is a necessary cause · B is a necessary cause · A and B are a sufficient cause cluster
An effect with three sufficient cause cluster U A B U A E U B E Three sufficient cause cluster of a disease · U is a necessary cause for the effect
Four types of causal relationships · Necessary and sufficient(A B) ®Without factor, disease does not develop ®Example: HIV ®Exposure X → Outcome Y ®Not very common ®E. g. , Lead is a necessary and sufficient cause of lead poisoning ®E. g. , Rabies virus is necessary and sufficient cause of human rabies ®It is NOT essential that a necessary and sufficient cause always produces the outcome
· Necessary and sufficient cause: All instances of the outcome are due to this cause, and this cause always produces the outcome. · Usually, when a cause is sufficient it is usually also necessary, for example the relation of measles virus infection to clinical measles, or rabies infection to clinical rabies. · HIV may possibly be a necessary and sufficient cause of AIDS, although this is looking increasingly unlikely, as we find longterm HIV positive individuals without AIDS. · The emphasis on necessary and sufficient causes is sometimes called deterministic causality
BOTH NECESSARY AND SUFFICIENT (e. g. HIV and AIDS) HAS DISEASE FREE OF DISEASE HAS EXPOSURE ALL NONE DOES NOT HAVE EXPOSURE NONE ALL
Four types of causal relationships · Necessary but not sufficient(AND) ®Multiple factors, including main factor, required ®Example: Development of tuberculosis requires M. tuberculosis and other factors, such as immunosupression, to cause disease ®Bacteria still necessary, but not sufficient to cause the disease ®Ex: carcinogenesis ®Exposure X + Other Causes → Outcome Y
Necessary cause: The cause must be present for the outcome to happen. However, the cause can be present without the outcome happening. Hepatitis B infection for is necessary for hepatocellular carcinoma; aspirin (probably) for Reyes syndrome. If outcomes are defining in terms of causes, the cause is necessary by definition. For example, the tubercle bacillus is necessary for tuberculosis by the definition of tuberculosis. Etiologic (as contrasted to manifestational) classification of diseases produces necessary causes.
NECESSARY CAUSE (e. g. the tubercle bacillus and tuberculosis) HAS DISEASE FREE OF DISEASE HAS EXPOSURE YES DOES NOT HAVE EXPOSURE NO YES
Four types of causal relationships · Neither sufficient nor necessary(and/or) ®Complex models of disease etiology ®Example: High fat diet and Heart disease, hypertension, diabetes, certain kinds of cancer ®More complex model, which probably most accurately represents the causality for most chronic disease ®Exposure X + Other Causes → Outcome Y; Exposure Z → Outcome Y
Four types of causal relationships · Sufficient but not necessary(OR) ®Factor can produce disease, but not necessary ®Example: Both radiation exposure or exposure to benzene are sufficient to cause leukemia, but neither are necessary if the other present ®Exposure X → Outcome Y; Exposure Z → Outcome Y
Sufficient cause: If the cause is present the outcome must occur. However, the outcome can occur without the cause being present.
SUFFICIENT CAUSE ( Rabies infection and death) HAS DISEASE FREE OF DISEASE HAS EXPOSURE YES NO DOES NOT HAVE EXPOSURE YES
Guidelines for judging whether an association is causal · Temporal relationship · · Strength of association Consideration of alternate explanations · Cessation of exposure · Dose response relationship · Specificity of the association · Biologic plausibility · Consistency with other knowledge Experimental evidence ·
Time Sequence / Temporality · Exposure of interest has to precede the outcome (by a period of time that biologically makes sense) · Easiest to establish in a cohort study · Smoking and lung cancer · REQUIRED!The only ironclad criterion · Sometimes this is hard to know, especially in crosssectional studies
· EXAMPLE Low serum cholesterol has been linked to increased risk of colon cancer in prospective cohort studies. But is a low serum cholesterol a cause of colon cancer, or does an early phase of colon cancer cause low cholesterol levels?
Temporal Relationship · Length of interval between exposure and disease very important · If the disease develops in a period of time too soon after exposure, the causal relationship is called into question True time course of relationship Exposure Latent period Disease In this case, the latent period is not long enough for disease to develop if caused by this exposure
Asbestos and Lung Cancer Well - established temporal relationship Asbestos Latent period of 10 - 20 yrs Lung Cancer New Study Asbestos Latent period of 3 yrs Lung Cancer In this case, the latent period is not long enough for lung cancer to develop if caused by exposure.
Guidelines for judging whether an association is causal · Temporal relationship · · Strength of association Consideration of alternate explanations · Cessation of exposure · Dose response relationship · Specificity of the association · Biologic plausibility · Consistency with other knowledge Experimental evidence ·
STRENGTH One reason for the importance is that any confounding variable must have a larger association with the outcome to be confounding. The larger the relative risk observed, the less likely it is that a confounder with an even larger relative risk is lurking in the background.
Strength of association Which odds ratio would you be more likely to infer causation from? OR #1: OR = 1. 4 95% CI = (1. 2 - 1. 7) OR#2: OR = 9. 8 95% CI = (1. 8 - 12. 3) OR#3: OR = 6. 6 95% CI = (5. 9 - 8. 1)
Strength of Association · The larger the relative risk or odds ratio, the higher the likelihood that the relationship is causal-OR>4 - RR>3 · However, care must be taken to examine confidence intervals and sample size ®If the confidence interval is wide (e. g. , 1. 8 - 22. 6), an OR of 12. 0 is less strong because we are less confident of the strength of the odds ratio
ROTHMAN · Common disease Vs rare disease ex: CVD and smoking · Counter example of strong non-causal association: down and birth rank · SO: strong association rule out hypothesis that the association is entirely due to weak unmeasured confounding or bias
Guidelines for judging whether an association is causal · Temporal relationship · · Strength of association Consideration of alternate explanations · Cessation of exposure · Dose response relationship · Specificity of the association · Biologic plausibility · Consistency with other knowledge Experimental evidence ·
Dose-Response Relationship · With increasing dose, there is increasing risk of disease · This is not considered necessary for a causal relationship, but does provide additional evidence that a causal relationship exists
Age-standardized death rates due to wellestablished cases of bronchogenic carcinoma
Fish Consumption and Risk of Stroke in Men (JAMA. 2002; 288: 3130 -3136)
Dose-Response · Difficulty: The presence of a doseresponse relationship doesn’t mean that the association is one of cause and effect. Could be, for example, due to confounding. · Smoking and hepatic cirrhosis (alcohol)
Dose-Response · Smoke more, higher CHD death rates · Smoking and death due to lung cancer · BMD and fracture · HBA 1 C and microangiopathy
ROTHMAN · Complementary component causes should be present (good coal)! · Different results can only be due to different methods or random errors · SO: consistency serves only to rule out hypothesis that the association is attributable to some factors that varies across studies
Guidelines for judging whether an association is causal · Temporal relationship · · Strength of association Consideration of alternate explanations · Cessation of exposure · Dose response relationship · Specificity of the association · Biologic plausibility · Consistency with other knowledge Experimental evidence ·
Biological Credibility / Plausibility · The belief in the existence of a cause and effect relationship is enhanced if there is a known or postulated biologic mechanism by which the exposure might reasonably alter the risk of developing the disease
Biological Credibility / Plausibility · OC use and circulatory disease (platelet adhesiveness; arterial wall changes) · Smoking and lung cancer (hundreds of carcinogens and promoters)
Biological Credibility / Plausibility Since what is considered biologically plausible at any given time depends on the current state of knowledge, the lack of a known or postulated mechanism does not necessarily mean that a particular association is not causal 40
Guidelines for judging whether an association is causal · Temporal relationship · · Strength of association Consideration of alternate explanations · Cessation of exposure · Dose response relationship · Specificity of the association · Biologic plausibility · Consistency with other knowledge Experimental evidence ·
Consideration of alternate explanations · If the investigators did not consider possible confounders and effect modifiers, the association is less likely to be causal · Requires a knowledge of the literature and known risk factors for the disease
Guidelines for judging whether an association is causal · Temporal relationship · · Strength of association Consideration of alternate explanations · Cessation of exposure · Dose response relationship · Specificity of the association · Biologic plausibility · Consistency with other knowledge Experimental evidence ·
Cessation of exposure · Upon elimination or reduction of exposure to the factor, the risk of disease declines · HOWEVER, in certain cases, the damage may be irreversible · Example: Emphysema is not reversed with the cessation of smoking, but its progression is reduced
Guidelines for judging whether an association is causal · Temporal relationship · · Strength of association Consideration of alternate explanations · Cessation of exposure · Dose response relationship · Specificity of the association · Biologic plausibility · Consistency with other knowledge Experimental evidence ·
SPECIFICITY Causality is enhanced if an exposure is associated with a specific disease, and not with a whole variety of diseases
SPECIFICITY EXAMPLE 1. Asbestos causes a specific lung disease, asbestosis, distinguishable from many other lung diseases. But low level lead exposure is associated with lower IQ rather than a distinguishable brain syndrome. Thus lead is more uncertain as a cause because of possible confounding with other causes of this rather non-specific effect, low IQ (e. g. SES).
Guidelines for judging whether an association is causal · Temporal relationship · · Strength of association Consideration of alternate explanations · Cessation of exposure · Dose response relationship · Specificity of the association · Biologic plausibility · Consistency with other knowledge Experimental evidence ·
CONSISTENCY Consistency can mean either: · Exact replication, as in the laboratory sciences, or · Replication under many different circumstances. In epidemiology, exact replication is impossible
Consistency with other knowledge · If a relationship is causal, the findings should be consistent with other data · Consistent observation of an association in different populations and within subgroups the population and with different study designs also lend s support to a real effect. · The association persists even with the most rigorous study designs and analysis · Very important criterion · Meta-analysis !
Guidelines for judging whether an association is causal · Temporal relationship · · Strength of association Consideration of alternate explanations · Cessation of exposure · Dose response relationship · Specificity of the association · Biologic plausibility · Consistency with other knowledge Experimental evidence ·
Experimental evidence • Experimental evidence is seldom available · Experimental Evidence ®Strengthens the case for a causal association ®RCTs and CTs • Virtually eliminate selection bias and confounding
ROTHMAN · Which types of evidence? · Logically is not a criteria but a test of hypothesis and unavailable most of the times · Can be much stronger than other test but interpretation is difficult
CAUSAL CRITERIA COMPARED SURGEON GENERAL SUSSER ASSOCIATION BRADFORD-HILL DIRECTION DOSE RESPONSE* EXPERIMENT TIME ORDER STRENGTH CONSISTENCY TIME ORDER** STRENGTH CONSISTENCY SPECIFICITY COHERENCE PREDICTIVE PERFORMANCE COHERENCE*** *Included under strength in other criteria. ** Temporality in Bradford-Hill.
Coherence with existing data Ancillary biological evidence that is coherent with the association might be helpful. For example, the effect of cigarette smoke on the bronchial epithelium of animals is coherent with an increased risk of cancer in human beings.
COHERENCE-Susser a. theoretical Compatible with pre-existing theory b. factual Compatible with pre-existing knowledge 1. biologic Compatible with current biological knowledge from other species or other levels of organization (e. g. cellular in humans) 2. statistical Compatible with a reasonable statistical model of the relationship of cause to effect (e. g. dose-response)
Analogy • Reasoning by analogy has sometimes caused harm. • Since thalidomide can cause birth defects, for instance, some lawyers (successfully) argued by analogy that Bendectin(an antiemetic widely used for nausea and vomiting in pregnancy) could also cause birth defects, despite evidence to the contrary.
Associations are observed Causation is inferred It is important to remember that these criteria provide evidence for causal relationships. All of the evidence must be considered and the criteria weighed against each other to infer the causal relationship
Assessing the relationship between factors in epidemiology
THANK YOU
- Slides: 70