Synthesis: Causal Inference EPIET Introductory Course, Lazareto, Menorca 2011 Kassiani Mellou, based on EPIET material.

Synthesis: Causal Inference

EPIET Introductory Course, Lazareto, Menorca 2011

Kassiani Mellou, based on EPIET material

2

How do we understand causality?

Intuitively?

3

4

5

How is cause defined?

“Antecedent event, condition, or characteristic that was necessary for the

occurrence of the disease event and without, the disease event either would not

have occurred at all or until some time later.”

Rothman KJ, Greenland: Causation and Causal Inference in Epidemiology, (Am J PH, 2005)

6

Cause in the context of epidemiology

Count and compare AND

Search for cause and effect

• Source of the outbreak?

• Risk factor for disease?

Why?

• Implement control measures

• Give recommendations

7

p<0.001Does a statistical association automatically mean that there

is a causal relationship?

RR = 89.795%CI = 82.5 – 91.4

8

Statistical association

• Causal ?

• Result of

• chance

• selection bias

• information bias

• confounding

9

Henle-Koch-Postulates (1890)

1. Pathogen must identified in ill person/animal

2. Pathogen must be culturable

3. Cultured pathogen should cause illness in test animal

4. Pathogen must be reisolated and found identical to original

10

Bradford Hill’s criteria (1965)1. Strength of Association

2. Consistency

3. Specificity

4. Temporality

5. Biological gradient (dose response)

6. Plausibility

7. Coherence

8. Experimental Evidence

9. Analogy

AB Hill: The Environment and Disease: Association or Causation?Proc Royal Soc Med 1965;58:295-300

11

Criteria for a Causal Relationship1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations have

been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunders publishers July 2008

12

Criteria for a Causal Relationship

1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations have

been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunder publishers July 2008

13

Temporal Relationship

Exposure must precede disease

Essential criterion for causality

Knowledge of:

• Latency period

• Incubation period

14

Criteria for a Causal Relationship

1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations have

been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunder publishers July 2008

15

Strength of Association

Strong associations are more likely being causal than weak ones.

Smoking > 20 cigarettes/day laryngeal carcinoma (RR 20)

BUT not all strong associations are causal…

16

Cases of Down Syndrome by Birth Order

17

Cases of Down Syndrome by Maternal Age Groups

18

Strength of Association

Strong associations are more likely being causal than weak ones.

Smoking > 20 cigarettes/day laryngeal carcinoma (RR 20)

BUT:Not all strong associations are causal…And weak associations do not rule out

causality…

19

Smoking and Lung cancer? Breast cancer? Passive smokingCigarette smoking and lung cancer

RR= ~ 10

Cigarette smoking and breast cancer

RR = ~ 1 -1.5

Passive smoking and lung cancer

RR = ~ 1.4

20

Criteria for a Causal Relationship

1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations have

been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunder publishers July 2008

21

Biologic Plausibility

Is consistent with current biological and medical common knowledge.

SmokingIngesting of chemicals and known

carcinogens DNA mutationslung cancer

22

Biologic Plausibility

Is consistent with current biological and medical common knowledge.

• Percivall Pott - scrotum cancer observed in chimney sweeps (1775)

• Peptic ulcers and Helicobacter pylori (1980s)

23

Criteria for a Causal Relationship

1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations have

been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunder publishers July 2008

24

Dose-response Relationship

Risk increases with more intense/more frequent exposure

But:• High dose at which any further increase has

no effect• Low dose may be that no response occurs or

can be measured

25

Criteria for a Causal Relationship

1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations have

been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunder publishers July 2008

26

Replication of findings

Findings found in:

• different populations

• by using different study designs

Jan Hendrik Schön – organic electronics

Hwang Woo-suk – stem cell research

27

Criteria for a Causal Relationship

1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations have

been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunder publishers July 2008

28

Effect of removing the exposure

A decrease in the outcome of interest is seen when the exposure is removed.

29

30

Criteria for a Causal Relationship

1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations

have been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunder publishers July 2008

31

Extent to which alternate explanations have been considered

Has adjustment been made for possible confounding?

32

”The Norwegian comedian Marve Fleksnes once stated: I am probably allergic to leather because

every time I go to bed with my shoes on, I wake up with a headache the next morning.”

33

Criteria for a Causal Relationship

1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations have

been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunder publishers July 2008

34

Specificity of the associationOne cause has one effect.

Asbestos exposure mesothelioma

abestosis

lung cancer

35

Rothman and GreenlandOne cause – one effect – simplistic and

not true

Most outcomes are as the result of many contributing causes

• Necessary

• Sufficient

• Probabilistic

36

Earlier head trauma

leading to equilibrium problems

Condition of the sidewalk

Weather conditions

Type of footwear

brittle bones

Source: Rothman KJ, Greenland: Causation and Causal Inference in Epidemiology, (Am J PH, 2005)

Use of cane to support

walking

37

Criteria for a Causal Relationship

1. Temporal relationship2. Strength of the association3. Biologic plausibility4. Dose–response relationship5. Replication of the findings6. Effect of removing the exposure7. Extent to which alternate explanations have

been considered8. Specificity of the association9. Consistency with other knowledge

L Gordis: Epidemiology 4th revised edition, W. Saunder publishers July 2008

38

Consistency with other knowledge

If an association is supported by the results of different disciplines

39

Summary

• Not a checklist! (don’t stop thinking)

• Beware of biologic plausibility

• Always aim for better evidence

• Association is not causality!!!

• Keep an open mind

• Remain critical

(… especially of your own studies)

Thank you for your attention!