Causal and non-causal associations

Biases, confounding, confusion

How to decide if a finding is real

Ora Paltiel October 26, 2014

Causation

Importance of studying causation

1. Clinical medicine

Advice to patients

Prevention

2. Social/environmental

Compensation

3. Population

Screening prevention

“Cause”

A factor which, if altered is followed by a change in the frequency or character of a disease.

Probabilistic approach: X is a cause of Y if the probability that Y occurs is increased as a consequence of X.

Problems: risk factor vs cause risk marker vs risk factor

Models

Triangle האפידמיולוגי המשולש

Host המאכסן

Agent המחלה הסביבה Environment מחולל

Web of causation

Association and cause

Explanation Finding

Association

Bias in selection Yes No

or measurement

Chance Likely Unlikely

Confounding Yes No

Cause Cause

Types of association

Positive vs inverse

Linear, exponential, quadratic (U or J shaped)

Two factors:

additive

multiplicative

Effect modification

Relationship between asbestos exposure (particle-years) and relative

risk of lung cancer

Wisconsin Card Sorting Test number of categories in early-onset, late-onset and control groups use of marijuana.

Fontes M A et al. BJP 2011;198:442-447

Annual mortality (per 1000 men) from all causes related to alcohol

consumption

Ischaemic heart disease

Age-standardized lung cancer death rates (per 100000 population) in relation to cigarette smoking and occupational

exposure to asbestos dust

Exposure toasbestos

History ofcigarettesmoking

Lung cancerdeath rate per

100000

No

Yes

No

Yes

No

No

Yes

Yes

11

58

123

602

0

2

4

6

8

Black Brown Blonde/red

Psoriatics

Controls

Prevalence of Actinic Keratoses in Psoriatics and Controls by Hair Colour

Effect modification

Odds

ratio

Prevalence of Actinic Keratoses in Psoriatics and Controls by propensity

to sunburn

Effect modification

0

1

2

3

4

5

Never Often/always

Psoriatics

ControlsOdds ratio

Association and cause

Explanation FindingAssociation

Bias in selection Yes Noor measurement

Chance Likely Unlikely

Confounding Yes No

Cause Cause

Definition of bias

“Any process at any stage of inference which tends to produce results or conclusions that differ systematically from the truth”

Stages of research in which bias can occur:

1. In reading-up on the field2. In specifying and selecting the study

sample3. In executing the experimental

maneuvre (or exposure)4. In measuring exposures and outcomes5. In analyzing the data6. In publishing the results

Membership bias

Membership in a group (the employed, joggers, etc…) may imply a degree of health which differs systematically from that of the general population.

Selection bias

Confirmation by randomized trial of observed effect in non-randomized trial: Salk vaccine for poliomyelitis

Type of study Poliomyelitis cases/total

(rate per 100,000)

Odds ratio

(95% CI)

Vaccine Control

Non-randomized

Randomized

60/231902

(26)

57/200745

(28)

391/725173

(54)

142/201229

(71)

0.55 (0.44-0.68)

0.43 (0.32-0.56)

External Validity

Confirmation by randomized trial of observed effect in non-randomized trial: Salk vaccine for poliomyelitis

Type of study Poliomyelitis cases/total

(rate per 100,000)

Odds ratio

(95% CI)

Vaccine Control

Non-randomized

Randomized

60/231902

(26)

57/200745

(28)

391/725173

(54)

142/201229

(71)

0.55 (0.44-0.68)

0.43 (0.32-0.56)

Cumulative mortality from verified colorectal cancer in the control

and screened groups

Cause specific mortality rates (per 1000 person years) in the intervention and control groups by faecal occult blood

testing

Social approval bias

Social approval bias

Intervention- letter, nutrition survey, benefits of fruits and veggies. 5-a day sticker

Control- letter, nutrition survey,

Referral bias

Admission rate (Berkson) bias

When combination of exposure and disease leads to higher rate of hospitalization, the relation between exposure and disease will become distorted in hospital based studies.

Measurement bias/ information bias esp.in case-control studies

a) presence of the outcome directly affects the exposure. E.g. endometrial cancer and estrogen

b) presence of the outcome affects the recollection of the exposure “recall bias”. E.g. birth defects and drug exposure

c) presence of the outcome affects the measurement or recording of the exposure.

Eg DVT and oral contraceptives

Association and cause

Explanation Finding

Association

Bias in selection Yes No

or measurement

Chance Likely Unlikely

Confounding Yes No

Cause Cause

P value

Association and cause

Explanation Finding

Association

Bias in selection Yes No

or measurement

Chance Likely Unlikely

Confounding Yes No

Cause Cause

Confounding

Confounding: coffee drinking, cigarette smoking, and coronary heart disease

EXPOSURE DISEASE

(coffee drinking) (heart disease)

CONFOUNDING

VARIABLE

(cigarette smoking)

Locations of potential bias in cohort Locations of potential bias in cohort studiesstudies

021657408021657408

021657408021657408

Association and cause

Explanation Finding

Association

Bias in selection Yes No

or measurement

Chance Likely Unlikely

Confounding Yes No

Cause? Cause

Criteria for causation

1. Is there evidence from true experimentation in humans?

2. Is the association strong?

3. Is the association consistent from study to study?

4. Is the temporal association correct?

5. Is there a dose-response gradient?

6. Does the association make biological sense?

7. Is the association specific?

(Adapted from Bradford Hill)

The isotretinoin dysmorphic syndromeLancet, March 3 1984

Sir, - we describe here a case of human teratogenicity that confirms the need for the drug isotretinoin to be avoided in pregnancy.

Post-mortem examination revealed a dysmorphic syndrome, with malformations of the ears, heart, and brain….This is the 1st case of human isotretinoin teratogenicity to be described in detail.

The isotretinoin teratogen syndrome

Two infants with prominent frontal bossing, hydrocephalus, microphthalmia, and small malformed, low-set, undifferentiated ears were born to mothers, who had taken isotretinoin in the first trimester of pregnancy. A Dandy-Walker malformation, microcephaly, hypertelorism, small ear canals, cleft palate, small mouth, and congenital heart disease were also observed. Isotretinoin is a potent teratogen in many. Maternal ingestion early in pregnancy, leads to a distinct clinical pattern of anomalies.

JAMA 1984; 251:3267-69.

Retinoic acid, an analogue of vitamin A, is known to be teratogenic in laboratory animals and has recently been implicated in a few clinical case reports. To study the human teratogenicity of this agent, we investigated 154 human pregnancies with fetal exposure to isotretinoin, a retinoid prescribed for severe recalcitrant cystic acne. The outcomes were 95 elective abortions, 26 infants without major malformations, 12 spontaneous abortions, and 21 malformed infants. A subset of 36 of the 154 pregnancies was observed prospectively.

Coherence

The outcomes in this cohort were 8 spontaneous abortions, 23 normal infants, and 6 malformed infants. Exposure to isotretinoin was associated with an unusually high relative risk for a group of selected major malformation (relative risk = 25.6; 95% CI 11.4-57.5).

Among the 21 malformed infants we found a characteristic pattern of malformation involving craniofacial, cardiac, thymic, and central nervous system structure.

Specificity

Further criterion:

• Reversibility -rarely applicable

Time sincestoppingsmoking(years)

Duration of smoking habits (years)

1-19 20-39 40-49 >50

Men01-45-9>10Women01-45-9>10

1.01.10.40.3

1.01.00.40.4

2.22.11.51.0

2.12.32.00.8

2.82.32.21.6

2.72.11.12.3

3.03.82.82.7

5.27.11.7

Relative risk for developing lung cancer by time since stopping smoking and total duration of

smoking habit

Level Source of EvidenceI Evidence obtained from at least one properly

designed randomized controlled trial

II-1 Evidence obtained from well-designed controlled trials without randomization

II-2 Evidence obtained from well-designed cohort or case-control analytic studies, preferably from more than one center or research group

II-3 Evidence obtained from multiple time series with or without the intervention. Dramatic results in uncontrolled experiments (such as the results of the introduction of penicillin treatment in the 1940s) could also be regarded as this type of evidence

III Opinions of respected authorities, based on clinical experience, descriptive studies, or reports of expert committees

Hierarchy of study methods to assess causation

Hierarchy of study methods to assess causation

Remember: Evidence based medicine means making decisions on BEST EVIDENCE,

not IDEAL or PERFECT evidence