Master Course based on Rothman: Epidemiology chapter 15

1
Master Coursebased onRothman Epidemiology
chapter 1-5

• Hein Stigum
• Presentation, data and programs at
• http//folk.uio.no/heins/

Dec-09
H.S.
1
2
Agenda

• Introduction
• Epidemiological thinking
• Concepts
• Causation
• Generalization
• Methods
• Measures
• Design
• Bias

Dec-09
H.S.
2
3
1. Introduction
Dec-09
H.S.
3
4
Epidemiology

• Study of exposure and disease
• Air pollution Heart disease
• Obesity Diabetes
• Education Exercise
• Questions
• How much exposure?
• How much disease?
• More disease among exposed?

5
Epidemiologic information

• Experiments
• Randomized Controlled Trials
• Observational data
• Registries
• Medical birth -, Cancer -, Patient registry
• Surveys
• Mother and Child Cohort
• Linking

Observational data ? Systematic errors
www.fhi.no
6
Downs syndrome
Age
Downs
Parity
Confounding
7
Reading problems

• Study
• 5000 adults
• Questionnaire
• 1000 respond
• 5 reading problems

Selection bias
8
Sexual partners

• Males report twice as many partners as females do

Information bias
9
2. Causation
Dec-09
H.S.
9
10
Component-, sufficient cause
Switch
wire
electricity
bulb
light
Pie Sufficient cause for light Parts
Component causes
11
Causal pies

• Component cause
• Sufficient cause
• Necessary cause
• Strength
• Interaction
• Induction time
• Attributable fraction

Dec-09
H.S.
11
12
Association versus cause

• Observe
• Smoking associated with Lung Cancer
• Infer cause

• Observe
• Yellow fingers associated with Lung Cancer
• Infer cause

13
Generalization
Dec-09
H.S.
13
14
Generalization

• Do the results apply outside the sample?
• Statistical generalization
• Prevalence of smoking among males, generalize to
  females?
• Representative sample
• Biological generalization
• Animal studies, generalize to humans?
• Information from outside the study
• Homogenous sample

Dec-09
H.S.
14
15
Summing up

• Disease
• Exposure ? Disease
• Observational data
• Cause can not be observed directly
• Generalize representative homogeneous

16
3. Measures
Dec-09
H.S.
16
17
Epidemiological measures

• Frequency
• prevalence
• incidence
• Association
• Risk difference
• Risk ratio
• Odds ratio
• Impact
• Attributable fraction

How much disease?
Dec-09
H.S.
17
18
Mathematical concepts

• Proportion
• Rate
• Odds

Risk, probability,
Km/h
Odds lives at one place in time
Dec-09
H.S.
18
19
Frequency measures
Dec-09
H.S.
19
20
Disease frequency
Theoretical concept
Estimator
!
Dec-09
H.S.
20
21
Disease frequency depicted
t
a
Risk time
Dec-09
H.S.
21
22
Prevalence example
Dec-09
H.S.
22
23
Incidence proportion example
1. If no loss to follow up
2. If loss to follow up
Dec-09
H.S.
23
24
Incidence rate (small cohort)
Assumption Can only get the disease once
Dec-09
H.S.
24
25
Incidence rate (large cohort)
1.
2.
3.
Dec-09
H.S.
25
26
Incidence of hip fracture, age 65
Incidence rate pr 10 000 person years
(Lofthus et al. 2001)
Dec-09
H.S.
26
27
Other measures

• Attack rate risk of infection
• Case fatality rate risk of death if ill

28
Epidemiological measures

• Frequency
• prevalence
• incidence
• Effect
• Risk difference
• Risk ratio
• Odds ratio
• Impact
• Attributable fraction

How much disease?
More disease among exposed?
Dec-09
H.S.
28
29
Effect measures
Dec-09
H.S.
29
30
Causal effect
Risk is not a measure of causal effect, need a
contrast.
Contrast
Counterfactual
Real
31
Causal effect 2

• Ideal
• Counterfactual same subjects same time
• Real
• Crossover same subjects diff. time
• Randomized exchangeable same time
• Observational adjust for confounding same
  time

32
Association measures

• More disease among exposed?
• Compare frequency among exposed1 and unexposed0

• Difference
• Risk Difference
• Ratio
• Risk Ratio
• Odds Ratio

0no effect
1no effect
1no effect
Dec-09
H.S.
32
33
RR and RD example
Disease lung cancer Exposure smoking
Dec-09
H.S.
33
34
OR example
Disease lung cancer Exposure smoking

• Why use OR?
• Trad. Case-Control
• Logistic regression

Dec-09
H.S.
34
35
Relative risk
Risk ratio rate ratio for short-term
risks. Both are termed relative risk
Dec-09
H.S.
35
36
Bullying, OR example

• Bullying in the nordic countries
• 17 114 children, 2 584 bullied

Dec-09
H.S.
36
37
Dec-09
H.S.
37
38
RR and OR depicted
Risks and Risk Ratio
Odds and Odds Ratio
Dec-09
H.S.
38
39
Epidemiological measures

• Frequency
• prevalence
• incidence
• Effect
• Risk difference
• Risk ratio
• Odds ratio
• Impact
• Attributable fraction

How much disease?
More disease among exposed?
How important?
Dec-09
H.S.
39
40
Impact measures
Dec-09
H.S.
40
41
Attributable fraction

• Among exposed
• In population

42
Attributable fraction example
43
Measures, Summing up

• Frequency
• prevalence
• incidence risk, rate
• Effect contrast R1 , R0
• RD
• RR
• OR ORRR (if rare disease)
• Impact of cases
• Afe , Afp (may sum gt100)

Dec-09
H.S.
43
44
4. Design
Dec-09
H.S.
44
45
True or false?

• It takes 2 to tango
• It takes 3 chords to play the blues
• It takes 4 numbers to be an epidemiologist

Dec-09
H.S.
45
46
The 2 by 2 table
.01.01.1 .01 .13
Add 100 to cell d .01.01.1 .005.125
Add 10 to cell c .01.01.05 .01.08
To increase power Cohort balance
exposure Case-Control balance disease
Dec-09
H.S.
46
47
3 examples

• Gender and Smoking
• Exercise and Coronary Heart Disease (CHD)
• Genes and Diabetes type 1

What design should we use?

• Considerations
• Disease rare / common
• Follow up time short / long
• Reversed causality ?
• Recall bias ?

Dec-09
H.S.
47
48
Cross-section
Dec-09
H.S.
48
49
Time
Dec-09
H.S.
49
50
Cross-sectional example
Pro fast and inexpensive Con reversed
causality
OK
Dec-09
H.S.
50
51
Cohort
Dec-09
H.S.
51
52
Disease frequency depicted
t
a
Risk time
Dec-09
H.S.
52
53
Cohort, Risk and Odds
Pro reliable Con costly, time consuming, loss
to follow up
Dec-09
H.S.
53
54
Cohort, Rate
Previous example (Risk and Odds)
Rate 3 years follow up time
Dec-09
H.S.
54
55
Case Control studies
Dec-09
H.S.
55
56
Gene-Diabetes
Full Cohort
Case-Control

• In practice Cases. 1-4 controls per case
• Sample controls independent of exposure
• Exposure back in time

Dec-09
H.S.
56
57
Gene-Diabetes
One Control per Case
Dec-09
H.S.
57
58
Case-Control studies

• Cohort studies
• Measure the exposure experiance of the entire
  population
• Case-Control studies
• Measure the exposure experiance of a sample of
  the source population of cases (base)
• Key assumption
• Sample controls independent of exposure (same
  sampling fraction)
• Prospective or retrospective

Dec-09
H.S.
58
59
Traditional Case-Control
Dec-09
H.S.
59
60
Design, Summing up
61
5. Bias
Dec-09
H.S.
61
62
Precision and validity

• Measures of populations
• precision - random error - statistics
• validity - systematic error epidemiology
• Lack of validity measures are biased
• type of bias
• direction of the bias

Dec-09
H.S.
62
63
Type of bias

• Selection bias
• Are those who answer different?
• Information bias
• Do they tell the truth?
• Confounding
• Is the association a cause?

Dec-09
H.S.
63
64
Selection bias
Dec-09
H.S.
64
Dec-09
H.S.
64
65
Sources of selection bias

• Selective response
• sexual survey
• Self selection
• Nevada atom test and leukemia
• Loss to follow up connected to disease
• air pollution and astma
• Healthy worker effect
• aluminium workers and lung disease

Dec-09
H.S.
65
66
Selection bias
Population
Sample
Respons
Responders
Non-responders
Outcome
Dec-09
H.S.
66
67
Information bias
Dec-09
H.S.
67
Dec-09
H.S.
67
68
Non-differential misclassification
True smoking
10 of smokers report no smoking

Non-differential RR?
Dec-09
H.S.
68
69
Other sources of information bias

• Not true
• males report more partners than females
• Not blinded
• passive smoking and astma
• Selective recall
• alcohol in pregnancy and malformations

Dec-09
H.S.
69
70
Confounding
Dec-09
H.S.
70
Dec-09
H.S.
70
71
Confounding

• Ideal
• Same subjects are both exposed and unexposed at
  the same time, (counterfactual)
• Practice
• As equal as possible
• Comparison bias
• Confounding

Dec-09
H.S.
71
72
Downs syndrome
Age
Downs
Parity
Confounding
73
Downs syndrom, logistic regression
Crude
Adjusted
Dec-09
H.S.
73
74
Bias, Summing up

• Random / Systematic error
• Systematict error ? bias in measure
• 3 types of bias
• Selection
• Information
• Confounding
• May remove bias in analysis
• New tool Causal models

75
Epidemiology, Summing up

• Study of exposure and disease
• Observe association, deduce cause
• Disease
• prevalence, incidence (risk, rate)
• Causal effect
• RD, RR, OR
• Design
• Cross-sectional, Cohort, Case-Control
• Observational ? bias,
• Selection, Information, Confounding