Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 60

The third factor

Effect modification
Confounding factor
FETP India

Competency to be gained
from this lecture
Identify and describe an effect modification
Eliminate a confounding factor

Key elements
Describing an effect modification
Eliminating a confounding factor

Stratification
Sub-groups can be defined according to
various characteristics in a population
Age
Sex
Socio-economic status

An association between a risk factor and an


outcome may be studied within these various
strata

Key elements
Describing an effect modification
Eliminating a confounding factor

Effect modification

Spotting effect modification


in a stratified analysis
Effect modification (= Interaction) occurs
when the answer about a measure of
association is:
it depends

Examples:
Efficacy of measles vaccine
Variation according to the age

Risk of myocardial infarction among women


taking oral contraceptives
Variation according to smoking habits
Effect modification

Describing an effect modification

Conduct crude analysis


Stratify data by suspected modifier
Observe the association strata by strata
Judge the heterogeneity of:
Odds ratios
Relative risks

Test a potential difference


Report the effect modification
Effect modification

Describing an effect modification

Conduct crude analysis


Stratify data by suspected modifier
Observe the association strata by strata
Judge the heterogeneity of:
Odds ratios
Relative risks

Test a potential difference


Report the effect modification
Effect modification

Describing an effect modification

Conduct crude analysis


Stratify data by suspected modifier
Observe the association strata by strata
Judge the heterogeneity of:
Odds ratios
Relative risks

Test a potential difference


Report the effect modification
Effect modification

Death from diarrhoea according to


breast- feeding, Brazil, 1980s
(Crude analysis)
Diarrhoea

Controls

Total

No breastfeeding 120

136

256

Breastfeeding

50

204

254

170

340

510

Total

Odds ratio: 3.6; 95% CI: 2.4- 5.5; p < 0.0001


Effect modification

Describing an effect modification

Conduct crude analysis


Stratify data by suspected modifier
Observe the association strata by strata
Judge the heterogeneity of:
Odds ratios
Relative risks
Test a potential difference
Report the effect modification
Effect modification

Death from diarrhoea according


to breastfeeding, Brazil, 1980s
Infants < 1 month of age
No breastfeeding
Breastfeeding
Total

Cases

Controls

Total

10

13

68

75

17

71

88

Cases

Controls

Total

110

133

243

43

136

179

153

269

422

Infants 1 month of age


No breastfeeding
Breastfeeding
Total

Describing an effect modification

Conduct crude analysis


Stratify data by suspected modifier
Observe the association strata by strata
Judge the heterogeneity of:
Odds ratios
Relative risks

Test a potential difference


Report the effect modification
Effect modification

Death from diarrhoea according


to breast feeding, Brazil, 1980s:
Analysis among infants < 1 month of age
Cases

Controls

Total

13

68

75

17

71

88

No breastfeeding 10
Breastfeeding
Total

Odds ratio: 32.4; 95% CI: 6- 203; p < 0.0001


Effect modification

Death from diarrhoea according


to breast feeding, Brazil, 1980s:
Analysis among infants 1 month of age
Cases

Controls

Total

No breastfeeding 110

133

243

Breastfeeding

43

136

179

153

269

422

Total

Odds ratio: 2.6; 95% CI: 1.7- 4.1; p < 0.0001


Effect modification

Describing an effect modification

Conduct crude analysis


Stratify data by suspected modifier
Observe the association strata by strata
Judge the heterogeneity of:
Odds ratios
Relative risks

Test a potential difference


Report the effect modification
Effect modification

Judge the heterogeneity of the


measures of association
To be a difference, a difference should make
a difference
Review public health implications

Odds ratios in the specific example:


Strata 1: OR = 32; 95% CI: 6.0- 200
Strata 2: OR = 2.6; 95% CI: 1.7- 4.1

Effect modification

Describing an effect modification

Conduct crude analysis


Stratify data by suspected modifier
Observe the association strata by strata
Judge the heterogeneity of:
Odds ratios
Relative risks

Test a potential difference


Report the effect modification
Effect modification

Woolfs test for heterogeneity


of the odds ratios
Statistical testing of the heterogeneity of the
odds ratios
Lacks statistical power
Calculation:
In statistical textbooks
In the softwares analysis output

Judgement is important
Effect modification

Handling heterogeneous
measures of association
O R s / R R s a re
d if f e r e n t a c r o s s st r a t a
O R s / R R s 9 5 % C . I.
d o n o t o v e rla p

O R s / R R s C . I.
d o o v e r la p

E f f e c t m o d if ic a t io n

U s e W o o l f 's t e s t

W o o l f 's t e s t s i g n i f i c a n t

W o o l f 's t e s t n o t s i g n i f i c a n t

E f f e c t m o d if ic a t io n

E f f e c t m o d if ic a t io n
u n l ik e ly
D is c u s s la c k o f p o w e r
o f W o l l f 's t e s t

Describing an effect modification

Conduct crude analysis


Stratify data by suspected modifier
Observe the association strata by strata
Judge the heterogeneity of:
Odds ratios
Relative risks

Test a potential difference


Report the effect modification
Effect modification

Conclusion of the Brazilian case-control


study on breastfeeding
and death from diarrhoea
The protective efficacy of breastfeeding is
more marked among infants under the age of
one month
This may correspond to a biological
phenomenon that must be reported as part
of the results

Effect modification

Reporting results in the presence of


an effect modification
Once the effect modification was detected
the study population is split
Results for the risk factor considered are
reported stratum by stratum

Effect modification

Vaccination against hepatitis B among


institutionalized children in Romania

Hepatitis B is highly endemic in Romania


Many children live in institutions
Institutionalized children are at higher risk
1995: Hepatitis B immunization initiated
1997: Evaluation through serologic survey

Effect modification

Hepatitis B vaccine efficacy among


institutionalized children over 6 months
of age *, Romania, 1997
HBV
Vaccine

Anti-HBc (+) Anti-HBc (-) RR


3 doses
15
383
0.48
< 3 doses
4
47
Ref.

95% C.I.
0.17-1.4

Vaccine efficacy, 52%, 95% CI 0-83%


* Born after implementation of routine vaccination

Effect modification

Hepatitis B vaccine efficacy among


institutionalized children over 6 months
of age *, by district, Romania, 1997
RR
61
11

3 doses
< 3 doses

322
36

Others District X

Anti-HBc (+) Anti-HBc (-)


3 doses
12
< 3 doses
1
3
3

95% C.I.
2.0
0.28-14
Ref.
0.12
Ref.

0.0-0.6

Wolf test for evaluation of interaction: p = 0.03


* Born after implementation of routine vaccination
Effect modification

Hepatitis B vaccine efficacy among


Romanian children in institutions:
Conclusions
The protective efficacy of hepatitis B
vaccine appears low overall
This overall low efficacy does not correspond
to a biological phenomenon
In fact, the efficacy is:
Normal in most districts (88%)
Low in district X

This points towards programme errors that


must be identified and prevented
Effect modification

Describing an effect modification:


Summary
The analysis plan:
Anticipates effect modifiers to collect data

The analysis:
Looks for effect modification to test it

The report:
Breaks down the population in strata to report
the effect modification

Effect modification

Key elements
Describing an effect modification
Eliminating a confounding factor

Confounding factor

What may explain an association


between a risk factor and an outcome?
?
?
?
?

Chance
Bias
Third factor
Causal association

Confounding factor

What may explain an association


between a risk factor and an outcome?
?
?
?
?

Chance
Bias
Third factor
Causal association

Confounding factor

Characteristics of a third,
confounding factor
Associated with the exposure
Without being a consequence of exposure

Associated with the outcome


Independently from the exposure
Exposure

Outcome

Confounding factor
Confounding factor

The nuisance introduced by


confounding factors
May simulate an association
May hide an association that does exist
May alter the strength of the association
Increased
Decreased

Confounding factor

Example of confounding factor

Apparent association
Exposure 1

Outcome

Confounding
factor

Confounding factor

Example of confounding factor (1)

Apparent association
Ethnicity

Pneumonia

Crowding

Confounding factor

Example of confounding factor (2)

Altered strength of association


Crowding
Pneumonia

Malnutrition

Confounding factor

Eliminating confounding in the


pneumonia example
Estimate the strength of the association
between malnutrition and pneumonia
Estimate the strength of the association
between crowding and pneumonia
Adjusted for the effect of malnutrition

Eliminate the confounding effect of crowding


on the false association between ethnicity
and pneumonia
Confounding factor

Controlling a confounding factor

Stratification
Restriction
Matching
Randomization
Multivariate analysis

Confounding factor

Controlling a confounding factor

Stratification
Restriction
Matching
Randomization
Multivariate analysis

Confounding factor

Adjustment to eliminate confounding


Examine strength of association across strata
Check for the absence of effect modification
If there is an effect modification, break in
various strata, report. End of the story

Observation of a strength of association:


Homogeneous across strata
Different from the crude measure

Calculate weighted average of stratumspecific measures of association


Confounding factor

Malaria and radio sets


Hypothesis: Could radio waves be a repellent
for female anopheles?
Cohort study on the risk factors for malaria
in an endemic area

Confounding factor

Incidence of malaria according to the


presence of a radio set,
Kahinbhi Pradesh
Crude data
Malaria No malaria
Radio

Total

80

440

520

No radio

220

860

1080

Total

300

1300

1600

RR: 0.7; 95% CI: 0.6- 0.9; p < 0.02


Confounding factor

Incidence of malaria according to the


presence of a radio set,
Kahinbhi Pradesh
Strata 1: Sleeping under a mosquito net
Malaria No malaria Total
Radio

30

370

400

No radio

50

630

680

Total

80

1000

1080

RR: 1.02; 95% CI: 0.7- 1.6; p < 0.97


Confounding factor

Incidence of malaria according to the


presence of a radio set,
Kahinbhi Pradesh
Strata 2: Sleeping without a mosquito net
Malaria No malaria
Total
Radio

50

70

120

No radio

170

230

400

Total

220

300

520

RR: 0.98; 95% CI: 0.8- 1.2; p < 0.95


Confounding factor

Mantel-Haenszel adjusted relative risk

aixL0i) / Ti]
RR M-H=
ci xL1i) /
T i]
Confounding factor

Malaria and radio sets:


Conclusion
No association between radio and malaria
within each strata
The new adjusted relative risk replaces the
crude one Apparent association
Radio sets

Malaria

Mosquito nets
Confounding factor

Mantel-Haenszel adjusted odds ratio

ai.di) / Ti]
OR M-H=

bi.ci) / Ti]
Confounding factor

Controlling a confounding factor

Stratification
Restriction
Matching
Randomization
Multivariate analysis

Confounding factor

Hepatitis B and blood transfusion in


Moldova
Hepatitis B virus infection is highly endemic
in Moldova
Routes of transmission are unknown
A case control study was initiated to assess
potential modes of transmission

Confounding factor

Acute hepatitis B and receiving a


transfusion in Moldova, 1994-1995
Cases

Controls

Total

Non-transfusion

69

189

258

Total

72

190

262

Transfusion

Odds ratio: 8.2; 95% CI : 0.8-220

Confounding factor

Acute hepatitis B and receiving a


transfusion in Moldova, 1994-1995
(According to receiving injections)
Injections

No injections

Case

Control

Total

No transfusion

22

28

Total

25

32

Transfusion

Case

Control

Total

No transfusion

47

183

230

Total

47

183

230

Transfusion

Odds ratio: Odds ratio: 0.8,


95% CI: 0.1-24.9

Confounding factor

Controlling a confounding factor

Stratification
Restriction
Matching
Randomization
Multivariate analysis

Confounding factor

Matching
Stratification conducted initially at the stage
of the study design of a case control study
Stratified analysis (matched) necessary

Confounding factor

Controlling a confounding factor

Stratification
Restriction
Matching
Randomization
Multivariate analysis

Confounding factor

Randomization
Distribution of exposure of interest at
random in the study population for a
prospective cohort
An association between an exposure and a
confounding factor will be:
Secondary to chance alone
Improbable

Confounding factor

Controlling a confounding factor

Stratification
Restriction
Matching
Randomization
Multivariate analysis

Confounding factor

Multivariate analysis
Mathematical model
Simultaneous adjustment of all confounding
and risk factors
Can address effect modification

Confounding factor

Taking into account a third factor


in practice
1.
2.
3.
4.
5.

Think of potential confounding factors


Collect accurate data on them
Conduct crude analysis
Stratify
Look for effect modification

Before
the study

Are the RR or OR different to each other?

6. If effect modification:

Report
Do not adjust

7. Control confounding factors through


adjustment

If applicable

During
the analysis

Analyzing a third factor


E x a m in e c ru d e O R / R R
E x a m in e O R s / R R s in e a c h s t r a t u m
Id e n t ic a l O R s / R R s a c r o s s st r a ta

D if f e r e n t O R s / R R s a c r o s s st r a ta

S t r a t a O R s / R R s s im ila r t o c r u d e
( C r u d e v a lu e f a lls b e t w e e n s t r a t a )

S t r a t a O R s / R R s d if fe r e n t f r o m c r u d e
(C ru d e v a lu e d o e s n o t fa ll b e tw e e n stra ta )

E f f e c t m o d if ic a t io n

T h ir d f a c t o r d o e s n o t p la y a r o le

C o n f o u n d in g f a c t o r

S t o p t h e a n a ly s is.
D O N O T a d ju st!

R e p ort O N E cru de O R /R R

A d j u s t u s in g t h e
M - H t e c h n iq u e

R e p o r t M U L T IP L E O R s / R R s
fo r e ac h stratu m

E l im in a t e t h e c o n f o u d in g
R e p ort O N E a d ju ste d O R / R R

Take-home messages
Describe effect modifications
The analysis must TEST for their occurrences

Control confounding factors


The analysis must ELIMINATE their influence

You might also like