Fictional
Scenario: HuGE
Case Study
Answers
and Bibliography
Answer 1
Construct Two by Two tables for genotype and exposure as follows
Genotype |
Case |
Control |
- |
20 |
60 |
C707T |
80 |
40 |
Odds Ratio (OR) = 6 (estimates relative risk)
Population Attributable Fraction = Frequency of cases (OR-1/OR) (Miettinen
formula)
0.8 * 5/6 = 0.66
|
|
Exposure |
Case |
Control |
No |
20 |
50 |
Yes |
80 |
50 |
Odds Ratio= 4
Population
Attributable Fraction 0.8 * ¾=0.60
|
|
Answer 2
Genotype |
Exposure |
Case |
Control |
OR |
PAF |
- |
- |
10 |
30 |
Ref |
Ref |
- |
+ |
10 |
30 |
1.0 |
0 |
+ |
- |
10 |
20 |
1.5 |
3.3% |
+ |
+ |
70 |
20 |
10.5 |
63.3% |
OR
with exposure alone is 1.0, genotype alone is 1.5 and both is
10.5
Based
on the above data. It looks like the combined effect of the
genotype and exposure is more than multiplicative (10.5 >
1.0 + 1.5 –1), strongly suggesting evidence of
gene-environment interaction. Furthermore, it does look like
the genotype alone or the exposure alone don’t really
increase the risk significantly and that most if not all their
main effects are due to their interaction (type I
interaction). IN fact, it looks like the combination of gene
variant and the exposure may account for about 63% cases in
this population. |
Answer 3
The
following findings enhance biological plausibility for causal
association
-
Strong
relative risks and also high attributable fraction
-
Biologic
plausibility (yes, strong because
of the biology and finding interactions)
-
Needs
replication in other studies (especially population-based)
Study
design: Cases highly selected may not reflect general population
of cases and controls may not be comparable to cases (possible
selection bias).
Because of frequency-matching, we may also need matched analysis
(not done here)
No
adjustment for potential confounding variables (diet, lifestyle,
etc.)
Answer 4
|
|
+ |
- |
|
Exposure |
+ |
70 |
10 |
|
|
- |
10 |
10 |
|
|
|
|
|
|
OR
case-only is 7
This
is an index of interaction on a multiplicative scale and
suggests that the combined effects of exposure and genotype is
more than multiplicative.
This
is a valid result only if we assume independence between
genotype and exposure in the population. We can test this by
repeating same anlaysis among controls
|
|
|
+ |
- |
|
Exposure |
+ |
20 |
30 |
|
|
- |
20 |
30 |
|
OR
case-only is 1
This
indicates that there is no association between exposure and
genotype in the general population which make results of
case-only analysis valid.
Case
only analysis:
Pros: quick and efficient, improves statistical power
Cons: needs independence assumption and cannot estimate
separate effects of exposures and genotype only interaction.
Also can only test for interaction on a multiplicative scale
Good
screening tool |
Answer 5
Obviously before issuing such recommendations,
these data need to be validated in other well designed population-based studies.
Even then, the bottom line will be: should we test people before taking
ImmuneBlast or ask everyone not to take ImmuneBlast regardless of their
genotype? Either way, we can prevent ImmuneBlast associated RA in the
population. It looks like 40% of the population (controls) have the susceptible
genotype to this exposure. Given that herbal preparation are over the counter
food additives,. Their safety and effectiveness have not gone through the usual
rigorous studies required by FDA. I would personally not favor a genotypic-based
intervention but across the board caution (if not banning) of ImmuneBlast by the
FDA. Obviously, more studies are needed before such a policy is arrived at.
Bibliography
1.
Commentary:
Facing the challenge of gene-environment interaction: the
two-by-four table and beyond.
Lorenzo D. Botto and Muin J. Khoury Am J Epidemiol (May
15, 2001)
2.
Epidemiology
Book chapter by M. Gwinn, M. J. Khoury - Genetics in Practice (in press)
3.
Genetic
Epidemiology Book chapter excerpted by M. Khoury, from Modern Epidemiology, Second
Edition (1998)
4.
Rothman KJ. Epidemiology: An Introduction. Oxford University Press,
New York, 2002.
home |
back to overview
|