Question
1: What was
the prevalence of the FVL genotype (at least one A allele)
among controls? Among cases?
Your
answer:
Because there were no
homozygotes among controls, the investigators combined data
for the G/A and A/A genotypes. They calculated the
following relative risks of venous thrombosis (estimated as
odds ratios):
- 7.9
(95% CI 3.2-19.4) for FVL
- 7.0
(95% CI 2.1-23.5) for FVL among OC nonusers
- 3.8
(95% CI 2.4- 6.0) for OC use.
The investigators also
reported the results of a stratified analysis to examine the
effect of OC use according to FVL genotype.
Factor V Leiden (FVL) and oral contraceptive (OC) use
among women with venous thrombosis and controls*
FVL |
|
|
+ |
- |
|
|
Cases |
Controls |
Cases |
Controls |
OC use |
+ |
25 |
10 |
84 |
63 |
|
- |
2 |
4 |
36 |
100 |
OR (95%
CI) |
|
5.0
(0.8-31.8) |
3.7 (2.2-6.1) |
*Adapted from Vandenbroucke et al., 1994.
“+” denotes
G/A or A/A genotype; “-” denotes G/G.
They
concluded that the relative risk of thrombosis in OC users
was similar regardless of FVL genotype and not different
from the overall relative risk associated with OC use-that
is, 5.0 »
3.7 »
3.8.
Question
2:
Do you
agree that the relative risk of venous thrombosis in OC
users was similar regardless of factor V genotype?
Your answer:
The
investigators reasoned that, because FVL did not appear to
modify the effect of OC use on risk of venous thrombosis, a
reasonable estimate of the joint effect was: ORFVL x
OROC use »
7 x 4 »
30. The odds ratio calculated
from the data, comparing women who had FVL and used OCs with
those who had neither risk factor, was 34.7 (95% CI 7.8 –
154).
Another way to present these
data is in a “two-by-four table.”[i]
This format lends itself to calculating measures of
association for genetic and environmental exposures and the
joint exposure, consistently using persons with neither
exposure as the reference group.
Factor
V Leiden (FVL) and oral contraceptive (OC) use among women
with venous thrombosis and controls* |
FVL† |
OC
use |
Cases |
Controls |
OR |
95%CI |
+ |
+ |
25 |
2 |
34.7 |
(7.8-310.0) |
+ |
- |
10 |
4 |
6.9 |
(1.8- 31.8) |
- |
+ |
84 |
63 |
3.7 |
(1.2- 6.3) |
- |
- |
36 |
100 |
ref |
|
Total |
|
155 |
169 |
|
|
*adapted from Botto
and Khoury, 2001
† “+” denotes G/A or A/A
genotype; “-” denotes G/G.
The
two-by-four table is also convenient for assessing
gene-environment interaction. Two common statistical
models of interaction are:
1.
additive, where ORge = ORg + ORe – 1, and
2.
multiplicative, where ORge = ORg x ORe
where
g denotes genotype and e denotes the environmental
factor. Inequality in either statement may be
interpreted as statistical evidence of interaction.
Of
course, biologic models of interaction can be more
complicated, depending on the number of genetic loci
involved, the dose of the environmental exposure, and the
interplay of their effects at the molecular level.
Question
3: What evidence do these data provide for or against
interaction between FVL and OC use in venous
thrombosis?
Your answer:
The
case-only study[ii]
is a nontraditional study design that has been suggested for
evaluating gene-environment interaction where only case data
are available (e.g., from a case series or registry), or
where sample sizes are too small for stratified
analysis. Under a multiplicative model of interaction
where genotype and exposure are independent in the
population,
OR
case-only = ORge / ( ORe x
ORg) = 1
where
OR
case-only measures the association between the
genotype and the exposure among cases. A departure
from unity indicates the presence of gene-environment
interaction.
We
can analyze the case data from the study by Vandenbroucke et
al. in the manner of a case-only study by constructing a
two-by-two table:
Association between factor V Leiden (FVL) and oral
contraceptive (OC) use
in women with venous thrombosis |
|
|
OC use |
|
|
|
|
+ |
- |
OR |
95% CI |
FVL*
|
+ |
25 |
10 |
1.1 |
0.5-2.5 |
|
- |
84 |
36 |
|
|
* “+” denotes G/A or A/A genotype; “-” denotes G/G.
Question
4: How does the result of this case-only analysis
compare with results of the case-control analysis?
Your answer:
From
their analysis, Vandenbroucke et al. concluded that “the
combined effect of these risk factors seems close to a
multiplication of the separate relative risks. In
terms of absolute effect, however, this means that the risk
of venous thrombosis among women who use OCs is much greater
when they carry the factor V Leiden mutation.”
The
investigators estimated the incidence (absolute risk) of
first venous thrombosis using data from the Leiden clinic,
which had a geographic source population of 109,824 women
aged 15-49 years (according to census data). In all,
117 cases of venous thrombosis occurred in this geographic
area during the 5 years of the study. Thus, the
estimated incidence of venous thrombosis among 15- to
49-year-old women was:
117 / (109,824 x 5) = 2.1 per 10,000 person-years
The
155 cases from all three areas were assumed to have arisen
from
155 / 2.1 / 10,000 » 740,000 person-years
Total
person-years were apportioned to the four exposure groups
according to the distribution of exposures among
controls. The results served as approximate
denominators for calculating incidence. For example,
women without FVL who did not use OCs accounted for 59% of
the control group and
0.59 x 740,000 = 437,870 person-years.
Question
5: What biases could be introduced by using this
approach to estimate person-years at risk for calculating
incidence ?
Your answer:
Estimated
population incidence of first venous thrombosis in women aged
15-49 years, according to presence of FVL and use of oral
contraceptives (OCs)*
|
Factor V
Genotype† |
OC use
|
Cases
|
Person-years (py)
|
per 10,000 py
|
G/G
|
no |
36 |
437,870
|
0.8 |
|
yes |
84 |
275,858
|
3.0 |
G/A or A/A
|
no |
10 |
17,515
|
5.7 |
|
yes |
25 |
8,757
|
28.5 |
*adapted from Vandenbroucke et al.,
1994
†G=normal allele, A=FVL allele
Question
6: What is the
risk difference associated with OC use in women without FVL?
In women with FVL?
Your answer:
Clearly, a woman who uses OCs
is at much higher risk for venous thrombosis if she has FVL,
raising the question whether women should be screened for
FVL before taking OCs. The investigators
addressed this issue in a second article.
Vandenbroucke JP, van der
Meer FJM, Helmerhorst FM, Rosendaal FR. Factor V
Leiden: should we screen OC users and pregnant women? BMJ 1996;313:1127-1130.
The authors considered the
“worst case” outcome, death from pulmonary
embolism. A U.S. study estimated that the case
fatality rate for venous thrombosis was 2% in persons aged
<40 years.
Estimated
population incidence of pulmonary embolism in women aged
15-49 years with FVL:
-
5.7 x 0.02 / 10,000 py = 0.11 / 10,000 py in OC
non-users
-
28.5 x 0.02 / 10,000 py = 0.57 / 10,000 py in OC users
Question
7: How many women with FVL would have to avoid OCs to
prevent one death from pulmonary embolism? How many
women would have to be screened to find this many women with
FVL?
Your answer:
The authors concluded that
about 400,000 women would have to be screened-and 20,000
carriers of FVL would have to avoid OCs-to prevent one death
from pulmonary embolism each year.
Question
8: What are
other potential costs and benefits to consider when deciding
whether to screen young women for FVL to prevent venous
thrombosis?
Your answer:
The authors discussed several
issues that introduce uncertainty in this kind of
analysis: imperfect estimates of FVL prevalence
and incidence of venous thrombosis, issues related to
screening test performance, costs associated with
alternative forms of contraception, and morbidity costs
associated with venous thrombosis (including risks
associated with anticoagulant therapy).
They concluded that routine
screening for FVL when prescribing OCs was “unlikely to
stand the competition for resources with other medical
screening and therapeutic interventions.” However,
they advised taking a family history of venous thrombosis to
help identify families with a “thrombophilic
tendency.” FVL will be found in half of these
families, who may also share one of the rarer thrombophilic
gene variants, or perhaps other sources of genetic
susceptibility that have yet to be identified.
[i]
Botto LD, Khoury MJ. Commentary: facing the
challenge of gene-environment interaction: the two-by-four table and beyond.
Am J Epidemiol 2001;1016-1020.
[ii]
Khoury MJ, Flanders WD. Nontraditional epidemiologic
approaches in the analysis of gene-environment
interaction: case-control studies without
controls. Am J Epidemiol 1996;144:207-213.
home
| back to overview | print
all answers | Reader's Forum
|