PAPERS
Delaying
childbearing:
effect
of
age
on
fecundity
and
outcome
of
pregnancy
Boukje
M
van
Noord-Zaadstra,
Caspar
W
N
Looman,
Hans
Alsbach,
J
Dik
F
Habbema,
Egbert
R
te
Velde,
Jan
Karbaat
Toeqepast
Natuurwetenschapp4
Onderzoek
Institute
(
Preventive
Health
Ca
Child
Health
Divisio
PO
Box
124,
2300
AC
Leiden,
The
Netherlands
Boukje
M
van
Noord
Zaadstra,
Msc,
epiden
Department
of
Publi
Health
and
Social
Medicine,
Erasmus
University
Medical
S
Rotterdam,
The
Netherlands
Caspar
W
N
Looman
statistician
J
Dik
F
Habbema,
PH
professor
of
medical
dec
sctences
Department
of
Obst
and
Gynaecology,
University
Hospital,
Utrecht,
The
Nether
Hans
Alsbach,
Msc,
medical
data
manager
Egbert
R
te
Velde,
MI
professor
of
reproductiv
medicine
Cryo
Biological
Laboratories,
Bijdor
Barendrecht,
The
Netherlands
Jan
Karbaat,
MD,
med
director
Correspondence
to:
Mrs
van
Noord-Zaad,
BMJ
1991;302:1361-5
Abstract
Objectives-To
study
the
age
of
the
start
of
the
fall
(critical
age)
in
fecundity;
the
probability
of
a
preg-
nancy
leading
to
a
healthy
baby
taking
into
account
the
age
of
the
woman;
and,
combining
these
results,
to
determine
the
age
dependent
probability
of
getting
a
healthy
baby.
Design-Cohort
study
of
all
women
who
had
entered
a
donor
insemination
programme.
Setting-Two
fertility
clinics
serving
a
large
part
of
The
Netherlands.
Subjects-Of
1637
women
attending
for
artificial
insemination
751
fulfilled
the
selection
criteria,
being
married
to
an
azoospermic
husband
and
nulliparous
and
never
having
received
donor
insemi-
nation
before.
Main
outcome
measures-The
number
of
cycles
before
pregnancy
(a
positive
pregnancy
test
result)
elijk
or
stopping
treatment;
and
result
of
the
pregnancy
of
(successful
outcome).
are,
Results-Of
the
751
women,
555
became
pregnant
and
461
had
healthy
babies.
The
fall
in
fecundity
was
estimated
to
start
at
around
31
years
(critical
age);
after
12
cycles
the
probability
of
pregnancy
in
a
woman
aged
>31
was
0-54
compared
with
0-74
in
a
tiologist
woman
aged
20-31.
After
24
cycles
this
difference
had
decreased
(probability
of
conception
0-75
in
c
women
>31
and
0-85
in
women
20-31).
The
prob-
ability
of
having
a
healthy
baby
also
decreased-by
3-5%
a
year
after
the
age
of
30.
Combining
both
these
chool,
age
effects,
the
chance
of
a
woman
aged
35
having
a
healthy
baby
was
about
half
that
of
a
woman
aged
25.
MSC,
Conclusion-After
the
age
of
31
the
probability
of
MSC,
conception
fails
rapidly,
but
this
can
be
partly
[D,
compensated
for
by
continuing
insemination
for
cision
more
cycles.
In
addition,
the
probability
of
an
adverse
pregnancy
outcome
starts
to
increase
at
about
the
same
age.
etrics
Introduction
rlands
Female
fecundity
(the
ability
to
conceive)
is
gener-
ally
acknowledged
to
decrease
with
increasing
age,
but
D,
the
beginning
of
the
fall
in
fecundity
has
not
been
)e
pinpointed
to
a
specific
age.
Such
information
is
of
importance
to
the
increasing
number
of
women
who
are
delaying
childbearing.
In
naturally
selected
popu-
lations
studying
the
decrease
in
fecundity
caused
by
biological
factors
is
confounded
by
diminished
sexual
activity
with
age
and
possibly
also
by
a
decrease
in
male
lical
fertility.
Schwartz
and
Mayaux
studied
the
age
effect
in
women
treated
by
artificial
donor
insemination.
Their
data
suggest
that
reduced
fecundity
starts
around
the
age
range
31-35,
but
their
follow
up
time
of
12
cycles
stra.
was
relatively
short.
Older
women
who
continue
treatment
for
a
much
longer
period
may
eventually
conceive.2
Moreover
there
is
evidence
that
advancing
maternal
age
has
an
adverse
effect
on
the
outcome
of
pregnancy
because
of
a
higher
abortion
and
perinatal
mortality.3
Our
study
of
a
cohort
of
women
receiving
donor
insemination
was
undertaken
to
examine
the
age
of
the
start
of
the
fall
in
fecundity
(critical
age),
the
prob-
ability
of
a
pregnancy
leading
to
a
healthy
baby
taking
into
account
the
age
of
the
woman,
and
the
age
dependent
probability
of
getting
a
healthy
baby,
by
combining
the
critical
age
and
the
probability
of
a
pregnancy
leading
to
a
healthy
baby.
Subjects
and
methods
Two
fertility
clinics
(A and
B)
serving
different
geographic
areas
participated
in
the
study.
Women
were
referred
to
the
clinics
by
a
general
practitioner
or
a
specialist.
The
source
population
consisted
of
all
women
who
attended
the
clinics
for
artificial
donor
insemination
between
January
1973
and
the
years
when
the
protocols
were
changed
from
treatment
with
fresh
semen
to
frozen
semen
(1980
for
clinic
A
and
1986
for
clinic
B).
A
total
of
1637
women
entered
the
artificial
insemi-
nation
programme
in
the
two
clinics,
1036
in
clinic
A
and
601
in
clinic
B.
We
studied
the
751
women,
444
from
clinic
A
and
239
from
clinic
B,
who
satisfied
the
selection
criteria-that
is,
women
who
were
married
to
an
azoospermic
husband,
were
nulliparous,
and
had
never
received
artificial
insemination
before.
Insemination
procedures
were
usually
conducted
in
every
subsequent
menstrual
cycle.
Women
were
inseminated
intracervically,
and
timing
of
insemina-
tion
was
estimated
on
the
basis
of
the
basal
body
temperature
chart
and
by
judging
the
quality
of
cervical
mucus.
On
average
the
number
of
insemina-
tions
per
cycle
was
three
for
clinic
A
and
two
for
clinic
B.
Fresh
semen
was
used
from
donors
aged
25-45
with
a
proved
fertility
(having
fathered
at
least
one
child)
and
with
sperm
properties
satisfying
the
World
Health
Organisation
criteria.
In
clinic
A
the
specialists
were
free
to
prescribe
sup-
plementary
treatment
such
as
induction
of
ovulation
by
clomiphene
if
a
woman
did
not
conceive
after
a
few
cycles.
In
clinic
B
induction
of
ovulation
was
largely
confined
to
women
who
proved
to
have
anovulatory
cycles
or
who
had
very
long
cycles
during
an
obser-
vation
period
before
treatment.
If
induction
of
ovu-
lation
was
started
the
treatment
was
continued
in
subsequent
insemination
cycles.
If
a
woman
did
not
conceive
hysterosalpingography
was
performed
after
the
sixth
cycle
and
laparoscopy
after
the
twelfth
cycle.
All
women
were
followed
until
the
end
of
treatment
with
artificial
insemination.
The
end
was
defined
as
the
result
of
each
woman's
last
insemination
cycle,
being
either
a
confirmed
pregnancy
(success)
or
stopping
treatment
without
pregnancy
(failure).
Only
first
con-
BMJ
VOLUME
302
8
JUNE
1991
1361
1-
0'
a)
0*
o
o.
0)
a)
QE6
0-
.0
0.
r
a)
a)
,;-
0
<%,
E
0
0
0
1(
ceptions
as
a
result
of
the
artificial
insemination
were
used
for
the
analysis.
Fifteen
women
who
did
not
report
the
results
of
their
last
insemination
cycle
were
recorded
as
a
failure
up
to
their
last
but
one
cycle.
If
menstruation
did
not
start
after
two
weeks
the
women
were
instructed
to
contact
the
clinic
for
a
pregnancy
test.
If
the
test
result
was
positive
the
woman
was
assumed
to
be
pregnant
and
was
referred
to
her
family
doctor.
Women
were
asked
to
report
the
outcome
of
pregnancy.
Successful
pregnancy
was
defined
as
a
pregnancy
leading
to
the
birth
of
a
healthy
child.
METHODS
OF
ANALYSIS
The
cumulative
probability
of
conception
by
in-
semination
cycle
was
calculated
using
Kaplan-Meier
estimai
groups
used
f
0
No
FIG
1-Cumulative
rate
of
pregnancy
accordii
(dotted
line
indicates
that
less
than
five
women
1
!
ct
(0
c
0)
a)
L-
a
CC
(-
41)
1
.
.
.
20
25
FIG
2-Pregnancy
rates
per
cycle
(hazard)
acc
as
1
-00.
Hazards
by
year
of
age
are
indicat
Example
of
use
offigure-women
aged
33
hac
has
in
each
cycle
75%
of
the
chance
of
a
womn
insemination
and
probability
of
conception.'
We
have
used
the
term
hazard
as
shorthand
for
the
more
informative
but
longer
term
"pregnancy
rate
per
cycle."
We
fitted
models
with
a
gradual
fall
of
fertility
(see
model
1.2
in
appendix)
and
a
model
that
estimates
a
constant
initial
rate
of
pregnancy,
an
age
at
which
the
decrease
begins
(critical
age),
and
the
rate
of
decrease
after
that
age
(see
model
1.3
in
appendix).
Though
model
1.2
describes
the
biological
pattern
of
the
fall
in
fertility,
model
1.3
would
answer
the
question
of
the
practising
clinician-namely,
at
what
age
does
the
fall
start?
The
dependency
of
the
probability
of
a
successful
pregnancy
on
age
was
analysed
by
logistic
regression
methods
(see
model
2.1
in
appendix).6
tes
for
the
two
clinics
separately
and
for
four
age
Results
;.4
Proportional
hazard
regression
analysis
was
FECUNDITY
or
analysing
the
relation
between
age
at
first
All
751
women
were
followed
until
the
end
of
treatment;
316
(7
1%)
women
in
clinic
A
and
239
(78%)
25-29
women
in
clinic
B
became
pregnant.
Only
15
women
--------
did
not
report
the
result
of
their
last
insemination
<25
cycle.
The
last
intake
of
women
treated
with
fresh
30-34
donor
semen
took
place
in
1985.
None
of
these
women
were
still
being
treated
at
the
end
of
1989.
The
women
in
clinic
B
were
on
average
1.5
years
older
than
those
in
>34
clinic
A;
this
difference
was
significant
(p<005).
There
was
a
systematic
difference
between
clinic
A
and
clinic
B
in
overall
cumulative
probability
of
conception.
The
cumulative
pregnancy
rates,
corrected
for
censor-
ing,
after
12
and
24
cycles
were
67%
and
80%
respectively
for
clinic
A
and
76%
and
91%
respectively
for
clinic
B.
Few
women
in
clinic
B
received
insemination
after
24
cycles;
clinic
A
continued
treatment
for
longer
and
after
35
cycles
had
a
cumulative
pregnancy
rate
of
90%.
The
systematic
difference
in
cumulative
pregnancy
,0
,5
,
rates
between
the
clinics
was
not
related
to
differential
1
5
20
25
30
35
40
treatment
with
age
by
clinic.
Therefore
all
751
women
of
insemination
cycles
were
used
for
the
analysis
of
the
effects
of
age.
ig
to
number
of
insemination
cycles
in
woen
of
different
ages
The
youngest
woman
entering
the
study
was
18
iwere
still
receiving
artificial
insemination)
years
old,
the
oldest
42.
Table
I
gives
the
age
distribution
of
all
the
women.
Figure
1
gives
the
curves
of
cumulative
probability
of
conception
for
the
four
age
*
1
Woman
groups.
The
cumulative
pregnancy
rates
after
12
cycles
were
0-75
(women
aged
<24),
0-72
(25-29),
0-72
*
10
Women
(30-34),
and
0
49
(>34).
The
probability
of
conception
for
each
age
group
was
also
calculated
with
the
*
100
Women
proportional
hazard
model
(see
model
1.1
in
appendix).
Hazard
fitedyritcalagThe
difference
in
probability
of
conception
between
Hazard
fitted
by
critical
age
model
the
oldest
and
the
other
age
groups
was
significant
Too
few
women
for
reliable
estimate
(likelihood
ratio
test).
The
oldest
age
group
had
about
half
the
hazard
(pregnancy
rate/cycle)
of
the
other
age
groups
(95%
confidence
interval
0-31
to
0-77).
In
a
more
detailed
analysis
we
estimated
the
preg-
nancy
chances
for
each
separate
year
of
age
(fig
2).
This
gave
a
dispersed
plot
as
many
points
were
based
on
a
small
number
of
observations.
The
general
shape,
however,
corroborates
the
use
of
a
model
in
which
the
\
probability
of
conception
falls
after
a
certain
age.
The
model
with
a
gradual
fall
seemed
to
fit
the
data
quite
well
(model
1.2).
At
age
31
the
probability
of
concep-
tion
was
95%
of
the
initial
level,
and
at
older
ages
the
fall
became
increasingly
steeper.
The
choice
of
95%,
however,
was
quite
arbitrary.
For
the
model
with
the
critical
age
(model
1.3)
the
fit
was
better
than
for
model
1.2.
Our
estimation
method
gave
a
critical
age
of
31
years.
After
the
age
of
31
the
chance
of
conception
per
,
*
*
cycle
fell
by
about
12%
each
year
of
age
(model
1.3).
30
35
40
To
assess
the
goodness
of
fit
we
investigated
three
aspects
of
the
regression
models.
Firstly,
the
models
Age
of
woman
1.2
and
1.3
should
be
adequate
for
describing
the
ording
to
age.
Mean
hazard
for
women
aged
20-30
was
scaled
relation
with
age.
Figure
2
shows
that
both
models
are
ted
by
blocks;
size
of
block
refers
to
number
of
observations
aie..
I
a
relative
hazard
of0-
75
indicating
that
a
woman
of
that
age
valid
in
this
respect.
The
critical
age
model
performed
an
aged
20-30
of
getting
pregnant
only
slightly
better.
Secondly,
the
proportional
BMJ
VOLUME
302
8
JUNE
1991
TABLE
I-Age
distribution
and
successful
conception
in
women
receiving
artificial
insemination
No
(%)
who
Age
No
of
became
(years)
women
pregnant
18-24
111
86(78)
25-29
390
298
(76)
30-34
201
146(75)
35-42
49
25(51)
Total
751
555
(74)
1
1362
0-15
FIG
3
-Basei
aged
under
3.
0
1
00
u
0
90-
a)
0
80j
a
n
7n
-
related
to
age
was
estimated
by
logistic
regression.
.
1
Pregnancy
Analysis
was
done
on
all
555
women
who
conceived,
0
Pregnancies
461
women
reporting
a
healthy
child
and
71
women
*
10
Pregnancies
reporting
adverse
outcomes.
The
23
women
with
-
*
unknown
outcome
were
first
analysed
as
successes,
100
Pregnancies
assuming
that
adverse
outcomes
would
have
been
reported.
There
was
a
significant
decrease
in
the
chance
of
having
a
healthy
child
after
the
age
of
30
(see
model
2.1
in
appendix).
If
a
pregnancy
occurred
*
women
aged
under
30
had
an
89%
chance
of
having
a
U
*
.
healthy
baby.
Thereafter
this
chance
decreased
by
*
^
about3
5%eachyear.
Eu
*
*There
was
no
significant
difference
between
clinic
A
E
y
*
.
and
clinic
B
in
the
probability
of
a
baby
being
healthy,
-
.
after
correction
for
the
difference
in
age
between
*
.
-
y
*
clinics
(hazard
ratio
1-01;
95%
confidence
interval
0
61
to
1
69).
No
significant
difference
was
found
in
successful
pregnancy
among
women
who
took
dif-
g
ferent
times
to
conceive
(hazard
ratio
1
01/completed
5
10
15
20
25
30
35
cycle;
0-96
to
1-06)
nor
between
women
with
preg-
No
of
insemination
cycles
nancies
that
ensued
after
induction
of
ovulation
and
line
chance
offirst
pregnancy
(hazard)
for
each
cycle
after
failures
in
previwus
cycles
for
women
those
with
pregnancies
occurring
during
spontaneous
I
attending
clinic
A
cycles
(hazard
ratio
1
03;
0
55
to
1-94).
We
plotted
the
probabilities
for
year
of
age
in
the
same
way
as
in
figure
2.
This
showed
the
critical
age
....................
>model
to
be
adequate
for
modelling
outcome
of
pregnancy.
Combining
the
chances
of
pregnancy
and
\"
-Pregnancy
regardless
of
a
pregnancy
being
successful,
gives
a
curve
of
the
\
of
outcome
falling
chance
of
successful
pregnancy
with
age
(fig
4).
060\
n
0'40
Pregnancy
resulting
N
0230
in
healthy
child
CZ
0.20-
>
010
cron
Sn
n
on
Age
of
woman
FIG
4-Rate
of
pregnancy
by
age
with
regard
to
outcome
hazards
should
be
stable
for
subsequent
insemination
cycles.
We
performed
the
analysis
for
the
first
three
cycles
and
for
later
cycles
separately.
In
both
subsets
the
start
of
the
fall
in
fertility
was
at
age
31.
This
proves
the
stability
to
be
adequate.
Thirdly,
according
to
the
models
the
difference
between
the
clinics
should
be
constant
with
age.
When
the
data
set
was
divided
into
clinics
the
critical
age
for
clinic
A
was
31
and
for
clinic
B
33
years.
This
difference,
however,
was
not
significant.
Figure
3
shows
how
the
chance
of
pregnancy
falls
in
each
subsequent
cycle
in
the
whole
group.
This
fall
results
from
the
selection
process
by
which
highly
fecund
women
tend
to
get
pregnant
earlier
than
less
fecund
women.
OUTCOME
OF
PREGNANCY
Of
the
555
women
who
conceived,
532
women
reported
a
result
of
pregnancy
and
23
(5%)
did
not.
Table
II
gives
the
reproductive
outcomes
according
to
age
group.
The
fall
in
the
probability
of
a
healthy
child
TABLE
iI-Outcome
of
pregnancy
according
to
age
No
(%)
of
women
aged:
Outcome
<25
25-29
30-34
>34
Total
Spontaneous
abortion
9
(12)
30
(10)
20
(14)
6
(25)
65
(12)
Stillbirth
1(0
3)
2
(1)
3
(1)
Congenitalanomaly
1(0-3)
1(07)
1(4)
3(1)
Healthy
child
69(88)
257
(89)
118
(84)
17
(71)
461
(87)
Known
outcome
78
(100)
289
(100)
141
(100)
24
(100)
532
(100)
Unknown
outcome
8
9
5
1
23
Total
86
298
146
25
555
Discussion
We
studied
the
effect
of
age
on
female
fecundity
and
outcome
of
pregnancy
in
women
receiving
donor
insemination.
By
excluding
confounding
variables,
such
as
diminished
sexual
activity
with
age
and
various
degrees
of
male
subfertility,
these
women
provide
a
better
opportunity
to
study
potentially
predictive
variables
with
regard
to
fecundity
and
outcome
than
do
naturally
selected
populations.78
To
prevent
selection
bias
toward
a
population
of
lower
fecundity,
only
women
married
to
azoospermic
husbands
and
who
had
not
been
treated
elsewhere
were
admitted
to
the
study.9
Although
we
cannot
fully
explain
the
difference
in
cumulative
conception
rates
between
clinics
A
and
B,
the
difference
in
policy
on
induction
of
ovulation
is
obvious.
As
cycles
with
and
without
ovulation
induced
by
clomiphene
were
not
randomly
compared
we
can
only
speculate
on
the
adverse
effect
of
clomiphene
in
women
with
normal
ovulatory
cycles.
10
We
added
a
model
with
a
continuous
fall
in
fecund-
ity
with
age
(model
1.2);
we
proved
this
to
be
no
improvement
on
the
critical
age
model
(1.3).
The
model
with
an
abrupt
start
of
fall
performed
slightly
better
and,
more
importantly,
determined
when
the
fall
in
fecundity
starts.
The
finding
of
a
critical
age
of
31
years
does
not
seem
to
agree
with
figure
1,
which
shows
a
significant
decrease
of
fecundity
only
in
women
older
than
34
and
not
in
those
aged
30-34.
However,
the
good
result
of
the
30-34
group
is
caused
by
the
fact
that
the
pregnancy
rates
in
women
aged
30
and
31
were
(possibly
by
chance)
rather
high
(fig
2).
We
could
not
account
for
the
effect
of
possible
confounders
such
as
smoking,
alcohol,
or
coffee
consumption
because
this
informa-
tion
was
not
systematically
available.
Therefore,
to
be
cautious
we
will
interpret
the
critical
age
to
be
around
31.
The
critical
age
around
31
for
decreasing
fecundity
falls
within
the
ranges
found
by
Howe
et
al
in
women
stopping
contraception,"
by
Schwartz
and
Mayaux
in
a
population
of
women
undergoing
donor
insemination,'
and
by
Menken
et
al
in
historical
populations.'2
The
critical
age
did
not
change
after
correcting
for
the
use
of
BMJ
VOLUME
302
8
JUNE
1991
a1)
0
C)
C:
co
c
a)
U)
0
CI
N
(C
a)
C
a)
Cm
m
0*10-
0*05-
AC.j
Qv
oli
1363
C
CI
C)
0.
03)
C)
a)
E
C0
>6
it)
E-
0O
0
0.
0.
>31
0
10-r
0
)
5
10
12
15
20
25
30
35
40
No
of
insemination
cycles
FIG
5-Cumulative
rate
of
pregnancy
in
women
aged
more
than
31
(n=
131)
and
31
oryounger
(n=620)
ovulation
induction,
nor
did
length
of
menstrual
cycle
affect
our
finding.
What
are
the
reasons
for
decreasing
female
fertility
with
age?
In
vitro
fertilisation
clearly
shows
that
the
number
of
oocytes
retrieved
and
the
rates
pregnancy
obtained
decrease
with
age.'3
14
Lower
pregnancy
rates
may
be
due
to
a
uterine
factor
interfering
with
implantation.
Reports
on
successful
oocyte
donation
in
women
over
40,
however,
suggest
that
oocyte
quality
rather
than
uterine
environment
is
the
limiting
factor
in
older
women.'5
Subtle
deterioration
of
oocytes
probably
starts
before
the
age
of
35.
Anovulation,
oligomenorrhoea,
or
cycle
irregularities
apparently
are
later
reflections
of
the
same
process
of
deterioration.
The
models
that
we
used
do
not
account
for
the
fact
that
during
a
series
of
insemination
cycles
the
ages
of
the
women
increase.
However,
the
age
effect
was
only
a
1%
fall
per
cycle
after
the
age
of
31.
Because
few
women
received
artificial
insemination
for
a
long
series
of
cycles
we
would
not
expect
including
aging
during
the
series
to
improve
the
models.
The
start
of
the
fall
around
age
31
means
that
women
older
than
31
will
take
longer
to
become
pregnant
(eventually)
than
would
younger
women.
We
divided
the
population
into
groups
younger
or
equal
and
older
than
31
(fig
5).
The
pattern
of
the
curve
suggests
that
a
policy
of
stopping
treatment
in
the
older
age
group
is
not
advisable;
treatment
for
longer
than
12
cycles
seems
to
be
worth
while.
After
12
cycles
the
pregnancy
rate
in
the
older
women
increased
from
54%
to
over
75%
at
24
cycles;
the
pregnancy
rate
in
the
younger
women
increased
in
the
same
period
from
74%
to
85%.
Because
of
the
small
number
of
women
completing
24
cycles
or
more,
however,
the
suggestion
in
our
data
that
older
women
will
eventually
have
the
same
pregnancy
rate
as
younger
women
must
be
interpreted
with
caution.
Apart
from
the
fecundity
we
also
found
that
the
chance
of
successful
pregnancy
(resulting
in
a
healthy
baby)
in
nulliparous
women
decreases
after
the
age
of
30.
Classifying
pregnancies
in
women
in
whom
the
outcome
was
unknown
as
unsuccessful
did
not
alter
this
conclusion.
Most
fertility
specialists
are
aware
of
the
fall
in
fecundity
and
the
chance
of
successful
pregnancy
with
increasing
age.
When
counselling
women
considering
delaying
childbearing
we
should
know
the
combined
effect
on
the
likelihood
of
giving
birth
to
a
healthy
child.
We
estimate
that
the
relative
chance
per
cycle
of
a
35
year
old
woman
giving
birth
to
a
healthy
baby
is
50%
that
of
women
of
25
(fig
4).
Older
women
who
do
not
get
pregnant
in
the
first
cycle
can
get
pregnant
in
one
of
the
next
cycles;
as
time
taken
to
conceive
was
not
related
to
outcome
of
pregnancy
the
differences
between
older
and
younger
women
in
the
cumulative
probability
of
having
a
healthy
child
will
become
smaller
after
every
subse-
quent
cycle.
Recently
Berkowitz
et
al
found
that
delayed
childbearing
poses
little,
if
any,
increased
risk
of
adverse
neonatal
outcome,'6
but
they
did
not
assess
spontaneous
abortions
and
chromosomal
abnor-
malities.
The
positive
association
of
unsuccessful
pregnancy
with
age
greater
than
30
in
our
study
was
largely
due
to
the
contribution
of
spontaneous
abortions.
Study
designs
ascertaining
the
reproductive
outcome
after
the
gestational
period
when
spontaneous
abortions
are
most
likely
to
occur
will
be
less
prone
to
show
an
age
effect.
Our
results
are
based
on
a
cohort
of
women
who
received
donor
insemination.
The
number
of
women
conceiving
in
this
population
is
known
to
be
lower
than
in
a
random
population.2
In
our
analysis,
however,
it
is
the
critical
age
and
the
decrease
with
age
in
the
probability
of
pregnancy
and
of
pregnancy
being
successful
that
are
at
issue.
We
cannot
find
any
reason
why
the
critical
age
in
this
population
would
differ
from
that
of
a
population
conceiving
naturally.
Women
receiving
artificial
insemination
still
represent
the
most
feasible
way
to
study
the
age
effect
on
female
fecundity
and
outcome
of
pregnancy.
The
question
of
how
long
women
can
wait
can
now
be
answered:
around
the
age
of
31
the
probability
of
pregnancy
in
nulliparous
women
starts
to
fall.
Older
women
can
get
pregnant,
but
at
a
slower
rate
than
younger
women.
Women
over
30
face
a
decreasing
chance
of
having
a
healthy
child.
We
thank
Dr
Donna
Baird
(Epidemiology
Branch,
National
Institute
for
Environmental
and
Health
Sciences,
Research
Triangle
Park,
North
Carolina,
USA)
and
Dr
Theo
Stijnen
(Department
of
Biostatistics,
Erasmus
University
Medical
School,
Rotterdam)
for
their
helpful
suggestions.
We
also
thank
Dr
Pauline
Verloove-Vanhorick
(Child
Health
Division,
TNO
Institute
of
Preventive
Health
Care,
Leiden)
and
the
staff
of
the
participating
clinics
for
their
support
and
Mrs
Ria
Huls
van
Vliet
(TNO
Institute
of
Preventive
Health
Care,
Leiden)
for
secretarial
help.
Appendix
The
regression
models
used
in
this
paper
were
estimated
using
GLIM
software.6
For
non-linear
models
(1.2
and
1.3)
only
estimates
can
be
calculated-that
is,
no
standard
errors.
The
three
models
for
the
Cox
regression
calculations
were
as
follows.
.
MODEL
FOR AGE
GROUPS
h=ke(12
L2+(3
L3+04.L4+x.K
Where:
h=Hazard
of
pregnancy
in
cycle
t-that
is,
the
probability
of
pregnancy
in
cycle
t
assuming
that
no
pregnancy
has
occurred
until
then.
k=Baseline
hazard
in
cycle
t
(not
modelled).
The
baseline
group
consists
of
women
from
age
group
1
and
clinic
A.
a=
Parameter
for
the
difference
between
the
hazards
for
age
group
i
and
age
group
1.
L-=Dummy
variable,
which
takes
the
value
1
when
the
woman
belongs
to
age
group
i
and
zero
otherwise.
x=
Parameter
for
the
difference
between
the
two
clinics.
K=Dummy
variable,
which
takes
the
value
1
when
the
woman
is
treated
by
clinic
B
and
zero
otherwise.
The
age
groups
are
<25
(1),
25-29
(2),
30-34
(3)
s35
(4).
The
baseline
hazard
(k)
was
found
to
fall
with
increasing
number
of
cycle.
It
reflects
a
selection
effect:
women
with
high
fecundity
get
pregnant
leaving
after
some
cycles
a
higher
proportion
of
women
with
average
low
fecundity
to
remain
in
the
population.
Parameter
estimates
are
t2=
-0
09,
CC3=
-0-07,
Ct4=
-0-71,
and
x=0
28.
When
the
number
of
age
groups
is
equal
to
the
number
of
different
ages
in
the
study
population
this
model
estimates
the
hazards
as
indicated
by
the
blocks
in
figure
2.
BMJ
VOLUME
302
8
JUNE
1991
1364
1.2
MODEL
WITH
AGE
AS
A
CONTINUOUS
VARIABLE
h=,XeBl±xK
Where:
,B=
Parameter
for
the
strength
of
the
effect
of
age.
iT=
Parameter
for
the
abruptness
of
the
start
of
fall.
X=Age
(now
a
continuous
explanatory
variable).
This
model
fits
a
smooth
curve
that
is
almost
constant
at
first
and
then
starts
to
fall
with
an
abruptness
depending
on
the
value
of
jT.
The
best
estimate
for
T
was
13.1;
addition
of
lower
order
polynomials
did
not
improve
the
fit,
which
means
that
a
rise
in
fecundity
of
younger
women
is
not
supported
by
our
data.
The
log
likelihood
was
3493.8.
1.3
CRITICAL
AGE
MODEL
h=
Xel.m+x.K
m=(l-yfl>y)
Where:
X=Critical
age
(age
where
the
fall
starts).
3=Rate
of
fall
after
the
critical
age.
m=
Number
of
years
older
than
critical
age,
or
zero
if
younger.
Only
discrete
values
of
the
critical
age
were
investigated.
The
value
with
the
maximum
likelihood
was
selected.
Therefore
standard
errors
could
not
be
calculated.
Figure
3
shows
the
baseline
hazards
(X).
The
relative
hazards
(eP`
m+xK)
are
indicated
by
the
line
in
figure
2.
The
estimates
were
y=31,
,B=-0136,
and
x=0272.
The
log
likelihood
was
3491.9
with
the
same
degrees
of
freedom
as
model
1.2.
The
estimated
hazard
of
a
certain
woman
in
a
certain
cycle
(h)
can
be
calculated
by
multiplication
of
the
baseline
hazard
(X),
depending
on
the
number
of
the
cycle
and
the
relative
hazard
(e3
1m+x
K)
depending
on
the
age
and
clinic.
We
realise
that
the
fact
that
our
population
consisted
of
a
mixture
of
fertile
and
infertile
women
is
theoretically
incom-
patible
with
the
use
of
the
Cox
regression
model.
17
The
mixed
character
of
the
population
should
lead
to
a
constant
shift
in
the
proportional
hazards
between
young
and
old
women
with
time
(insemination
cycle).
However,
modelling
interaction
of
the
parameters
with
time
(that
is,
assuming
13
and
y
are
different
in
a
first
and
second
period
of
time)
does
not
lead
to
a
significantly
better
fit.
We
therefore
conclude
that
the
theoretical
incompatibility
does
not
cause
problems
in
this
particular
analysis.
2.1
LOGISTIC
MODEL
FOR
PROBABILITY
OF
SUCCESSFUL
PREGNANCY
e
a+Bi.m+x.K
1
+ea+B.m+x.K
Where
p
is
the
probability
of
successful
pregnancy.
The
other
symbols
have
the
same
meaning
as
in
model
1.3.
The
estimator
for
difference
between
clinics
(x)
was
not
significant.
The
parameters
of
the
model
without
x
were
ct=2.
13,
P=
-0-25,
and
y=30.
1
Schwartz
D,
Mayaux
MJ.
Female
fecundity
as
a
function
of
age.
N
Engl3r
Med
1982;306:404-6.
2
Bongaarts
J.
Infertility
after
age
30:
a
false
alarm.
Fam
Plann
Perspect
1982;14:75-8.
3
Stein
ZA.
A
woman's
age:
childbearing
and
child
rearing.
Am
J7
Epidemiol
1985;121:327-43.
4
Kaplan
EK,
Meier
P.
Non-parametric
estimation
from
incomplete
observa-
tions.Journal
of
the
American
Statistics
Association
1958;53:457-8
1.
5
Cox
DR.
Regression
models
and
life
tables.
Journal
of
the
Royal
Statistical
Society
1972;34(series
B):
187-202.
6
Aitkin
MD,
Anderson
AO.
Statistical
modelling
in
GLIM.
Oxford:
Clarendon
Press,
1989.
7
Spira
A.
The
decline
of
fecundity
with
age.
Maturitas
1988;suppl
1:15-22.
8
Van
Noord-Zaadstra
BM,
Karbaat
J,
te
Velde
ER,
Habbema
JDF,
van
der
Maas
PJ.
The
study
of
risk
habits
in
reproductive
and
perinatal
epidemiologic
research:
the
use
of
a
donor
inseminated
population
of
women.
Paediatr
Perinat
Epidemiol
1989;3:11-8.
9
Emperaire
JC,
Gauzere-Soumireu
E,
Audebert
AJM.
Female
fertility
and
donor
insemination.
Fertil
Steril
1982;37:90-3.
10
Yoshimara
Y,
Hosoi
Y,
Atlas
SJ,
Wallach
EE.
Effect
of
clomiphene
citrate
on
in
vitro
ovulated
ova.
Fertil
Steril
1986;45:800-4.
11
Howe
G,
Westhoff
C,
Vessey
M,
Yeates
D.
Effects
of
age,
cigarette
smoking,
and
other
factors
on
fertility:
findings
in
a
large
prospective
study.
BM3'
1985;290:
1697-1700.
12
Menken
J,
Trussell
J,
Larsen
U.
Age
and
infertility.
Science
1986;233:
1389-95.
13
Harrison
KL,
Breen
TM,
Hennesy
JF,
et
al.
Patient
age
and
success
in
a
human
IVF
programme.
Aust
N
ZJ
Obstet
Gynaecol
1989;29:326-8.
14
Medical
Research
Intemational
Society
for
Assisted
Reproductive
Technology,
the
American
Fertility
Society.
In
vitro
fertilization-embryo
transfer
(IVF-ET)
in
the
United
States:
1989
results
from
the
IVF-ET
registry.
Fertil
Steril
1991;55:14-23.
15
Sauer
MV,
Paulson
RJ,
Lobo
RA.
A
preliminary
report
on
oocyte
donation
extending
reproductive
potential
to
women
over
40.
N
Engl
J
Med
1990;323:
1-60.
16
Berkowitz
GS,
Skovron
ML,
Lapinski
RH,
Berkowitz
RL.
Delayed
child-
bearing
and
the
outcome
of
pregnancy.
N
Engl3r
Med
1990;322:659-64.
17
Lamb
EJ,
Hagen
N,
Pauker
SG.
The
mean
interval
to
conception:
a
measure
of
utility
for
the
analysis
of
decisions
involving
fertility.
Am
J
Obstet
Gynecol
1989;160:
1470-8.
(Accepted
18
April
1991)
Office
of
Population
Censuses
and
Surveys,
London
WC2B
6JP
Anna
McCormick,
FFPHM,
senior
medical
statistician
BMJ
1991;302:1365-7
Unrecognised
HIV
related
deaths
Anna
McCormick
Abstract
Objectives-To
establish
whether
follow
up
of
deaths
from
selected
HIV
related
causes
could
increase
the
number
of
cases
of
HIV
infection
reported
to
the
Public
Health
Laboratory
Service
Communicable
Disease
Surveillance
Centre
(CDSC),
and
to
estimate
the
proportion
of
deaths
among
HIV
positive
men
that
occurred
in
men
who
were
not
known
to
be
HIV
positive
at
the
time
of
death
by
the
person
who
signed
the
death
certificate.
Design-Follow
up
of
draft
death
entries
received
by
the
Office
of
Population
Censuses
and
Surveys
on
which
one
of
11
medical
or
external
causes
likely
to
be
related
to
HIV
was
stated;
letters
were
sent
to
the
people
who
signed
the
certificates.
The
respondents
were
invited
to
report
men
known
to
have
been
HIV
positive
who
were
not
already
on
the
CDSC
register.
Setting-England
and
Wales.
Subjects-Men
aged
15-54
who
died
in
February
1989
to
July
1989
with
one
of
the
11
selected
HIV
related
diseases
as
cause
of
death
on
their
death
certificates.
Main
outcome
measures-Number
of
men
reported
to
the
CDSC
as
a
result
of
this
follow
up;
estimate
of
excess
deaths
due
to
an
HIV
related
cause;
estimate
of
the
proportion
of
excess
deaths
that
occurred
in
those
who
were
not
known
to
be
HIV
positive
at
the
time
of
death.
Results-Replies
were
received
for
473
deaths
(86%).
Forty
were
for
men
known
to
have
been
HIV
positive,
31
of
whom
had
been
reported
to
CDSC
by
the
time
they
died;
six
were
subsequently
reported.
The
respondent
did
not
know
that
the
deceased
was
HIV
positive
for
20
(35%)
of
the
57
excess
deaths
in
men
for
whom
one
of
the
medical
causes
was
stated
and
41
(93%)
of
the
44
excess
deaths
in
men
for
whom
one
of
the
external
causes
was
stated.
Conclusion-Follow
up
of
death
registrations
is
not
an
efficient
way
of
increasing
the
number
of
cases
of
HIV
infection
reported
to
CDSC.
Between
35%
and
60%
of
HIV
positive
people
for
whom
certain
causes
are
stated
may
be
dying
without
HIV
positivity
having
been
diagnosed.
There
may
be
implications
for
those
caring
for
people
with
these
conditions
and
those
who
carry
out
postmortem
examinations.
Introduction
On
p
1375
I
report
that
mortality
from
95
selected
causes
increased
by
25%
between
1984
and
1989
among
single
men
aged
15-54
and
there
is
evidence
that
BMJ
VOLUME
302
8
JUNE
1991
1365
