American Journal of Epidemiology
Copyright O 1997 by The Johns HopWns University School of Hygiene and Public Health
All rights reserved
Vol. 145, No. 9
Printed In U.SA.
Environmental Factors, Reproductive History, and Selective Fertility in
Farmers' Sibships
Petter Kristensen,1 Lorentz M. Irgens,2 and Tor Bjerkedal3
In a national study of births to farmers in Norway, grain farming was associated with short gestational age
(21-24 weeks). An impact of selective fertility and maternal heterogeneity on the association was suspected
but could not be assessed further in a traditional birth-based design. Thus, analyses based on the mother as
the observational unit were performed. A total of 45,969 farmers with a first birth in 1967-1981 were followed
for subsequent births and perinatal mortality. A perinatal loss increased farmers' likelihood to continue to
another pregnancy, but this selective fertility was less dominant than in the general population due to a higher
baseline fertility. The effect of the mother's reproductive history on the grain farming-midpregnancy delivery
association was analyzed in 59,338 farmers with more than one single birth in 1967-1991. A history of preterm
birth (<37 weeks) in previous or subsequent pregnancies both was an independent determinant of midpregnancy delivery and also increased the effect of grain exposure. Nongrain farmers with a history of only
term births had 1.3 midpregnancy deliveries per 1,000 births; grain farmers with a history of only term births
had 1.8 cases per 1,000 (odds ratio (OR) 1.4, 95% confidence interval (Cl) 1.0-1.9); nongrain farmers with a
history of preterm birth had 6.8 cases per 1,000 (OR 5.5, 95% Cl 4.0-7.6), whereas grain farmers with a history
of preterm birth had 13.7 cases per 1,000 (OR 11.0, 95% Cl 7.7-15.9). Selective fertility had only a marginal
impact on the association. The study demonstrates that a maternally based design can contribute in the
assessment of joint effects of environmental and maternal factors. Am J Epidemiol 1997;145:817-25.
agriculture; effect modifiers (epidemiology); environmental exposure; epidemiologic methods; fertility;
gestational age; reproductive history
tend to be overrepresented at higher birth orders. Because a previous loss is associated with perinatal loss
in subsequent pregnancies, this selection process will
inflate fetal and infant mortality at high birth orders, as
observed in cross-sectional studies with the birth as the
observation unit (1). The impact of selective fertility
on birth order-specific perinatal mortality can be estimated in a longitudinal design with the mother, rather
than the birth, as the unit of observation; this has been
quantified in the Norwegian population by Skjaerven et
al. (1). A mother-based design has also been suggested
in the study of environmental reproductive hazards
(1). Provided that an adverse reproductive effect of an
environmental hazard is stronger among women with
a history of perinatal loss (i.e., stronger than an additive effect), the effect of exposure may be exaggerated
by selective fertility (1). The impact of a greater number of births among the exposed women can be controlled in a cross-sectional design by standardizing for
birth order, but this will not be adequate to assess the
effect of potential heterogeneity present in higher birth
orders. It is important to clarify such potential interaction between an environmental agent and an endogenous factor, not only to avoid the bias of selective
In a study of perinatal outcomes in births to fanners,
we found that birth in midpregnancy (gestation weeks
21-24) was associated with grain farming; the results
support the hypothesis of a labor-inducing effect of
mycotoxins from grain field fungi. In a traditional
analysis based on the pregnancy as the unit of observation, we were unable to account for the mother's
reproductive history, although we suspected a selective fertility effect as well as biologic heterogeneity
since the association between grain farming and midpregnancy delivery was stronger for high birth orders
and multiple pregnancies (unpublished results).
The tendency for a woman to replace a perinatal loss
causes selective fertility (1). In populations in which
birth control is common, women with a previous loss
Received for publication June 18, 1996, and accepted for publication December 5, 1996.
Abbreviations: Cl, confidence interval; OR, odds ratio.
1
National Institute of Occupational Health, Oslo, Norway.
2
Medical Birth Registry of Norway, University of Bergen, Norway.
3
Section for Preventive Medicine, Institute of General Practice
and Community Medicine, University of Oslo, Norway.
Reprint requests to Dr. Petter Kristensen, National Institute of
Occupational Health, P.O.B. 8149 Dep, N-0033 Oslo, Norway.
817
818
Kristensen et al.
fertility but also to understand the mechanisms involved.
In the present report, births to farmers were established with the mother as the unit of observation. The
extent and impact of selective fertility on perinatal
mortality in farmers were compared with results reported for the general population of Norway (1).
Another purpose was to examine the influence of the
mother's previous and subsequent reproductive history
on the association between the exposure factor in grain
and the duration of pregnancy. To our knowledge, the
impact of the mother's reproductive history on the
adverse effects of environmental exposures in a longitudinal mother-based design has not been reported.
Therefore, we considered it important to compare results in the cross-sectional (birth-based) and longitudinal design.
MATERIALS AND METHODS
The linkage between and within registers used in
this study was made possible by the unique national
identification number assigned to all residents of Norway. All farm holders born after 1924 were identified
in the five agricultural or horticultural censuses performed by Statistics Norway between 1969 and 1989.
The farmers' spouses were identified in the Central
Population Register. The resulting file of farmers and
spouses was linked to the Medical Birth Registry of
Norway, identifying 113,949 women born after 1924
with a total of 192,417 births in 1967-1991. The
Medical Birth Registry of Norway contains information on all births since 1967 with 16 or more completed weeks of gestation (2). The birth records include identification of both parents and information on
maternal health, pregnancy, and birth. The registry
also includes maternal records, which are aggregates
of all birth records pertaining to each mother.
The agricultural censuses provided the basis for the
exposure indicators that were assigned to the births.
The census information closest to the time of conception was linked to each birth. Twenty-six percent of
the births occurred in families that cultivated grain.
We also used the annual reports from the Norwegian
Grain Corporation for the harvest years between 1966
and 1991, which provide information on climate and
grain crop quality in the main grain regions (southeast
and middle Norway), where 95 percent of the grain is
produced. The corporation characterized the harvest
quality for each season and each region into three
levels (high, medium, and poor), which indicated a
low, medium, and high probability of field fungi
growth and mycotoxin formation in grain. These categories were assigned to each birth according to the
location of the farm and the date of birth.
Selective fertility in farmers was compared with that
of the general population, as reported earlier (1); the
analysis file was therefore established according to
Skjaerven et al. (1). By means of the mother's identity,
we created maternal records by adding subsequent
births to 45,969 single first births in 1967-1981. Maternal records including multiple births were excluded.
The maternal records were allocated to cohorts according to the year of the first birth: Mothers whose first
birth was in 1967-1971 were followed for subsequent
births through 1979; first births in 1972-1976 were
followed through 1984; and those with a first birth in
1977-1981 were followed through 1989. The followup period of the latest cohort was 5 years longer than
that of Skjaerven and coworkers (1). Follow-up was
restricted to birth orders less than five. A birth was
classified as a perinatal death if it ended as a late
abortion (gestational age 16-27 weeks), a stillbirth
(gestational age 28-46 weeks or unknown), or early
neonatal death (the first week of life). The outcome
was expressed as perinatal mortality (perinatal deaths
per 1,000 at risk). Fertility was based on the occurrence of all subsequent births registered in the Medical
Birth Registry. Birth order was defined from the mother's report at the first delivery and from the date of
birth for the remainder. All births (n = 108,524) were
stratified according to birth order, the outcome of the
previous birth(s), and the period of the first birth. The
proportion of women who continued to another pregnancy is referred to as the "continuation proportion."
The continuation proportion of women with no perinatal death was considered as baseline fertility and
was compared with women who had one or more
previous perinatal deaths as a ratio. The continuation
ratio is a measure of the strength of selective fertility.
Birth order-specific and total perinatal mortality were
adjusted for the effects of selective fertility according
to Skjaerven et al. (1, appendix 1). Continuation proportions, continuation ratios, and the birth order- and
period-specific impact of selective fertility on perinatal mortality were compared for farmers and previously published (1) results for the general population.
In the study of the impact of reproductive history on
the effect of the grain factor, we established maternal
records of single births of known gestational age and
birth orders lower than five in birth sibships that
included more than one birth in 1967-1991. Thus, a
total of 59,338 maternal records with 140,568 births
was left for further analysis; 46,064 records included
birth order one, whereas the remaining 13,274 mothers
had additional births before 1967. The exposure indicators used were grain farming (yes/no) and grain
harvest quality (high, medium, poor; restricted to
southeast and middle Norway). The outcome variable
Am J Epidemiol
Vol. 145, No. 9, 1997
Environmental Factors and Reproductive History
was a delivery in weeks 21-24 of gestation (yes/no).
The indicator of maternal exposure susceptibility was
one or more preterm births (<37 gestation weeks) in
previous or subsequent pregnancies (termed "preterm
history" or "term history"). In the classification of the
susceptibility factor, information on gestational age of
single births beyond birth order four was also used.
All analyses were performed with the Epicure statistical software program (3) and were based on birth
order-specific contingency tables produced in the
DATAB procedure. In the study of the grain factor, we
used odds ratio (OR) estimates as an approximation of
relative risk. We stratified the contingency tables and
fitted logistic regression models for period, region,
type of municipality (agricultural or not), and maternal
age. As these factors did not add substantially to the
interpretation of the results, the crude OR estimates
are presented throughout. However, summary results
across birth order were adjusted for birth order. Approximate 95 percent confidence intervals (CIs) served
as measures of the stability of the point estimates.
RESULTS
Selective fertility
The fertility for various strata of birth order and
previous perinatal outcome is shown in table 1. The
baseline continuation proportion (for farmers without
a perinatal death) decreased from 0.88 to 0.44 and 0.20
after the first, second, and third births, respectively.
For all birth orders, women with a previous loss were
more likely to go on to another pregnancy; the continuation ratio was only 1.05 after first births, increasing to 1.8-1.9 after second births and 2.0-3.8 after
third births.
Perinatal mortality in the different strata are also
shown in table 1. For birth orders two to four, there
was a sharp contrast, with a low perinatal mortality for
women with no previous loss and two- to eightfold
increased perinatal mortality among women with a
previous loss. For birth order four, some of the strata
included few mothers; for all fourth birth mothers with
a previous loss, the perinatal mortality was 43.6 per
1,000 as compared with 16.0 per 1,000 for mothers
with no loss from any of the three previous births.
We examined grain farmers separately for all strata
of birth order and reproductive history: They had
lower continuation proportions and slightly higher
perinatal mortality than in all farmers in almost all
strata (data not shown). The baseline continuation
proportions for grain farmers decreased from 0.83
after first births to 0.14 after third births, and selective
fertility was stronger than in all farmers except after
the first birth. Grain farmers with no previous loss had
Am J Epidemiol
Vol. 145, No. 9, 1997
819
a continuation ratio of 2.1 after second births (all farmers,
1.8) and 2.6 after third births (all farmers, 2.2).
The continuation proportions for all farmers, stratified on period of the first birth, are presented in table
2. The continuation proportions were not very different from the total in any of the three periods, and there
was no clear time trend. For women with no previous
loss, mothers with a first birth in 1967-1971 had only
moderately higher continuation proportions of additional births than mothers who had a first birth in the
later periods; and mothers with their first birth in
1977-1981 had marginally higher continuation proportions than mothers with a first birth in 1972-1976.
The perinatal mortality declined over time, as shown
in table 3. For birth order one, the main decrease was
between 1972-1976 and 1977-1981; for birth orders
two and three, mortality decreased markedly between
the first and the second periods, but not between the
second and the third periods; for birth order four,
mortality decreased steadily during the entire period.
To adjust for the effect of selective fertility on perinatal mortality, adjusted mortality was calculated,
based on the assumption that mothers with a history of
earlier loss had the same continuation proportions as
mothers with no previous loss (1, appendix 1). The
lower adjusted mortality obtained was close to the
observed values for second births in all time strata. For
birth order three, the observed mortality was 4.4 percent higher than the adjusted values; this increase was
restricted to the first two periods. In contrast, the
difference between observed and adjusted mortality
was 15.1 percent for birth order four, as the relative
difference increased from 7 percent in the first period
to 34.4 percent in the last period.
Effects of exposure to grain
hi total, 266 mothers experienced 277 midpregnancy
deliveries (2.0 per 1,000); six nongrain farmers and
five grain farmers had two midpregnancy deliveries.
The proportions show a slight U-shaped pattern by
birth order (table 4). The proportion of midpregnancy
deliveries among grain farming mothers was increased
by 56 percent; this increase was moderate (OR 1.311.53) for birth orders one through three and nearly
fourfold for birth order four.
Stratification on previous or subsequent reproductive history revealed, as expected, that the proportion
of midpregnancy deliveries differed by preterm history. For different birth orders, nongrain farmers had
1.0-1.5 cases per 1,000 in the term history stratum and
3.8-8.2 cases per 1,000 in the preterm history stratum
(table 4). The grain farmers with a term history had a
moderate increase of midpregnancy deliveries (1.8 per
1,000; OR 1.41, 95 percent CI 1.04-1.92); birth order
i
50
p
in
cf
45,969
L
(45,049)
D
(920)
0.88
0.88
0.92
Reference
1.05
Continuation Continuation
proportion
ratio
13.8
12.7
62.5
PNM
•D, perinatal loss; L, live birth surviving the first week.
Total
20.0
PNM
Outcome
(no.)
40,389
LL
(39,038)
LD
(503)
0.45
0.44
0.83
0.79
0.79
DD
(53)
DL
(795)
Continuation
proportion
Outcome
(no.)
Second births
Reference
1.9
1.8
1.8
Continuation
ratio
12.7
11.5
33.5
25.4
95.2
PNM
LLL
(17,004)
18,293
LLD
(198)
LDL
(404)
LDD
(14)
DLL
(615)
DLD
(16)
0.21
0.20
0.55
0.39
0.43
0.39
0.56
0.74
0.75
DDD
(4)
DDL
(38)
Continuation
proportion
Outcome
(no.)
Third births
Reference
2.8
2.0
2.2
2.0
2.9
3.8
3.8
Continuation
ratio
3,873
3,322
109
158
238
28
No.
19.9
16.0
55.9
50.6
37.8
111.1
35.7
PNM
Fourth births
TABLE 1. Perinatal mortality (PNM) per 1,000, continuation proportions, and continuation ratios by outcome* of previous births among mothers In farming In Norway,
1967-1989, and with a first birth in 1967-1981 (n = 45,969)
Environmental Factors and Reproductive History
TABLE 2. Continuation proportion by period of the first birth
(1967-1971,1972-1976,1977-1981) and by outcome* ol
previous births among farmers in Norway, 1967-1989 (n o
45,969)
Rrel birth
(no. at mothers)
Second
birth
DD0.68
DL 0.79
LD0.82
DDD; DDL
DLD; DLL;
LDD; LDL;
LLD
0.41
L0.89
LL0.47
LLL
0.21
1972-1976(15,534)
DD 0.96
D0.94
DL 0.80
LD 0.84
DDD; DDL;
DLD; DLL;
LDD; LDL;
LLD
0.44
L0.87
LL0.41
LLL
0.18
DD0.96
DL 0.79
LD 0.85
DDD; DDL
DLD; DLL;
LDD; LDL;
LLD
19721976
19771381
TnJol
IUUU
One
Observed PNM*
22.5
20.4
14.8
20.0
Two
Adjusted PNM
Observed PNM
Increaset (%)
15.7
15.8
0.1
11.6
11.6
0.8
13.0
13.1
0.3
13.7
13.8
0.4
Three
Adjusted PNM
Observed PNM
Increaset (%)
16.1
16.8
4.9
8.4
8.9
5.9
9.4
9.4
<0.1
12.2
12.7
4.4
Four
Adjusted PNM
Observed PNM
Increaset (%)
21.3
22.8
7.0
15.2
18.7
22.8
11.2
15.0
34.4
17.3
19.9
15.1
• PNM, perinatal mortality per 1,000.
t Difference between observed and adjusted PNM is expressed
as the percentage that the adjusted PNM is Increased, due to
selective fertility, to reach the observed PNM.
1977-1981 (10,664)
D0.92
Year of drat birth
19671971
order
1967-1971 (19,771)
D0.91
TABLE 3. Perinatal mortality adjusted for the effects of
selective fertility, by birth order and year of first birth among
women In farming In Norway (n = 45,969)
n,^_
Third
birth
821
0.45
with a maternal term history (from 29 to 19 percent)
than for births of mothers with a preterm history (from
LL0.44
LLL
0.19
27 to 23 percent). To assess the effect of selective
* D, perinatal loss; L, live birth surviving the first week.
fertility, we recalculated the OR estimates for grain
farming on the assumption that the proportion of births
to mothers exposed to grain was similar to that of the
four was the only stratum with a substantial increase.
Grain farmers with a preterm history had 13.73 mid- birth order one-term history stratum, irrespective of
birth order and preterm history status. The summary
pregnancy deliveries per 1,000; the risk was doubled
OR in the term history stratum changed from 1.41
(OR 1.99, 95 percent CI 1.31-3.01) and was strongest
(table 4) to 1.44, and that in the preterm history strafor birth orders one and four. The combined effects of
tum remained unchanged.
preterm history and exposure to grain were also calculated in a model with interaction terms, using the
Analysis of the effect of harvest quality revealed an
nongrain and term history category as reference. The
overall moderate effect for high and medium harvest
OR estimate in the grain-term history stratum was as quality, whereas the poor harvest quality had 4.99
shown in table 4, 5.49 (95 percent CI 3.99-7.55) in the
midpregnancy deliveries per 1,000 and a threefold
nongrain-preterm history stratum and 11.04 (95 per- increased OR (table 5). Results stratified on reproduccent CI 7.65-15.93) in the grain-preterm history stratum.
tive history indicated a dose-response pattern among
mothers with a term history, whereas mothers in the
The results in table 4 are based on information on
exposure from the census closest in time to the con- grain-preterm history stratum had a substantial increase in risk, irrespective of harvest quality, although
ception. Analyses were also performed on the basis of
the medium and poor quality levels were based on
information on exposure for the first recorded birth, or
if the mother was ever exposed to grain, or if she was small numbers.
The gestational week distribution for all births was
exposed at all births. Since 89 percent of those ever
calculated for the four groups classified by reproducexposed were always exposed, this provided little adtive history and grain farming (figure 1). As expected,
ditional information, and the results were only marthe two groups with a preterm history had considerginally different from those presented in table 4 (data
ably shorter gestation duration than the groups without
not shown).
a preterm history. The distributions of the preterm
The number of births to mothers exposed to grain in
history groups appeared to be bimodal, including a
the different strata in table 4 reveals that grain farming
peak in the left-hand tail in weeks 21-22. This early
affected 29 percent of first births and only 20 percent
peak
was particularly high in the grain-preterm history
of fourth births. The decline was stronger for births
L0.87
Am J Epidemiol
Vol. 145, No. 9, 1997
1
8
CO
p
1
5;
5;
a
c
2,204
678
21,475
6,705
31,062
23,679
7,383
6.84
13.73
1.27
1.79
1.97
1.71
2.68
Per
1,000
1.69
2.57
40
19
12
6
5.44
8.85
1.30
1.94
1.90
59
28
13
Per
1,000
No.
Third births
Cases
56
38
121
62
100
277
177
No.
1.63
1.49
1.53
OR
1.99
1.41
1.56
OR*
0.61-4.36
0.77-2.87
0.88-2.63
95% Cl
1.31-3.01
1.04-1.92
1.22-2.00
95% Cl*
1,041
309
2,104
8,900
12,354
9,941
2,413
Births
2,117
780
30,546
12,621
46,064
32,663
13,401
Births
OR, odds ratio; Cl, confidence interval,
t Only term births (237 weeks of gestation) In previous or subsequent pregnancies.
t Preterm births In previous or subsequent pregnancies.
Preterm history
No grain
Grain
Term history
No grain
Grain
Total
No grain
Grain
Births
8,183
2,767
95,041
34,577
Term hlstoryt
No grain
Grain
Preterm hlstoryt
No grain
Grain
140,568
103,224
37,344
Total
No grain
Grain
Births
Cases
Total
4
5
9
7
12
13
25
No.
17
14
47
25
103
64
39
No.
1.54
1.98
2.24
1.96
2.91
Per
1,000
1.01
3.33
2.02
1.31
4.97
Per
1,000
3.84
16.18
Cases
Fourth births
8.03
17.95
Cases
First births
4.26
3.30
3.82
OR
2.26
1.29
1.49
OR
1.14-15.98
1.23-8.87
1.74-8.38
95% Cl
1.11-4.60
0.79-2.09
1.00-2.22
95% Cl
1,000
2,821
34,120
13,147
51,088
36,941
14,147
Births
23
13
37
17
90
60
30
No.
1.08
1.29
1.76
1.62
2.12
Per
1,000
8.15
13.00
Cases
Second births
1.60
1.19
1.31
OR
0.81-3.18
0.67-2.11
0.84-2.03
95% Cl
TABLE 4. Odds ratios for mldpregnancy delivery (gestatlonal weeks, 21-24) of grain farming by birth order and reproductive history, restricted to 59,338 mothers In
farming with more than one single birth In Norway, 1967-1991
Environmental Factors and Reproductive History
823
TABLE 5. Odds ratios for midpregnancy delivery (gestatlonal weeks 21-24 ) by harvest quality In grain
farming and reproductive history among farmers In Norway, 1967-1991*
Cases
Category
No.
No.
Total
No grain
High harvest quality
Medium harvest quality
Poor harvest quality
Term history}
No grain
High harvest quality
Medium harvest quality
Poor harvest quality
Preterm history§
No grain
High harvest quality
Medium harvest quality
Poor harvest quality
Per
1,000
Crude
ORt
95% Clt
103,224
20,590
8,184
4,207
177
51
22
21
1.71
2.48
2.69
4.99
1
1.44
1.57
2.88
Reference
1.06-1.97
1.00-2.44
1.83-^.53
95,041
121
27
15
16
1.27
1.41
1.98
4.11
1
1.11
1.55
3.19
Reference
0.73-1.69
0.91-2.66
1.89-5.38
56
6.84
16.04
19,094
7,581
3,890
8,183
1,496
603
317
24
7
5
11.61
15.77
1
2.32
1.69
2.28
Reference
1.43-3.76
0.77-3.72
0.9O-5.73
• Analysis restricted to women with more than one single birth during 1967-1991; harvest quality classified
only for grain farmers in southeast and middle Norway,
t OR, odds ratio; Cl, confidence Interval.
$ Only term births (237 weeks of gestation) in previous or subsequent pregnancies.
§ Preterm blrth(s) In previous or subsequent pregnancies.
group; during weeks 20-25, the relative difference
was larger between the grain-preterm history group
and the nongrain-preterm history group than the difference between the latter and the two no preterm
history groups. Whereas a clear effect of exposure to
grain was evident and extended until week 32 in the
preterm history group, the effect in the term history
groups was not impressive and included only weeks 22
and 23.
DISCUSSION
Selective fertility
As in the general population (1), female farmers in
Norway with a previous perinatal loss tend to have
higher continuation proportions. The perinatal mortality for specific strata of previous reproductive history
was comparable to that of the general population (1)
and was increased by several times for mothers who
had experienced an earlier loss in comparison with
those who had not. Thus, the higher perinatal mortality
observed at high birth orders might be due partly to
selective fertility.
In some respects, female farmers differ from the
general population. The main difference is a higher
probability of having a second and third birth among
mothers who had no previous loss (baseline continuation proportion). Whereas the baseline proportion in
the general population was 0.76 after the first birth and
0.30 after the second birth (1), the corresponding proAm J Epidemiol
Vol. 145, No. 9, 1997
portions among farmers were 0.88 and 0.44; farmers
also had a higher continuation proportion after a third
birth, but the difference was smaller (0.20 vs. 0.17
(I))The differences were due mainly to a different secular fertility trend; whereas the general population had
a marked decline in continuation proportions from
1967 through 1984 (1), such a trend appeared to be
virtually absent in the farming population. The differences in selective fertility between the general population and farmers can be reasonably explained by
sibship size desire (4).
As a consequence, the observed perinatal mortality
among farmers can be explained to a lesser degree by
selective fertility, especially for second and third
births. Under the assumption that the fertility for all
women was equal to women with no previous loss, the
total observed perinatal mortality in the general population was increased by about 2 percent due to selective fertility (1), compared with 1 percent among the
farmers.
In the general population, the part of the observed
perinatal mortality explained by selective fertility increased markedly in relative (but not absolute) terms
between 1967 and 1984 (1). We found no such increase for second and third births among farmers, but
there was an increase over time for fourth births.
In reproductive epidemiologic studies, it is important to recognize different reproductive patterns across
824
Kristensen et al.
N o grain, term history
Grain, term history
N o grain, preterm history
Grain, preterm history
g
XI
5en
o
CD
2>
16
18
20
22
24
26
28
30
32
34
Gestational age (weeks)
FIGURE 1. The left-hand tail of the distribution curves for gestational age in all births to farmers in Norway, 1967-1991. Births are divided
Into four groups according to the mother's reproductive history of previous and subsequent births (preterm history or not) and grain
cultivation on the parents' farm. No grain, term history, n = 103,224; grain, term history, n = 37,344; no grain, preterm history, n = 8,183;
grain, preterm history, n = 2,767.
population groups. These differences also should be
taken into account when the impact of selective fertility is assessed.
Effect of exposure to grain
The results suggest heterogeneity in the effect of
grain farming on midpregnancy delivery, depending
on whether the mother had a preterm delivery in
previous or subsequent pregnancies. The effect was
stronger in both absolute and relative terms for mothers with a preterm history than for those without.
Examination of the impact of regional and seasonspecific climate conditions and grain quality, which
was used as an indicator of mycotoxin level, showed
additional differences between mothers with and
without a preterm history: For births of mothers with
a term history, the results indicate a dose-response
pattern for decreasing harvest quality, whereas a more
uniform effect was seen on births to mothers with a
preterm history. The credibility of the modifying effect of a preterm history on the grain-midpregnancy
delivery association is strengthened by the distributions of gestation duration shown in figure 1; the effect
of exposure to grain in the preterm history stratum not
only was stronger but also extended over gestational
weeks 20-32.
A biologically plausible explanation is that a maternal factor that increases the probability of preterm
birth also increases the susceptibility to the exposure
factor in grain farming. The results are in accordance
with the mycotoxin hypothesis: The risk is dependent
on the mycotoxin level for mothers without the maternal susceptibility factor, whereas mothers with this
factor are at increased risk even at low levels. This
interpretation is also consistent with an increased effect of the grain factor in multiple pregnancies, which
could be considered a pregnancy-specific susceptibility factor. These interpretations must be evaluated
with caution, however, as only crude proxies of true
exposures were available; and even if some mycotoxins induce labor and are strong reproductive toxicants
in domestic animals (5), no measurements of occupational exposures are available. Other interpretations
are also possible, e.g., that the exposure factor had a
long-term effect on the mother. We were not able to
explore this further because few grain farmers (11
percent) changed into other types of farming.
Am J Epidemiol
Vol. 145, No. 9, 1997
Environmental Factors and Reproductive History
We considered the possibility that the grainmidpregnancy delivery association was biased by selective fertility. Grain farmers had a reproductive pattern closer to that of the general population than other
farmers, and selective fertility was stronger than for all
farmers. Intuitively, we believe that selective fertility
could not be responsible: The grain effect was present
in all birth orders, and first births were influential in
the summary results. The proportion of grain farmers
declined with increasing birth order, but more so for
births to mothers with a term history than to those with
a preterm history. The combined bias that could result
was assessed by recalculating the data in table 4 under
the assumption that the proportion of births to grain
farmers did not change. This procedure resulted in
only marginal changes in the OR estimates.
The preterm history variable is likely to be an imperfect indicator of true exposure susceptibility.
Besides, complete reproductive histories were not
available, so that an unknown fraction of those classified as having a term history had preterm births
before 1967 or after 1991. Misclassification of covariates will bias the exposure-outcome association and
may induce false heterogeneity (6). However, it can be
shown that covariate misclassification will attenuate
true heterogeneity given that the misclassification of
the covariate is nondifferential with regard to exposure
and outcome and that the exposure occurrence is similar in different strata of the susceptibility indicator (as
in our situation). Therefore, false heterogeneity is not
a likely explanation of the differing grain effect in the
term and preterm history strata. However, it is likely
that misclassification of the susceptibility indicator
inflates the true risk estimate of grain farming in the
term history stratum. The gestational age distribution
(figure 1) could suggest that the limited grain effect in
the term history stratum was fortuitous. This suggestion is strengthened by the fact that the interpretation
of an effect that involves only a few gestational weeks
is biologically obscure; however, the clear doseresponse gradient with harvest quality contradicts this.
Use of the mother as the observation unit offers
some advantages over the traditional birth-based study
in research on environmental exposures and adverse
reproductive events. The main advantage is that the
relations between susceptibility factors linked to reproductive history and environmental exposures are
easier to study. The relation between such effect modifiers and exposures is important to disclose the true
nature of the exposure effect and will provide an
additional opportunity for mechanistic interpretations
of the effect. Studies restricted to first births have been
proposed to eliminate problems due to selective fertility (1); however, the heterogeneity of adverse reproAm J Epidemiol
Vol. 145, No. 9, 1997
825
ductive events in primiparous women will not be identified in studies based on first births. Only the
inclusion of subsequent births will provide this information (i.e., the impact of later preterm births on the
proportion of midpregnancy deliveries at first births,
table 4). Also, the effect of selective fertility can be
controlled in studies based on mothers. However, due
to the small total impact of selective fertility on perinatal mortality, selective fertility is unlikely to have
much impact on the association between environmental exposures and adverse reproductive outcomes.
Maternally based studies may involve some problems related to their establishment and administration.
It is also easier to run into power problems; mothers
with only one identified birth will not provide information on reproductive history, and inclusion of very
high birth orders adds to the computational problems.
Studies restricted to single births will not reveal effects
in multiple pregnancies, as in the present population.
In conclusion, there are arguments in favor of performing analyses based on both the birth and the
mother.
ACKNOWLEDGMENTS
This work was supported by the Research Council of
Norway (grant 103542/110).
The authors thank Professors Petter Laake and Tor
Norseth for their valuable supervision and advice at all
stages of the study; Anne S. Bye for file preparation and
linking in Statistics Norway; Ole-Henrik Edland for file
linking and preparation in the Medical Birth Registry; Dag
Sandli at Norwegian Grain Corporation for classification of
the grain harvests; and Elisabeth Heseltine for revising the
manuscript.
REFERENCES
1. Skjsrven R, Wilcox AJ, Lie RT, et al. Selective fertility and
the distortion of perinatal mortality. Am J Epidemiol 1988;
128:1352-63.
2. Bjerkedal T. The medical birth registry of Norway. In: Mednick SA, Baert AE, eds. Prospective longitudinal research: an
empirical basis for the primary prevention of psychosocial
disorders. Published on behalf of the World Health Organization Regional Office for Europe. Oxford: Oxford University
Press, 1981.
3. Preston DL, Lubin JH, Pierce DA, et al. Epicure. Seattle, WA:
Hirosoft International Corporation, 1993.
4. Kravdal 0. Who has a third child in contemporary Norway? A
register-based examination of sociodemographic determinants. Oslo-Kongsvinger: Central Bureau of Statistics of Norway, 1990. (Report 90/6).
5. Hood RD, Szczech GM. Teratogenicity of fungal toxins and
fungal-produced antimicrobial agents. In: Keeler RF, Tu AT,
eds. Handbook of natural toxins. Vol. 1: Plant and fungal
toxins. New York, NY: Marcel Dekker, Inc., 1983:201-35.
6. Greenland S. The effect of misclassification in the presence of
covariates. Am J Epidemiol 1980; 112:564-9.