Original Contribution

Inammation and Triglycerides Partially Mediate the Effect of Prepregnancy

Body Mass Index on the Risk of Preeclampsia
Lisa M. Bodnar
1,2
, Roberta B. Ness
1,2,3,4
, Gail F. Harger
1,2
, and James M. Roberts
1,2,3
1
Department of Epidemiology, Graduate School of Public Health, University of Pittsburgh, Pittsburgh, PA.
2
Magee-Womens Research Institute, Pittsburgh, PA.
3
Department of Obstetrics, Gynecology, and Reproductive Sciences, School of Medicine, University of Pittsburgh,
Pittsburgh, PA.
4
Department of Medicine, School of Medicine, University of Pittsburgh, Pittsburgh, PA.
Received for publication March 22, 2005; accepted for publication June 28, 2005.
The objective of this study was to quantify the mediating role of inammation and triglycerides in the association
between prepregnancy body mass index (weight (kg)/height (m)
2
) and preeclampsia. The authors conducted a
nested case-control study of 55 preeclamptic women and 165 pregnant controls fromthe Pregnancy Exposures and
Preeclampsia Prevention Study (Pittsburgh, Pennsylvania, 19972001). Serum samples collected at 20 weeks
gestation were analyzed for levels of C-reactive protein and triglycerides. The adjusted odds ratio (AOR) from a
multivariable conditional logistic regression model assessing the total effect of body mass index on preeclampsia
risk was compared with the AOR from the same model after results were controlled for C-reactive protein, triglyc-
erides, and confounding factors (direct-effects model). The percentage of the total effect that was mediated through
inammation and triglycerides was calculated as 100 [ln(direct-effects AOR)/ln(total-effects AOR)]. In the total-
effects model, 4- and 8-unit increases in body mass index were associated with 1.7-fold (95% condence interval
(CI): 1.3, 2.3) and 2.9-fold (95% CI: 1.6, 5.2) increases in preeclampsia risk, whereas in the direct-effects model,
these AORs were 1.4 (95%CI: 1.0, 1.9) and 2.0 (95%CI: 1.0, 3.8), respectively. Inammation was a more important
mediator than triglycerides. These ndings suggest that approximately one third of the total effect of body mass
index on preeclampsia risk is mediated through inammation and triglyceride levels.
body mass index; C-reactive protein; inammation; obesity; pre-eclampsia; pregnancy; triglycerides
Abbreviations: AOR, adjusted odds ratio; BMI, body mass index; CRP, C-reactive protein.
Preeclampsia is a multisystemic, pregnancy-specic dis-
order that is diagnosed by new-onset hypertension and pro-
teinuria after 20 weeks gestation. It is a leading cause of
maternal and infant morbidity and mortality worldwide (1),
yet its etiology remains unclear. Numerous studies with vary-
ing denitions of preeclampsia have shown that high mater-
nal prepregnancy body mass index (BMI; weight (kg)/height
(m)
2
) is a strong, modiable risk factor for preeclampsia
(27). Using the same denition of preeclampsia as in the
current study, we previously reported that the risk of pre-
eclampsia rose strikingly from a prepregnancy BMI of 15
to a BMI of 35, such that, compared with a BMI of 21, the
risk of preeclampsia was approximately doubled at a BMI of
26, tripled at a BMI of 30, and halved at a BMI of 18 (2).
Two commonly hypothesized mechanisms underlying the
BMI-preeclampsia relation are hyperlipidemia and inam-
mation, but judging from published reports, neither these
factors nor others have been formally tested as mediators.
Correspondence to Dr. Lisa M. Bodnar, A742 Crabtree Hall, Graduate School of Public Health, University of Pittsburgh, 130 DeSoto Street,
Pittsburgh, PA 15261 (e-mail: Bodnar@edc.pitt.edu).
1198 Am J Epidemiol 2005;162:11981206
American Journal of Epidemiology
Copyright 2005 by the Johns Hopkins Bloomberg School of Public Health
All rights reserved; printed in U.S.A.
Vol. 162, No. 12
DOI: 10.1093/aje/kwi334
Advance Access publication November 3, 2005

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

3
1
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

Overweight is associated with alterations in lipid concen-
trations and an activation of inammatory markers (8, 9),
and both of these metabolic abnormalities are characteristic
of preeclamptic pregnancies before the onset of clinically
evident disease (10, 11). Dyslipidemia and an exaggerated
inammatory response in preeclampsia are thought to con-
tribute to widespread endothelial dysfunction and the sub-
sequent maternal syndrome (12). If adiposity is causally
important in preeclampsia, it may contribute to the patho-
physiology in one of ve ways: 1) solely through an exag-
gerated inammatory response; 2) solely through lipid
abnormalities; 3) through both of these pathways only;
4) through one or both of these pathways while also having
a direct, residual effect on one or more alternative pathways
not mediated by either inammation or hyperlipidemia; or
5) solely through its residual direct effect on alternative
pathways. Our objective was to estimate the proportion of
the effect of BMI on preeclampsia risk that is mediated by
triglyceride concentrations and inammation, as well as the
direct effect on pathways not involving triglycerides and
inammation.
MATERIALS AND METHODS
The Pregnancy Exposures and Preeclampsia Prevention
Study, a prospective cohort study of the pathogenesis of
preeclampsia, was conducted at Magee-Womens Hospital in
Pittsburgh, Pennsylvania (19972001) (13). Eligible women
were those aged 1444 years with a singleton pregnancy
who were planning to deliver at Magee-Womens Hospital.
Women were recruited at 16 weeks of pregnancy after
providing informed, written consent. Atotal of 2,981 women
enrolled in the study (72 percent response rate). An interview-
administered questionnaire at enrollment collected data on
maternal sociodemographic characteristics, medical history,
and prepregnancy behaviors. Nonfasting blood samples were
collected at times of usual blood draws for clinical indica-
tions (initial visit, 1618 weeks, 2629 weeks, and predeliv-
ery). Medical records were abstracted to obtain data on
prepregnancy weight and height, blood pressures and urinary
protein measurements throughout gestation, use of hyperten-
sive medicines, antepartum and delivery events, and neona-
tal outcomes. The study was approved by the institutional
review board of Magee-Womens Hospital.
We conducted a nested case-control study among the
1,198 women in the Pregnancy Exposures and Preeclampsia
Prevention Study who had had no previous deliveries at
more than 20 weeks, no terminations (spontaneous or elec-
tive) in the index pregnancy, and no preexisting medical
conditions (e.g., pregestational diabetes mellitus, chronic
hypertension). We selected women whose index pregnancy
was their rst pregnancy, because the etiology of preeclamp-
sia may differ by parity (14). Of the 59 preeclamptic cases in
this cohort, 55 had an available banked blood sample taken
at 20 weeks gestation and were therefore eligible for this
analysis. We chose 20 weeks as the cutpoint because it pre-
cedes the clinical onset of disease. We individually matched
the 55 cases (3:1) to 165 nonpreeclamptic controls on the
basis of maternal race/ethnicity (non-Hispanic White, non-
Hispanic Black, other), age (within 3 years of the cases
age), gestational age at blood sampling (within 3 weeks of
the cases gestational age), and smoking status at enrollment
(yes, no). Women who quit smoking before enrollment were
considered nonsmokers.
Denition of study variables
Primary exposure variable: prepregnancy BMI. Prepreg-
nancy BMI was based on measured height and maternal self-
report of prepregnancy weight at the initial visit. Prepreg-
nancy BMI was categorized as underweight (BMI <18.5),
normal weight (BMI 18.524.9), overweight (BMI 25.0
29.9), or obese (BMI 30.0) (15).
Primary outcome variable: preeclampsia. Preeclampsia
was dened as gestational hypertension and proteinuria
and return of all abnormalities to normal levels by 12 weeks
postpartum (16). Our research denition emphasized spec-
icity in the diagnosis of preeclampsia. Gestational hyper-
tension was dened as systolic blood pressure persistently at
or above 140 mmHg and/or diastolic blood pressure persis-
tently at or above 90 mmHg for the rst time after 20 weeks
gestation. We determined blood pressure as the average of
the last ve pressures obtained after hospital admission for
delivery, before administration of medications or clinical
perturbations that would alter blood pressure. Proteinuria
was dened as the excretion of more than 300 mg of protein
in 24 hours, a random urine sample of 2 protein, a cathe-
terized urine sample of 1 protein, or a protein:creatinine
ratio greater 0.3. While the gold standard for the quantica-
tion of urinary protein is 24-hour urine collection, 24-hour
urine samples are infrequently collected in clinical practice.
Rather, dipsticks and catheterized samples are commonly
used. We based proteinuria on a level of 2 rather than
1 for a random urine sample so as to minimize misclassi-
cation caused by subjective visual determination of color
on the dipstick. When available, the protein:creatinine ratio,
which is independent of urine concentration, was used.
Protein:creatinine ratio correlates well with 24-hour urinary
protein in most studies (17). This denition of proteinuria
has been used in other large-scale epidemiologic studies of
preeclampsia (18, 19). Blood pressure and protein measure-
ments at 6 weeks postpartum were used to determine
whether abnormalities had been resolved. If abnormalities
persisted, women returned for reevaluation at 12 weeks
postpartum.
Mediating variables: C-reactive protein and triglycerides.
Nonfasting serum samples obtained at 20 weeks gestation
were stored in aliquots at 80C until they were assayed
for C-reactive protein (CRP), a sensitive index of systemic
inammation (20), and triglycerides, an indicator of lipid
status which, if elevated, is a marker of hyperlipidemia (21).
Serum CRP was measured by Quest Diagnostics, Inc. (Pitts-
burgh, Pennsylvania), using a high-sensitivity immunotur-
bidimetric assay. The detection limit of the assay was 0.1
mg/dl. For the 3.6 percent of the sample (n 8) with CRP
concentrations below the detectable limit, we imputed 0.1
mg/dl. The inter- and intraassay variabilities at Quests lab-
oratories were 2.9 percent and 1.8 percent, respectively.
Serum triglyceride level was analyzed enzymatically using
Body Mass Index and Preeclampsia 1199
Am J Epidemiol 2005;162:11981206

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

3
1
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

a triglyceride-GPO (glycerophosphate oxidase) reagent set
from Pointe Scientic, Inc. (Canton, Michigan). The inter-
and intraassay variabilities in our laboratory were 4.7 per-
cent and 5.6 percent, respectively.
Covariates. Maternal race/ethnicity was self-reported.
Data on maternal education (<12, 12, or >12 years), marital
status (married, unmarried), and income in the year before
the index pregnancy were also available. Poverty index ratio
(0130 percent, 131299 percent, or 300 percent) was de-
ned as total household income divided by the year-specic
poverty threshold (22). Women were classied as current
smokers or nonsmokers at enrollment on the basis of self-
report. Women who had quit smoking were included as non-
smokers. Women were classied as either users or nonusers
of multivitamin/mineral supplements in the 6 months before
conception. For assessment of physical activity, women
were asked whether they had usually engaged in leisure-
time physical activity during the year before the index preg-
nancy and, if so, to categorize the usual intensity of this
activity as low, medium, or high.
Statistical analysis
To estimate the proportion of the BMI effect that was
mediated by inammation and triglycerides, we took several
analytical steps. First, we estimated the total effect of BMI
on the risk of preeclampsia. We dened the total effect as the
effect of all potential pathways by which BMI could affect
preeclampsia risk (e.g., inammation, triglycerides, other
lipids, oxidative stress, insulin resistance, etc.). To estimate
the total effect, we tted a conditional, multivariable logistic
regression model that adjusted for confounders of the BMI-
preeclampsia relation (total-effects model; see Appendix).
Second, we estimated the effect of BMI that was directly
mediated through pathways not involving inammation.
To do this, we included CRP and confounders of the CRP-
preeclampsia relation in the conditional logistic total-effects
model (23) to t direct-effects model 1 (Appendix). Be-
cause hyperlipidemia also causes inammation, this BMI
effect also represents the direct effect of BMI that is not
mediated by the effect of elevated triglyceride concentra-
tions on inammation. Longitudinal triglycerides and CRP
measurements, which we lacked, were needed to isolate the
effect of inammation that was independent of hypertri-
glyceridemia. Third, we estimated the direct effect of BMI
that was exerted through pathways not mediated by either in-
ammation or triglycerides. We included CRP, triglycerides,
and confounders of the CRP-preeclampsia and triglyceride-
preeclampsia relations in the conditional logistic total-effects
model to t direct-effects model 2 (Appendix). Unlike
direct-effects model 1, this model attempts to eliminate all
pathways by which triglycerides may mediate the BMI-
preeclampsia relation (through inammation or otherwise).
Finally, after tting each model and obtaining adjusted odds
ratios (AORs), we calculated the percentage of the BMI
effect that was direct (i.e., not mediated through pathways
involving inammation or triglycerides) as [ln(direct-effects
AOR)/ln(total-effects AOR)] 3 100. Because BMI was a
continuous variable in our analysis, the percentage of the
effect that was direct was calculated more simply as (direct-
effects coefcient/total-effects coefcient) 3 100. We may
infer that the remaining proportion (100 percentage direct
effect) is the percentage of the BMI effect that is mediated
through triglycerides and inammation.
A directed acyclic graph (24, 25) (gure 1) was used to
determine potential confounders entered into each model.
Potential confounders included maternal race/ethnicity, age,
smoking status, gestational age at blood sampling, educa-
tion, marital status, income, prepregnancy multivitamin/
mineral use, and prepregnancy physical activity. Effect mod-
ication by race and gestational age of blood sampling were
tested separately in each of the three models using likeli-
hood ratio tests (p < 0.15). Covariates that did not satisfy
our a priori change-in-estimate criterion (a change in the
AOR of more than 8 percent) were considered not to be
inuential and were removed from the models. Confounders
were tested separately in each of the three models. BMI was
linear in the logit of preeclampsia and thus was specied as
a linear term in the models. CRP, triglycerides, and gesta-
tional age at blood sampling were curvilinear in the logit of
preeclampsia and were specied as spline terms.
RESULTS
Preeclamptic women were not signicantly different
from controls in terms of age, race/ethnicity, smoking status,
gestational age at blood sampling, marital status, education,
income, multivitamin use, or physical activity (table 1).
However, the distribution of prepregnancy BMIs varied by
preeclampsia status. Approximately 32 percent of controls
were overweight or obese, whereas 56 percent of preeclamp-
tic women were dened as such. Mean CRP concentration
and the proportion of women with markedly elevated CRP
concentrations (26) were signicantly higher in preeclamp-
tic women than in controls; triglyceride concentrations and
the percentages of women with high triglyceride levels (27)
were not signicantly different.
Overweight and obese women tended to have higher CRP
concentrations at 20 weeks gestation than their leaner
counterparts (table 2). In fact, BMI had a positive, linear
relation with log CRP concentration. After adjustment for
race/ethnicity, marital status, education, smoking status, and
age in a multiple linear regression model, a 5-unit increase
in prepregnancy BMI was associated with a 46 percent rise
in CRP (95 percent condence interval: 33, 61). In contrast,
there was no signicant association between BMI and tri-
glycerides after adjustment for the covariates mentioned
above.
After adjustment for marital status and gestational age of
blood sampling, prepregnancy BMI was strongly associated
with preeclampsia risk in the total-effects model (table 3). A
4-unit rise in BMI was associated with a 70 percent increase
in the risk of preeclampsia; an 8-unit increase in BMI nearly
tripled the risk. Direct-effects model 1, which theoretically
eliminated the inammatory pathway as a mediator of the
BMI-preeclampsia relation, showed attenuated AORs. In
this model, after adjustment for marital status, gestational
age of blood sampling, and prepregnancy multivitamin/
mineral use (a confounder of the CRP-preeclampsia
1200 Bodnar et al.
Am J Epidemiol 2005;162:11981206

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

3
1
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

relation), a 4-unit rise in BMI was associated with a 40
percent increase in preeclampsia risk, and an 8-unit rise
approximately doubled the risk. The reduction in the AORs
in direct-effects model 1 indicated that approximately 69
percent of the BMI effect was directly mediated by path-
ways not involving inammation (likelihood ratio test of
direct-effects model 1 vs. total-effects model: p 0.08).
We can estimate that approximately 31 percent of the resid-
ual effect may be mediated by an exaggerated inammatory
response. Adjustment for triglycerides and the aforemen-
tioned confounders, and thus theoretical elimination of both
the inammatory pathway and the triglyceride pathway, fur-
ther reduced the AORs only slightly. Approximately 64 per-
cent of the BMI effect went directly through pathways not
mediated by triglycerides or inammation (likelihood ratio
test: direct-effects model 2 vs. total-effects model, p 0.03;
direct-effects model 2 vs. direct-effects model 1, p < 0.08).
We can therefore estimate that approximately 36 percent of
the residual effect may be mediated through inammation
and triglyceride levels. We did not observe any effect modi-
cation on the multiplicative scale by race/ethnicity or sample
gestational age.
DISCUSSION
Our results conrm previous ndings that prepregnancy
BMI is a strong independent risk factor for preeclampsia,
and they extend previous ndings by demonstrating that
inammation and triglyceride levels at 20 weeks may be
important mediators of the BMI-preeclampsia association.
Inammation, however, was clearly the more inuential
mediator in our population. Our data suggested that after
adjustment for measured confounders, 31 percent of the
effect of prepregnancy BMI on the risk of preeclampsia
was mediated by a heightened inammatory response and
that 36 percent of the effect was mediated by both a height-
ened inammatory response and increased triglyceride lev-
els. The importance of CRP as a mediator in our population
was not unexpected given that CRP concentrations at 20
weeks were strongly associated with prepregnancy BMI in
multivariable analysis and with preeclampsia in bivariable
analysis, whereas triglyceride concentrations were not sig-
nicantly related to the exposure or outcome. Because mean
triglyceride concentrations were within the range of normal
in the vast majority of cases and controls, it is unlikely that
the effect of inammation that we observed was attributable
to altered triglyceride levels. However, we could not directly
separate these effects because we lacked longitudinal CRP
and triglyceride measurements.
Our results agree with those of studies that have shown
a heightened inammatory response as measured by serum
CRP (28, 29) and proinammatory cytokine (30, 31) levels
before clinically evident preeclampsia, though not all inves-
tigators have reported positive associations between CRP
and preeclampsia incidence (32, 33). Our ndings are in-
consistent with previous reports showing that women who
subsequently develop preeclampsia evidence higher fasting
triglyceride concentrations than controls in the rst and
second trimesters, long before clinical manifestations of
the disorder (3437). The discrepancies we noted in our
triglyceride data may be partially explained by our use of
nonfasting blood samples, our wide range of gestational
ages for study samples (though this was controlled for in our
FIGURE 1. Directed acyclic graph of the mediating role of inammation and triglycerides in the association between prepregnancy body mass
index (BMI; weight (kg)/height (m)
2
) and preeclampsia, Pregnancy Evaluation and Preeclampsia Prevention Study, 19972001. The graph
corresponds to the assumption of no unmeasured confounders given control for measured factors. x, y, and z represent vectors of measured
confounders that may be associated with BMI, lipids, and inammation, respectively. u denotes a vector of unmeasured covariates. Arrows
(directed edges) represent causal effects. Lack of directed edges from u to BMI, lipids, or inammation represents the unveriable assumption of
no unmeasured confounders associated with BMI, lipids, or inammation, conditional on the x, y, and z. If there were directed edges from u to
BMI, lipids, or inammation, u would be considered a vector of unmeasured confounders.
Body Mass Index and Preeclampsia 1201
Am J Epidemiol 2005;162:11981206

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

3
1
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

TABLE 1. Characteristics of pregnant women by case/control status, Pregnancy Exposures and
Preeclampsia Prevention Study, Pittsburgh, Pennsylvania, 19972001
Normal pregnancy
(n 165)
Preeclampsia
(n 55)
p value*
Mean or %
SDy or
95% CIy
Mean or %
SD or
95% CI
Mean age (years) 25.4 6.1 25.3 6.1 0.89
Maternal race/ethnicity (%)
Non-Hispanic White 78.2 78.2 1.0
Non-Hispanic Black 18.2 18.2
Other 3.6 3.6
Smoking status during pregnancy (%)
Nonsmoker 62.4 61.8 0.93
Smoker 37.6 38.2
Mean gestational age (weeks) at
blood sampling 13.8 4.1 13.6 4.4 0.84
Prepregnancy body mass indexz (%)
<18.5 6.7 5.5 0.01
18.524.9 61.2 38.2
25.029.9 18.2 32.7
30 13.9 23.6
Marital status (%)
Unmarried 67.3 60.0 0.32
Married 32.7 40.0
Years of education (%)
<12 13.3 12.7 0.49
12 23.0 30.9
>12 63.6 56.4
Poverty index ratio (%)
<130 47.3 46.9 0.62
131299 21.8 16.3
300 30.9 36.7
Multivitamin use 6 months before
conception (%)
No 44.2 56.4 0.11
Yes 55.8 43.6
Leisure-time physical activity in the
year before conception (%)
No physical activity 59.4 63.6 0.14
Yes, low-intensity 4.9 10.9
Yes, moderate- or high-intensity 35.8 25.5
Mean{ C-reactive protein level (mg/dl) 0.33 0.30, 0.38 0.49 0.37, 0.64 <0.01
Markedly elevated# C-reactive
protein level (%) 10.3 23.6 <0.05
Mean{ triglyceride level (mg/dl) 91.3 84.9, 98.1 97.5 84.5, 112.5 0.38
Elevated** triglyceride level (%) 6.7 9.1 0.55
* Continuous variables were compared by t test, and categorical variables were compared by chi-squared test for
homogeneity.
y SD, standard deviation; CI, condence interval.
z Weight (kg)/height (m)
2
.
Dened as total household income divided by the year-specic poverty threshold (22).
{ Geometric mean with 95% condence interval.
# Dened as a C-reactive protein level greater than 1.0 mg/dl (26).
** Dened as a triglyceride level greater than 200 mg/dl (27).
1202 Bodnar et al.
Am J Epidemiol 2005;162:11981206

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

3
1
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

TABLE 3. Total and direct effects of prepregnancy body mass index* on the risk of preeclampsia (n
220), Pregnancy Exposures and Preeclampsia Prevention Study, Pittsburgh, Pennsylvania, 19972001
Total-effects
modely
Direct-effects
model 1z
Direct-effects
model 2
AOR{ 95% CI{ AOR 95% CI AOR 95% CI
Increase (kg/m
2
) in prepregnancy BMI{
0 1.0# 1.0# 1.0#
2 1.3 1.1, 1.5 1.2 1.0, 1.4 1.2 1.0, 1.4
4 1.7 1.3, 2.3 1.4 1.0, 2.0 1.4 1.0, 1.9
6 2.2 1.5, 3.4 1.7 1.1, 2.9 1.7 1.0, 2.7
8 2.9 1.6, 5.2 2.1 1.1, 4.0 2.0 1.0, 3.8
10 3.8 1.9, 7.8 2.5 1.1, 5.7 2.4 1.1, 5.3
C-reactive protein concentration (mg/dl)**
0.5 1.0# 1.0#
1.0 1.6 1.0, 2.5 1.7 1.0, 2.8
1.5 2.4 1.0, 6.1 2.9 1.1, 7.6
2.0 2.6 0.9, 7.8 3.0 1.0, 9.5
Triglyceride concentration (mg/dl)yy
50 1.0#
100 0.5 0.2, 1.2
150 0.7 0.3, 1.8
200 1.0 0.3, 3.1
Direct-effects analysis
BMI coefcient 0.134 (0.036)zz 0.092 (0.041) 0.086 (0.041)
Percentage of effect that was direct 69 64
* Weight (kg)/height (m)
2
.
y Conditional logistic regression model adjusting for gestational age at blood sampling and marital status.
z Conditional logistic regression model adjusting for gestational age at blood sampling, marital status, C-reactive
protein concentration, and prepregnancy multivitamin/mineral use.
Conditional logistic regression model adjusting for gestational age at blood sampling, marital status, C-reactive
protein concentration, triglyceride concentration, and prepregnancy multivitamin/mineral use.
{ AOR, adjusted odds ratio; CI, condence interval; BMI, body mass index.
# Reference category.
** Based on a linear spline with a knot at 1.8 mg/dl.
yy Based on a linear spline with a knot at 90 mg/dl.
zz Numbers in parentheses, standard error.
Calculated as (direct-effects BMI coefcient/total-effects BMI coefcient) 3 100.
TABLE 2. Geometric mean* C-reactive protein and triglyceride concentrations by prepregnancy body mass indexy category,
Pregnancy Exposures and Preeclampsia Prevention Study, Pittsburgh, Pennsylvania, 19972001z
Geometric mean
Prepregnancy body mass index category
Underweight
(<18.5) (n 14)
Normal weight
(18.524.9) (n 122)
Overweight
(25.029.9) (n 48)
Obese
(30.0) (n 36)
Mean 95% CI Mean 95% CI Mean 95% CI Mean 95% CI
C-reactive protein (mg/dl) 0.25 0.16, 0.38 0.29 0.25, 0.33 0.48 0.37, 0.64{ 0.74 0.56, 0.97#
Triglycerides (mg/dl) 85.0 62.4, 115.9 90.4 82.8, 98.6 95.3 82.1, 110.7 101.5 88.2, 116.8
* Mean values were compared using analysis of variance with Bonferroni correction.
y Weight (kg)/height (m)
2
.
z Results from multiple linear regression analysis (with adjustment for race/ethnicity, marital status, education, smoking status, and age) were
as follows: 1) a 5-unit increase in prepregnancy BMI was associated with a 46 percent rise in C-reactive protein concentration (95% CI: 33, 61);
2) a 5-unit increase in prepregnancy BMI was associated with a 5.4-mg/dl increase in triglyceride concentration (95% CI: 1.1, 11.9).
CI, condence interval.
{ Signicantly greater than normal weight (p < 0.05).
# Signicantly greater than underweight and normal weight (each p < 0.05).
Body Mass Index and Preeclampsia 1203
Am J Epidemiol 2005;162:11981206

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

3
1
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

BMI-preeclampsia multivariable models), and/or our pre-
dominance of mild, late-onset preeclampsia cases. A previ-
ous study found that hypertriglyceridemia before 20 weeks
gestation increased the risk of early-onset preeclampsia
but had no effect on late-onset disease (34). Lipid level may
be a more important mediator of the BMI-preeclampsia as-
sociation in other populations in which triglyceride concen-
trations are signicantly elevated before the onset of the
disease among cases as compared with controls.
Few studies have examined the association between pre-
gravid overweight and dysregulation of metabolic and in-
ammatory pathways during pregnancy. CRP concentrations
in the third trimester among women who are obese at 1012
weeks gestation (BMI 27.7) were reported in one study to
be double those of leaner women (BMI <27.7)results
consistent with our data (38). In contrast to our ndings,
these investigators reported signicantly higher fasting tri-
glyceride concentrations in obese women compared with
lean women (38).
Importantly, we observed that approximately two thirds
of the effect of prepregnancy BMI on preeclampsia risk was
mediated by pathways that are independent of inammation
and triglyceride levels. Future research will be needed to
delineate the other mechanisms explaining this important
association. Possible intermediates include insulin resis-
tance, oxidative stress, other lipids, and endothelial dysfunc-
tion. Lifestyle factors such as inadequate dietary intake or
physical inactivity during pregnancy may contribute to this
association as well, either through these metabolic distur-
bances (i.e., by increasing oxidative stress) or through other
means.
Our results are predicated on several important assump-
tions that, if not upheld, may have biased our results. We
assumed that CRP is a valid measure of systemic inamma-
tion and that triglyceride concentration is a valid marker of
lipid abnormalities at 20 weeks gestation. Both of these
markers have been used in past studies of preeclampsia and
cardiovascular health and should have been appropriate in
our population. Although CRP was initially recognized as a
nonspecic indicator of inammation, recent data suggest
that specic biologic functions of CRP may link it to the
pathogenesis of atherosclerosis, includingfacilitating macro-
phage low density lipoprotein uptake, binding the phospho-
choline of oxidized low density lipoprotein, up-regulating
the expression of adhesion molecules in endothelial cells,
and inhibiting endothelial nitric-oxide synthase expression
in aortic endothelial cells (39). These mechanisms are prob-
ably relevant to preeclampsia, which has many of the same
pathophysiologic features as atherosclerosis (1). We chose
triglycerides over other lipid markers because past studies
have reported that differences in triglyceride levels between
preeclamptics and controls have tended to be larger than
differences in other lipids (3437). Nevertheless, we cannot
rule out the possibility that inclusion of another biomarker of
lipid status or inammation in the direct-effects models
would have yielded different results.
We also assumed that the direct effects are multiplicative
and that we accurately measured all confounders of the
BMI-preeclampsia, CRP-preeclampsia, and triglyceride-
preeclampsia associations. However, we lacked data on die-
tary intake, access to health care, and genetic factors that
may predispose people to obesity, inammation, elevated
triglyceride levels, and preeclampsia, and some of our co-
variates may have been measured with error. We encourage
other investigators to test the robustness of our ndings
using covariate-rich data sets from populations with varying
proportions of preeclampsia subtypes based on severity, ges-
tational age of onset, and the presence of fetal growth re-
striction. The use of different biomarkers of inammation
and lipid status would be particularly illuminating.
Understanding mechanisms underlying the BMI-
preeclampsia relation is of great public health importance.
Although reducing body weight before conception may
lower the risk of preeclampsia, weight loss during pregnancy
is not recommended (40). Other interventions must be iden-
tied to reduce the risk of preeclampsia among women who
enter pregnancy with a high BMI. If our aforementioned as-
sumptions are correct and our results are conrmed by other
investigators, these data suggest that interventions designed
to reduce inammation and triglyceride concentrations may
be effective in lowering the risk of preeclampsia among
overweight and obese women at the start of pregnancy.
ACKNOWLEDGMENTS
This work was partially supported by grants 2PO1
HD30367 and 5MO1 RR00056 from the National Institute
of Child Health and Human Development. Dr. Lisa Bodnar
was supported by grant K12 HD43441 from the National
Institute of Child Health and Human Development.
The authors thank Drs. Stephen Cole, Miguel Herna`n, and
Enrique Schisterman for guidance regarding their analytical
approach. They also thank the staff of the Pregnancy
Exposures and Preeclampsia Prevention Study for their
dedication to this project.
Conict of interest: none declared.
REFERENCES
1. Roberts JM. Pregnancy-related hypertension. In: Creasy RK,
Resnik R, eds. Maternal-fetal medicine. Philadelphia, PA:
W B Saunders Company, 1998:83372.
2. Bodnar LM, Ness RB, Markovic N, et al. The risk of pre-
eclampsia rises with increasing prepregnancy body mass
index. Ann Epidemiol 2005;15:47582.
3. Eskenazi B, Fenster L, Sidney S. A multivariate analysis of
risk factors for preeclampsia. JAMA 1991;266:23741.
4. Mittendorf R, Lain KY, Williams MA, et al. Preeclampsia:
a nested, case-control study of risk factors and their interac-
tions. J Reprod Med 1996;41:4916.
5. OBrien TE, Ray JG, Chan WS. Maternal body mass index and
the risk of preeclampsia: a systematic overview. Epidemiology
2003;14:36874.
6. Sibai BM, Ewell M, Levine RJ, et al. Risk factors associated
with preeclampsia in healthy nulliparous women. The Calcium
for Preeclampsia Prevention (CPEP) Study Group. Am J
Obstet Gynecol 1997;177:100310.
1204 Bodnar et al.
Am J Epidemiol 2005;162:11981206

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

3
1
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

7. Thadhani R, Stampfer MJ, Hunter DJ, et al. High body mass
index and hypercholesterolemia: risk of hypertensive disorders
of pregnancy. Obstet Gynecol 1999;94:54350.
8. Chan DC, Barrett HP, Watts GF. Dyslipidemia in visceral obe-
sity: mechanisms, implications, and therapy. Am J Cardiovasc
Drugs 2004;4:22746.
9. Visser M, Bouter LM, McQuillan GM, et al. Elevated
C-reactive protein levels in overweight and obese adults.
JAMA 1999;282:21315.
10. Redman CW, Sacks GP, Sargent IL. Preeclampsia: an exces-
sive maternal inammatory response to pregnancy. Am J
Obstet Gynecol 1999;180:499506.
11. Hubel CA. Dyslipidemia, iron, and oxidative stress in pre-
eclampsia: assessment of maternal and feto-placental interac-
tions. Semin Reprod Endocrinol 1998;16:7592.
12. Roberts JM. Endothelial dysfunction in preeclampsia. Semin
Reprod Endocrinol 1998;16:515.
13. Ness RB, Markovic N, Bass D, et al. Family history of hy-
pertension, heart disease, and stroke among women who de-
velop hypertension in pregnancy. Obstet Gynecol 2003;
102:136671.
14. Ness RB, Roberts JM. Heterogeneous causes constituting the
single syndrome of preeclampsia: a hypothesis and its impli-
cations. Am J Obstet Gynecol 1996;175:136570.
15. Clinical guidelines on the identication, evaluation, and treat-
ment of overweight and obesity in adultsthe evidence report.
Obes Res 1998;6(suppl 2):51S209S.
16. Report of the National High Blood Pressure Education Pro-
gram Working Group on High Blood Pressure in Pregnancy.
Am J Obstet Gynecol 2000;183(suppl):S122.
17. Neithardt AB, Dooley SL, Borensztajn J. Prediction of
24-hour protein excretion in pregnancy with a single voided
urine protein-to-creatinine ratio. Am J Obstet Gynecol 2002;
186:8836.
18. Levine RJ, Esterlitz JR, Raymond EG, et al. Trial of Calcium
for Preeclampsia Prevention (CPEP): rationale, design, and
methods. Control Clin Trials 1996;17:44269.
19. Levine RJ, Hauth JC, Curet LB, et al. Trial of calcium to
prevent preeclampsia. N Engl J Med 1997;337:6976.
20. Kluft C, de Maat MP. Sensitive markers of inammation make
it possible to study the chronic process: the rise of interest in
low levels of C-reactive protein. Vascul Pharmacol 2002;39:
99104.
21. Austin MA, Hokanson JE, Edwards KL. Hypertriglycer-
idemia as a cardiovascular risk factor. Am J Cardiol 1998;81:
7B12B.
22. Bureau of the Census, US Department of Commerce. Poverty
in the United States: 1990. Series P-60. Washington, DC:
Bureau of the Census, 1991.
23. Cole SR, Hernan MA. Fallibility in estimating direct effects.
Int J Epidemiol 2002;31:1635.
24. Greenland S, Pearl J, Robins JM. Causal diagrams for epide-
miologic research. Epidemiology 1990;10:3748.
25. Pearl J. Causal diagrams for empirical research. Biometrika
1995;82:66988.
26. Pearson TA, Mensah GA, Alexander RW, et al. Markers of
inammation and cardiovascular disease: application to clin-
ical and public health practice: a statement for healthcare
professionals from the Centers for Disease Control and
Prevention and the American Heart Association. Circulation
2003;107:499511.
27. Expert Panel on Detection, Evaluation, and Treatment of High
Blood Cholesterol in Adults. Executive summary of the third
report of the National Cholesterol Education Program (NCEP)
Expert Panel on Detection, Evaluation, and Treatment of High
Blood Cholesterol in Adults (Adult Treatment Panel III).
JAMA 2001;285:248697.
28. Qiu C, Luthy DA, Zhang C, et al. A prospective study of
maternal serum C-reactive protein concentrations and risk of
preeclampsia. Am J Hypertens 2004;17:15460.
29. Wolf M, Kettyle E, Sandler L, et al. Obesity and preeclampsia:
the potential role of inammation. Obstet Gynecol 2001;98:
75762.
30. Williams MA, Farrand A, Mittendorf R, et al. Maternal second
trimester serum tumor necrosis factor-alpha-soluble receptor
p55 (sTNFp55) and subsequent risk of preeclampsia. Am J
Epidemiol 1999;149:3239.
31. Sacks GP, Studena K, Sargent K, et al. Normal pregnancy and
preeclampsia both produce inammatory changes in periph-
eral blood leukocytes akin to those of sepsis. Am J Obstet
Gynecol 1998;179:806.
32. Djurovic S, Clausen T, Wergeland R, et al. Absence of en-
hanced systemic inammatory response at 18 weeks of ges-
tation in women with subsequent pre-eclampsia. BJOG 2002;
109:75964.
33. Levine RJ, Qian C, Leshane ES, et al. Two-stage elevation
of cell-free fetal DNA in maternal sera before onset of pre-
eclampsia. Am J Obstet Gynecol 2004;190:70713.
34. Clausen T, Djurovic S, Henriksen T. Dyslipidemia in early
second trimester is mainly a feature of women with early onset
pre-eclampsia. BJOG 2001;108:10817.
35. Chappell LC, Seed PT, Briley A, et al. A longitudinal study
of biochemical variables in women at risk of preeclampsia.
Am J Obstet Gynecol 2002;187:12736.
36. Gratacos E, Casals E, Sanllehy C, et al. Variation in lipid levels
during pregnancy in women with different types of hyperten-
sion. Acta Obstet Gynecol Scand 1996;75:896901.
37. Lorentzen B, Endresen MJ, Clausen T, et al. Fasting serum
free fatty acids and triglycerides are increased before 20 weeks
of gestation in women who later develop preeclampsia.
Hypertens Pregnancy 1994;13:1039.
38. Ramsay JE, Ferrell WR, Crawford L, et al. Maternal obesity
is associated with dysregulation of metabolic, vascular, and
inammatory pathways. J Clin Endocrinol Metab 2002;87:
42317.
39. Black S, Kushner I, Samols D. C-reactive protein. J Biol Chem
2004;279:4848790.
40. Institute of Medicine, National Academy of Sciences. Nutri-
tion during pregnancy. Washington, DC: National Academy
Press, 1990.
APPENDIX
We tted the following logistic regression models by maximizing the conditional likelihood.
Total-effects model:
logit Prpreeclampsia
it
1jA
it
a
it
; X
it
x
it
b
1i
a
it
b
2
x
it
b
3
:
Body Mass Index and Preeclampsia 1205
Am J Epidemiol 2005;162:11981206

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

3
1
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

Direct-effects model 1:
logit Prpreeclampsia
it
1jA
it
a
it
; X
it
x
it
; B
it
b
it
; Y
it
y
it
k
1
a
it
k
2
x
it
k
3
b
it
k
4
y
it
k
5
:
Direct-effects model 2:
logit Prpreeclampsia
it
1jA
it
a
it
; X
it
x
it
; B
it
b
it
; Y
it
y
it
; C
it
c
it
; Z
it
z
it

d
1i
a
it
d
2
x
it
d
3
b
it
d
4
y
it
d
5
c
it
d
6
z
it
d
7
:
Above, i 1, 2, . . ., 55 denotes independent groups; t 1, 2, 3, 4 denotes the observations for the ith group; A denotes body
mass index; B denotes C-reactive protein; C denotes triglycerides; x denotes a vector of measured confounders for the body
mass index-preeclampsia association; y denotes a vector of measured confounders for the C-reactive proteinpreeclampsia
association; and z denotes a vector of measured confounders for the triglyceride-preeclampsia association.
1206 Bodnar et al.
Am J Epidemiol 2005;162:11981206

b
y

g
u
e
s
t

o
n

J
a
n
u
a
r
y

3
1
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m