DOI: 10.1111/j.1471-0528.2007.01483.x
Systematic review
www.blackwellpublishing.com/bjog
Accuracy of body mass index in predicting
pre-eclampsia: bivariate meta-analysis
JS Cnossen,a MMG Leeflang,b EEM de Haan,a BWJ Mol,c JAM van der Post,c
KS Khan,d G ter Riete
a Department of General Practice, b Department of Clinical Epidemiology and Biostatistics and c Department of Obstetrics and Gynaecology,
Academic Medical Center, Amsterdam, the Netherlands d Department of Obstetrics and Gynaecology, Birmingham Women’s Hospital,
Birmingham, UK e Horten Center, University of Zurich, Zurich, Switzerland
Correspondence: Dr JS Cnossen, Department of General Practice, Academic Medical Center, Meibergdreef 15, 1100 DD, Amsterdam,
the Netherlands. Email [email protected]
Accepted 2 July 2007. Published OnlineEarly 28 September 2007.
Objective The objective of this study was to determine the
accuracy of body mass index (BMI) (pre-pregnancy or at booking)
in predicting pre-eclampsia and to explore its potential for clinical
application.
Design Systematic review and bivariate meta-analysis.
Setting Medline, Embase, Cochrane Library, MEDION, manual
searching of reference lists of review articles and eligible primary
articles, and contact with experts.
Population Pregnant women at any level of risk in any healthcare
setting.
Methods Reviewers independently selected studies and extracted
data on study characteristics, quality, and accuracy. No language
restrictions.
Main outcome measures Pooled sensitivities and specificities
(95% CI), a summary receiver operating characteristic curve, and
corresponding likelihood ratios (LRs). The potential value of BMI
was assessed by combining its predictive capacity for different
prevalences of pre-eclampsia and the therapeutic effectiveness
(relative risk 0.90) of aspirin.
Results A total of 36 studies, testing 1 699 073 pregnant women
(60 584 women with pre-eclampsia), met the selection criteria. The
median incidence of pre-eclampsia was 3.9% (interquartile range
1.4–6.8). The area under the curve was 0.64 with 93% of
heterogeneity explained by threshold differences. Pooled estimates
(95% CI) for all studies with a BMI ‡ 25 were 47% (33–61) for
sensitivity and 73% (64–83) for specificity; and 21% (12–31) and
92% (89–95) for a BMI ‡ 35. Corresponding LRs (95% CI) were
1.7 (0.3–11.9) for BMI ‡ 25 and 0.73 (0.22–2.45) for BMI < 25,
and 2.7 (1.0–7.3) for BMI ‡ 35 and 0.86 (0.68–1.07) for BMI < 35.
The number needed to treat with aspirin to prevent one case of
pre-eclampsia ranges from 714 (no testing, low-risk women) to 37
(BMI ‡ 35, high-risk women).
Conclusions BMI appears to be a fairly weak predictor for pre-
eclampsia. Although BMI is virtually free of cost, noninvasive, and
ubiquitously available, its usefulness as a stand-alone test for risk
stratification must await formal cost-utility analysis. The findings
of this review may serve as input for such analyses.
Keywords Accuracy, body mass index, likelihood ratio, meta-
analysis, pre-eclampsia, sensitivity and specificity.
Please cite this paper as: Cnossen J, Leeflang M, de Haan E, Mol B, van der Post J, Khan K, ter Riet G. Accuracy of body mass index in predicting pre-eclampsia:
bivariate meta-analysis. BJOG 2007;114:1477–1485.
Introduction
Pre-eclampsia is a major cause of maternal and perinatal
morbidity and mortality.1–3 It has a long preclinical phase
before signs and symptoms become manifest in the second
half of pregnancy. Prediction of pre-eclampsia is important
because increased monitoring and administration of early
preventative treatment with, for example, aspirin can be more
selectively targeted at high-risk women.4
The exact aetiology of pre-eclampsia is still unknown. Several risk factors have been identified including obesity.5 The
World Health Organization (WHO) estimates that currently
more than 1 billion adults are overweight, of whom at least
300 million are clinically obese.6 From 1999 to 2003, in the
total delivery population, the proportion of women who were
overweight or obese pre-pregnancy increased from 37.1 to
40.5%.7 This worldwide increase in overweight is likely to
cause a rise in pre-eclampsia and other complications, such
as gestational diabetes.8 The body mass index (BMI) is an
international standard for obesity measurement adjusting
bodyweight for height (weight [kg]/height squared [m2])
and an indicator of nutritional status.
ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology
1477
Cnossen et al.
We conducted a systematic review to determine the accuracy of BMI (pre-pregnancy or at booking) to predict preeclampsia and to explore its potential for clinical application.
We meta-analysed the data using a bivariate regression analysis, accounting for (negative) correlation between sensitivity
and specificity, and derived likelihood ratios (LRs) thereof.9–12
Methods
Study identification and selection
We performed an electronic search targeting all procedures
used for the prediction of pre-eclampsia. We searched Medline, Embase, Cochrane Library, and MEDION (www.
mediondatabase.nl/) from inception to April 2006. The search
strategy13 consisted of MeSH or keyword terms related to preeclampsia combined with methodological filters for identification of studies on diagnostic tests and aetiology.14–16 We
checked reference lists of review articles and eligible primary
studies to identify cited articles not captured by electronic
searches and contacted experts. Reference Manager 10.0 databases were established incorporating results of all searches.
We selected studies in a three-stage process. First, one
reviewer scrutinised titles and/or abstracts of all citations.
Second, to ensure independent duplicate selection for this
review, we performed a search based on keywords ‘*weight*
OR body mass* OR obesity OR adipos* OR fat OR quetelet*’
within the Reference Manager database for citations to be
scrutinised by a second reviewer. We obtained full manuscripts of all citations that were selected by at least one of
the reviewers. Final in/exclusion decisions were made after
independent and duplicate examination of the full manuscripts of selected citations. We included studies if they
reported numerical data allowing construction of a 2 · 2 table
cross-classifying BMI test results (pre-pregnancy or at
booking) and occurrence of pre-eclampsia. We included
studies examining pregnant women at any level of risk in
any healthcare setting. Language restrictions were not applied.
Any disagreements were resolved by consensus or by a third
reviewer.
For each included paper, two reviewers independently
extracted data on clinical and methodological study characteristics and on test accuracy. Study characteristics consisted
of women’s risk classifications, characteristics of the index test
and of the reference standard, and details of the outcome,
such as onset and severity of pre-eclampsia.
Assessment of study quality
Two reviewers independently assessed all included manuscripts for study quality according to an adapted version of
QUADAS.17 We considered consecutive or random entry of
pregnant women to a cohort as ensuring an appropriate spectrum of patients. Patient selection criteria were considered
appropriate when the study reported on parity, singleton/
1478
multiple pregnancies, diabetes mellitus, and chronic hypertension. We deemed the reference standard for pre-eclampsia
appropriate if it combined persistent systolic blood pressure ‡
140 mmHg and/or diastolic blood pressure ‡ 90 mmHg with
proteinuria ‡ 0.3 g/24 hours or ‡ 1+ dipstick (30 mg/dl in
a single urine sample) new after 20 weeks of gestation. Similarly, we deemed the reference standard for superimposed
pre-eclampsia appropriate when it combined the development of proteinuria ‡ 0.3 g/24 hours or ‡ 1+ dipstick after
20 weeks of gestation in chronically hypertensive women.18,19
A definition of pre-eclampsia according to recent (from the
year 2000 onwards) international guidelines was also considered appropriate.18–20 We also assessed if details of the index
test (kg/m2, pre-pregnancy or before 20 weeks) and reference
test (measurement device, position of patient, and Korotkoff
phase), and a description of reasons for withdrawals were
adequate.
Data synthesis
For each study, we constructed a 2 · 2 table cross-classifying
BMI test results and the occurrence of pre-eclampsia. We
used receiver operating characteristic (ROC) plots to visualise
data. We used a bivariate meta-regression model to calculate
pooled estimates of sensitivity and specificity for several BMI
cutoff values and to fit a summary ROC (sROC) curve. This
method has been extensively described elsewhere.9–12 Briefly,
rather than using a single outcome measure per study, such as
the diagnostic odds ratio, the bivariate model preserves the
two-dimensional nature of diagnostic data in a single model.
This model incorporates the correlation that may exist
between sensitivity and specificity within studies due to possible differences in threshold between studies. The bivariate
model uses a random effects approach for both sensitivity and
specificity, allowing for heterogeneity beyond chance due to
clinical or methodological differences between studies. In
addition, the model acknowledges the difference in precision
by which sensitivity and specificity have been measured in
each study. This means that studies with a larger number of
women with pre-eclampsia receive more weight in the calculation of the pooled estimate of sensitivity, while studies with
more women without pre-eclampsia are more influential in
the pooling of specificity. We calculated a correlation coefficient (r) to show the strength and shape of the correlation
between sensitivity and specificity, and hence the amount of
variance that can be explained due to threshold differences
(r2). Despite several cutoff values for BMI, for statistical reasons, each study was represented once in the sROC analysis.
This was achieved by randomly sampling one 2 · 2 table from
each study, ensuring that all areas of the sROC curve were
represented. For clinical usefulness, we derived LRs from the
estimated sensitivities and specificities.
We defined all subgroup analyses, except one (prepregnancy or at booking), a priori requiring that at least 80%
ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology
Body mass index and prediction of pre-eclampsia
of studies reported clearly on a particular item and each
subgroup contained at least three studies. Data on threshold,
pre-pregnancy or booking BMI, severity and onset (before or
after 34 weeks) of pre-eclampsia, incidence (cutoff 4%), study
design, consecutive entry, prospective data collection, and
adequacy of reference standard were considered one by one
in the subgroup analyses. First, we entered the subgroup characteristic as a covariate in the model. If homogeneity between
studies with and without the characteristic was rejected
(P < 0.10), the subgroup analysis was performed.
The bivariate models were fitted using the Proc NLMixed
(SAS 9.1 for Windows [SAS Institute Inc, Cary, NC, USA]).
Potential for clinical application
At a (fixed) relative risk (RR) reduction of aspirin of 10%
(RR = 0.90),4 the absolute treatment effect becomes greater
as the absolute risk rises. For a woman with a risk of preeclampsia of 40%, aspirin reduces it to 36%, whereas for
a woman who has a 4% risk, this is reduced to 3.6%. Testing
makes sense if treatment is conditional on obtaining a positive
test result. Therefore, the impact of risk stratification using
BMI depends on the shift that testing brings about in the pretest probability distribution.21 We illustrated this impact with
an example of decision-making in clinical practice about the
use of aspirin in women at risk of pre-eclampsia, which has
been shown to be an effective preventive treatment.4 We calculated the risk, and hence the therapeutic benefits associated
with BMI test results from sensitivity and specificity. The
number needed to treat (NNTreat) is the number of women
that one needs to treat (with aspirin) to prevent one case of
pre-eclampsia and is calculated by 1/(probability after testing
positive – probability after treatment).
Results
Figure 1 summarises the process of literature identification
and selection. Thirty-six studies met the selection criteria.22–57
Four papers were found in reference lists, of which one was
included. The two papers from Sebire et al.46,47 reported on
the same cohort, and therefore, we considered it as one study.
We excluded studies for a variety of reasons. The main reasons for exclusion were no distinction between pre-eclampsia
and pregnancy-induced hypertension, insufficient data to
construct a 2 · 2 table, and reporting weight or weight gain
or a different weight-index test instead of a BMI.
Table 1 summarises study characteristics. We included 23
cohort studies, 11 case–control studies, and one randomised
trial. A total of 1 699 073 pregnant women were included, of
whom 60 584 (3.6%) developed pre-eclampsia. The median
incidence of pre-eclampsia in cohort studies was 3.9% (interquartile range 1.4–6.8). For calculation of the median incidence, we excluded case–control studies that did not report
the incidence in their source population. The number of
women analysed in cohort studies ranged from 28127 to
561 770,56 and in case–control studies from 5437 to 6747.28
Twenty-seven studies were performed in Western countries.
Eleven studies were prospectively designed. Sixteen studies
excluded all women with diabetes and/or chronic hypertension.
Potentially relevant citations identified from electronic searches to capture primary articles
on all tests used in the prediction of pre-eclampsia
n = 17–847
Potentially relevant citations on BMI in the prediction of pre-eclampsia
n = 1–307
References excluded after screening titles and/ or abstracts
n = 1–170
Primary articles retrieved for detailed evaluation
- fromelectronic searches
- from reference lists
Articles excluded
- no BMI calculated
- insufficient data toconstruct 2 x 2 table
- no pre-pregnancy BMI versus pre-eclampsia outcome
- reviews/ letters/ comments/ editorials
- pre-eclampsia not separate outcome/
pregnancy induced hypertension only
- matching on BMI
- gestational age unclear
- data duplication
- other
n = 133
n=4
n = 22
n = 20
n = 13
n=5
n = 101
n = 25
n=3
n=1
n=7
n=5
Primary articles included in meta-analysis
n = 36
Figure 1. Process from initial search to final inclusion for studies on BMI in the prediction of pre-eclampsia.
ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology
1479
1480
USA
South America
Peru
France
USA
Japan
Netherlands
Taiwan
USA
USA
USA
Denmark
USA
Sweden
USA
Peru
USA
Denmark
Sweden
USA
Carr28 (2005)
Conde-Agudelo56 (2000)
Enquobahrie29 (2005)
Galtier-Dereure30 (1995)
Goldman31 (1991)
Ioka32 (2003)
Knuist33 (1998)
Lee34 (2000)
Mittendorf35 (1996)
Mostello36 (2002)
Murai37 (1997)
Nohr38 (2005)
Ogunyemi39 (1998)
Ostlund40 (2004)
Qiu41 (2004)
Qiu57 (2006)
Ramos42 (2005)
Rode43 (2005)
Ros44 (1998)
Rudra45 (2005)
USA
Bianco25 (1998)
USA
China
Baulon24 (2005)
Bowers27 (1999)
Denmark
Basso23 (2004)
USA
USA
Baeten22 (2001)
Bodnar26 (2005)
Country
First author (year)
Prospective cohort from
birth register
Prospective nested case–control
Retrospective case–control
Retrospective cohort
Retrospective cohort
Retrospective cohort from birth
register
Retrospective case–control
Retrospective cohort
Retrospective case–control
Retrospective case–control from
birth register
Retrospective case–control
Prospective cohort from national
birth cohort
Prospective cohort
Retrospective case–control from
birth certificates
Retrospective cohort from
birth register
Retrospective case–control
Retrospective case–control
Prospective cohort
Prospective cohort
Prospective cohort
Retrospective cohort
Prospective cohort
Retrospective cohort
Retrospective cohort from
birth register
Retrospective cohort from
birth register
Retrospective cohort
Type of study
757
559
189
22 658
8092
2418
430 852
582
54
54 505
29 735
2741
4702
312
168
305
488
2080
561 770
6747
281
1179
11 926
11 921
59 074
96 384
Number of
women analysed
n.r.
n.r.
n.r.
8.3
2.4
5.2
2.8
7.9
n.r.
2.1
1.4
n.r.
5.9
n.r.
n.r.
n.r.
17.4
1.4
5.1
n.r.
5.8
4.8
3.7
3.9
0.5
6.8
Incidence
PET (%)
Table 1. Study characteristics for individual studies on the predictive accuracy of BMI in predicting pre-eclampsia
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
First visit
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
Pre-pregnancy
First visit
Pre-pregnancy
Pre-pregnancy
Gestational
age (BMI)
n.r.
(continued)
IN: normotensive women
IN: women with pre-eclampsia and normotensive controls
IN: singleton pregnancies; EX: fetal anomalies
IN: singleton pregnancies; EX: delivery , 37 weeks
IN: nulliparous women; EX: ,34 years
IN: singleton pregnancies, black low-income women;
EX: delivery , 37 weeks of gestation
IN: singleton pregnancies; EX: DM
IN: normotensive women, singleton pregnancies
IN: participation in telephone interview
IN: Peruvian women
EX: hepatic/cardiac/renal failure, DM
IN: gestational diabetes
Not reported
IN: nulliparous women, singleton pregnancies;
EX: DM, CH, renal disease, obstetric anomaly
IN: normotensive women; EX: fetal malformations
IN: singleton pregnancies
IN: singleton pregnancies
IN: normotensive women, singleton pregnancies;
EX: CH, renal/cardiovascular disease
IN: age 20–34 years; EX: multiple pregnancies,
extremes of age, BMI between 27 and 34,
missing height or pre-pregnancy weight
IN: normotensive women, singleton pregnancies;
EX: pre-existing medical condition, positive
toxicology screen and HIV
IN: normotensive women, singleton pregnancies;
EX: delivery before 36 weeks, pre-existing chronic
maternal illness (HIV)
IN: normotensive women, singleton pregnancies;
EX: fetal demise , 24 weeks
IN: all pregnancies of at least 20 weeks of gestation
IN: singleton pregnancies
IN: nulliparous women
Population
Cnossen et al.
ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology
CH, chronic hypertension; DM, diabetes mellitus; EX, excluded; IN, included; n.r., not reported; PET, pre-eclampsia.
10–14 weeks
,9 weeks
Weiss54 (2004)
Yamamoto55 (2001)
USA
Japan
2.4
n.r.
16 102
224
Prospective cohort
Case–control
Pre-pregnancy
At booking
17 211
448
Prospective cohort
Retrospective case–control
Thadhani52 (1999)
van Hoorn53 (2002)
USA
Australia
0.5
n.r.
3446
Cohort
Suzuki51 (2000)
Japan
8.6
5.3
0.4
Steinfeld49 (2000)
Stone50 (1994)
USA
USA
2424
19 034
Pre-pregnancy
IN: unselected population
IN: normotensive nulliparous women; EX: medical or
obstetric complications, know fetal complications
IN: singleton pregnancies
IN: normotensive women, singleton pregnancies;
EX: less severe cases, DM, renal/autoimmune/
cardiovascular diseases
IN: normotensive women; EX: monochorionic
twins and systemic illnesses
IN: normotensive women
IN: primigravid, singleton pregnancies;
EX: DM, CH and medication use
IN: unselected population, singleton pregnancies
IN: normotensive women; EX: first visit after
9 weeks of gestation, CH, DM and cardiovascular
disease
At booking
First visit
(13–21 weeks)
Pre-pregnancy
Pre-pregnancy
0.8
7.6
325 395
4310
Retrospective cohort
Prospective randomised controlled
trial
Retrospective cohort
Retrospective cohort
UK
USA
Sebire46,47 (2001)
Sibai48 (1997)
First author (year)
Table 1. (Continued)
Country
Type of study
Number of
women analysed
Incidence
PET (%)
Gestational
age (BMI)
Population
Body mass index and prediction of pre-eclampsia
Twenty-eight studies reported pre-pregnancy BMI, and seven
studies reported BMI at booking. We excluded one study for
the meta-analysis because this study excluded women with
a BMI between 27 and 34.25
Figure 2 summarises study quality. Patient recruitment,
selection criteria, and reasons for withdrawal were poorly
reported. Eleven studies used a definition of pre-eclampsia
according to current international standards. Fifteen studies
reported a definition of pre-eclampsia according to former
guidelines. Details of how the reference test was executed were
seldom reported. Only one study reported whether women
received treatment to prevent pre-eclampsia.48
Figure 3 shows the sROC curve for the included studies. The
area under the curve (AUC) was 0.64, which was similar to the
AUC representing all data points for all studies (AUC 0.63).
The correlation coefficient was –0.965, indicating that 93% of
heterogeneity could be explained due to differences in threshold for a positive test. Subgroup analysis on pre-pregnancy
BMI versus BMI at booking yielded statistically significant
differences (P < 0.0001), showing slightly better performance
for pre-pregnancy BMI. However, their clinical relevance seems
doubtful (sensitivitypre-pregnancy 48.1% (47.9–48.2) versus
sensitivityat booking 44.9% (44.4–45.5) and specificitypre-pregnancy
61.4% (61.4–61.5) versus specificityat booking 58.5% (58.5–58.6).
The discrepancy between statistical significance and clinical
relevance is due to some of the studies having very large sizes.
Below, we report the overall results. Incidence of pre-eclampsia,
study design, data collection, and adequacy of reference standard
did not further reduce heterogeneity. Analyses on clinical characteristics and consecutive entry were hindered by a lack of
reporting. We used cutoff values that were most often used
and reported by the Institute of Medicine (IOM) and WHO
for categorisation.6,58 Pooled estimates (95% CI) for all studies
with a BMI ‡ 25 (18 studies) were 47% (33–61) for sensitivity
and 73% (64–83) for specificity. For a BMI ‡ 30 (19 studies),
these estimates were 19% (19–20) and 90% (88–93), and for
a BMI ‡ 35 (4 studies), the estimates were 21% (12–31) and
92% (89–95). Pooling data on BMI ‡ 30 required an adjusted
Patient spectrum
10
Selection criteria
12
Appropriate reference test
11
20
5
23
35
Adequate description reference test 2 1
Withdrawals explained
0%
8
16
Adequate description index test
32
12
20%
23
40%
60%
80% 100%
Compliance with quality item
yes
no
unclear
Figure 2. Summary of study quality. Data are presented as 100%
stacked bars. Numbers in stacks represent numbers of studies.
ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology
1481
Cnossen et al.
Table 2 shows that the probability of pre-eclampsia after
positive BMI testing increases strongly in moderate (say incidence 7%) and high-risk (say 12%) populations to 18 and
29%, respectively, for women with a BMI ‡ 35. These latter
posttest probabilities correspond with NNTsaspirin of 62 and
37, respectively. BMI results < 35 lead to posttest probabilities
of 1.2, 5.9, and 10.7% in populations with a baseline risk of
1.4, 6.8, and 12.2%, respectively. These numbers, in turn,
correspond to NNTsaspirin of 830, 170, and 94.
1.0
Sensitivity
0.8
0.6
0.4
Discussion and conclusion
0.2
0.0
0.0
0.2
0.4
0.6
0.8
1.0
1-specificity
Figure 3. sROC curve for individual studies on the accuracy of BMI in
predicting pre-eclampsia.
model of the bivariate analysis resulting in a very narrow confidence interval for sensitivity. Therefore, we also calculated estimates for BMI ‡ 29 (seven studies, IOM classification), which
were 22% (12–32) for sensitivity and 89% (82–95) for specificity. The corresponding LRs (95% CI) were 1.7 (0.3–11.9) for
BMI ‡ 25 and 0.73 (0.22–2.45) for BMI < 25, 1.9 (0.3–11.9) for
BMI ‡ 29 and 0.88 (0.61–1.29) for BMI < 29, and 2.7 (1.0–7.3)
for BMI ‡ 35 and 0.86 (0.68–1.07) for BMI < 35.
Based on a huge body of evidence, BMI (at any cutoff), prepregnancy or at booking, appears to be a fairly weak predictor
for pre-eclampsia. Although BMI is virtually free of cost, noninvasive, and ubiquitously available, its usefulness as a standalone test for risk stratification must await formal cost-utility
analysis. The findings of this review may serve to populate
such a model.
Based on the estimates in this review, among low-risk pregnancies without testing, one needed to treat 714 women with
aspirin to prevent one case of pre-eclampsia. By contrast, in
high-risk women, who in addition have a BMI ‡ 35, one
needed to treat only 37 women to prevent one case. This
may seem impressive. However, it is not entirely clear how
clinicians may decide on the exact risk level a woman has
before modifying it further using BMI. Note also that a single
test with modest values of sensitivity implies that many with
a negative test result will develop pre-eclampsia and thus, in
this example, will not benefit from what appears to be the
Table 2. Illustration of potential impact of BMI testing at two cutoff values among pregnant women at different prevalences and numbers
of women needed to treat with aspirin to prevent one case of pre-eclampsia
Test result
No test, no treatment**
Prior risk
(prevalence of PET, %)*
1.4
6.8
12.2
No test, treat all**
1.4
6.8
12.2
BMI ‚ 35; sensitivity 21%; specificity 92%
Test all, treat test positives
1.4
6.8
12.2
BMI ‚ 25; sensitivity 47%; specificity 73%
Test all, treat test positives
1.4
6.8
12.2
Probability of PET after
testing positive (%)
Treatment effect
(RR)
Probability of PET
after treatment (%)
NNTreat
1.4
6.8
12.2
—
—
1.4
6.8
12.2
—
0.90
714
147
82
3.6
16.1
26.7
0.90
3.2
14.5
24.0
278
62
37
2.4
11.3
19.5
0.90
2.2
10.2
17.6
417
88
51
PET, pre-eclampsia.
*Interquartile ranges of median prevalence in this review used as low and moderate prior risk. 12.2% was used as high prior risk.
**Numbers are equal for all tests regardless of threshold and sensitivity and specificity.
1482
ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology
Body mass index and prediction of pre-eclampsia
cheap and safe treatment of aspirin. The moderate positive
and negative LRs of a BMI > 35 underscore these conclusions.
In 2003, O’Brien et al.59 reviewed 13 studies and also reported
on the predictive power of BMI for pre-eclampsia. However,
they pooled predictive values, an accuracy parameter that is
generally considered too sensitive to variations in prevalence.60 Duckitt and Harrington,61 in their review, covering
different tests at booking, included ten studies on BMI and
reported different summary estimates, without explaining
why they selected a RR of pre-eclampsia of 2.47 in their
abstract. In addition, their methodology is flawed by the
adoption of a cause-effect framework discarding studies with
baseline incomparability, although this is unimportant in
(noncausal) studies on prediction.62 Finally, a pooled RR
(of 2.47) does not allow clinicians to update their pre-test
pre-eclampsia probabilities for women below the cutoff value,
since such women remain on their prior probability. In contrast, we pooled sensitivities and specificities, which are less
susceptible to variations in prevalence60 and are more useful
in determining whether the posttest probability increases or
decreases for the various BMI categories.
The mechanism by which obesity causes pre-eclampsia is
not completely understood, although establishing a causal
link is not particularly relevant for purposes of developing
a predictive test. Obesity has a strong link with insulin resistance and predisposes for type II diabetes and gestational
diabetes. Both obesity and diabetes are associated with
a higher risk of cardiovascular diseases, and hence preeclampsia.5,63,64 Hyperinsulinaemia may directly predispose
to hypertension by increased renal sodium reabsorption and
stimulation of the sympathetic nervous system. Another
possible explanation is that insulin resistance and/or associated hyperglycaemia impair endothelial function.5,64 Studies
that include a high proportion of women with diabetes
or chronic hypertensive may thus find higher estimates for
test accuracy of BMI. In contrast, smoking reduces the risk
of pre-eclampsia and has no effect on the risk of gestational
diabetes.65,66 Smokers usually have a lower BMI than nonsmokers, however, this effect is reduced when they quit
smoking.67 Our estimates of predictive accuracy may be
biased due to the inability to adjust for these factors. However, this drawback holds for all meta-analyses focusing on
evaluation of a single test, whereas in clinical practice more
information is available.
Identification of primary studies on pre-eclampsia and
BMI in database searches is difficult. In primary studies,
BMI is often reported as a secondary ‘exposure’ in a table
showing some general population characteristics. Electronic
retrieval is possible only when BMI is mentioned in the title,
abstract, or as a keyword. Factors that may influence the predictive accuracy of BMI are poor reporting of selection criteria, preventive treatment, and self-reported weight and
height. Self-reported weight in particular may be underesti-
mated in women68 and may result in a lower BMI (more test
negatives), thus lowering sensitivity and raising specificity.
However, a subgroup analysis on pre-pregnancy BMI versus
BMI at booking did not indicate this to be a strong phenomenon, although statistically significant. Due to lack of reporting on early and late onset, and on severity of pre-eclampsia,
we were unable to estimate the predictive accuracy of BMI for
severe early onset pre-eclampsia, which has considerable
maternal and perinatal morbidity and mortality.2,3
Future research should focus on undertaking high-quality
primary studies of test accuracy including all risk indicators
that clinicians normally use. This will enable future reviewers
to focus on meta-analysis of adjusted accuracy parameters or,
even better, on meta-analysis using individual patient data.69
The proper role of BMI among other tests in rational risk
stratification strategies for pre-eclampsia must await the
results from such analyses ideally incorporating the costs
and benefits of preventive treatment.
Funding
This study is supported by the UK NHS Health Technology
Assessment Programme, project code 01/64/04. j
References
1 Khan KS, Wojdyla D, Say L, Gulmezoglu AM, Van Look PF. WHO
analysis of causes of maternal death: a systematic review. Lancet
2006;367:1066–74.
2 Zhang J, Meikle S, Trumble A. Severe maternal morbidity associated
with hypertensive disorders in pregnancy in the United States.
Hypertens Pregnancy 2003;22:203–12.
3 Buchbinder A, Sibai BM, Caritis S, MacPherson C, Hauth J, Lindheimer
MD, et al. Adverse perinatal outcomes are significantly higher in severe
gestational hypertension than in mild preeclampsia. Am J Obstet Gynecol 2002;186:66–71.
4 Askie LM, Duley L, Henderson-Smart DJ, Stewart LA; PARIS Collaborative Group. Antiplatelet agents for prevention of pre-eclampsia:
a meta-analysis of individual patient data. Lancet 2007;369:1791–8.
5 Sibai B, Dekker G, Kupferminc M. Pre-eclampsia. Lancet 2005;365:
785–99.
6 Puska P, Nishida C, Porter D. WHO global strategy on diet, physical
activity and health: obesity and overweight. 2003 [www.who.int/
dietphysicalactivity/media/en/gsfs_obesity_pdf]. Accessed 12 February
2007.
7 Yeh J, Shelton JA. Increasing prepregnancy body mass index: analysis
of trends and contributing variables. Am J Obstet Gynecol 2005;
193:1994–8.
8 Dietl J. Maternal obesity and complications during pregnancy. J Perinat
Med 2005;33:100–5.
9 Arends LR. Multivariate meta-analysis: modelling the heterogeneity.
Mixing apples and oranges: dangerous or delicious. Thesis. Rotterdam:
Erasmus University; 2006;119–56.
10 Reitsma JB, Glas AS, Rutjes AW, Scholten RJ, Bossuyt PM, Zwinderman
AH. Bivariate analysis of sensitivity and specificity produces informative
summary measures in diagnostic reviews. J Clin Epidemiol 2005;58:
982–90.
ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology
1483
Cnossen et al.
11 Van Houwelingen HC, Zwinderman KH, Stijnen T. A bivariate approach
to meta-analysis. Stat Med 1993;12:2273–84.
12 Van Houwelingen HC, Arends LR, Stijnen T. Advanced methods in
meta-analysis: multivariate approach and meta-regression. Stat Med
2002;21:589–624.
13 Cnossen JS, van der Post JA, Mol BW, Khan KS, Meads CA, Ter RG.
Prediction of pre-eclampsia: a protocol for systematic reviews of test
accuracy. BMC Pregnancy Childbirth 2006;6:29.
14 Yale University School of Medicine. Finding the evidence on PubMed.
2003 [http://info.med.yale.edu/library/reference/publications/pubmed].
Accessed 10 December 2003.
15 Bachmann LM, Coray R, Estermann P, Ter Riet G. Identifying diagnostic
studies in MEDLINE: reducing the number needed to read. J Am Med
Inform Assoc 2002;9:653–8.
16 Haynes RB, Wilczynski N, McKibbon KA, Walker CJ, Sinclair JC.
Developing optimal search strategies for detecting clinically sound
studies in MEDLINE. J Am Med Inform Assoc 1994;1:447–58.
17 Whiting P, Rutjes AW, Dinnes J, Reitsma J, Bossuyt PM, Kleijnen J.
Development and validation of methods for assessing the quality of
diagnostic accuracy studies. Health Technol Assess 2004;8:iii, 1–234.
18 Brown MA, Lindheimer MD, de Swiet M, Van Assche A, Moutquin JM.
The classification and diagnosis of the hypertensive disorders of pregnancy: statement from the International Society for the Study of Hypertension in Pregnancy (ISSHP). Hypertens Pregnancy 2001;20:IX–XIV.
19 Report of the National High Blood Pressure Education Program
Working Group on High Blood Pressure in Pregnancy. Am J Obstet
Gynecol 2000;183:S1–S22.
20 Brown MA, Hague WM, Higgins J, Lowe S, McCowan L, Oats J, et al. The
detection, investigation and management of hypertension in pregnancy:
full consensus statement. Aust N Z J Obstet Gynaecol 2000;40:139–55.
21 Khan KS, Chien PF. Seizure prophylaxis in hypertensive pregnancies:
a framework for making clinical decisions. Br J Obstet Gynaecol 1997;
104:1173–9.
22 Baeten JM, Bukusi EA, Lambe M. Pregnancy complications and
outcomes among overweight and obese nulliparous women. Am J
Public Health 2001;91:436–40.
23 Basso O, Wilcox AJ, Weinberg CR, Baird DD, Olsen J. Height and risk of
severe pre-eclampsia. A study within the Danish National Birth Cohort.
Int J Epidemiol 2004;33:858–63.
24 Baulon E, Fraser WD, Piedboeuf B, Buekens P, Xiong X. Pregnancyinduced hypertension and infant growth at 28 and 42 days postpartum. BMC Pregnancy Childbirth 2005;5:14–20.
25 Bianco AT, Smilen SW, Davis Y, Lopez S, Lapinski R, Lockwood CJ.
Pregnancy outcome and weight gain recommendations for the
morbidly obese woman. Obstet Gynecol 1998;91:97–102.
26 Bodnar LM, Ness RB, Markovic N, Roberts JM. The risk of preeclampsia
rises with increasing prepregnancy body mass index. Ann Epidemiol
2005;15:475–82.
27 Bowers D, Cohen WR. Obesity and related pregnancy complications in
an inner-city clinic. J Perinatol 1999;19:216–19.
28 Carr DB, Epplein M, Johnson CO, Easterling TR, Critchlow CW. A
sister‘s risk: family history as a predictor of preeclampsia. Am J Obstet
Gynecol 2005;193:965–72.
29 Enquobahrie DA, Sanchez SE, Muy-Rivera M, Qiu C, Zhang C,
Austin MA, et al. Hepatic lipase gene polymorphism, pre-pregnancy
overweight status and risk of preeclampsia among Peruvian women.
Gynecol Endocrinol 2005;21:211–17.
30 Galtier-Dereure F, Montpeyroux F, Boulot P, Bringer J, Jaffiol C. Weight
excess before pregnancy: complications and cost. Int J Obes Relat
Metab Disord 1995;19:443–8.
31 Goldman M, Kitzmiller JL, Abrams B, Cowan RM, Laros RK Jr. Obstetric
complications with GDM. Effects of maternal weight. Diabetes
1991;40(Suppl 2):79–82.
1484
32 Ioka A, Tsukuma H, Nakamuro K. Lifestyles and pre-eclampsia with
special attention to cigarette smoking. J Epidemiol 2003;13:90–5.
33 Knuist M, Bonsel GJ, Zondervan HA, Treffers PE. Risk factors for preeclampsia in nulliparous women in distinct ethnic groups: a prospective
cohort study. Obstet Gynecol 1998;92:174–8.
34 Lee CJ, Hsieh TT, Chiu TH, Chen KC, Lo LM, Hung TH. Risk factors
for pre-eclampsia in an Asian population. Int J Gynaecol Obstet 2000;
70:327–33.
35 Mittendorf R, Lain KY, Williams MA, Walker CK. Preeclampsia.
A nested, case-control study of risk factors and their interactions.
J Reprod Med 1996;41:491–6.
36 Mostello D, Catlin TK, Roman L, Holcomb WL Jr, Leet T. Preeclampsia
in the parous woman: who is at risk? Am J Obstet Gynecol 2002;
187:425–9.
37 Murai JT, Muzykanskiy E, Taylor RN. Maternal and fetal modulators of
lipid metabolism correlate with the development of preeclampsia.
Metabolism 1997;46:963–7.
38 Nohr EA, Bech BH, Davies MJ, Frydenberg M, Henriksen TB, Olsen J.
Prepregnancy obesity and fetal death: a study within the Danish
National Birth Cohort. Obstet Gynecol 2005;106:250–9.
39 Ogunyemi D, Hullett S, Leeper J, Risk A. Prepregnancy body mass
index, weight gain during pregnancy, and perinatal outcome in a rural
black population. J Matern Fetal Med 1998;7:190–3.
40 Ostlund I, Haglund B, Hanson U. Gestational diabetes and preeclampsia. Eur J Obstet Gynecol Reprod Biol 2004;113:12–16.
41 Qiu C, Luthy DA, Zhang C, Walsh SW, Leisenring WM, Williams MA.
A prospective study of maternal serum C-reactive protein concentrations and risk of preeclampsia. Am J Hypertens 2004;17:154–60.
42 Ramos GA, Caughey AB. The interrelationship between ethnicity and
obesity on obstetric outcomes. Am J Obstet Gynecol 2005;193:
1089–93.
43 Rode L, Nilas L, Wojdemann K, Tabor A. Obesity-related complications
in Danish single cephalic term pregnancies. Obstet Gynecol 2005;
105:537–42.
44 Ros HS, Cnattingius S, Lipworth L. Comparison of risk factors for
preeclampsia and gestational hypertension in a population-based
cohort study. Am J Epidemiol 1998;147:1062–70.
45 Rudra CL, Williams MA. BMI as a modifying factor in the relations
between age at menarche, menstrual cycle characteristics, and risk
of preeclampsia. Gynecol Endocrinol 2005;21:200–5.
46 Sebire NJ, Jolly M, Harris J, Regan L, Robinson S. Is maternal
underweight really a risk factor for adverse pregnancy outcome? A
population-based study in London. BJOG 2001;108:61–6.
47 Sebire NJ, Jolly M, Harris JP, Wadsworth J, Joffe M, Beard RW,
et al. Maternal obesity and pregnancy outcome: a study of
287,213 pregnancies in London. Int J Obes Relat Metab Disord
2001;25:1175–82.
48 Sibai BM, Ewell M, Levine RJ, Klebanoff MA, Esterlitz J, Catalano PM,
et al. Risk factors associated with preeclampsia in healthy nulliparous
women. The Calcium for Preeclampsia Prevention (CPEP) Study Group.
Am J Obstet Gynecol 1997;177:1003–10.
49 Steinfeld JD, Valentine S, Lerer T, Ingardia CJ, Wax JR, Curry SL.
Obesity-related complications of pregnancy vary by race. J Matern
Fetal Med 2000;9:238–41.
50 Stone JL, Lockwood CJ, Berkowitz GS, Alvarez M, Lapinski R,
Berkowitz RL. Risk factors for severe preeclampsia. Obstet Gynecol
1994;83:357–61.
51 Suzuki S, Yoneyama Y, Sawa R, Shin S, Araki T. Clinical usefulness of
maternal body mass index in twin pregnancies. Hypertens Pregnancy
2000;19:273–9.
52 Thadhani R, Stampfer MJ, Hunter DJ, Manson JE, Solomon CG,
Curhan GC. High body mass index and hypercholesterolemia: risk of
hypertensive disorders of pregnancy. Obstet Gynecol 1999;94:543–50.
ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology
Body mass index and prediction of pre-eclampsia
53 van Hoorn J, Dekker G, Jeffries B. Gestational diabetes versus obesity
as risk factors for pregnancy-induced hypertensive disorders and fetal
macrosomia. Aust N Z J Obstet Gynaecol 2002;42:29–34.
54 Weiss JL, Malone FD, Emig D, Ball RH, Nyberg DA, Comstock CH, et al.
Obesity, obstetric complications and cesarean delivery rate–a populationbased screening study. Am J Obstet Gynecol 2004;190:1091–7.
55 Yamamoto S, Douchi T, Yoshimitsu N, Nakae M, Nagata Y. Waist to hip
circumference ratio as a significant predictor of preeclampsia, irrespective of overall adiposity. J Obstet Gynaecol Res 2001;27:27–31.
56 Conde-Agudelo A, Belizan JM, Lindmark G. Maternal morbidity and
mortality associated with multiple gestations. Obstet Gynecol 2000;
95:899–904.
57 Qiu C, Sanchez SE, Larrabure G, David R, Bralley JA, Williams MA.
Erythrocyte omega-3 and omega-6 polyunsaturated fatty acids and
preeclampsia risk in Peruvian women. Arch Gynecol Obstet 2006;
274:97–103.
58 Institute of Medicine: Subcommittee on Nutritional Status and weight
Gain During Pregnancy IoMU. Subcommittee on Dietary Intake and
Nutritional Supplements During Pregnancy. Nutrition During Pregnancy: Part I, Weight Gain. Part II, Nutrient Supplements. Washington,
DC: National Academy Press, 1990.
59 O’Brien TE, Ray JG, Chan WS. Maternal body mass index and the risk of
preeclampsia: a systematic overview. Epidemiology 2003;14:368–74.
60 Altman DG. Practical Statistics for Medical Research, 1st edn. London:
Chapman & Hall, 1997.
61 Duckitt K, Harrington D. Risk factors for pre-eclampsia at antenatal booking: systematic review of controlled studies. BMJ 2005;330:565–71.
62 Cnossen JS, Leeflang MM, Mol BW, van der Post JAM, Khan KS, Gupta
JK, ter Riet G. Risk factors for pre-eclampsia at antenatal booking:
prediction or causality? 2006[http://bmj.bmjjournals.com/cgi/eletters/
330/7491/565]. Accessed 12 February 2007.
63 McTigue K, Larson JC, Valoski A, Burke G, Kotchen J, Lewis CE, et al.
Mortality and cardiac and vascular outcomes in extremely obese
women. JAMA 2006;296:79–86.
64 Seely EW, Solomon CG. Insulin resistance and its potential role in
pregnancy-induced hypertension. J Clin Endocrinol Metab 2003;88:
2393–8.
65 Xiong X, Wang FL, Davidge ST, Demianczuk NN, Mayes DC, Olson DM,
et al. Maternal smoking and preeclampsia. J Reprod Med 2000;45:
727–32.
66 Cnattingius S, Lambe M. Trends in smoking and overweight during
pregnancy: prevalence, risks of pregnancy complications, and adverse
pregnancy outcomes. Semin Perinatol 2002;26:286–95.
67 Adams KF, Schatzkin A, Harris TB, Kipnis V, Mouw T, BallardBarbash R, et al. Overweight, obesity, and mortality in a large prospective cohort of persons 50 to 71 years old. N Engl J Med 2006;
355:763–78.
68 Kuczmarski MF, Kuczmarski RJ, Najjar M. Effects of age on validity
of self-reported height, weight, and body mass index: findings from
the Third National Health and Nutrition Examination Survey, 1988–1994.
J Am Diet Assoc 2001;101:28–34.
69 Khan KS, Bachmann LM, Ter Riet G. Systematic reviews with individual
patient data meta-analysis to evaluate diagnostic tests. Eur J Obstet
Gynecol Reprod Biol 2003;108:121–5.
ª 2007 The Authors Journal compilation ª RCOG 2007 BJOG An International Journal of Obstetrics and Gynaecology
1485