Prediction of preterm birth in symptomatic women using decision tree modeling for biomarkers.

Abstract

The objective of the study was to use recursive partitioning (RP) to identify gestational age-specific and threshold values for infectious and endocrine biomarkers of imminent delivery.
RP was developed using a previously collected data set and then applied to a prospectively collected cohort of women in threatened preterm labor. Predictors of preterm birth were considered, including white blood cell count (WBC), corticotrophin-releasing hormone (CRH), cortisol, and maternal age.
At 22-27 weeks' gestation, WBC of greater than 12,000/mL was the most accurate predictor of delivery within 48 hours; at 28-31 weeks' gestation, CRH of greater than 684 pg/mL was the most accurate predictor; and at 32-26 weeks' gestation, CRH and maternal age were the most important variables.
These results indicate that maternal WBC greater than 12,000/mL prior to 28 weeks' gestation and CRH beyond 28 weeks are the most accurate biomarkers in predicting preterm birth within 48 hours. RP assists in establishing clinically relevant and gestational age-specific threshold levels for these variables.

Full text

Page 1

Prediction of preterm birth in symptomatic women
using decision tree modeling for biomarkers
Jacquelyn L. Hill, MSc; M. Karen Campbell, PhD; Guang Yong Zou, PhD; John R. G. Challis, PhD;
Gregor Reid, PhD; Hiroshi Chisaka, MD; Alan D. Bocking, MD
OBJECTIVE: The objective of the study was to use recursive partition-
ing (RP) to identify gestational age–specific and threshold values for
infectious and endocrine biomarkers of imminent delivery.
STUDY DESIGN: RP was developed using a previously collected data
set and then applied to a prospectively collected cohort of women in
threatened preterm labor. Predictors of preterm birth were considered,
including white blood cell count (WBC), corticotrophin-releasing hor-
mone (CRH), cortisol, and maternal age.
RESULTS: At 22-27 weeks’ gestation, WBC of greater than 12,000/mL
was the most accurate predictor of delivery within 48 hours; at 28-31
weeks’ gestation, CRH of greater than 684 pg/mL was the most accu-
rate predictor; and at 32-26 weeks’ gestation, CRH and maternal age
were the most important variables.
CONCLUSIONS: These results indicate that maternal WBC greater than
12,000/mL prior to 28 weeks’ gestation and CRH beyond 28 weeks are
the most accurate biomarkers in predicting preterm birth within 48
hours. RP assists in establishing clinically relevant and gestational
age–specific threshold levels for these variables.
Key words: corticotrophin-releasing hormone, infection, preterm
labor, recursive partitioning
Cite this article as: Hill JL, Campbell MK, Zou GY, et al. Prediction of preterm birth in symptomatic women using decision tree modeling for biomarkers. Am J
Obstet Gynecol 2008;198:468.e1-468.e9.
P
arefewmethodsavailabletoreliablypre-
dict true preterm labor in women who
present with symptoms of labor. Cur-
retermbirthcontinuestobeamajor
challenge in obstetrics,1and there
rentlytranscervicalultrasoundmeasure-
ments and/or cervicovaginal fetal fi-
bronectin levels are the most commonly
used diagnostic tools.2These techniques
bothhavehighnegativepredictivevalues
butrelativelylowpositivepredictiveval-
ues. Consequently, many women and
theirfetusesareexposedunnecessarilyto
tocolytic drugs and corticosteroids and
are admitted to hospital. These admis-
sions and treatments create emotional
and financial stress on the mother and
family as well as significant financial
coststosociety.Weandothershavepre-
viously shown that maternal cortico-
tropin-releasing hormone (CRH) levels
are higher in symptomatic women who
give birth within 48 hours, compared
with those who do not.3,4
Our previous prospective study in-
volving218symptomatic
showedthat,atdifferentgestationalages,
different factors were associated with
preterm birth within 48 hours.5White
blood cell count (WBC), a marker of
subclinical infection, was the only vari-
able shown to be predictive at 22-27
weeks, supporting previously published
worksuggestiveofthegestationalagede-
pendence (particularly less than 32
weeks) of the relationship between
markersofinfection
women
and preterm
birth.6-8Beyond 28 weeks’ gestation,
CRH and adrenocorticotropic hormone
werepredictorsofbirthwithin48hours,
indicating an early activation of the fetal
hypothalamic-pituitary-adrenal
Additionally, at 32-36 weeks’ gestation,
demographic and lifestyle variables were
significantly associated with preterm
birth.
For the purposes of clinical decision
making, it is useful to identify threshold
levels of biomarkers, above which pre-
term birth is probable. Recursive parti-
tioning (RP)to
trees9,10has been increasingly utilized to
identify accurate biomarker cutpoints
for obstetrical conditions including pre-
term birth.11-13Prediction rules derived
from decision tree analysis and based on
critical thresholds of variables are sim-
pler and more clinically applicable than
the results of more conventional analy-
ses. For example, logistic regressions
(LR) can treat predictive variables as
continuous or categorical and can esti-
mateeffectswithstatisticalaccuracy,but
the interpretation of parameters are
more abstract and do not as readily pro-
vide decision-making thresholds. How-
ever, RP analysis has limitations, includ-
ing the potential for spurious findings
because the findings are highly depen-
axis.
producedecision
From the Departments of Epidemiology and
Biostatistics (Ms Hill and Drs Campbell and
Zou), Obstetrics and Gynaecology (Dr
Campbell), Surgery and Immunology, and
Microbiology, and the Canadian R&D
Centre for Probiotics, Lawson Health
Research Institute (Dr Reid), the University
of Western Ontario, London, and the
Departments of Physiology and Obstetrics
& Gynaecology, University of Toronto,
Toronto (Drs Challis, Chisaka, and
Bocking), ON, Canada, and the Tohoku
University Graduate School of Medicine,
Sendai, Japan (Dr Chisaka).
Presented at the 26th Annual Scientific
Meeting of the American Gynecological and
Obstetrical Society, Chicago, IL, Sept. 26-29,
2007.
Received June 4, 2007; revised Nov. 7, 2007;
accepted Jan. 11, 2008.
Reprints not available from the authors.
This study was supported by grants from the
Canadian Institutes of Health Research.
0002-9378/$34.00
© 2008 Mosby, Inc. All rights reserved.
doi: 10.1016/j.ajog.2008.01.007
AGOSPapers
www.AJOG.org
468.e1
American Journal of Obstetrics & Gynecology APRIL 2008

Page 2

dent on the data set used in the analy-
sis.14We know of no previous reports
describing the use of classification trees
to predict preterm birth among symp-
tomatic women with the biomarkers of
interest in this study.
The objectives of this study were: (1)
toexplore,byRP,therelationshipofges-
tational age–specific endocrine, infec-
tion, and other variables with delivery
within 48 hours using our previously
published data,5(2) to compare results
derived by LR and RP; and (3) in a new
prospectivecohortofsymptomaticpreg-
nant women, to apply the threshold val-
uesforinfectiousandendocrinebiomar-
kers developed in the previous cohort
using RP. We were also interested in the
potentialutilityofassessmentofcervico-
vaginal fetal fibronectin measurements,
in these women because at the time of
this study, there was minimal informa-
tion regarding its predictive value for
preterm birth in symptomatic women.
MATERIALS AND METHODS
This study was approved by the Univer-
sity of Western Ontario Review Board
forResearchInvolvingHumanSubjects.
Briefly, RP was developed using a previ-
ously collected data set and then applied
to a prospectively collected cohort
termed validation study.
Derivation data and analysis
Data for 201 of the 218 women in our
earlier study5, excluding 17 for missing
data, were partitioned with respect to
preterm birth within 48 hours of admis-
sionornot,bythealgorithmofAtkinson
and Therneau,15,16implemented in S
Plus 6.1 (Seattle, WA). The resulting
classification tree graphically summa-
rizesthepredictors,inorderoftheirdis-
criminating power relative to each
other.17,18The method also selects the
optimum threshold level for each pre-
dictor. The highest-risk subgroups are
easilyidentifiedinthetreebythecombi-
nations of predictors.11Analysis as-
signed equal misclassification costs for
false positives as for false negatives and
used the Gini index for splitting nodes.
For all 3 gestational age categories, the
following predictor variables were con-
sidered: WBC, cortisol, cervical dilata-
tion2-4cm,CRH,multiparous,smoker,
maternalage,contractionsatadmission,
and cervical effacement of at least 50%.
Maternal age, endocrine variables, and
WBC were continuous scale variables,
whereascervical
smoking, contractions, and cervical ef-
facement were binary variables.
InadditiontodevelopingtheRPalgo-
rithm, for 32-36 weeks’ gestation only,
sample size permitted the further evalu-
ationoftheuseofthesecutpointsinaLR
analysis.
dilatation, parity,
Validation data
Two hundred seven women with single-
ton pregnancies, who presented with
threatened preterm labor to St Joseph’s
Health Care London, a tertiary perinatal
center in London, Canada, from April
1999 to July 2002, were recruited to the
study after obtaining informed consent.
Sample size was based on our previous
study with an anticipated birth rate
within 48 hours of presentation of 30%.
Exclusionsweremultifetalgestations,fe-
tal anomalies, maternal diabetes melli-
tus, abruptio placenta, preeclampsia, in-
trauterine growth restriction, cervical
dilatation greater than 4 cm, ruptured
membranes, and clinical signs of infec-
tion (WBC greater than 18,000/mL).
A sterile speculum examination was
performed in all women prior to digital
examination and cervicovaginal swabs
were taken to determine the presence of
bacterialvaginosis(BV)byNugentscor-
ing.19In women who were eligible, cer-
vicovaginal swabs were taken for deter-
mination of fetal fibronectin levels using
aDacronswab.Swabswereimmediately
placed in buffer provided by the manu-
facturer(Adeza
Sunnyvale, CA). Measurements were
madeusingacommerciallyavailableen-
zyme-linkedimmunosorbentassay(Ad-
ezaBiomedicalCorp).Resultswerecon-
sideredpositiveat50ng/mLorgreater.20
Maternal blood samples were ob-
tained at the time of recruitment, and
WBCwasdeterminedfromwholeblood
before centrifugation to collect plasma.
Aliquots of plasma were immediately
frozen and stored at -80oC for later de-
BiomedicalCorp,
termination of CRH and cortisol by ra-
dioimmunoassay.4Intra- and interassay
coefficients of variation for CRH previ-
ously measured were 12% and 6.5%, re-
spectively.Interassaycoefficientsofvari-
ation for cortisol measurements were
approximately 11%. CRH and cortisol
valueswerenormalizedusinggestational
age appropriate norms from an earlier
studyofsubjectswhogavebirthatterm4
using the Standardization Procedure of
SAS (version 8.02, Cary, NC). Gesta-
tionalagewasdeterminedbyacombina-
tion of last menstrual period and early
ultrasound examination.
RESULTS
Derivation study
Usingthederivationdataat22-27weeks’
gestational age, RP analysis indicated
that WBC was the most accurate predic-
tor of preterm delivery within 48 hours
(Figure 1). A threshold of the 48th per-
centile of these data (12,000/mL) was
important and, for those with WBC
12,000/mL or greater, further differenti-
ation was provided by the next most ac-
curate predictor, cortisol 55th percentile
or greater(202
women at this gestational age exceeded
both thresholds and, of these, 10 (71%)
delivered within 48 hours. In contrast,
none of the 21 women with WBC less
than 12,000/mL delivered within 48
hours.Competitivevariablesforthefirst
predictive split in the tree, in lieu of
WBC, were cortisol and cervical dilata-
tion. At higher levels of WBC, competi-
tive variables for the second split, in lieu
of cortisol, were maternal age and WBC
(at a different cutpoint).
For women of 28-31 weeks’ gesta-
tional age, CRH was the most accurate
predictor of birth within 48 hours. Of 8
women exceeding the CRH threshold of
the 83rd percentile (684 pg/mL), 5
(63%) gave birth within 48 hours as
compared with 4 of the 43 (9%) of
women with CRH less than 684 pg/mL
(Figure2).Competitivevariables,inlieu
of CRH, were WBC, cortisol, cervical ef-
facement, and maternal age.
For gestational ages 32-36 weeks (Fig-
ure 3), CRH was the most predictive
variable with a first split at a threshold
ng/mL).Fourteen
www.AJOG.org
AGOS Papers
APRIL 2008 American Journal of Obstetrics & Gynecology 468.e2

Page 3

valueofthe61stpercentileforthisgesta-
tional age category (1368 pg/mL). For
those with CRH less than 1368 pg/mL,
maternal age (30 years or older) was the
next splitting variable. For those with
CRH 1368 pg/mL or greater, cortisol
18th percentile or greater (152 ng/mL)
was the next splitting variable. As indi-
cated in Figure 3, women with CRH lev-
els of 1368 pg/mL or greater and cortisol
levels of greater than 152 ng/mL had an
83% chance of delivering within 48
hours. Women with CRH levels of less
than 1368 pg/mL but maternal age 30
yearsorolderhada75%chanceofdeliv-
ering within 48 hours. The low-risk
groups were younger women with low
CRH or women of any age for whom
CRH, but not cortisol, was elevated.
Table 1 presents the results of LR for
the derivation data, within the 32-36
weeks’ gestational age group, incorpo-
rating all 3 variables identified by recur-
sive partitioning algorithm of Atkinson
and Therneau (RPART). The indepen-
dent effects of CRH of 1368 pg/mL or
greater,cortisolof152ng/mLorgreater,
and maternal age of 30 years or older
were to increase the odds of birth within
48 hours by factors of 5.1, 3.7, and 4.7,
respectively.
Validation study
Of the 207 symptomatic women re-
cruited to the validation study, 101
(49%)wereadmittedtothehospitalwith
a diagnosis of threatened preterm labor.
Forty-six(22%)receivedacourseofglu-
cocorticoids and 13 (6.3%) received to-
colytics. Forty-eight of these women
(23%) gave birth less than 37 weeks, and
of those, 17 (8.2%) gave birth within 48
hours. One hundred fifty women were
previously pregnant, and of these, 46
(30.7%) had a previous preterm birth.
Demographicandclinicalcharacteristics
of the women enrolled are provided in
Table 2. Nugent scores were available in
185 women and the prevalence of BV
was 37.3%. There was no difference in
Nugent score between women who gave
birth within 48 hours, compared with
those who did not (Table 3).
For all gestational age groups in the
validationstudy,thereishighsensitivity,
goodspecificity,butlowpositivepredic-
tive values because of the low prevalence
of the outcome in the data set (Table 4).
At 22-27 weeks’ gestation, all of the sub-
jects delivering within 48 hours had pe-
ripheral WBC counts of 12,000/mL or
more, resulting in a highly sensitive test.
None of the subjects with WBC counts
less than this threshold delivered within
48hours,resultinginverygoodnegative
predictive value.
Cortisol was not included in deter-
mining test accuracy in this gestational
age group because a large proportion of
records were missing cortisol data. For
28-31 weeks’ gestation, all of the women
fromthetestsetwhodeliveredwithin48
hours had CRH values above the thresh-
old,indicatingahighlysensitivetest.The
majority of women who did not deliver
within 48 hours (85%) had CRH levels
below the threshold of 684 pg/mL, indi-
cating a highly specific predictor of pre-
term birth in symptomatic women. Be-
cause only 20% of women with CRH
valuesabove684pg/mLdeliveredwithin
48hours,CRHhadpoorpositivepredic-
tive value for this sample of women but
had excellent negative predictive value.
FIGURE 1
Decision tree for 22-27 weeks’ gestational
age in the derivation data set
cortisol < 202 ng/mL
Delivery <48h = 2 (25%)
Delivery ≥48h = 6 (75%)
Delivery <48h = 10 (71%)
Delivery ≥48h = 4 (29%)
WBC < 12,000/mLWBC ≥ 12,000/mL
Delivery <48h = 12 (28%)
Delivery ≥48h = 31 (72%)
Delivery <48h = 0 (0%)
Delivery ≥48h = 21 (100%)
cortisol ≥ 202 ng/mL
Delivery <48h = 12 (55%)
Delivery ≥48h = 10 (45%)
Shaded boxes are terminal nodes in which the majority of subjects delivered within 48 hours.
Nonshaded boxes are terminal nodes in which the majority of subjects delivered more than 48 hours
after presentation.
Hill.Predictionofpretermbirthusingdecisiontreemodelingforbiomarkers.AmJObstetGynecol2008.
FIGURE 2
Decision tree for 28-31
weeks’ gestational age
in the derivation data set
CRH < 684 pg/mL
CRH ≥ 684 pg/mL
Delivery <48h = 5 (63%)
Delivery ≥48h = 3 (37%)
Delivery <48h = 9 (18%)
Delivery ≥48h = 42 (82%)
Delivery <48h = 4 (9%)
Delivery ≥48h = 39 (91%)
Shaded boxes are terminal nodes in which the
majority of subjects delivered within 48 hours.
Nonshaded boxes are terminal nodes in which
the majority of subjects delivered more than 48
hours after presentation.
Hill.Predictionofpretermbirthusingdecisiontree
modelingforbiomarkers.AmJObstetGynecol2008.
AGOS Papers
www.AJOG.org
468.e3
American Journal of Obstetrics & Gynecology APRIL 2008

Page 4

The prediction algorithm for the
32-36 weeks’ gestational age group was
applied using CRH as the primary vari-
able with and without maternal age as a
secondary variable. The sensitivity was
high (80%), and when using CRH alone
as a predictive variable, the specificity
was high (84%). Again, the low preva-
lence of the outcome is responsible for
the low positive predictive value.
Fetal fibronectin levels were ob-
tained in only 104 of the women en-
rolled in the validation study because
the remainder of women were ineligi-
ble for this test because of a recent pel-
vic examination or vaginal bleeding.
Fetal fibronectin was positive in 8
women and negative in 96. The posi-
tive and negative predictive values for
delivery within 48 hours based on a
positive fetal fibronectin level were
50% and 98%, respectively.
COMMENT
Theseresultsareconsistentwithourpre-
vious findings that maternal CRH levels
are predictive of preterm birth within 48
hours of presentation in symptomatic
women at 32 weeks’ gestation and be-
yond.Thisstudyrestrictedthesampleto
women with intact membranes and
therefore excludes a major confounder
in our previous studies (ruptured mem-
branes). We used RP for the first time to
select cutpoints of biological markers to
predictpretermdeliveryinsymptomatic
women.
At 22-27 weeks’ gestation, maternal
blood WBC count of 12,000/mL or
more, a marker of subclinical infection
or inflammation, was the most effective
indicator of preterm birth. Because sub-
jects with very high WBC were already
excludedfromthisstudy,thisrepresents
FIGURE 3
Decision tree for 32-36 weeks’ gestational age in the derivation data set
CRH < 1368 pg/mL CRH ≥ 1368 pg/mL
Delivery <48h = 53 (50%)
Delivery ≥48h = 54 (50%)
Delivery <48h = 23 (36%)
Delivery ≥48h = 41 (64%)
Maternal Age < 30 yrs
Maternal Age ≥ 30 yrs
Cortisol < 152 ng/mLCortisol ≥ 152 ng/mL
Delivery <48h = 30 (70%)
Delivery ≥48h = 13 (30%)
Delivery <48h = 12 (75%)
Delivery ≥48h = 4 (25%)
Delivery <48h = 1 (13%)
Delivery ≥48h = 7 (87%)
Delivery <48h = 29 (83%)
Delivery ≥48h = 6 (17%)
Delivery <48h = 11 (23%)
Delivery ≥48h = 37 (77%)
Shaded boxes are terminal nodes in which the majority of subjects delivered within 48 hours. Nonshaded boxes are terminal nodes in which the majority
of subjects delivered more than 48 hours after presentation.
Hill.Predictionofpretermbirthusingdecisiontreemodelingforbiomarkers.AmJObstetGynecol2008.
TABLE 1
Results of using multiple logistic regression to predict preterm
birth at 32-36 weeks’ gestational age (n ? 100 because
of casewise deletion for missing data on individual variables)
Dichotomized variable
CRH
..............................................................................................................................................................................................................................................
CortisolP ? .036
..............................................................................................................................................................................................................................................
Maternal ageP ? .002
..............................................................................................................................................................................................................................................
Area under the curve of receiver-operating curve using the test data ? 0.77.
CI, confidence interval.
Variable fit
P ? .001
Odds ratio (90% CI)
5.1 (2.3, 11.3)
3.7 (1.3, 10.9)
4.7 (2.1, 10.9)
Hill.Predictionofpretermbirthusingdecisiontreemodelingforbiomarkers.AmJObstetGynecol2008.
TABLE 2
Demographic and clinical
characteristics of the 207 women
recruited to the validation study
Characteristic
Married/common law
...........................................................................................................
Completed high school
...........................................................................................................
Completed college/university
...........................................................................................................
Admitted to hospital
...........................................................................................................
Primiparous
...........................................................................................................
Smoker
...........................................................................................................
Glucocorticoid administration
...........................................................................................................
Caesarian delivery
...........................................................................................................
Treatment with tocolytics
...........................................................................................................
Of the 150 multiparous women, 46 (30.7%) had a
history of preterm birth.
Number
(%)
184 (88.9)
170 (82.0)
128 (62.0)
101 (49.0)
57 (27.5)
55 (26.8)
46 (22.0)
37 (18.0)
13 (6.3)
Hill.Predictionofpretermbirthusingdecisiontree
modelingforbiomarkers.AmJObstetGynecol2008.
www.AJOG.org
AGOS Papers
APRIL 2008 American Journal of Obstetrics & Gynecology 468.e4

Page 5

a change within the normal range. Evi-
dence shows that births of very young
gestational age are associated with infec-
tion,bothclinicallyandsubclinically.6,21
Another study compared amniotic
fluid WBC to maternal blood WBC with
respect to outcome of histological cho-
rioamnionitis in laboring women with
intact membranes and thus chose a
higher cutpoint (14,700/mL) to distin-
guish women with the outcome, based
on a compromise between sensitivity
and specificity.22WBC values at this
higherlevelorhigherresultedinasignif-
icantly shorter amniocentesis to delivery
interval than those women with WBC
less than 14,700/mL. There is consider-
able evidence that there is a generalized
activation of the immune system during
labor consistent with these findings.
Our findings not only support previ-
ous parametric multivariable analysis
but also add to this by identifying a
threshold WBC value of 12000/mL,
above which the probability of preterm
delivery is high in this sample. This is of
relevance to clinical decision making.
The high negative predictive value of
WBC when less than 12000/mL is reflec-
tive of the overall low prevalence of pre-
term birth in the sample. WBC less than
12000/mL indicates a low probability of
delivering preterm and a low risk of in-
trauterineor maternal
infection.
At28-31weeks’gestation,theRPalgo-
rithm selected CRH of greater than 684
pg/mL as the most accurate predictor of
birthwithin48hours.Atthisgestational
age,anegativetestresultalsohasimpor-
subclinical
tance for clinical decision making. Test-
ing of the CRH threshold in the valida-
tionsetshowedhighdiagnosticaccuracy
andnegativepredictivevalueamongthis
sample of women. There were however,
ahighnumberoffalsepositivesattribut-
able to the low prevalence of the out-
come of interest. The CRH threshold of
greater than 684 pg/mL identified by RP
may therefore be useful as a diagnostic
tool only for deciding which women
should not receive tocolytic and glu-
cocorticoid therapy at these gestational
ages.
A positive test value may not be as in-
formative. The presence of subclinical
infection is also an important factor for
pretermbirthat28-31weeks’gestational
age because WBC was a competitor for
the first predictive split of the RP tree,
but it is not of incremental predictive
value once a prediction has been gener-
ated using CRH.
At 32-36 weeks’ gestation, WBC as a
marker of infection is not predictive of
preterm birth, a finding consistent with
our previous multivariable analyses.5
The most predictive variable was CRH
with a threshold of 1368 pg/mL or
greater, indicating a 70% risk of birth
within 48 hours. Analysis of the deriva-
tion data indicated that prediction
TABLE 3
Nugent scores for 207 subjects in the validation study
BV (Nugent
score)(n ? 207)
Normal (0-3)55 (29.7%)
..............................................................................................................................................................................................................................................
Intermediate (4-6)61 (33.0%)
..............................................................................................................................................................................................................................................
BV (7-10)69 (37.3%)
..............................................................................................................................................................................................................................................
Missing22
All subjectsDelivery within
48 h (n ? 17)
4 (33.3%)
Delivery greater
than 48 h (n ? 190)
51 (29.5%)
3 (25.0%)58 (33.5%)
5 (41.7%) 64 (37.0%)
5 17
Hill.Predictionofpretermbirthusingdecisiontreemodelingforbiomarkers.AmJObstetGynecol2008.
TABLE 4
Frequencies of preterm birth within 48 h and performance of RP algorithm
in both the data used to derive the RP algorithm and the validation data
Delivery within 48 h (%)
Accuracy and predictive values
of cutpoints applied to validation data
Gestational age
at presentation
22-27ksa
................................................................................................................................................................................................................................................................................................................................................................................
28-31 weeks9 (18.0) 2 (3.6)
................................................................................................................................................................................................................................................................................................................................................................................
32-36 wksb
53 (50.0)12 (11.3)
................................................................................................................................................................................................................................................................................................................................................................................
32-36 wksc
................................................................................................................................................................................................................................................................................................................................................................................
Total with outcome74 (37.0) 17 (8.2)
................................................................................................................................................................................................................................................................................................................................................................................
Total in set201 207
................................................................................................................................................................................................................................................................................................................................................................................
Number of women from derivation data who presented with threatened preterm labor were 43, 51, and 107 at gestational ages of 22-27, 28-31, and 32-36 weeks, respectively. For the validation
data, there were 45, 56, and 106 women, respectively.
PPV, positive predictive value; NPV, negative predictive value.
aThe RP algorithm was applied using only WBC as predictive variable.
bThe RP algorithm was applied using CRH and age as predictive variables
cThe RP algorithm was applied using only CRH as predictive variables.
Derivation
data
12 (28.0)
Validation
data
3 (6.7)
Sensitivity
(%)
100
Specificity
(%)
71
PPV
20
NPV
100
10085 20 100
80 5422 94
80 844496
Hill.Predictionofpretermbirthusingdecisiontreemodelingforbiomarkers.AmJObstetGynecol2008.
AGOS Papers
www.AJOG.org
468.e5
American Journal of Obstetrics & Gynecology APRIL 2008