Keywords predictive factors - preterm delivery - maintenance tocolysis
Preterm birth accounts for half of the childhood neurodevelopmental disabilities and
almost 75% of perinatal deaths occur in infants born before 37 weeks' gestation.[1 ]
[2 ] Although approximately 75% of women presenting with threatened preterm labor remain
initially undelivered after an initial course of tocolytics of 48 hours, their risk
of preterm delivery after this period is still increased; 65% of women deliver before
37 weeks.[3 ] Unfortunately, the risk is difficult to estimate for the individual woman. Previously,
several factors such as short cervical length and positive fetal fibronectin (fFN)
have been shown to be predictors of early delivery in pregnant women.[4 ]
[5 ] It is important to identify women who will deliver within 1 week because women with
a high risk may benefit from prolonged hospitalization in a tertiary center[6 ] and other management options for preterm labor. Since preterm birth is multifactorial,[7 ] it is likely that a single test alone cannot predict preterm birth accurately.
In the present study, we assessed which demographic and clinical characteristics,
results of vaginal examination and laboratory variables are predictive factors for
delivery within 7 days in women with threatened preterm labor who had not delivered
within 48 hours after initial treatment.
Materials and Methods
Setting
This is a secondary analysis of the APOSTEL-II trial (Assessment of Perinatal Outcome
with Sustained Tocolysis in Early Labor), performed between June 2008 and February
2010. Women with threatened preterm labor between 26+0 and 32+2 weeks gestational age were randomly allocated to maintenance tocolysis with nifedipine
or placebo. At that point, women had already been treated with tocolytics for 48 hours
to complete a course of corticosteroids. Both the randomized controlled trial and
the secondary analysis were approved by the Institutional Review Board of the participating
hospitals. The design and main results have been previously published.[8 ]
[9 ] All participants gave informed consent. Because the trial has shown that maintenance
therapy is ineffective in prolonging pregnancy and improving perinatal outcome, both
women with maintenance tocolytic therapy and women with placebo were included in the
analysis. Also women refusing randomization, but consenting follow-up of their data
(the nonrandomized group) were included in the present study. Data were entered in
a database by research nurses and midwives and validation of the data was performed
by the lead author of this article.
Outcome
The outcome variable of primary interest of our prediction models was delivery within
7 days after initial 48 hours of arrest of preterm labor.
Predictors under Study
Based on the literature[10 ]
[11 ]
[12 ]
[13 ] and expert experience, we identified candidate predictors for delivery within 7
days after arrest of threatened preterm labor. Candidate predictors were maternal
age, ethnicity, education level, body mass index, history of preterm birth before
32 weeks and before 37 weeks, multifetal gestation, premature prelabor rupture of
membranes (PPROM), vaginal bleeding at study entry, Group B Streptococcus status,
C-reactive protein (CRP) at study entry, fFN at study entry, dilatation at study entry
(digital exam), and cervical length at study entry (ultrasound). A combination of
parity and a history of preterm birth were categorized into multiparous women with
a prior birth ≥ 37 weeks' gestation (reference), nulliparous women, multiparous women
with a prior birth < 32 weeks, and multiparous women with a prior birth 32 to 37 weeks.
We developed two separate models, one for women with PPROM (model 1) in whom the variables
dilatation, cervical length, and fFN had not been assessed, and one for women without
PPROM (model 2) which included these variables.
Data Analysis
Associations between the candidate predictors and delivery within 7 days were analyzed
with logistic regression analysis. Although generally not recommended,[14 ] we performed a preselection based on the univariable analyses p -value (<0.20) to retain a reasonable number of events per variable in the multivariable
model.[15 ]
Maternal age, body mass index, gestational age, CRP, dilatation, and cervical length
were analyzed as continuous variables. Linearity of their association with the outcome
was assessed using cubic spline analyses.[16 ] In case of no linearity, variables were transformed with logistic transformation
or the addition of a quadratic term according to the shape of their plots. All other
variables were dichotomous. To correct for the allocated intervention in the original
trial, we also included intervention as a variable in the analysis.
Various subjects had missing values, ranging from 0% missing values in maternal age
to 60% in fFN in women without PPROM. Because missing values could be selectively
missing, complete case analysis may yield to biased results.[17 ]
[18 ]
[19 ] Hence, before performing the analyses, the missing values were imputed using multiple
imputation (10 times). The imputation model included all potential predictors as well
as the outcome of interest.[16 ]
[20 ]
[21 ]
[22 ]
In prognostic model research, there is a chance of finding spurious predictors and
overestimated regression coefficients.[16 ]
[20 ]
[23 ] Such overfitted models will create too extreme and optimistic predictions when applied
in new cohorts. To assess the degree of overfitting or optimism in this study, we
(internally) validated the models using bootstrapping techniques.[24 ] This yielded a shrinkage factor, with which the regression coefficients were multiplied
(uniformly shrunken) to adjust for overfitting and optimism. All analyses including
the bootstrapping techniques were performed in R version 2.10.0 (The R Foundation
for Statistical Computing, 2009, Vienna, Austria).
The ability of the two models to discriminate between women who delivered within or
beyond 7 days was quantified with the area under the receiver operating characteristic
curve (c -statistic). Calibration was assessed by comparing the predicted probabilities with
the observed frequencies of delivery within 7 days. The agreement between the observed
proportions of delivery within 7 days and the predicted risks was studied with calibration
plots,[16 ]
[25 ] which provided additional insight in the distribution of the predicted outcome incidences.
Results
In the APOSTEL-II trial, 636 women were eligible for participation, of whom 406 women
gave informed consent for randomization between maintenance tocolysis with nifedipine
(201 women) and placebo (205 women) ([Fig. 1 ]). The other 230 women refused randomization but gave informed consent for follow-up
of their medical data. There was no loss to follow-up in the randomization group,
while eight women were lost to follow-up in the nonrandomization group.
Fig. 1 Trial profile of the APOSTEL-II trial (Assessment of Perinatal Outcome with Sustained
Tocolysis in Early Labor).
Baseline characteristics for the total cohort of 628 women for complete cases (n = 30) and for cases with one or more missing variable (n = 598) are shown in [Appendix 1 ]. Values after imputation are displayed in [Table 1 ]. Delivery within 1 week after arrest of threatened preterm labor occurred in 151
women (24%), 61 of 144 (42%) women with PPROM and 90 of 484 (19%) women without PPROM
(p < 0.001). This indicates that PPROM is a major predictive factor for delivery within
7 days. Some variables were not linear with the outcome. For women without PPROM,
maternal age and CRP were transformed with logistic transformation.
Appendix 1
Baseline demographics and clinical characteristics for complete cases and cases with
at least one missing value
Total study population n = 628
Women with PPROM n = 144
Women without PPROM n = 484
Complete cases n = 24
Incomplete cases[a ] n = 120
p -Value
Complete cases n = 6
Incomplete cases n = 478
p -Value
Age (y)[b ]
30.7 ± 4.4
30.8 ± 5.2
0.90
31.2 ± 6.3
29.3 ± 5.3
0.38
Non-Caucasian ethnicity
6 (25)
26 (22)
1.0
0 (0)
75 (16)
1.0
Low educational level
16 (67)
29 (24)
0.65
4 (67)
139 (29)
0.59
Nulliparous
12 (50)
63 (53)
1.0
4 (67)
273 (57)
0.96
Prior preterm birth < 32 wk
3 (13)
12 (10)
1.0
0 (0)
63 (13)
0.73
Prior preterm birth < 37 wk
5 (21)
25 (21)
1.0
0 (0)
103 (22)
0.43
Body mass index
25.2 ± 5.8
24.3 ± 5.4
0.46
21.1 ± 1.8
23.1 ± 4.3
0.25
Multifetal gestation
4 (17)
24 (20)
0.93
2 (33)
104 (22)
0.86
Vaginal bleeding
8 (33)
24 (20)
0.24
2 (33)
84 (18)
0.64
Laboratory examination at study entry
C-reactive protein (g/L)
8 (4–46)
10 (3–29)
0.77
7 (4–17)
23 (8–30)
0.32
Streptococcus Group B positive
4 (17)
18 (15)
0.43
1 (17)
54 (11)
1.0
Fibronectin status positive
–
–
–
3 (50)
45 (9.4)
< 0.001
Vaginal examination at study entry
Dilatation at study entry
–
–
–
1 (0–2)
1 (0–2)
0.75
Cervical length at study entry, cm
–
–
–
23 (15–30)
24 (4–31)
0.76
Randomized
0.006
0.10
No
1 (4)
42 (35)
0 (0)
179 (38)
Yes, placebo
9 (38)
39 (32)
4 (67)
153 (32)
Yes, nifedipine
14 (58)
39 (32)
2 (33)
146 (31)
Delivery < 7 d
7 (29)
51 (43)
0.77
2 (33)
80 (17)
0.62
Abbreviation: PPROM, premature prelabor rupture of membranes.
a Incomplete cases are cases with at least one missing value.
b Data are mean ± standard deviation, median (interquartile range) or number (%).
Table 1
Baseline demographics and clinical characteristics for the total study cohort
Total study population (n = 628)
Value after imputation
Age (y)[a ]
29.7 ± 5.3
Non-Caucasian ethnicity
117 (19)
Low educational level[b ]
368 (59)
Parity and prior preterm birth
Prior birth ≥37 wk
146 (23)
Nulliparous
353 (56)
Prior preterm birth < 32 wk
74 (12)
Prior preterm birth 32–37 wk
55 (9)
Body mass index[c ]
22.5 (20.4–26.4)
Multifetal gestation
135 (21)
PPROM
144 (23)
Vaginal bleeding
118 (19)
Laboratory examination at study entry
C-reactive protein (g/L)
8 (3–24)
Streptococcus Group B positive
139 (22)
Fibronectin status positive
189 (30)
Vaginal examination at study entry
Dilatation at study entry
1 (0–2)
Cervical length at study entry, mm
23 (15–31)
Randomized
No
222 (35)
Yes, placebo
205 (33)
Yes, nifedipine
201 (32)
Delivery < 7 d
151 (24)
Abbreviation: PPROM, premature prelabor rupture of membranes.
a Data are mean ± standard deviation, median (interquartile range) or number (%).
b Low educational level is defined as primary, secondary, or lower professional school
as highest finished education.
c The body mass index is weight (kg) divided by square height (m2 ).
[Table 2 ] summarizes the baseline characteristics of the women who had PPROM at inclusion
for women who delivered within 1 week versus those who delivered beyond that week.
The results of the univariable and multivariable analyses for all women with PPROM
are shown in the same table. In the univariable analysis, variables related to delivery
within 7 days in women with PPROM were nulliparity (odds ratio [OR], 3.6; 95% confidence
interval [CI], 1.6–8.5) and prior preterm birth 32 to 37 weeks (OR, 3.5; 95% CI, 1.0–12
as compared with a prior birth ≥ 37 weeks). After backward selection, nulliparity,
prior preterm birth < 32 and 32 to 37 weeks, and vaginal bleeding were included in
the model.
Table 2
Univariable and multivariable analyses for the prediction of delivery within 7 days
after successful 48 hours treatment of threatened preterm labor in women with PPROM
Women with PPROM (n = 144, 23%)
Delivery < 7 d
Delivery > 7 d
Univariable analysis
Multivariable analysis
Odds ratio (95% CI)[a ]
p -Value
Beta coefficient
Odds ratio (95% CI)
Characteristic
n = 61 (43%)
n = 83 (58%)
Age (y)[b ]
31.4 ± 5.4
30.4 ± 4.7
1.04 (0.98–1.12)
0.21
Non-Caucasian ethnicity
13 (21)
21 (25)
0.80 (0.35–1.81)
0.59
Low educational level
34 (56)
52 (63)
0.75 (0.31–1.83)
0.53
Parity and prior preterm birth
Prior birth ≥ 37 wk
10 (16)
31 (37)
Reference
Nulliparous
41 (66)
35 (42)
3.63 (1.56–8.47)
0.003
1.02
2.77 (1.15–6.65)
Prior preterm birth < 32 wk
3 (5)
10 (12)
0.93 (0.21–4.06)
0.92
− 0.015
0.99 (0.22–4.39)
Prior preterm birth 32–37 wk
8 (13)
7 (8)
3.54 (1.03–12.2)
0.046
0.99
2.70 (0.76–9.58)
Body mass index (kg/m2 )
22.8 (20.5–25.3)
24.0 (20.5–28.6)
0.96 (0.89–1.03)
0.53
Multifetal gestation
14 (23)
14 (17)
1.49 (0.65–3.42)
0.34
Vaginal bleeding
17 (28)
15 (18)
1.73 (0.78–3.82)
0.18
0.57
1.77 (0.75–4.17)
C-reactive protein (g/L)
10 (3–31)
9 (3–30)
1.00 (0.99–1.02)
0.77
Streptococcus Group B positive
15 (24)
20 (24)
0.98 (0.34–2.83)
0.97
Randomized
No
23 (38)
20 (24)
Reference
Yes, placebo
19 (31)
29 (35)
0.55 (0.24–1.28)
0.17
Yes, nifedipine
19 (31)
34 (41)
0.48 (0.21–1.11)
0.085
Abbreviation: PPROM, premature prelabor rupture of membranes.
a Averaged over the 10 imputation sets using Rubin rules. Intercept − 1.0760. c -statistic 0.68 (0.60–0.77). Coefficients were shrunken with an average shrinkage
factor 0.72.
b Data are mean ± standard deviation, median (IQR) or number (%). Percentages may not
sum to 100 because of rounding. Absolute numbers are based on the mean of 10 imputations.
[Table 3 ] shows the baseline characteristics of the women without PPROM at inclusion, also
divided in women who delivered within 1 week and women who delivered beyond 1 week.
In the univariable analysis, variables related to delivery within 1 week were vaginal
bleeding (OR, 4.6; 95% CI, 2.8–7.8), positive fFN status (OR, 14.97; 95% CI, 5.1–44),
dilatation (OR, 1.9; 95% CI, 1.5–2.4), cervical length (OR, 0.4; 95% CI, 0.3–0.5),
and placebo study medication (OR, 2.0; 95% CI, 1.1–3.5). After backward selection,
maternal age, vaginal bleeding, positive fFN status, and cervical length were included
in the model. Both multivariable models showed moderate to good discriminative ability,
with c -statistics of 0.68 (95% CI, 0.60–0.77) for the model with PPROM and 0.89 (95% CI,
0.84–0.93) for the model without PPROM. Calibration plots of both models are shown
in [Fig. 2a ], [b ] and show good agreement between predicted risk and observed proportions, which indicates
good calibration.
Fig. 2 (a) Calibration plot of model 1 (women with PPROM) with the observed risk of delivery
within 7 days by predicted probabilities of delivery within 7 days. The dots indicate
the deciles with 95% confidence intervals of women with similar predicted risk. The
histograms indicate the frequencies across the predicted probabilities. (b) Calibration
plot of model 2 (women without PPROM) with the observed risk of delivery within 7
days by predicted probabilities of delivery within 7 days. The dots indicate the deciles
with 95% confidence intervals of women with similar predicted risk. The histograms
indicate the frequencies across the predicted probabilities.
Table 3
Univariable and multivariable analyses for the prediction of delivery within 7 days
after successful 48 hours treatment of threatened preterm labor in women without PPROM
Women without PPROM (n = 484, 77%)
Delivery < 7 days
Delivery > 7 days
Univariable analysis
Multivariable analysis
Odds ratio (95% CI)[a ]
p -Value
Beta coefficient
Odds ratio (95% CI)
Characteristic
n = 90 (19%)
n = 394 (81%)
Age (y)[b ]
30.9 ± 4.6
29.0 ± 5.4
0.72 (0.46–1.12)[c ]
0.14
0.063
1.07 (1.00–1.13)
Non-Caucasian ethnicity
15 (17)
68 (17)
0.95 (0.51–1.79)
0.88
Low educational level
45 (50)
237 (60)
0.68 (0.40–1.16)
0.15
Parity and prior preterm birth
Prior birth ≥ 37 wk
18 (20)
87 (22)
Reference
Nulliparous
59 (65)
219 (56)
1.29 (0.72–2.32)
0.39
Prior preterm birth < 32 wk
7 (8)
54 (14)
0.63 (0.25–1.60)
0.33
Prior preterm birth 32–37 wk
6 (7)
34 (9)
0.85 (0.31–2.33)
0.76
Body mass index (kg/m2 )
21.6 (20.2–24.4)
22.3 (20.4–24.8)
0.96 (0.90–1.03)
0.29
Multifetal gestation
21 (23)
86 (22)
1.07 (0.62–1.85)
0.82
Vaginal bleeding
36 (40)
50 (13)
4.64 (2.77–7.79)
< 0.001
1.43
4.20 (2.07–8.52)
C-reactive protein (g/L)
10 (4–25)
7 (3–21)
1.14 (0.86–1.51)[c ]
0.16
Streptococcus Group B positive
23 (25)
81 (21)
1.31 (0.72–2.41)
0.38
Fibronectin status positive
59 (66)
130 (33)
14.9 (5.08–43.7)
< 0.001
1.83
6.23 (2.15–18.0)
Dilatation (cm)
2 (1–3)
1 (0–1)
1.93 (1.52–2.44)
< 0.001
Cervical length (mm)
12 (7–18)
24 (16–32)
0.36 (0.25–0.52)
< 0.001
− 0.68
0.50 (0.34–0.75)
Randomized
No
24 (27)
155 (39)
Reference
Yes, placebo
37 (41)
120 (30)
1.99 (1.13–3.51)
0.02
Yes, nifedipine
29 (32)
119 (30)
1.55 (0.86–2.81)
0.15
Abbreviation: PPROM, premature prelabor rupture of membranes.
a Averaged over the 10 imputation sets using Rubin rules.
b Data are mean ± standard deviation, median (IQR) or number (%). Percentages may not
sum to 100 because of rounding. Absolute numbers are based on the mean of 10 imputations.
c Log transformed. Intercept − 3.8334. c -statistic 0.89 (0.84–0.93). Coefficients were shrunken with an average shrinkage
factor 0.92.
Discussion
In this study, we investigated if women at increased risk of delivery within 7 days
after arrest of threatened preterm labor could be identified from certain antepartum
characteristics. Our results from the multivariable analysis show that in women with
PPROM, the relevant predictive variables are nulliparity, previous preterm delivery < 32
and 32 to 37 weeks' gestation, and vaginal bleeding. In women without PPROM, predictive
variables were maternal age, vaginal bleeding, positive fFN status, and cervical length.
The analytic models show moderate discriminative capability for women with PPROM and
good discriminative capability for women without PPROM.
Using the multivariable associations, it is possible to calculate the risk of delivery
within 7 days after arrest of threatened preterm labor, the next formula can be used:
p = 1 / [1 + exp (− 1 ×− 3.8334 + 1.43 × blood loss + 0.063 × log.age + 1.83 × fFN
pos + − 0.68 × cervical length)] for women without PPROM; and
p = 1 / [1 + exp (− 1 × − 1.076 + 0.57 × blood loss + 1.02 × nulliparity + − 0.015 ×
prior preterm birth <32 weeks + 0.99 × prior preterm birth 32–37 weeks)] for women
with PPROM.
Most studies have concentrated on screening early in pregnancy and on the outcome
of preterm delivery < 32 to 37 weeks.[26 ]
[27 ]
[28 ]
[29 ]
[30 ] Identifying patients at risk of preterm delivery should be considered differently
at each stage of pregnancy. For example, early in pregnancy history of preterm birth
and ethnicity are indicators for preterm delivery.[26 ]
[28 ] In midpregnancy, fFN detection and cervical length are associated with preterm delivery.[27 ]
[29 ]
[30 ] In symptomatic patients, fFN and cervical length improved identification of women
with a low risk to deliver spontaneously within 7 days.[31 ] In general, sensitivity and specificity of these predictive factors are fairly low.
We concentrated on women who did not deliver after initial therapy for threatened
preterm labor because it may affect their management with regard to prolonged admission
or discharge after initial medical treatment.
Several methodological aspects of the study deserve consideration: study population,
missing values, unexpected results, over-, and underestimation.
We included both randomized and nonrandomized women in the study. Although this might
raise concern about heterogeneity, we aimed to perform an analysis for all patients
with arrested preterm labor—whether they participate in a randomized trial or not—to
exclude the Hawthorn effect from these results.[32 ] We feel we could do this because the intervention of maintenance tocolysis was not
effective in prolonging pregnancy and improving perinatal outcome in the original
trial.
We performed our study in all 10 Dutch tertiary care centers, which indicates good
representation for the Dutch population. From the population, 4.3% was of African
ethnicity, and 14.7% was non-Caucasian non-African. African ethnicity is a well-known
risk factor for preterm delivery,[26 ]
[33 ] which we did not identify in our study. This is probably attributed to the fact
that the incidence of African ethnicity in the study was low.
We did not include smoking in our analyses because smoking as a risk factor for preterm
birth in the literature mostly included both spontaneous and medically indicated preterm
births combined,[34 ]
[35 ]
[36 ] and we feel that delivery within 7 days after arrest of threatened preterm labor
is mostly based on only spontaneous preterm births.
We encountered missing values, for example, in fFN testing 60% of the values were
missing. fFN testing was not standard in the Netherlands at the time of this trial,
and women had to give separate informed consent for performing this test. To prevent
loss in statistical power, we imputed missing values, which is superior to complete
case analysis.[19 ]
We expected women with a prior preterm birth to have an increased risk of delivery
within 7 days after arrest of threatened preterm labor in the current pregnancy in
women without PPROM.[28 ] We observed that this was not the case in our study. The unexpected finding may
have been caused by intervention effects or selection bias.[37 ] As women with a prior preterm delivery may be treated earlier in the process of
threatened preterm labor than women without a prior preterm delivery, it is possible
that this led to a seemingly more effective treatment of threatened preterm labor,
by starting treatment in the latent phase of labor instead of the acceleration phase
of labor. Also, these women have more risk to delivery early, for example, in the
first 48 hours after admission for threatened preterm labor. In that case, they were
not even included in our trial. We cannot exclude the possibility of selection bias
because collection of data on women who refuse randomization and refuse follow-up
of their data (nonparticipants) is not allowed.
We observed slight over- and underestimation of risk for delivery within 1 week, as
shown in [Fig. 2a, b ]. For the sum of variables, there is a tendency for slight overestimation of low
predicted risk and for slight underestimation of high predicted risk ([Fig. 2a, b ]). The switch from overestimation in low predicted risk to underestimation in high
predicted risk was at approximately 50% for women with PPROM and approximately 20%
for women without PPROM. This is due to the low number of cases in the higher risk
group of women without PPROM, which suggests that PPROM is a major risk factor for
delivery within 1 week.
Women with initial arrest of threatened preterm labor remain at risk for delivery
within 7 days. PPROM and vaginal bleeding in the current pregnancy are relevant predictive
factors in all women, as are maternal age, cervical length, and fFN in women without
PPROM and nulliparity, prior preterm birth < 32 weeks, and prior preterm birth 32
to 37 weeks in women with PPROM. Most risk factors for delivery within 1 week after
arrest of preterm labor are nonadjustable, for example, maternal age and history of
preterm birth. Even so, it is of clinical use to know whether a woman is at high or
low risk of delivery within 1 week, to determine the necessary level of care. Although
women at low risk can be observed in secondary care or home care, women with a high
risk may benefit from prolonged admission in a tertiary center.
Authors' Contributions
J.A.M.P., F.K.L., and B.W.J.M. contributed to the design of the randomized trial.
All authors participated in recruitment of participants, and collected data. C.R.
and E.S. analyzed and interpreted the data. C.R. drafted the article. All authors
critically reviewed the report. All authors have seen and approved the final version.
Funding Source
The randomized trial was funded by ZonMw, the Netherlands Organization for Health
Research and Development healthcare efficiency program (grant number 80–82310–98–08210).
The funder had no role in study design; collection, analysis, and interpretation of
data; writing the report; or the decision to submit for publication.
