RESULTS
A total of 1000 patients were screened for study eligibility and a
cohort of 772 were included in final analysis (Figure S1). Demographic
and clinical characteristics were compared between patients with
sonographically visualizable (present) and non-visualizable (absent) CGA
(Table 1). The majority of patients were White (68%) with an average
maternal age of 33 years, and an average CL on screening TVUS of four
centimeters. Patients with no visualizable CGA were more likely to be
parous (prior term births 1.2 vs 0.6, p=0.07), but this was not a
statistically significant difference. The CGA absent group was more
likely to have had progesterone supplementation during pregnancy (17%
vs 4%, p=0.04); this composite includes use of vaginal or intramuscular
progesterone formulations, which did not differ by CGA group when
examined individually (vaginal n(%): 6 (0.8) vs 1 (5.6) among CGA
present vs absent, p=0.397) (intramuscular n(%): 24 (3.2) vs 2 (11.1)
among CGA present vs absent, p=0.237). Otherwise, the sonographically
present and absent CGA groups were overall similar. Notably, there was
no difference in history of prior spontaneous PTB, in GA at time of
TVUS, or in reported CL.
Of the patients included in final analysis, 18 (2.3%) were found to
have sonographically absent CGA, and the overall rate of spontaneous PTB
in the cohort was 2.6% (Table 2). Compared to those with a visualizable
CGA, patients with absent CGA were significantly more likely to have a
spontaneous PTB <37 weeks (2% vs 28%, p<0.001), a
difference that persisted for PTB <34 and <32 weeks
(p<0.001). Second-trimester CGA visibility rates were 98% and
75% for patients that delivered at term and preterm, respectively
(p<0.001). Neonates born to mothers with visually absent CGA
were more likely to have a lower birthweight (3124 vs 3358g, p=0.03),
but did not vary by mode of delivery or by postnatal disposition to the
newborn nursery or a neonatal ICU. Of note, a post-hoc evaluation of GA
at delivery among the screened patients who were ultimately excluded
from the final study analysis revealed an additional 53 births
<37 weeks, for a collective overall PTB rate of 7.1% among
the 1000 patients initially screened.
The best performing and most clinically relevant model for prediction of
PTB <37 weeks was a multivariable model with five variables:
total parity, GA at TVUS, CL at TVUS, any supplementation with
progesterone, and absence of CGA on TVUS. Multicollinearity between the
included predictor variables was found to be absent using Spearman’s
correlation, as all rho coefficients were very low and none
statistically significant. Performance of the final models were
illustrated via ROCs, calculating odds ratios and the AUC for each
(Figure 2). The full model (inclusive of both CL and CGA evaluation)
demonstrated a greater AUC compared to the reduced model (inclusive of
CL evaluation only) (AUC 0.83 vs 0.742, respectively)22. Further, the likelihood ratio test demonstrated
improved model fit for the full model compared to the reduced model (p
<0.001) 23.
Agreement between reviewers for qualitative evaluation of CGA (visually
absent or present glands at screening TVUS) was assessed via Cohen’s
Kappa statistic, which was adjusted for the low prevalence of CGA
absence and the rare outcome of PTB. The prevalence-adjusted
bias-adjusted kappa (PABAK) was 0.89 (bias index 0.055, prevalence index
0.905), reflecting strong agreement between reviewers.
To quantify sonographically visible CGA when present, gland length and
width were measured on an optimal sagittal cervical image. The
visualized width of the CGA ranged from 0.12 cm at narrowest to 2.5 cm
at widest, with an average width of 0.9 cm. Visualized CGA length was
3.4 cm on average. Mathematically estimated area of the CGA was
calculated by multiplying the measured CGA length and widths. Unlike the
dichotomous finding of a sonographically present or absent CGA,
quantitative measures of CGA did not significantly differ between the
patients who delivered at term and those who delivered preterm
<37 weeks (Figure S2), nor did they differ significantly
compared to those who delivered at <34 weeks or <32
weeks. Quantitative CGA measures also did not demonstrate clinically
significant association with the continuous outcome of GA at delivery as
correlation coefficients were very low (approximately zero), though
statistical significance was achieved for CGA length and narrowest width
(p=0.018 and p=0.025, respectively).