The impact of cut-off values on the prevalence of short cervical length in pregnancy

Abstract

Background: A prior study suggested that implementing a cut-off value of ≤30 mm for a short cervical length (CL) could potentially introduce selection bias and alter the distribution of CL measurements. As such, the objective of this study is to evaluate how CL distribution and incidence of short CL are affected when using different cut-off values for a short CL.

Study design: This is a secondary analysis of the Quadruple P (QP) Screening study; a prospective cohort study that included low-risk patients with singleton pregnancies undergoing fetal anomaly scan at 18-22 weeks of gestation, including a CL measurement. Patients with a short cervix, defined as ≤35 mm, were subsequently counseled for the QP trial; a randomized controlled trial (RCT) comparing progesterone to cervical pessary for the prevention of preterm birth. If participation to the RCT was refused, patients with a CL ≤25 mm were advised to use progestogen. The primary objective of this current study was to assess the normal distribution of CL across the entire cohort and to assess the incidence of short CL when using the cut-off values of ≤35 and ≤25 mm. Normal distributions for CL were simulated based on mean and standard deviation(SD) of the original data. The Kolmogorov-Smirnov test was used to evaluate the distribution of the CL measurements. Moreover, to evaluate the motives behind ultrasound measurements around the cut-off value, sonographers were asked to fill out a qualitative questionnaire.

Results: The total cohort included 19.171 eligible participants who underwent CL measurement, with a mean CL of 43.9 mm (±8.1 SD). The distribution of all CL observed measurements deviated significantly from the normal distribution (p < 0.001). A total of 1.852 (9.7%) patients had short CL ≤35 mm, which was significantly lower than expected when compared to the simulated normal distribution (n = 2.661, 13.9%; p < 0.001). The incidence of short CL ≤25 mm in our cohort statistically differed from the simulated normal distribution (238, 1.2% vs 177, 0.9%; p=0.003). When comparing our data to the simulated normal distribution, the difference in distributions is most pronounced when examining the difference between 35 and 36 mm. Results of the questionnaire reveal sonographers claimed not to be influenced by a cut-off value for study participation or progesterone treatment.

Conclusion: This study demonstrates that using any cut-off value for a short CL influences the incidence and distribution of CL. When using a cut-off value of ≤35 mm for study inclusion, the incidence of measurements of a short CL is lower than the anticipated incidence compared to a normal distribution. However, when using a cut-off value of ≤25 mm for progesterone treatment, the frequency of CL measurements is higher than expected below this threshold compared to a normal distribution. This study highlights the risk of introducing selection bias, most likely unintentionally, when cut-off values for short CL are used, regardless of the specific value chosen. Therefore healthcare providers should measure the CL with caution if essential decisions depend on a specific cut-off value.

Keywords: Cervical length; Cut-off value; Selection bias; Threshold value.