Infl uence of cut - off value on the prevalence of short cervical length

Objective: To assess the distribution of cervical length (CL) and explain the low prevalence of short CL ≤ 30 mm measured in a large cohort of women without a history of preterm birth; The Triple P study (NTR-2078). Methods: CL was measured in a cohort study with the aim to investigate the predictive capacity of cervical length measurement, to identify women at increased risk for preterm birth. Women with short CL were included in a randomized clinical trial evaluating the effect of progesterone compared to placebo. In total 57 centres and 20,234 women participated in this study. Cut off value for a short CL was defined as ≤ 30 mm based on existing literature. The distribution of CL was assessed for each individual centre and measurements were compared between levels of care: primary (29 ultrasound centres), secondary (21 general hospitals) and tertiary care institutions (7 university medical centres). Comparison was also performed for centres with a low, intermediate and high volume of CL measurements. Furthermore the distributions before and after publicly addressing the prevalence of short CL were analysed (12.284 vs.7.950 women). measurements of who (1.8%) had a short cervix. Mean cervical length was 44.2 mm (SD 7.8 mm). A 'dip' was observed in the 20 – 30 mm CL window, defined by the ratio of CL measurements between the original and expected data being below 50%. The 'dip' seemed to be present in the 89% of participating centres. All levels of care and low, intermediate and high-volume centres had a 'dip' below 30 mm. A significant difference was found in the distributions over time before and after publicly addressing the low prevalence of short cervix (1.7% and 2.0% of the measurements were 30 mm, respectively (p<0.001)). Conclusions: A cutoff value was used to include women with a short cervix in a randomised clinical trial which was embedded in a cohort study. We suggest that the use of a pre-defined cutoff value influenced the distribution of the CL measurements. Since the measurement is not blinded, non-blinded assessors might favour the control intervention, inducing selection bias. This might have resulted in too little measurements around the cutoff value. Other trials using similar designs could benefit from this observation and take precaution to avoid selection bias.