Development and validation of predictive models for QUiPP App v.2: tool for predicting preterm birth in asymptomatic high‐risk women

• Published in: Ultrasound in obstetrics & gynecology 2020-03, Vol.55 (3), p.348-356
• Main Authors: Watson, H. A., Seed, P. T., Carter, J., Hezelgrave, N. L., Kuhrt, K., Tribe, R. M., Shennan, A. H.
• Format: Article
• Language: English
• Subjects: Algorithms; App; Biomarkers; Birth; Calibration; cervical length; fetal fibronectin; Fetuses; Fibronectin; Gestation; Prediction models; Pregnancy; Premature birth; preterm birth; Risk analysis; risk assessment; Risk communication; Risk factors; Secondary analysis; Statistical analysis; Survival; Survival analysis; Training; Ultrasonic imaging; Ultrasound; Womens health
• ISSN: 0960-7692; 1469-0705
• DOI: 10.1002/uog.20401

ABSTRACT Objectives Accurate mid‐pregnancy prediction of spontaneous preterm birth (sPTB) is essential to ensure appropriate surveillance of high‐risk women. Advancing the QUiPP App prototype, QUiPP App v.2 aimed to provide individualized risk of delivery based on cervical length (CL), quantitative fetal fibronectin (qfFN) or both tests combined, taking into account further risk factors, such as multiple pregnancy. Here we report development of the QUiPP App v.2 predictive models for use in asymptomatic high‐risk women, and validation using a distinct dataset in order to confirm the accuracy and transportability of the QUiPP App, overall and within specific clinically relevant time frames. Methods This was a prospective secondary analysis of data of asymptomatic women at high risk of sPTB recruited in 13 UK preterm birth clinics. Women were offered longitudinal qfFN testing every 2–4 weeks and/or transvaginal ultrasound CL measurement between 18 + 0 and 36 + 6 weeks' gestation. A total of 1803 women (3878 visits) were included in the training set and 904 women (1400 visits) in the validation set. Prediction models were created based on the training set for use in three groups: patients with risk factors for sPTB and CL measurement alone, with risk factors for sPTB and qfFN measurement alone, and those with risk factors for sPTB and both CL and qfFN measurements. Survival analysis was used to identify the significant predictors of sPTB, and parametric structures for survival models were compared and the best selected. The estimated overall probability of delivery before six clinically important time points (