Prediction of spontaneous onset of labor at term using clinical, ultrasound, and biochemical data: A multicenter prospective observational study (the <scp>PREDICT</scp> study)

INTRODUCTION: The objective of the study was to develop a predictive model for spontaneous onset of labor between 39 and 41 weeks' gestation using clinical, ultrasound and biochemical features. MATERIAL AND METHODS: We conducted a multicenter, prospective, observational study in two university hospitals in Switzerland. Women with singleton pregnancies in cephalic presentation and intact membranes who opted for expectant management until late term were eligible for the study. Predictors collected at 39 weeks included maternal characteristics, cervical ultrasound measurements and biochemical markers. Competing risks survival regression models were developed, and predictive performance was assessed using time-dependent, receiver operating characteristic curves, calibration plots, and the Brier score. The main outcome measure was spontaneous onset of labor or prelabor rupture of membranes occurring before 41 weeks' gestation. RESULTS: A total of 429 women were recruited. Main outcome occurred in 72.0% of participants. Fourteen percent of women underwent labor induction at or beyond 41 weeks and another 14.0% required earlier induction for medical reasons. The final predictive model included maternal age, body mass index, prior vaginal delivery, cervical length, and a positive fetal fibronectin test. The model showed an area under the curve of 0.71-0.72 and good calibration. Using a dual cutoff approach to predict spontaneous labor or prelabor rupture of membranes within 7 days, participants were classified into three groups: 12.5% in a low-probability group (predicted probability ≤14.9%), 74.7% in an intermediate group (14.9%-56.8%), and 12.8% in a high-probability group (>56.8%). CONCLUSIONS: The model provides individualized probability estimates for spontaneous labor onset and may support shared decision-making in term pregnancies. Predictive accuracy was moderate, but good calibration suggested clinical utility.