Prediction models for spontaneous preterm birth in asymptomatic low-risk singleton pregnancies: a systematic review

Objective: To systematically review prediction models developed for spontaneous preterm birth (sPTB) in asymptomatic low-risk singleton pregnancies and to evaluate their methodological quality and clinical applicability. Materials and methods: A systematic search of PubMed, Scopus, Embase and Cochrane Library was conducted until 20 August 2025, following PRISMA guidelines (PROSPERO: CRD420251129960). Studies reporting development or validation of prediction models for sPTB in asymptomatic low-risk singleton pregnancies were included. Data extraction was performed using the CHARMS checklist, and methodological quality was assessed using the PROBAST tool. Due to substantial heterogeneity in predictors, modeling approaches, and outcome definitions, findings were summarized narratively. Results: Among 16,475 screened records, three prospective cohort studies met the inclusion criteria. Sample sizes ranged from 135 to 1,107, with sPTB rates of 4.4–11.5%. All models were developed using logistic regression, demonstrating moderate to strong apparent discrimination (AUC 0.781–0.94). One model combined uterocervical angle (UCA) ≥99° and cervical length (CL) ≤33.8 mm with odds ratio of 24.3 (95% CI:13.9–42.4), another used the UCA/CL ratio (cutoff 3.09°/mm) with AUC of 0.78, and a third integrated CL with quantitative fetal fibronectin, and glutamate-acetate-D-lactate with AUC of 0.94. CL was a consistent predictor across studies. Marked heterogeneity in predictors, modeling strategies and reporting was observed across studies, limiting comparability. All studies were assessed as having a high risk of bias, primarily within the analysis domain, and none performed internal or external validation. Conclusion: Evidence on prediction models for sPTB in asymptomatic low-risk singleton pregnancies remains limited and heterogeneous. None of the existing models demonstrated sufficient methodological rigor or validation for clinical application. Current evidence is insufficient, and robust multicenter studies with standardized predictors, transparent reporting and proper validation are needed to develop reliable prediction models.