External validation of QUiPP App in three independent European cohorts of symptomatic women

Objective: To validate externally the QUantitative Innovation in Predicting Preterm birth (QUiPP) App v.2 for the prediction of spontaneous preterm birth (sPTB) in symptomatic women attending tertiary care in Europe. Methods: The QUiPP App v.2 was validated in three independent datasets: a prospective European multicenter cohort across five countries (n = 452), a retrospective single-center cohort in The Netherlands (n = 581) and a retrospective single-center cohort in Belgium (n = 399). The cohorts consisted of pregnant women between 23 and 34 weeks' gestation with symptoms of threatened preterm birth attending a tertiary care hospital between 2012 and 2023. We calculated risk estimates using the QUiPP App v.2 by inputting quantitative fetal fibronectin (qfFN) and/or cervical length (CL) measurement, in addition to other risk factors. As a result of the absence of a fibronectin detection kit in the Belgian cohort, only the QUiPP model based on CL could be validated in this dataset. The European cohort had no missing cases, but for the Dutch cohort, only complete cases were analyzed due to missing data. For the Belgian cohort, we statistically corrected for patients lost to follow-up using inverse probability of censoring weighting. Discrimination was assessed using receiver-operating-characteristics (ROC)-curve analysis of three QUiPP models (qfFN alone, CL alone and CL plus qfFN) for the risk of sPTB at six predefined timepoints: within 1, 2 and 4 weeks after testing, and at < 30, < 34 and < 37 weeks' gestation. Sensitivity, specificity and positive and negative likelihood ratios were calculated using risk thresholds of 5%, 10% and 15%. Model calibration was assessed to evaluate the agreement between expected and observed outcomes. Results: The predictive performance of the QUiPP App v.2 for sPTB within 1 week after testing had an area under the ROC curve (AUC) of 0.84 (95% CI, 0.79–0.89) and 0.74 (95% CI, 0.66–0.83) in the European and Dutch cohorts, respectively, using the combined model of CL plus qfFN, and 0.80 (95% CI, 0.75–0.85) in the Belgian cohort using the CL-only model. Predictive performance was greater for shorter-term outcomes, specifically sPTB < 30 weeks, compared with longer-term outcomes, such as sPTB < 37 weeks. The highest AUC (0.91 (95% CI, 0.86–0.95)) was achieved by the model using CL plus qfFN for the prediction of sPTB < 30 weeks in the European cohort. Calibration was excellent for patients with negligible risk; however, for women at greater risk of sPTB, the risk was generally underestimated compared with the observed event rate. Conclusions: The QUiPP App v.2 offers reassurance for patients with low predicted risk of sPTB and has greater predictive performance for shorter-term, compared with longer-term, outcomes. Despite significant differences in prevalence from the original QUiPP dataset, the model combining CL plus qfFN and the qfFN-only model perform reasonably well. Statistical correction for patients lost to follow-up in the dataset comprising censored and uncensored patients improved the discriminative ability of the CL predictor.