Development and temporal validation of a multimodal post-test risk stratification model for pregnancies with abnormal prenatal findings

To develop and temporally evaluate a multimodal post-test classification framework integrating prenatal imaging and genetic findings in pregnancies with abnormal prenatal findings. We retrospectively screened 826 pregnancies undergoing routine prenatal screening or diagnostic testing. After exclusion of 8 pregnancies with insufficient follow-up information and 72 with incomplete data on the final model predictors, 746 pregnancies were included in the final complete-case analysis, including a normal pregnancy group (n = 641) and an abnormal pregnancy group (n = 105). Clinical characteristics and testing results were compared between groups. According to the enrollment period, 546 pregnancies were used for model development and internal validation, and 200 pregnancies from the same institution during a later time period were used as a temporally separated internal validation cohort. Variables independently associated with abnormal pregnancy status classification were identified using multivariable logistic regression, and a nomogram was constructed. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Ultrasound abnormalities (OR = 74.36, 95% CI: 21.75-254.30), abnormal karyotypes (OR = 12.59, 95% CI: 5.55-28.56), and pathogenic CNVs (OR = 19.47, 95% CI: 9.35-40.54) were independently associated with abnormal pregnancy status classification within this post-test framework. The framework demonstrated favorable apparent classification performance, with AUCs of 0.917 (bootstrap 95% CI: 0.867-0.961) in the training set and 0.906 (bootstrap 95% CI: 0.822-0.977) in the testing set. However, these findings should be interpreted cautiously because the included variables were closely related to the criteria used for outcome classification, which may have led to optimistic estimates of apparent performance. Calibration was acceptable, and DCA suggested potential net benefit across threshold probabilities of 0.05-0.85. In the temporally separated internal validation cohort, the framework achieved an AUC of 0.969 (95% CI: 0.918-1.000), with a sensitivity of 0.950 and a specificity of 0.972; however, this result should be interpreted within the context of a single-center post-test clinical workflow and requires further multicenter evaluation. The multimodal post-test classification framework integrating ultrasound, karyotype, and CNV findings showed favorable apparent classification performance in this cohort of pregnancies with abnormal prenatal findings. Its potential value lies in supporting structured post-test risk communication and multidisciplinary discussion, particularly in cases with complex or discordant prenatal findings. This framework should not be interpreted as a primary screening tool or as an independent predictor of the natural course of pregnancy. Further prospective multicenter evaluation is needed before broader clinical application.