Real-Time VBAC Prediction Using the Abo Yousif Neural Framework: A Dual-Layer Clinical Model

Abstract

Background: VBAC is a difficult clinical decision in poor rural and remote environments. But the calculators available for use have fixed variables and don’t dynamically change themselves. Thus, we present here the Abo Yousif Model, where dynamic neural-clinical architecture is applied for dynamic data of mothers for the prediction of VBAC success and changing labor parameters.

Methods: 360 cases in a multicenter retrospective cohort of four tertiary hospitals in Sudan analyzed. The model uses a transformer-architecture approach of a dual model with two predictive layers: (1) stationary maternal characteristics such as age, BMI, parity, prior VBAC and (2) dynamic labor inputs such as cervical dilation, fetal station and oxytocin dose and timing. The DEC-index (Dynamic Effacement-Cervix Index), VSPC (VBAC Success Probability Curve) and the other prominent innovations we propose offer live interpretability.

Results: The model achieved AUC of 0.91, 88% accuracy, 85% sensitivity, and 90% specificity. Clinical relevance was confirmed through SHAP analysis, with cervical progression and oxytocin response as the chief predictors. The VSPC provided a picture of the VBAC probability over time, which enabled patients to help the clinician know what to do, and so to be able to make better decisions accordingly.

Conclusion: It can be concluded in that the Abo Yousif Model offers a clinically interpretable and real-time prediction of VBAC, and it outperformed those traditional calculators. Its dynamic indices/two-layer structure applicable for labour ward in LMICs. Further validation, mobile testing is required to extend its usefulness. Integration into mobile platforms and electronic health record (EHR) systems could improve labor ward decision-making, particularly in low-resource settings.