Development and Validation of a Clinical Prognostic Risk Score for Stillbirth in Ethiopia: A Cohort Study Using PMA Data

ABSTRACT Background Stillbirth remain a major public health concern, particularly in low‐resource settings. Early identification of women at risk can significantly improve neonatal outcomes. This study aimed to Development and Validation of a Clinical Prognostic Risk Score for Stillbirth in Ethiopia: A Cohort Study Using PMA Data. Methods A predictive model was developed using Performance Monitoring for Action (PMA) data from women. The data were exported to R version 4.4.3 for cleaning and analysis. Fifteen candidate predictors were initially considered and selected using Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression with 10‐fold cross‐validation. Model performance was evaluated in terms of discrimination (area under the curve [AUC]), calibration, Brier score, and the Spiegelhalter test. Clinical usefulness was assessed using decision curve analysis (DCA). Internal validation was performed using bootstrap resampling. Finally, a simplified risk score and nomogram were developed to facilitate clinical application. Results Predictors retained in the final model included prolonged labor, membrane rupture before 9 months, membrane rupture > 24 h during delivery, and partner encouragement for ANC. The model demonstrated excellent discrimination (AUC = 0.905) and calibration (optimism‐corrected slope = 1.005; Brier score = 0.0072; Spiegelhalter p  = 1.00). The simplified risk score (range 0–10) showed strong discrimination, with high scores strongly associated with stillbirth (LR 159.7 for scores 9–10). Using a cutoff ≥ 2, the model achieved 83% sensitivity, 93% specificity, 15% PPV, and 100% NPV. Conclusion The four‐factor prediction model, along with its simplified risk score and nomogram, accurately identifies pregnancies at high risk of stillbirth in Ethiopian women. By providing individualized risk estimates, the tool can support clinical decision‐making and help prioritize interventions for those most at risk. External validation is needed before it can be widely applied in practice.