Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Presentation Icons

Additional registration fee required

Faculty have requested this content not be shared outside of the session

CME Credit Offered

Abstract Award

Recording available 2/16-5/2

AIUM Credit

Foundation Awardee

Poster Icons

Abstract Award

Foundation Awardee

Poster Session 3

Category: Clinical Obstetrics

Poster Session 3

(981) Stillbirth at term: even advanced artificial intelligence cannot predict the unpredictable

Thursday, February 12, 2026

10:30 AM - 12:00 PM

Submitting Author(s)

MB

Martina Benuzzi, MD

Obstetrics and Gynecology Unit, Department of Medical and Surgical Sciences for Mothers, Children and Adults, University of Modena and Reggio Emilia, Azienda Ospedaliero-Universitaria Policlinico, Modena, Italy.
Modena, Emilia-Romagna, Italy

Presenting Author(s)

Tetsuya Kawakita, MD, MS

Associate professor
EVMS Department of Obstetrics and Gynecology, Macon and Joan Brock Virginia Health Sciences at Old Dominion University, Norfolk, VA, United States
Norfolk, Virginia, United States

Coauthor(s)

IE

Ismail El Moudden, DSc, PhD

EVMS Department of Obstetrics and Gynecology, Macon and Joan Brock Virginia Health Sciences at Old Dominion University, Norfolk, VA, United States
Norfolk, Virginia, United States
FM

Francesca Monari, MD

Obstetrics and Gynecology Unit, Department of Medical and Surgical Sciences for Mothers, Children and Adults, University of Modena and Reggio Emilia, Azienda Ospedaliero-Universitaria Policlinico, Modena, Italy.
Modena, Emilia-Romagna, Italy
Beatrice Melis, MD

Obstetrics and Gynecology Unit, Department of Medical and Surgical Sciences for Mothers, Children and Adults, University of Modena and Reggio Emilia, Azienda Ospedaliero-Universitaria Policlinico, Modena, Italy.
Modena, Emilia-Romagna, Italy
GG

Gloria Guariglia, MD

Obstetrics and Gynecology Unit, Department of Medical and Surgical Sciences for Mothers, Children and Adults, University of Modena and Reggio Emilia, Azienda Ospedaliero-Universitaria Policlinico, Modena, Italy.
Modena, Emilia-Romagna, Italy
Isabella Neri, MD, PhD

Assistant Professor
Obstetrics and Gynecology Unit, Department of Medical and Surgical Sciences for Mothers, Children and Adults, University of Modena and Reggio Emilia, Azienda Ospedaliero-Universitaria Policlinico, Modena, Italy.
Modena, Emilia-Romagna, Italy
AL

Antonio La Marca, MD, PhD

Obstetrics and Gynecology Unit, Department of Medical and Surgical Sciences for Mothers, Children and Adults, University of Modena and Reggio Emilia, Azienda Ospedaliero-Universitaria Policlinico, Modena, Italy.
Modena, Emilia-Romagna, Italy
Fabio Facchinetti, MD, PhD (he/him/his)

Professor
University of Modena and Reggio Emilia, Modena, Italy
Modena, Emilia-Romagna, Italy

Objective:
To assess whether advanced machine learning (ML) models can accurately predict stillbirth at term (≥37 weeks), using clinically derived risk scores and addressing extreme outcome imbalance.

Study Design:
This was a stratified analysis of 302,193 term pregnancies (from 325,860 total), including 322 stillbirths (1.07/1,000) during the period 2014–2023. Data were obtained from CEDAP (“Certificato di Assistenza al Parto”), the Italian birth registry that collects comprehensive perinatal information on all births occurring in Emilia-Romagna, a Northern region of Italy. Three domain-specific risk scores (maternal characteristics, obstetric history, and pregnancy factors) were combined into a composite risk score to train Random Forest and XGBoost models. Class imbalance (936:1) was addressed with cost-sensitive weighting. Models were trained using 5-fold cross-validation and evaluated using AUC-ROC, sensitivity, specificity, PPV, and NPV.

Results:
Despite rigorous modeling, both algorithms showed limited performance. Random Forest yielded AUC 0.605, sensitivity 39.1%, PPV 0.24%; XGBoost performed similarly (AUC 0.626, sensitivity 40.6%, PPV 0.23%). NPV was >99.9% but 99.7% of positive predictions were false positives. The top five predictors (gestational age, pre-pregnancy BMI, maternal age, weight and height) explained 72% of model variance. Gestational age alone contributed 48%, despite the term-only cohort. Model performance fell below clinical utility thresholds (AUC < 0.70).

Conclusion:
Stillbirth at term remains largely unpredictable, even with advanced artificial intelligence and rich clinical data. These results highlight a biological knowledge gap, not a technical one—especially regarding placental function in late gestation. In order to meaningfully reduce stillbirth risk, future research must shift from static data modeling to dynamic, physiologically-based investigation.