External Validation of a Multivariable Prediction Model for Placenta Accreta Spectrum

Area of Science:

• Obstetrics and Gynecology
• Maternal-Fetal Medicine
• Medical Informatics

Background:

• Placenta accreta spectrum (PAS) is a critical obstetric complication associated with severe hemorrhage and mortality.
• Accurate predelivery prediction of PAS is vital for optimizing delivery location and surgical planning.
• This study aimed to validate the Weiniger prediction model for PAS in two distinct US tertiary care cohorts.

Purpose of the Study:

• To validate the diagnostic performance of a previously developed prediction model for placenta accreta spectrum (PAS).
• To assess the model's discrimination, calibration, and clinical utility across two different patient cohorts.
• To evaluate the impact of spectrum bias on the model's applicability in diverse clinical settings.

Main Methods:

• Two distinct cohorts of patients were analyzed: Cohort A (N=253) with risk factors/ultrasound features, and Cohort B (N=99) referred for ultrasound findings of PAS.
• Model performance was evaluated using c-statistics for discrimination, calibration metrics (intercept, slope, curves), and decision curve analysis for clinical usefulness.
• The outcome variable was a confirmed surgical and/or pathological diagnosis of PAS.

Main Results:

• The Weiniger model demonstrated acceptable discrimination in Cohort A (c-statistic=0.728) and excellent discrimination in Cohort B (c-statistic=0.866).
• The model tended to underestimate PAS risk and showed evidence of overfitting in both validation cohorts.
• Decision curve analysis indicated clinical benefit in Cohort A (threshold >0.25) but not in Cohort B.

Conclusions:

• The Weiniger PAS prediction model's performance is influenced by population case-mix, emphasizing the importance of considering spectrum bias.
• The model demonstrated clinical utility in populations with significant heterogeneity in PAS risk factors and ultrasound features at a threshold probability >25%.
• Further refinement may be needed to improve generalizability across diverse patient populations presenting with suspected PAS.