Objective
To systematically review and meta-analyze predictive models for perinatal mortality, including stillbirths from 28 weeks gestation and early neonatal deaths within six days, focusing on calibration and discrimination metrics.
Methods
We conducted a comprehensive search across databases like PubMed and Scopus from inception to April 1, 2025, targeting observational studies that developed or validated predictive models for perinatal mortality reporting at least one discrimination metric, such as the Area Under the Curve (AUC) or C-statistic. Two reviewers independently screened studies and assessed bias using the PROBAST tool. Meta-analyses were performed utilizing a random-effects model with heterogeneity assessed via the I2 statistic.
Results
Sixteen studies were included, representing 8 553 805 neonates. Eight studies focused on stillbirths, five on early neonatal deaths, and three on both. Pooled AUC estimates ranged from 0.78 to 0.86, with higher discrimination in internally validated models. Calibration was reported in 11 studies, but varied in quality, with one study rated high risk of bias.
Conclusion
This meta-analysis is the first to synthesize predictive models specific to stillbirths and early neonatal mortality. While internal performance metrics are promising, significant shortfalls in external validation and generalizability remain. Standardized methodologies and thorough external validations are crucial for reliable perinatal risk prediction.
PROSPERO registration number
CRD42025638383.
扫码关注我们
求助内容:
应助结果提醒方式:
