Prediction Models for Severe Maternal Outcomes in Non-ICU Settings: Systematic Review

Numerous risk assessment tools for severe maternal outcomes (SMOs) exist, but systematic evaluations of prediction models for non-ICU settings are lacking, where the majority of antenatal and intrapartum care is provided.

To systematically review and critically appraise prognostic models for predicting SMOs in non-ICU settings.

PubMed, Embase, Cochrane Library, Web of Science and CINAHL were searched from June 1, 2009, to November 22, 2024.

Studies developing, validating or updating multivariable prognostic models for SMOs in non-ICU settings were included. Eligible populations were pregnant women and those within 42 days postpartum.

Two reviewers independently conducted screening, data extraction and quality assessment using CHARMS and PROBAST AI. A narrative synthesis was performed due to study heterogeneity.

Thirteen studies reporting 15 evaluations of 13 distinct prognostic models were included. Study populations encompassed pregnant or postpartum women in non-ICU settings, including general obstetric admissions and specific high-risk cohorts (acute fatty liver of pregnancy, liver dysfunction, pulmonary hypertension and cardiac arrest). Models demonstrated moderate to good discrimination (C-statistics 0.74-0.95), yet all 15 evaluations were assessed as high risk of bias due to inadequate sample sizes, inappropriate variable selection, poor missing data management, arbitrary categorisation of continuous predictors, temporal misalignments between predictor measurement and intended use, and insufficient internal validation. Only five models underwent external validation, and none reported calibration performance.

Findings underscore the need for enhanced methodological rigour, models tailored to specific gestational stages, comprehensive outcome definitions and dynamic risk assessment, thereby facilitating their feasible integration into clinical practice.