Development and Internal Validation of a Clinical Decision Support Nomogram for Predicting Neonatal Intensive Care Unit Admission in Late-Onset Preeclampsia

Technical Summary: Development and Internal Validation of a Clinical Decision Support Nomogram for Predicting NICU Admission in Late-Onset Preeclampsia

Problem Statement
Late-onset preeclampsia (≥34 weeks gestation) presents a significant clinical challenge. While often less severe than early-onset disease, it accounts for a large proportion of preeclampsia cases and is associated with meaningful neonatal morbidity. A critical decision point occurs intrapartum or immediately post-delivery, where clinicians must determine the need for Neonatal Intensive Care Unit (NICU) admission. Current prediction models often rely on early-pregnancy screening or do not align with the specific time horizon and available data (e.g., labor events, birth weight) required for immediate neonatal triage in late-onset cases. There is a need for a transparent, individualized risk estimation tool that utilizes routinely collected variables available at the time of delivery to support resource planning and team preparedness.

Methodology
The study employed a retrospective cohort design involving 560 women diagnosed with preeclampsia who delivered at ≥34 weeks gestation at Fujian Maternity and Child Health Hospital between 2016 and 2020.

• Data Partitioning: Participants were randomly allocated into a training cohort (n=392) and an internal test cohort (n=168) at a 7:3 ratio.
• Predictor Selection: A broad set of candidate predictors was considered, including maternal demographics, comorbidities (diabetes, anemia), prenatal care initiation (specifically by 24 weeks), intrapartum events (fever, fetal distress, labor duration), treatments (magnesium sulfate), and neonatal variables (birth weight).
• Model Development:
  1. Feature Selection: Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression was applied to the training cohort to screen candidate predictors and prevent overfitting.
  2. Multivariable Modeling: Selected variables were entered into a multivariable logistic regression model.
  3. Visualization: The final model was converted into a nomogram and implemented as a standalone web-based calculator.
• Validation and Evaluation:
  • Discrimination: Assessed using the Area Under the Receiver Operating Characteristic Curve (AUC).
  • Calibration: Evaluated via calibration plots, calibration slope, and intercept.
  • Overall Performance: Measured using the Brier score.
  • Clinical Utility: Assessed using Decision Curve Analysis (DCA) to determine net benefit across threshold probabilities.
  • Internal Validation: Bootstrap validation (1,000 resamples) was performed to calculate optimism-corrected statistics.
• Reporting: The study adhered to TRIPOD principles and updated TRIPOD+AI guidance.

Key Contributions

• Targeted Prediction Window: Unlike early-pregnancy screening tools, this model is specifically designed for the intrapartum or immediate post-delivery decision point, integrating variables that only become available during labor (e.g., labor duration, intrapartum fever, birth weight).
• Multivariable Integration: The model synthesizes diverse data points, including maternal comorbidities, prenatal care history, intrapartum stressors, and neonatal metrics, rather than relying on isolated predictors.
• Clinical Tool Implementation: The authors developed a user-friendly web calculator and a nomogram, translating complex regression coefficients into a transparent format for clinical handover and resource planning.

Results

• Outcome Prevalence: NICU admission occurred in 91 of 560 neonates (16.3%).
• Final Predictors: The multivariable model retained nine predictors:
  • Positive Associations: Diabetes (OR 2.53), Magnesium Sulfate Treatment During Pregnancy (MSTDP) (OR 2.31), and a trend for intrapartum fever (OR 3.70, p=0.058).
  • Negative Association: Prenatal care initiated by 24 weeks (OR 0.48).
  • Strongest Predictor: Birth weight (OR 1.00 per gram, p<0.001).
  • Non-Significant/Unstable: Anemia categories, fetal distress (due to sparse events), and labor stage durations were not independently significant in the final model, though some were retained for clinical interpretability or LASSO selection.
• Performance Metrics:
  • Training Cohort: AUC of 0.814 (95% CI 0.747–0.881).
  • Internal Test Cohort: AUC of 0.767 (95% CI 0.671–0.863).
  • Bootstrap Validation: Optimism-corrected C-statistic of 0.770, calibration slope of 0.844, and Brier score of 0.097.
  • Test Cohort Validation: Brier score of 0.131 and calibration slope of 0.863.
• Clinical Utility: Decision Curve Analysis indicated that the nomogram provides a greater net benefit than "treat-all" or "treat-none" strategies across relevant threshold probabilities.

Significance and Claims
The authors claim that this study provides a practical, internally validated tool for estimating the risk of NICU admission specifically for neonates born to women with late-onset preeclampsia. The primary significance lies in its utility as a decision support system rather than a diagnostic rule. The model is intended to:

1. Organize routinely available delivery-period data into a transparent risk estimate.
2. Facilitate earlier notification of the neonatal team.
3. Support anticipatory monitoring and more deliberate allocation of NICU resources.

The paper explicitly maintains a modest stance regarding its current scope. It acknowledges that the model is based on a single-center retrospective design and that NICU admission rates are influenced by institutional protocols and staffing. Consequently, the authors state that external validation, calibration updating, and prospective workflow evaluation are required before the tool can be recommended for routine clinical implementation. The model is not intended to replace neonatal assessment but to augment the preparation process at the time of delivery.