Risk factors and nomogram for the prediction of intracranial hemorrhage in very preterm infants

Abstract

Aims: This study aims to identify important risk factors for intracranial hemorrhage (ICH) in very preterm infants at our institution and develop a predictive nomogram for early detection of ICH.

Methods: We retrospectively analyzed neonates with a gestational age (GA) under 32 weeks, admitted to the neonatal intensive care unit from March 2022 to July 2023. Infants were categorized into two groups based on ultrasound findings and assessed for thirteen variables including gender, GA, birth weight (BW), acidosis, among others. We used multivariate logistic regression analysis to build a prediction model and identify independent risk factors for ICH. We build a prediction model by assigning 241 cases to the training set and 103 to the validation set (ratio 7:3).

Results: Among 344 very preterm infants, the incidence of ICH was 36.9% (89 cases) in training set. Significant differences were observed in gestational age, birth weight, antenatal corticosteroids, mechanical ventilation days, and acidosis between cases and controls. Logistic regression analysis identified gestational age (OR = 0.674), antenatal corticosteroids (OR = 0.257), acidosis (OR = 2.556), and mechanical ventilation mechanical ventilation days(OR = 0.257) as independent risk factors for ICH. The C-index of the training and validation sets was 0.814 (95% CI: 0.762-0.869) and 0.784 (95% CI: 0.693-0.875), respectively. According to decision curve analysis, our model outperformed the "None" and "All" baseline lines over a wide range of risk thresholds (0.12-0.92).

Conclusion: Acidosis and mechanical ventilation are independent risk factors for ICH in very preterm neonates, while higher gestational age and antenatal corticosteroid use are protective. The nomogram developed from these four factors demonstrates strong predictive accuracy and calibration, which can aid clinicians in identifying preterm infants at high risk for ICH and facilitate early diagnosis and management.

Keywords: Intracranial hemorrhage; Nomogram; Prediction model; Risk factors; Very preterm neonates.

Fig. 1

patients flow chart

Fig. 2

Percentage of neonates by gestational age in case and control groups

Fig. 3

Nomogram for IVH prediction in very preterm neonates. GA: gestational age; ACS: antenatal corticosteroid; MechVent: mechanical ventilation; 1 stand for no and 2 stand for yes in ACS, Acidosis

Fig. 4

a ROC curve in training set Fig. 4b ROC curve in validation set. B ROC curves. ROC : receiver operating characteristic; AUC: area under the ROC curve

Fig. 5

Calibration curve. In this curve, Mean absolute error = 0.021, Mean squared error = 0.00084, 0.9 Quantile of absolute error = 0.049

Fig. 6

Decision curve. In this curve, our model is above the None and All lines within a relatively large range (0.12–0.92), indicating that the model has good clinical application value