Application of bedside real-time ultrasound-guided umbilical venous catheterization in neonates and development of an XGBoost-based complication prediction model: a retrospective study

To assess the application value of bedside real-time ultrasound (US) for umbilical venous catheterization (UVC) in neonates and evaluate its safety and predictive factors for complications.

This was a single-arm, retrospective, observational study conducted at a single institution, without a direct internal control group (such as X-ray guided placement) for comparison. A total of 132 neonates who received UVCs using real-time US guidance with color Doppler were included. Clinical variables, including success rates and complications within 2 weeks post-insertion, were captured. Factors contributing to complications were identified using Multivariate Logistic Regression. An XGBoost model and SHAP analysis were employed to predict and interpret complication risks.

The success rate for one-time catheter insertion was 95.45%, the mean indwelling time was 7.31 ± 0.27 days, and the overall complication rate was 15.15%, with accidental catheter dislodgement occurring in 9.09% of cases. Independent risk factors for complications included non-standard catheterization, preterm birth, lower birth weight, and a lower 1 min Apgar score. The XGBoost model performed well in predicting complications, with SHAP analysis identifying non-standard technique and birth weight as predominant influences.

Bedside real-time US is shown to be a safe and accurate procedure for UVC placement in neonates. While this study did not include a direct comparison group, the high success and low complication rates align with literature suggesting that US can improve accuracy and potentially decrease radiation exposure compared to routine X-ray verification. The observed improvements are inferred through comparisons with external literature; therefore, further randomized controlled trials are needed to quantify its superiority over X-ray guidance.