Enhanced neonatal surgical site infection prediction model utilizing statistically and clinically significant...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Pediatric Surgery

Background:

• Neonatal surgical site infections (SSI) pose a significant clinical challenge.
• Accurate prediction models are crucial for preventing SSIs in neonates.
• Machine learning offers advanced methods for analyzing complex healthcare data.

Purpose of the Study:

• To develop and validate a predictive model for neonatal SSI.
• To compare the efficacy of different statistical and machine learning algorithms.
• To identify key variables for accurate SSI prediction in neonates.

Main Methods:

• Utilized the 2012-2015 National Surgical Quality Improvement Program-Pediatric dataset.
• Developed and validated models using logistic regression (LR) and random forest classification (RFC).
• Created a hybrid model combining clinical knowledge with significant RFC variables.

Main Results:

• Included 16,842 neonates with 542 (4%) identified SSIs.
• All models demonstrated similar predictive performance (Area Under Curve ~0.65-0.68).
• The hybrid model achieved comparable predictability using significantly fewer variables (18).

Conclusions:

• The hybrid model offers a parsimonious and effective approach to predicting neonatal SSI.
• Machine learning algorithms can uncover novel predictive factors, improving clinical models.
• This study highlights the potential of integrating machine learning into neonatal surgical care.