The best machine learning algorithm for building surgical site infection predictive models: A systematic review and...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Predictive Modeling for Infections

Background:

• Surgical site infections (SSIs) pose a significant clinical challenge.
• Numerous machine learning (ML) algorithms exist for SSI prediction, but their comparative performance is unclear.

Purpose of the Study:

• To conduct a network meta-analysis comparing the predictive performance of various ML algorithms for SSIs.
• To identify the optimal ML algorithm for SSI prediction.

Main Methods:

• Systematic search of multiple databases (MEDLINE, EMBASE, CINAHL, Web of Science, Cochrane Library) up to November 2023.
• Inclusion of diagnostic accuracy trials using ML for SSI prediction models.
• Performance evaluation using Relative Diagnostic Odds Ratio (RDOR) and superiority index (SI).

Main Results:

• Models incorporating surgical type outperformed those without (RDOR 2.71).
• Models using mixed structured and textual data were superior to those using only structured data (RDOR 8.70).
• Boosted Classifiers showed the best overall prediction for mixed data models (SI 6.17), while Support Vector Machine excelled for structured data (SI 4.70).

Conclusions:

• ML algorithms leveraging both structured and textual data offer optimal performance for SSI prediction.
• Boosted Classifiers emerge as a leading algorithm for predicting surgical site infections.