Evaluating the accuracy of machine learning in predicting postoperative flap complications: A meta-analysis

Area of Science:

• Medical informatics
• Surgical outcomes research
• Artificial intelligence in healthcare

Background:

• Flap surgery is crucial for reconstruction but carries risks of postoperative complications.
• Accurate prediction of flap failure is essential for timely intervention and improved patient outcomes.
• Machine learning (ML) offers potential for enhancing predictive accuracy in surgical settings.

Purpose of the Study:

• To systematically review and meta-analyze the sensitivity and specificity of ML models in predicting complications after flap surgery.
• To evaluate the overall performance of various ML algorithms in identifying postoperative flap complications.
• To identify specific ML models with superior predictive capabilities.

Main Methods:

• A comprehensive search of five major databases (MEDLINE, PubMed, EMBASE, EMCARE, Google Scholar) was conducted.
• Five studies, analyzing 7734 patients with 10 ML models, were included after screening 49 records.
• Data on sensitivity, specificity, accuracy, and complication rates were extracted and analyzed using the QUADAS-2 tool.

Main Results:

• The pooled sensitivity of ML models was 41.9% (95% CI: 41.0%-42.7%), and pooled specificity was 78.6% (95% CI: 78.2%-79.1%).
• Gradient boosting (GB) models exhibited the highest specificity (84.6%).
• Artificial neural network models demonstrated the highest sensitivity (49.8%).

Conclusions:

• ML models show promising high specificity in predicting flap failure, particularly GB models.
• The relatively low pooled sensitivity suggests limitations in current ML models for early complication detection.
• Further research and model refinement are needed to enhance the sensitivity of ML algorithms for flap surgery complications.