Machine learning for predicting outcomes, complications and resource utilisation after hip arthroscopy: A systematic...

Area of Science:

• Orthopaedic Surgery
• Medical Informatics
• Health Services Research

Background:

• Machine learning (ML) is increasingly adopted in orthopaedic surgery for outcome prediction.
• The specific utility of ML in hip arthroscopy remains under-evaluated.
• Existing literature quality for ML in this field requires critical appraisal.

Purpose of the Study:

• To evaluate ML models for predicting outcomes, complications, and resource use in hip arthroscopy.
• To compare the performance of ML models against traditional statistical methods.
• To critically assess the quality and reporting standards of studies on ML in hip arthroscopy.

Main Methods:

• A systematic literature search was performed across major databases (PUBMED, MEDLINE, EMBASE, Cochrane).
• Studies utilizing ML for clinical prediction in hip arthroscopy were included.
• Data extraction focused on study design, outcomes, model details, validation, and adherence to TRIPOD guidelines.

Main Results:

• Seventeen studies (37,549 patients) were analyzed, focusing on clinical outcomes (47.1%), adverse events (47.1%), and resource utilization (5.9%).
• Median AUC for clinical outcomes was 0.66 and for adverse events was 0.7; resource utilization showed R² of 0.8.
• Most studies showed variable adherence to TRIPOD guidelines (average 12.3/22 criteria met), with few comparisons favoring ML over traditional regression.

Conclusions:

• ML models demonstrate variable performance in predicting hip arthroscopy outcomes and complications.
• Comparisons with traditional regression methods often show no significant difference.
• Improved transparency and adherence to reporting guidelines (e.g., TRIPOD) are crucial for future ML research in hip arthroscopy.