Artificial intelligence and machine learning is successful in predicting clinical outcomes after hip arthroscopy for femoroacetabular impingement syndrome.

Abstract

Purpose: To systematically review the current literature regarding the role of artificial intelligence and machine learning in predicting and optimising clinical outcomes following hip arthroscopy.

Methods: A systematic review of the PubMed, Cochrane, and EMBASE databases was completed in December 2024. Studies were included if they assessed the application of AI/ML to clinical outcomes of hip arthroscopy. Exclusion criteria were imaging-only studies, non-English publications, conference abstracts, review articles and meta-analyses. Extracted data included study characteristics, input features, algorithm types, sample sizes, and model performance. Descriptive statistical analysis was performed due to data heterogeneity.

Results: Sixteen studies met inclusion criteria, covering applications across prediction of intraoperative findings (n = 1), prediction of post-operative outcomes (n = 5), prediction of patient-reported outcomes (n = 7) and prediction of revision (n = 3). Input features commonly utilised included demographics, imaging data, preoperative patient-reported outcomes (PROs), and comorbidities. Supervised learning models were the most widely applied, including logistic regression, random forests, support vector machines (SVMs), and artificial neural networks (ANNs). Performance metrics demonstrated robust predictive ability, with AUC values ranging from 0.66 to 0.94 and accuracy rates exceeding 75% in most studies. Applications included predicting revision surgery risk, prolonged opioid use, postoperative satisfaction, and time to return to sport. Imaging-based algorithms, particularly leveraging MRI data, showed promise for surgical planning and diagnostic precision.

Conclusions: AI and ML show significant promise in enhancing outcome prediction and patient stratification in hip arthroscopy. Future research should prioritise the standardisation of datasets, external validation, and interpretability to facilitate clinical translation.

Level of evidence: Level V.