Machine learning-based prediction models for chronic postsurgical pain after orthopaedic surgery: a scoping review

Abstract

Chronic postsurgical pain (CPSP) is a burdensome complication following orthopaedic surgery. Machine learning (ML) offers potential for improved prediction; however, the methodological landscape remains unsynthesized. This scoping review mapped ML-based CPSP prediction models in orthopaedics, summarizing algorithms, performance, predictors, and methodological quality. A systematic search was conducted across PubMed, Embase, Web of Science, Cochrane Library, CNKI, Wanfang, and VIP databases from inception to June 1st, 2026, following Joanna Briggs Institute (JBI) guidance for scoping reviews. Data extraction covered ML algorithms, predictive performance, predictors, and methodological quality domains including hyperparameter tuning, validation strategies, and interpretability approaches. Fifteen studies with 18,801 participants were included. The incidence of CPSP ranged from 7.2 to 39.6%. Fifteen ML algorithms were identified; tree‑based ensembles predominated, with XGBoost (n = 8) and Random Forest (n = 7) most frequent. Model discrimination varied widely, with AUCs ranging from 0.53 to 0.948. However, substantial heterogeneity in CPSP definitions and validation strategies limited the comparability of reported performance. True external validation using geographically or institutionally independent cohorts was performed in only one study, temporal validation in five, and the remaining nine relied solely on internal validation. Forty‑eight unique predictors were identified, categorized into preoperative, intraoperative, and postoperative groups. Commonly reported candidate predictors included age, preoperative pain intensity, functional scores, HADS score, inflammatory markers, and length of hospital stay. ML-based CPSP prediction in orthopaedics is rapidly growing but methodologically nascent. Standardized definitions, rigorous methodology, external validation, and explainable ML are needed to facilitate clinical translation.