Systematic review and meta-analysis of machine learning-based prediction models for readmission risk after total hip...

Postoperative readmission is a critical metric after total hip and knee arthroplasty (THA/TKA). While machine learning (ML) models for predicting readmission risk are proliferating, the stability of their performance and the robustness of their methodology remain contentious. This study aimed to systematically review and quantitatively synthesize the evidence base for ML-based readmission prediction after THA/TKA.

A systematic search was conducted across PubMed, Embase, Cochrane Library, and Web of Science from inception to December 31, 2025. Studies developing or validating ML models for THA/TKA readmission were included. Model performance (C-statistic/AUC) was extracted. Study quality was assessed using the PROBAST + AI tool. A multivariate random-effects meta-analysis was performed to pool C-statistics and quantify heterogeneity, with subgroup analyses stratified by study design, surgery type, prediction timeframe, and algorithmic class.

Fifteen studies (57 distinct models) were included. The pooled C-statistic was 0.76 (95% CI: 0.71-0.81). However, extreme heterogeneity (I ² = 99.9%) rendered this point estimate of limited clinical utility; the 95% prediction interval (0.38-0.94) highlighted profound outcome unpredictability. Subgroup analyses revealed significant moderators: single-center models showed optimistically higher performance (0.86) compared to multicenter models (0.65), and THA-specific models yielded higher estimates than TKA-specific models, although these findings were derived from few studies and should be interpreted as exploratory. Advanced ML algorithms did not demonstrate consistent superiority over traditional logistic regression. Crucially, the PROBAST + AI assessment identified a high risk of bias in the majority of studies, primarily due to analytical shortcomings and a universal lack of model recalibration.

The current body of evidence for ML-based readmission prediction after THA/TKA is characterized by extreme heterogeneity and high methodological bias, severely constraining clinical utility. The inability to pool calibration metrics represents a critical evidence gap. Future research must prioritize multi-institutional validation, stringent adherence to reporting standards (e.g., TRIPOD-AI), and the mandatory transparent reporting of both discrimination and calibration metrics to realize any potential clinical benefit.

https://www.crd.york.ac.uk/PROSPERO/view/1305608, identifier CRD420261305608.