EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Joint‐specific measures improve risk adjustment in total knee arthroplasty: A machine learning approach

Dirk Müller iD, Amna Gillani iD, Michael T Hirschmann iD, Igor Lazic iD, Florian Hinterwimmer iD, Anabel Arber, Florian Krüger, Georg Matziolis iD, Rüdiger von Eisenhart-Rothe iD

Knee Surgery, Sports Traumatology, Arthroscopy · 2026-07 · 34 (7) · 2661-2672

DOI: 10.1002/ksa.70480 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Knee Surgery, Sports Traumatology, Arthroscopy, online erschienen am 15.06.2026. Verfasst von Dirk Müller und 8 weiteren Autor:innen. Beteiligte Einrichtung: Department of Orthopaedic Surgery, TUM Klinikum Rechts der Isar, TUM School of Medicine and Health, Technical University of Munich, Munich, Germany. Hauptthemen laut MEDLINE: Arthroplasty, Replacement, Knee, Machine Learning, Postoperative Complications, Risk Adjustment. Die Arbeit nennt 51 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Purpose

Accurate risk adjustment in total knee arthroplasty (TKA) is essential for outcome prediction and quality assessment. Most existing prediction models rely solely on patient demographics and comorbidities and do not account for joint‐specific pathology. This study evaluated whether incorporating radiographic and clinical joint‐specific parameters improves machine learning (ML)–based risk adjustment and prediction of postoperative complications and residual pain following TKA.

Methods

A retrospective analysis was performed on 1207 primary TKA procedures conducted at a single academic centre between 2018 and 2022. Three outcomes at 1 year were analysed: residual pain (Visual Analogue Scale [VAS] ≥ 4), any complications and major complications. Predictor variables included patient‐related factors (e.g., age, body mass index, American Society of Anesthesiologists score, comorbidities) and joint‐specific parameters (e.g., limb alignment, range of motion) derived from preoperative radiographs and clinical assessment. Binary classification models were developed using a stacked gradient‐boosting ensemble combining XGBoost and CatBoost. For each outcome, models using patient‐specific variables alone were compared with models incorporating both patient‐ and joint‐specific variables. Model performance was evaluated using accuracy, sensitivity, specificity and area under the receiver operating characteristic curve (AUC).

Results

Incorporating joint‐specific parameters significantly improved prediction of complications. For major complications, the combined model achieved an AUC of 0.74 compared with 0.66 using patient variables alone. For any complications, the AUC increased from 0.64 to 0.72. No improvement was observed for predicting residual pain. The most influential joint‐specific predictors included prior septic surgery, large bone defects, Kellgren–Lawrence Grade < 3, prior ligament reconstruction and preoperative knee flexion < 70°.

Conclusion

Inclusion of joint‐specific features improved ML‐based prediction of postoperative complications following TKA, but did not improve prediction of residual pain. These findings suggest that joint‐specific parameters may enhance risk adjustment for postoperative complications in TKA. Level of Evidence Level III retrospective cohort study.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Arthroplasty, Replacement, KneeMachine LearningPostoperative ComplicationsRisk AdjustmentKnee JointHumansRetrospective StudiesFemaleAgedMalePredictive Learning ModelsMiddle AgedPostoperative PainBoosting Machine Learning AlgorithmsClassification AlgorithmsPain MeasurementPrediction Algorithms

Stichwörter der Autor:innen

complicationfeature importancemachine learningrisk adjustmenttotal knee arthroplasty (TKA)

Autor:innen und Institutionen

  • Department of Orthopaedic Surgery, TUM Klinikum Rechts der Isar, TUM School of Medicine and Health, Technical University of Munich, Munich, Germany.
  • Institute for AI and Informatics in Medicine, TUM Klinikum Rechts der Isar, TUM School of Medicine and Health, Technical University of Munich, Munich, Germany.
  • Department of Orthopaedic Surgery and Traumatology, Kantonsspital Baselland, Bruderholz, Switzerland.
  • Department of Orthopaedic Surgery, Waldkliniken Eisenberg, Jena University Hospital, Friedrich Schiller University Jena, Eisenberg, Germany.
  • AE-German Society for Endoprosthetics, Freiburg, Germany.

ORCID-Profile: Dirk Müller, Amna Gillani, Michael T Hirschmann, Igor Lazic, Florian Hinterwimmer, Georg Matziolis, Rüdiger von Eisenhart-Rothe

Zitieren

APA

Müller, D., Gillani, A., Hirschmann, M. T., Lazic, I., Hinterwimmer, F., Arber, A., Krüger, F., Matziolis, G., & von Eisenhart-Rothe, R. (2026). Joint‐specific measures improve risk adjustment in total knee arthroplasty: A machine learning approach. Knee Surgery, Sports Traumatology, Arthroscopy, 34(7), 2661-2672. https://doi.org/10.1002/ksa.70480

Vancouver

Müller D, Gillani A, Hirschmann MT, Lazic I, Hinterwimmer F, Arber A, et al. Joint‐specific measures improve risk adjustment in total knee arthroplasty: A machine learning approach. Knee Surgery, Sports Traumatology, Arthroscopy. 2026;34(7):2661-2672. doi:10.1002/ksa.70480

BibTeX
@article{Mller2026,
  author = {Müller, Dirk and Gillani, Amna and Hirschmann, Michael T and Lazic, Igor and Hinterwimmer, Florian and Arber, Anabel and Krüger, Florian and Matziolis, Georg and von Eisenhart-Rothe, Rüdiger},
  title = {Joint‐specific measures improve risk adjustment in total knee arthroplasty: A machine learning approach},
  journal = {Knee Surgery, Sports Traumatology, Arthroscopy},
  year = {2026},
  volume = {34},
  number = {7},
  pages = {2661--2672},
  doi = {10.1002/ksa.70480}
}
Kostenlos starten →