Development and Pre-validation of a Machine Learning-based Prediction Algorithm for Early Functional Recovery in Patients Undergoing Hip and Knee Replacement Surgery

Official title

Development and Pre-validated Multiple Variable Prediction Model Using Machine Learning for Early Functional Recovery After Joint Replacement Surgery.

Overview

The goal of this observational study is to develop and pre-validate a machine learning algorithm to predict early recovery of mobility in patients undergoing hip or knee joint replacement surgery. The primary research question is: Can a machine learning model accurately classify patients with faster versus slower recovery of autonomous mobility in the first days after joint replacement surgery? Patients who have undergone elective hip or knee arthroplasty and received post-operative physiotherapy will have their clinical and perioperative data collected retrospectively (2020-2023) and prospectively (March 2026-December 2027). The algorithm will be trained on retrospective data and tested prospectively to evaluate its predictive performance for early mobilization and length of hospital stay.

Detailed description

This observational study aims to develop and pre-validate a machine learning algorithm to predict early mobility recovery and hospital length of stay in patients undergoing elective hip or knee arthroplasty. The study includes a retrospective phase (2020-2023) using existing clinical and physiotherapy data, and a prospective phase (March 2026-December 2027) to validate the model in routine clinical practice.

Data Collection and Outcomes:

Mobility recovery: assessed by the ability to ascend and descend three steps within the first four postoperative days, recorded in the physiotherapy diary and electronic health record.

Length of stay: considered regular if discharged by the fifth postoperative day; longer stays are defined as prolonged.

Predictors: Baseline demographics (age, sex, BMI, ASA score, preoperative hemoglobin) and clinical/perioperative characteristics (type of surgery and anesthesia, initiation of physiotherapy, pain level, urinary catheter use, orthostatic intolerance).

Sample Size: 943 patients total (600 retrospective, 343 prospective), based on model development requirements and AUROC estimation.

Data Analysis: The dataset will be split into training, validation, and test sets. Multiple supervised learning algorithms (e.g., logistic regression, random forest, gradient boosting) will be compared. Model performance will be evaluated using AUROC, sensitivity, specificity, precision, F1-score, and calibration. Missing data will be handled with imputation or native algorithm methods when supported.

Model Validation: Prospective data will be used to assess model discrimination and calibration, and to identify potential temporal or clinical biases. Retraining may be performed using combined datasets to improve generalizability.

Study Flow: Retrospective patients identified via hospital records; prospective patients identified on the first postoperative physiotherapy session, provided with study information, and consented. Predictive results are stored in a separate registry inaccessible to treating clinicians.

Participating Centers:

IRCCS Istituto Ortopedico Rizzoli, Bologna - patient enrollment. Complex Structure of Medical Physics, Arcispedale S. Maria Nuova - data analysis and AI modeling.

Interventions

• Other Predictive Model for Early Mobility Recovery and Length of Stay
  Application of a machine learning-based predictive algorithm to retrospectively and prospectively analyze clinical and perioperative data in patients undergoing hip or knee arthroplasty, without influencing clinical decision-making.

Primary outcome measures

• Area under the receiver operating characteristic curve (AUROC) for discrimination ability of the machine learning predictive model [Time frame: Through study completion, an average of 2 years]

Secondary outcome measures (2)

• Calibration of the machine learning predictive model assessed by calibration plots [Time frame: through study completion, an average of 2 years]
• Predictive performance of the machine learning model assessed by precision and F1-score [Time frame: Through study completion, an average of 2 years]

Eligibility criteria

Inclusion criteria

• Adults aged 18 years or older
• Patients underwent elective hip or knee arthroplasty.
• Patients for whom postoperative physiotherapy was initiated.

Exclusion criteria

• Patients who underwent surgery for oncologic disease, femoral fracture, or revision joint arthroplasty.
• Patients for whom postoperative physiotherapy was not provided due to postoperative complications
• clinical data are unavailable.

Criteria are shown verbatim from the registry (in English). Final eligibility is always assessed by the study center.

Healthy volunteers: No

Study design

Observational model

Cohort

Study locations

Italy · 2 centers

• SAITeR IRCCS Istituto Ortopedico Rizzoli — Bologna
• Azienda U.S.L. - IRCCS di Reggio Emilia — Reggio Emilia

Publications

• Ribbons K, Cochrane J, Johnson S, Wills A, Ditton E, Dewar D, Broadhead M, Chan I, Dixon M, Dunkley C, Harbury R, Jovanovic A, Leong A, Summersell P, Todhunter C, Verheul R, Pollack M, Walker R, Nilsson M. Biopsychosocial based machine learning models predict patient improvement after total knee arthroplasty. Sci Rep. 2025 Feb 10;15(1):4926. doi: 10.1038/s41598-025-88560-w. PMID 39929870
• de Hond AAH, Steyerberg EW, van Calster B. Interpreting area under the receiver operating characteristic curve. Lancet Digit Health. 2022 Dec;4(12):e853-e855. doi: 10.1016/S2589-7500(22)00188-1. Epub 2022 Oct 18. No abstract available. PMID 36270955
• Hamel MB, Toth M, Legedza A, Rosen MP. Joint replacement surgery in elderly patients with severe osteoarthritis of the hip or knee: decision making, postoperative recovery, and clinical outcomes. Arch Intern Med. 2008 Jul 14;168(13):1430-40. doi: 10.1001/archinte.168.13.1430. PMID 18625924
• Gandhi R, Wasserstein D, Razak F, Davey JR, Mahomed NN. BMI independently predicts younger age at hip and knee replacement. Obesity (Silver Spring). 2010 Dec;18(12):2362-6. doi: 10.1038/oby.2010.72. Epub 2010 Apr 8. PMID 20379147
• Corbacioglu SK, Aksel G. Receiver operating characteristic curve analysis in diagnostic accuracy studies: A guide to interpreting the area under the curve value. Turk J Emerg Med. 2023 Oct 3;23(4):195-198. doi: 10.4103/tjem.tjem_182_23. eCollection 2023 Oct-Dec. PMID 38024184
• Baklola M, Reda Elmahdi R, Ali S, Elshenawy M, Mohamed Mossad A, Al-Bawah N, Mohamed Mansour R. Artificial intelligence in disease diagnostics: a comprehensive narrative review of current advances, applications, and future challenges in healthcare. Ann Med Surg (Lond). 2025 May 26;87(7):4237-4245. doi: 10.1097/MS9.0000000000003423. eCollection 2025 Jul. PMID 40851938

Identifiers

NCT: NCT07333560 · 641/2025/Oss/IOR

Primary sources (government registries)

• ClinicalTrials.gov (US NIH)

View this study on ClinicalTrials.gov ↗