A Retrospective Machine Learning Analysis to Predict 3-Month Nonunion of Unstable Distal Clavicle Fracture Patients Treated with Open Reduction and Internal Fixation

Aug 2, 2026

Upload by Miles

A retrospective study evaluates machine learning methods to predict the 3-month nonunion risk in patients with unstable distal clavicle fractures treated with open reduction and internal fixation. Data were collected from Nanjing Luhe People’s Hospital between January 2015 and May 2023, with 3-month nonunion defined by disappeared fracture line and continuous callus. Five ML models were trained and tested after LASSO refinement, then assessed with AUROC, AUPRC, calibration and classification metrics.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Background

Methods

Results

Conclusion

Introduction

Therapeutics and Clinical Risk Management downloaded from [https://www.dovepress.com/](https://www.dovepress.com/) For personal use only. Therapeutics and Clinical Risk Management Open Access Full Text Article ORIGINAL RESEARCH A Retrospective Machine Learning Analysis to Predict 3-Month Nonunion of Unstable Distal Clavicle Fracture Patients Treated with Open Reduction and Internal Fixation Changke Ma 1 , *, Wei Lu2 , 3 , *, Limei Liang4 , *, Kaizong Huang 3 , Jianjun Zou 3 , 5 1Department of Orthopaedics, Nanjing Luhe People’s Hospital, Yangzhou University, Nanjing, People’s Republic of China; 2School of Basic Medicine and Clinical Pharmacy, China Pharmaceutical University, Nanjing, People’s Republic of China; 3Department of Pharmacology, Nanjing First Hospital, Nanjing Medical University, Nanjing, People’s Republic of China; 4Department of Rehabilitation, Nanjing Luhe People’s Hospital, Yangzhou University, Nanjing, People’s Republic of China; 5Department of Pharmacy, Nanjing First Hospital, China Pharmaceutical University, Nanjing, People’s Republic of China *These authors contributed equally to this work Correspondence: Kaizong Huang; Jianjun Zou, Email [kzhuang@nju.edu.cn](kzhuang@nju.edu.cn); [zoujianjun100@126.com](zoujianjun100@126.com) | Background: This retrospective study aims to predict the risk of 3-month nonunion in patients with unstable distal clavicle fractures (UDCFs) treated with open reduction and internal fixation (ORIF) using machine learning (ML) methods. ML was chosen over traditional statistical approaches because of its superior ability to capture complex nonlinear interactions and to handle imbalanced datasets.
Methods: We collected UDCFs patients at Nanjing Luhe People’s Hospital (China) between January 2015 and May 2023. The unfavorable outcome was defined as 3-month nonunion, as represented by disappeared fracture line and continuous callus. Patients meeting inclusion criteria were randomly divided into training (70%) and testing (30%) sets. Five ML models (logistic regression, random forest classifier, extreme gradient boosting, multi-layer perceptron, and category boosting) were developed. Those models were selected based on univariate analysis and refined using the Least Absolute Shrinkage and Selection Operator (LASSO) . Model performance was evaluated using AUROC, AUPRC, accuracy, sensitivity, specificity, F1 score, and calibration curves.
Results: A total of 248 patients were finally included into this study, and 76 (30.6%) of them had unfavorable outcomes. While all five models showed similar trends, the CatBoost model achieved the highest performance (AUROC = 0.863, AUPRC = 0.801) with consistent identification of the risk factors mentioned above. The SHAP values identified the CCD as the significant predictor for assessing the risk of 3-month nonunion in patients with UDCFs within the Chinese demographic.
Conclusion: The refined model incorporated four readily accessible variables, wherein the CCD, HDL levels, and blood loss were associated with an elevated risk of nonunion. Conversely, the application of nerve blocks, including postoperative block, was correlated with a reduced risk. Our results suggest that ML, particularly the CatBoost model, can be integrated into clinical workflows to aid surgeons in optimizing intraoperative techniques and postoperative management to reduce nonunion rates.
Keywords: distal clavicle fracture, machine learning, prediction, nonunion | | --- | | Introduction
Distal clavicle fractures constitute a prevalent type of upper limb fracture, representing 28% of all adult clavicle fractures.1 Among these, 10% to 52% are classified as displaced and unstable, attributed to the multidirectional forces exerted by the robust trapezius, latissimus dorsi, and pectoral muscles.2 Recent research indicates that the nonunion rate for unstable distal clavicle fractures (UDCFs) subjected to nonoperative treatment can reach up to 30% .3,4 Co

FAQs

What outcome does the study aim to predict?

The study predicts the risk of 3-month nonunion in unstable distal clavicle fracture patients treated with ORIF. Nonunion is defined by disappeared fracture line and continuous callus.

Which machine learning models were developed and compared?

Five models were built: logistic regression, random forest classifier, extreme gradient boosting, multi-layer perceptron, and CatBoost. Model selection used univariate analysis and refinement with LASSO.

What variables were associated with a higher or lower nonunion risk?

The refined model links CCD, HDL levels, and blood loss to increased nonunion risk. Use of nerve blocks, including postoperative block, correlated with reduced risk.

Aug 2, 2026

Upload by Miles