Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Predictive Model Development to Identify Failed Healing in Patients after Non-Union Fracture Surgery

Cedric Donié, Marie K. Reumann, Tony Hartung, Benedikt J. Braun, Tina Histing, Satoshi Endo, Sandra Hirche

TL;DR

This study addresses the critical problem of predicting failed healing after first non-union revision surgery in long bones. Using TRUFFLE, a 797-patient dataset, it compares three classifiers—XGBoost, SVM, and logistic regression—on 356 engineered features with careful handling of missing data and cohort split, emphasizing calibration and threshold analysis. All models achieve about 70% sensitivity, with XGBoost delivering the highest specificity (~66%), statistically outperforming the others, and showing robust performance across thresholds as reflected by the Unified Performance Measure. The results demonstrate feasibility of risk stratification on a limited, single-center clinical dataset and point to multi-center validation for broader clinical adoption and personalized treatment planning.

Abstract

Bone non-union is among the most severe complications associated with trauma surgery, occurring in 10-30% of cases after long bone fractures. Treating non-unions requires a high level of surgical expertise and often involves multiple revision surgeries, sometimes even leading to amputation. Thus, more accurate prognosis is crucial for patient well-being. Recent advances in machine learning (ML) hold promise for developing models to predict non-union healing, even when working with smaller datasets, a commonly encountered challenge in clinical domains. To demonstrate the effectiveness of ML in identifying candidates at risk of failed non-union healing, we applied three ML models (logistic regression, support vector machine, and XGBoost) to the clinical dataset TRUFFLE, which includes 797 patients with long bone non-union. The models provided prediction results with 70% sensitivity, and the specificities of 66% (XGBoost), 49% (support vector machine), and 43% (logistic regression). These findings offer valuable clinical insights because they enable early identification of patients at risk of failed non-union healing after the initial surgical revision treatment protocol.

Predictive Model Development to Identify Failed Healing in Patients after Non-Union Fracture Surgery

TL;DR

This study addresses the critical problem of predicting failed healing after first non-union revision surgery in long bones. Using TRUFFLE, a 797-patient dataset, it compares three classifiers—XGBoost, SVM, and logistic regression—on 356 engineered features with careful handling of missing data and cohort split, emphasizing calibration and threshold analysis. All models achieve about 70% sensitivity, with XGBoost delivering the highest specificity (~66%), statistically outperforming the others, and showing robust performance across thresholds as reflected by the Unified Performance Measure. The results demonstrate feasibility of risk stratification on a limited, single-center clinical dataset and point to multi-center validation for broader clinical adoption and personalized treatment planning.

Abstract

Bone non-union is among the most severe complications associated with trauma surgery, occurring in 10-30% of cases after long bone fractures. Treating non-unions requires a high level of surgical expertise and often involves multiple revision surgeries, sometimes even leading to amputation. Thus, more accurate prognosis is crucial for patient well-being. Recent advances in machine learning (ML) hold promise for developing models to predict non-union healing, even when working with smaller datasets, a commonly encountered challenge in clinical domains. To demonstrate the effectiveness of ML in identifying candidates at risk of failed non-union healing, we applied three ML models (logistic regression, support vector machine, and XGBoost) to the clinical dataset TRUFFLE, which includes 797 patients with long bone non-union. The models provided prediction results with 70% sensitivity, and the specificities of 66% (XGBoost), 49% (support vector machine), and 43% (logistic regression). These findings offer valuable clinical insights because they enable early identification of patients at risk of failed non-union healing after the initial surgical revision treatment protocol.
Paper Structure (12 sections, 2 equations, 6 figures)

This paper contains 12 sections, 2 equations, 6 figures.

Table of Contents

  1. Introduction
  2. Methods
  3. Patient cohort
  4. Data preprocessing
  5. Model development
  6. Model evaluation
  7. Performance metrics
  8. Statistics for model comparison
  9. Model calibration
  10. Ablation studies
  11. Results and Discussion
  12. Conclusion

Figures (6)

  • Figure 1: Timeline of the clinical treatment of long bone non-unions
  • Figure 2: Confusion matrix for XGBoost with a threshold of 0.26. This threshold was chosen to guarantee at least 70% sensitivity.
  • Figure 3: Empirical cumulative distribution functions of XGBoost, SVM, and logistic regression. Each classifier is trained 300 times on randomly sampled 80% of the training data. A lower curve indicates stochastic dominance.
  • Figure 4: UPM is only slightly affected by the chosen decision threshold. UPM is calculated with different thresholds above which a prediction is rated positive.
  • Figure 5: Calibration display for XGBoost based on the test data. The true class is shown in the scatter plot. This scatter plot is smoothed using LOWESS.
  • ...and 1 more figures