Extremity Soft Tissue Sarcoma Reconstruction Nomograms: A Clinicoradiomic, Machine Learning-Powered Predictor of...

Area of Science:

• Orthopedic Oncology
• Surgical Oncology
• Machine Learning in Medicine

Background:

• Limb-sparing extremity soft-tissue sarcoma (eSTS) resections present challenges in selecting appropriate wound closure techniques.
• Predicting postoperative complications is crucial for optimizing patient outcomes and surgical planning.

Purpose of the Study:

• To develop and validate Sarcoma Reconstruction Nomograms (SARCON), a machine learning tool for predicting adverse outcomes following eSTS resection.
• To provide probabilistic estimates of complications based on chosen reconstructive modalities.

Main Methods:

• A retrospective cohort of 316 limb-sparing eSTS resections was analyzed, integrating clinical and radiomic data.
• Machine learning classifiers (Logistic Regression with Lasso, Naïve Bayes, FasterRisk) were trained to predict surgical site infections, wound dehiscence, seroma, and minor/major complications.
• Model performance was rigorously evaluated using cross-validation and a dedicated test set.

Main Results:

• Logistic Regression with Lasso regularization demonstrated superior performance, with area under the receiver operator curves ranging from 0.83 to 0.93.
• The study analyzed outcomes including surgical site infections (12%), wound dehiscence (16%), seroma formation (8.5%), minor complications (34%), and major complications (25%).
• SARCON successfully generated nomograms for predicting these adverse events.

Conclusions:

• SARCON provides surgeons with valuable probabilistic insights into potential complications associated with different reconstructive choices.
• This tool facilitates informed decision-making, enabling optimization of reconstructive strategies for eSTS patients.
• The developed nomograms enhance the ability to anticipate and mitigate postoperative adverse events.