Computed tomography-based nutritional associated nomogram on machine learning predicts survival outcomes in patients...

Area of Science:

• Oncology
• Radiology
• Nutritional Science

Background:

• Soft-tissue sarcomas (STS) are aggressive cancers with high recurrence and mortality despite surgery.
• Patient nutritional status and body composition are critical determinants of cancer outcomes.
• Machine learning can integrate complex data for improved survival predictions.

Purpose of the Study:

• To determine the prognostic significance of nutrition-associated factors in surgically treated STS patients.
• To develop a predictive model for nutritional assessment using survival analysis and machine learning.

Main Methods:

• Retrospective analysis of 638 STS patients undergoing surgical excision (2009-2018).
• Collected nutrition-associated indicators from blood tests and CT scans.
• Employed random survival forest (RSF) and random forest analyses for outcome prediction and nomogram construction.

Main Results:

• RSF identified stage, hospital duration, subtype, BMI, and tumor size as key predictors of overall survival (OS).
• RSF-based nomograms demonstrated strong calibration and discriminative ability for predicting clinical outcomes.
• Nutritional indexes showed prognostic value for various clinical outcomes.

Conclusions:

• Multiple nutritional assessment indexes are valuable for predicting clinical outcomes in STS.
• Long-term surveillance and management of nutritional status are crucial for STS patients.