Implementing a Machine Learning Strategy to Predict Pathologic Response in Patients With Soft Tissue Sarcomas Treated...

Area of Science:

• Oncology
• Pathology
• Medical Statistics

Background:

• Neoadjuvant chemotherapy (NAC) is increasingly used for locally advanced, high-risk soft tissue sarcomas.
• A standardized definition and prognostic impact of good histologic response (GHR) after NAC are currently lacking.

Purpose of the Study:

• To identify histologic features in post-NAC surgical specimens that independently predict metastatic relapse-free survival (MFS).
• To define a robust criterion for GHR using a machine learning approach.
• To evaluate the prognostic value of GHR compared to existing models.

Main Methods:

• Retrospective analysis of 175 soft tissue sarcoma patients treated with NAC.
• Quantitative histopathologic analysis of surgical specimens.
• Multimodel, multivariate survival analysis including Cox regression and random survival forests to define GHR (<5% stainable tumor cells).
• Comparison of prognostic models using concordance indices and Monte-Carlo cross-validation.

Main Results:

• Machine learning models converged on GHR defined as <5% stainable tumor cells.
• Five-year MFS probability was 100% for GHR patients vs. 73% for non-GHR patients (P=.0122).
• The prognostic model incorporating GHR significantly outperformed standard models and the SARCULATOR nomogram (c-index 0.72 vs. 0.57/0.54).

Conclusions:

• Histologic response to NAC is a significant predictor of MFS in soft tissue sarcoma.
• The defined GHR criterion (<5% tumor cells) offers improved prognostic accuracy.
• GHR may serve as a valuable endpoint for future neoadjuvant therapy studies.