Machine Learning-Guided Survival Prediction and Treatment Sequencing in Advanced Soft Tissue Sarcoma Beyond Second-Line Therapy: A Retrospective Cohort Study.

INTRODUCTION: Evidence guiding the optimal sequencing of later-line systemic therapy in advanced soft tissue sarcoma (STS) remains limited. The aim of this study was to identify prognostic factors for overall survival (OS) beyond second-line treatment and to explore therapy sequencing in routine clinical practice. METHODS: A total of 90 patients with advanced STS receiving third-line or later systemic therapy were retrospectively analyzed. Extreme gradient boosting (XGBoost) was used to identify clinical predictors of 1-year OS. The most influential variables were subsequently evaluated in multivariable Cox models for OS from the start of third- and fourth-line therapy. Sequencing analyses compared OS according to the line of administration of commonly used later-line agents. RESULTS: In the third-line cohort (n = 88), inferior OS was independently associated with progression on second-line therapy (hazard ratio [HR] 2.31, p = 0.005). In contrast, lipo-/leiomyosarcoma histology was associated with improved survival (HR 0.37, p = 0.002), as was a time to progression ≥ 12 months on first-line therapy (HR 0.43, p = 0.007). Findings were largely consistent in the fourth-line cohort (n = 57). Sequencing analyses suggested sustained activity of trabectedin in later lines (p = 0.023), greater benefit of earlier pazopanib use (p = 0.022), and no significant impact of treatment line for gemcitabine + docetaxel combination therapy (p = 0.12). CONCLUSIONS: Machine learning-guided variable selection identified clinically relevant predictors of survival in later-line STS. Prior treatment response and histology strongly influence outcomes, and exploratory sequencing analyses suggest differential timing effects across systemic therapy agents.