AI-assisted systematic review and pooled analysis of anthracycline- versus taxane-based chemotherapy in advanced or metastatic angiosarcoma.

Abstract

11522 Background: Comparative evidence for first-line anthracycline- (Ant) vs taxane-based (Tax) chemotherapy in advanced angiosarcoma (AS) is limited by small, heterogeneous retrospective studies. We performed an AI-assisted systematic review and pooled analysis to overcome fragmented reporting and evaluate efficacy and safety across clinically relevant subgroups, including cutaneous and radiation-induced AS (RIAS). Methods: A PubMed search (2000–2025) identified 11 eligible studies, 6 providing quantitative class-specific outcomes. An AI-assisted workflow (GPT-5) harmonized heterogeneous datasets, resolved denominator inconsistencies, and enabled supervised reconstruction of missing subgroup numerators. Endpoints included objective response rate (ORR), disease control rate (DCR), survival outcomes, and toxicity. ORR and DCR were compared using two-proportion z-tests. PFS, OS, and safety outcomes were summarized descriptively based on reported medians, due to incomplete and heterogeneous reporting that precluded pooling. Results: Among 426 evaluable patients (Ant n = 245; Tax n = 181), Tax demonstrated higher pooled ORR (44.8% vs 29.0%, p = 0.001) and DCR (63.7% vs 52.6%, p = 0.018). Benefits were marked in cutaneous AS (ORR 50.8% vs 31.5%, p = 0.017) and RIAS (41.8% vs 30.2%, p = 0.036), while non-cutaneous AS showed comparable ORR (25.3% vs 27.8%, p = 0.68). Reported median PFS (6.2 vs 5.0 months) and OS (14.3 vs 12.2 months) numerically favored Tax. Safety analysis, enabled by AI integration of fragmented datasets, showed that Tax was mainly associated with low-grade neuropathy and mild myelosuppression. Ant toxicity reporting was sparse but included fatal neutropenic events, indicating greater hematologic risk. Conclusions: Tax shows superior antitumor activity in cutaneous AS and RIAS with favorable tolerability. Anthracyclines remain an option for selected patients, although higher hematologic toxicity is expected. These findings, derived through AI-assisted reconstruction of incomplete datasets, highlight both the therapeutic relevance of Tax in AS and the potential of AI to enhance evidence synthesis in rare cancers.