CD47–SIRPα as a Therapeutic Target in Sarcoma: A Systematic Review of Expression, Preclinical, and Clinical Evidence

Background/Objectives: Sarcoma comprises mesenchymal malignancies for which PD-1/PD-L1 checkpoint inhibitors have produced few durable responses. The CD47–SIRPα innate checkpoint suppresses macrophage phagocytosis through a “don’t eat me” signal and is the most clinically advanced myeloid checkpoint, but its evidence base in sarcoma has not been systematically assembled or appraised. Methods: A PRISMA 2020-compliant systematic review (PROSPERO CRD420261452040) was undertaken; three bibliographic databases and three trial registers, without language restriction, were searched on 14 September 2026. Expression, preclinical and clinical studies were appraised and graded separately. Synthesis followed SWiM guidelines and no pooled effect estimate was calculated. Results: Of 391 records identified, 50 studies met the eligibility criteria as separate analysis units (45 journal articles, 4 conference abstracts, 1 registry record), appraised across expression (20), preclinical (38) and clinical (18, including 2 interventional trials) domains. CD47 expression was subtype-dependent and bimodal, consistent in chordoma and angiosarcoma, frequently negative in leiomyosarcoma and Ewing sarcoma, and contested in undifferentiated pleomorphic sarcoma; reported positivity in osteosarcoma varied six-fold between cohorts, and the assay was often incompletely described. In preclinical models, CD47 blockade consistently increased phagocytosis and combination regimens outperformed single-agent use, but few studies used patient-derived or humanized models. Only two interventional trials enrolled sarcoma patients; both were single-arm, one was terminated without posted results, and nine of ten pharmacological strategies had no sarcoma-specific clinical data. Certainty was low for expression and patient-cohort evidence, and very low for preclinical and interventional evidence. Conclusions: The CD47–SIRPα axis is biologically well characterized and expressed in a definable subset of sarcomas, but has not been tested in these tumors in a design capable of detecting a treatment effect. Four of the fifty records were conference abstracts, which carry a higher risk of reporting bias and non-publication. The rate-limiting step lies in the assay and the trial architecture rather than in the biology: a harmonized immunohistochemical standard and a biomarker-selected trial with an internal comparator are prerequisites for an informative next study.