外周免疫指标预测炎性乳腺癌的病理完全缓解和残余癌负荷

Background: Circulating immune profiles offer a noninvasive approach to monitor treatment response in breast cancer. Inflammatory breast cancer (IBC), although clinically distinct, shares resistance and immune evasion features among each other. To identify immune escape mechanisms and early predictors of response in IBC, we profiled longitudinal peripheral immune markers and evaluated their association with pathologic complete response (pCR) and residual cancer burden (RCB) across treatment regimens and subtypes and developed composite immune indices to improve prediction under class imbalance. Methods: Peripheral blood samples (n=93) from patients treated with neoadjuvant or induction therapy for IBC were analyzed by multiparameter flow cytometry and annotated for subtype, regimen, pCR, and RCB. Group comparisons used Mann-Whitney U (pCR) and Kruskal-Wallis (RCB). Predictive models included logistic regression and random forest classifiers with class-weighting. Composite indices included CD8 Exhaustion, CD8 Memory, NK-ADCC scores derived from z-scaled components. Model performance was assessed using 5-fold cross-validation with AUC, PR-AUC, and calibration. Planned extensions include leave-one-regimen validation, temporal analyses, and external evaluation using a pre-registered pipeline. Results: Univariate analyses identified significant associations between pCR and the CD8 Exhaustion Index (p≈0.007) and CD8 Memory Index (p≈0.038), with NK-ADCC trending (p≈ 0.078). Logistic regression and random forest models incorporating combined indices improved AUC from ~0.70 when individually associated, to ~0.92 ± 0.07 (5-fold CV), albeit with wide uncertainty due to only up to 5 pCR events. Individual markers linked to pCR included Stem_Memory_CD8, NKG2A_ADCC, and CTLA4_CD8, while CD56hi NK and NKG2A +CD56hi NK distinguished RCB gradients (p<0.05 before correction). No markers remained significant after FDR adjustment given sample size. Conclusions: Peripheral immune signatures, particularly combined CD8 exhaustion, memory indices, NK ADCC were associate with higher pCR and lower RCB. Composite indices outperform single markers and capture the balance between immune activation and suppression. Ongoing cross-cohort validation will assess generalizability across regimens and subtypes. Severe class imbalance (~5/93 pCR) and small subgroup sizes constrain precision of our analysis. Findings are hypothesis-generating and warrant external validation in larger prospective cohorts.