Identification of an efficient immunohistochemical panel for precise differential diagnosis of sebaceous carcinoma of eyelid.

Sebaceous carcinoma (SC) is a prevalent neoplasm of the eyelid, ranking second to basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) in incidence. In some Asian regions, the incidence of SC is as high as 50% with a high risk of invasion and recurrence. Diagnosis of SC is still challenging owing to its atypical histological features and variable immunohistochemical profiles, especially in poorly differentiated ones. Therefore, it is critical to identify more precise immunohistochemical makers for the prediction of SC. In this study, we performed a double-blind, retrospective analysis using tissue microarrays to evaluate the diagnostic accuracy of TRPS1, GATA3, and adipophilin in a cohort with 98 SCs, 56 BCCs, and 55 SCCs. Receiver operating characteristic curves analysis and DeLong's test were used to confirm the applicability of these markers. Binary logistic regression was applied to construct a cancer classification model, with model validity confirmed through likelihood ratio tests (Δ-2LL = 255.937, p < 0.001) and pseudo-R² metrics (Cox & Snell R² = 0.706, Nagelkerke R² = 0.803, McFadden R² = 0.578). Variable contributions were assessed via hierarchical likelihood ratio tests and Wald analyses, revealing significant discriminative power of TRPS1 score (OR = 524.410, p < 0.001) and other predictors. Our results revealed that the combination of above three markers significantly enhanced the diagnostic performance compared to individual ones. Notably, TRPS1 and GATA3 demonstrated superior discriminatory power in distinguishing SC from BCC and SCC to adipophilin. In poorly differentiated SCs, TRPS1 exhibited a markedly higher diagnostic efficacy than GATA3 and adipophilin. In addition, TRPS1 exhibited higher AUC values than GATA3 and adipophilin in differentiating primary and recurrent SCs, although the difference was not significant. Collectively, our findings indicate that the TRPS1/GATA3/adipophilin-based model demonstrated excellent discriminatory efficacy for eyelid malignancies (Kappa = 0.784, micro-average F1 = 0.882, AUC-ROC = 0.964 [95%CI:0.934-0.994]), achieving clinical-grade performance.