Distinguishing lupus lymphadenitis from Kikuchi disease based on clinicopathological features and C4d immunohistochemistry

Abstract

Objectives: Distinguishing Kikuchi disease (KD) from lupus lymphadenitis (LL) histologically is nearly impossible. We applied C4d immunohistochemical (IHC) stain to develop diagnostic tools.

Methods: We retrospectively investigated clinicopathological features and C4d IHC staining in an LL-enriched development cohort (19 LL and 81 KD specimens), proposed risk stratification criteria and trained machine learning models, and validated them in an external cohort (2 LL and 55 KD specimens).

Results: Clinically, we observed that LL was associated with an older average age (33 vs 25 years; P=0.005), higher proportion of biopsy sites other than the neck [4/19 (21%) vs 1/81 (1%); P=0.004], and higher proportion of generalized lymphadenopathy compared with KD [9/16 (56%) vs 7/31 (23%); P=0.028]. Histologically, LL involved a larger tissue area than KD did (P=0.006). LL specimens exhibited more frequent interfollicular pattern [5/19 (26%) vs 3/81 (4%); P=0.001] and plasma cell infiltrates (P=0.002), and less frequent histiocytic infiltrates in the necrotic area (P=0.030). Xanthomatous infiltrates were noted in 6/19 (32%) LL specimens. Immunohistochemically, C4d endothelial staining in the necrotic area [11/17 (65%) vs 2/62 (3%); P<10-7], and capillaries/venules [5/19 (26%) vs 7/81 (9%); P=0.048] and trabecular/hilar vessels [11/18 (61%) vs 8/81 (10%); P<10-4] in the viable area was more common in LL. During validation, both the risk stratification criteria and machine learning models were superior to conventional histological criteria.

Conclusions: Integrating clinicopathological and C4d findings could distinguish LL from KD.

Keywords: C4d; Kikuchi-Fujimoto disease; artificial intelligence; biopsy; histopathology; immunohistochemistry; lymphadenopathy; machine learning; necrotizing lymphadenitis; systemic lupus erythematosus.