Development and validation of a cytokine-based diagnostic model for Kikuchi-Fujimoto disease

The diagnosis of Kikuchi-Fujimoto disease (KFD) is challenging, which is mainly based on typical histopathological findings from a lymph node biopsy. We aimed to develop and validate a predictive model for the diagnosis of KFD based on cytokine profiles and other laboratory findings.

This study included pediatric patients with KFD and controls with other febrile diseases in a retrospective training cohort (n = 673) and a prospective validation cohort (n = 116). Clinical data, laboratory parameters, and serum cytokines were collected. Variables with P < 0.05 in univariate analysis and acceptable collinearity were entered into multivariable logistic regression with stepwise selection. A nomogram was constructed based on the final model. Model performance was evaluated using ROC analysis, calibration curves, Hosmer-Lemeshow test, Brier score, and decision curve analysis.

The final model included age, WBC, neutrophil, hemoglobin, platelets, ESR, IL-10, and IFN-γ/IL-6, with IFN-γ/IL-6 modeled as a 4-knot restricted cubic spline. The derived nomogram showed excellent performance in the training cohort (AUC = 0.989, sensitivity 97.5%, specificity 94.4%) and maintained robust discrimination in the external validation cohort (AUC = 0.971, sensitivity 94.1%, specificity 93.9%). Calibration was good in both cohorts (Hosmer-Lemeshow P = 0.99 and 0.19, external calibration slope 0.88), with low Brier scores (0.025 and 0.060). The model achieved high predictive accuracy, with 0.966 (κ = 0.836) in the training cohort and 0.931 (κ = 0.831) in the external cohort.

This study proposes the first validated diagnostic model for KFD, demonstrating excellent performance and highlighting the IFN-γ/IL-6 ratio as a key predictor. With further validation, it may serve as a practical decision-support tool to guide clinical management.