Development and validation of an early risk-stratification model for hemophagocytic lymphohistiocytosis in severe fever with thrombocytopenia syndrome

OBJECTIVE: To develop and validate a model for early risk stratification of secondary hemophagocytic lymphohistiocytosis (HLH) in patients with severe fever with thrombocytopenia syndrome (SFTS). METHODS: This retrospective cohort included adults with laboratory-confirmed SFTS admitted to The First Affiliated Hospital with Nanjing Medical University between January 2019 and July 2024. Predictor variables were derived from clinical and laboratory data obtained within 3 days after virologic confirmation, corresponding to the predefined early-evaluation window of the study. HLH status (binary outcome) was defined using the entire-course HScore (≥170), calculated from the worst available values over the clinical course; HLH-2004 criteria and early-window HScore distributions were summarized descriptively to provide transparent outcome accounting. Twenty-eight candidate predictors entered LASSO with Boruta refinement, and retained variables were used to construct a multivariable logistic model and nomogram. Model performance was evaluated by discrimination, calibration, and decision-curve analysis in the derivation cohort, an internal validation set, and an external cohort of 60 patients from The First Affiliated Hospital with Anhui Medical University. RESULTS: Among 249 patients (152 non-HLH, 97 HLH), HLH was associated with higher peak temperature, longer fever, more lymphadenopathy/splenomegaly and neurological symptoms, and more severe thrombocytopenia, hypertriglyceridemia, hypofibrinogenemia, hyperferritinemia, higher viral load, elevated muscle/liver enzymes and LDH, and coagulopathy (all p < 0.05). LASSO-Boruta identified six routinely available predictors-peak temperature, splenomegaly, fever duration, triglycerides, fibrinogen, and ferritin. The model showed LR χ² = 214.82 (p < 0.0001), R² = 0.784, C-index = 0.962, Dxy = 0.923, with near-perfect calibration in derivation. In internal validation, discrimination remained near-perfect (AUC 0.997, 95% CI 0.989-1.000); mild miscalibration was corrected by intercept-and-slope recalibration, and decision curves showed net benefit across wide thresholds. External validation (n = 60) confirmed excellent discrimination (AUC 0.907, 95% CI 0.835-0.980), slight miscalibration resolved by recalibration, and preserved net benefit across most thresholds. CONCLUSIONS: A simple model based on early clinical and laboratory variables supports risk stratification for HLH in SFTS and may facilitate closer monitoring, repeated HLH assessment, and timely individualized management.