Analysis of mortality risk factors and a nomogram prediction model for 81 cases of Epstein-Barr virus-associated hemophagocytic lymphohistiocytosis

Objective To investigate the factors influencing mortality in children with Epstein-Barr virus-associated hemophagocytic lymphohistiocytosis and develops a predictive model to assess mortality risk. Methods We retrospectively analyze the clinical data and laboratory results of 81 children diagnosed with Epstein-Barr virus-associated hemophagocytic lymphohistiocytosis during their initial hospitalization at the Children’s Hospital of Soochow University from January 2019 to November 2024, which includes 12 deceased cases. We employ univariate logistic regression and least absolute shrinkage and selection operator regression to screen variables. We subsequently perform cross-validation and variable dimensionality reduction. The final regression model facilitates the construction of a predictive nomogram to evaluate the discrimination ability and clinical net benefit of the proposed approach. Results Stepwise screening via univariate logistic regression, leave-one-out cross-validation, and Firth-penalized logistic regression identifies activated partial thromboplastin time and direct bilirubin as independent risk factors for mortality ( P < 0.05). The constructed nomogram demonstrates strong discrimination capability, yielding an area under the receiver operating characteristic curve of 0.891 with a 95% confidence interval of 0.730 to 0.995. The calibration curve shows high consistency between predicted and actual observed probabilities, reflecting a Brier score of 0.072. Decision curve analysis indicates that the model provides a positive net clinical benefit within a specific risk threshold range. Conclusion The predictive nomogram constructed using least absolute shrinkage and selection operator regression and logistic regression effectively evaluates the mortality risk in patients with Epstein-Barr virus-associated hemophagocytic lymphohistiocytosis and demonstrates promising clinical utility.