Predicting six-month mortality in adult hemophagocytic lymphohistiocytosis with machine learning: a prognostic approach utilizing laboratory data.

Abstract

Background: Hemophagocytic lymphohistiocytosis (HLH) is associated with high mortality rates. This study was conducted to develop and validate a predictive model for adult HLH patients at high risk of six months mortality using machine learning (ML) algorithms.

Methods: The study utilized univariate analysis and LASSO regression, incorporating eleven ML algorithms, to perform a comprehensive analysis of both admission and discharge variables. Model performance was assessed using metrics such as AUC, F1 score, Kaplan Meier (KM) curves, calibration curves and decision curve analysis (DCA).

Results: A total of 136 patients meeting the HLH-2004 criteria were included in this study. Among them, 59 patients received chemotherapy or other cytotoxic drugs, while the remaining 77 patients underwent symptomatic treatment. The median age was 52 years among nonsurvivors and 48 years among survivors, with 47 (42.6%) males in the nonsurvivor group and 31 (47.6%) in the survivor group. Age and nine discharge variables were identified as the most significant features for model construction. Random forest (RF) algorithm demonstrated superior predictive capabilities, achieving an AUC of 1.00, accuracy of 0.98 and F1 score of 0.98 in the training cohort, and an AUC of 0.89, accuracy of 0.85 and F1 score of 0.85 in the validation cohort. The top five predictors were all discharge variables (ferritin, white blood cell (WBC), albumin (ALB), platelet (PLT) and direct bilirubin (DB)).

Conclusion: The predictive model developed in this study provides a valuable tool for clinicians to early identify high-risk HLH patients, thereby enabling more targeted and effective interventions.