Research Article: Integration of Boruta algorithm and latent class analysis for risk factors of 30-day mortality in pediatric hemophagocytic lymphohistiocytosis based on peripheral blood indicators
Abstract:
To screen key peripheral blood indicators based on the Boruta feature selection algorithm, construct a latent class analysis (LCA) model, identify clinically heterogeneous subtypes of pediatric hemophagocytic lymphohistiocytosis (HLH), and evaluate the 30-day mortality risk following diagnosis.
A retrospective cohort study design was employed, enrolling 133 children diagnosed with HLH at the Children's Medical Center of the Affiliated Hospital of Guangdong Medical University between January 1, 2015, and December 30, 2024. Demographic characteristics and laboratory indicators at admission were collected. Patients were categorized into a non-survivor group (29 cases) and a survivor group (104 cases) based on their 30-day survival outcomes after diagnosis. Univariate logistic regression analysis was initially conducted for preliminary variable screening, followed by the Boruta algorithm to eliminate feature noise and identify key predictors. An LCA model was constructed based on the screened variables for subgroup classification, and model fit was evaluated using Akaike information criterion (AIC), Bayesian information criterion (BIC), and entropy, among other metrics. Additionally, Shapley Additive exPlanations (SHAP) analysis, importance scoring, and restricted cubic spline (RCS) models were employed to interpret key variables and explore nonlinear relationships.
After dual screening through univariate analysis and the Boruta algorithm, six key predictive variables were identified: activated partial thromboplastin time (APTT), mean corpuscular hemoglobin (MCH), central nervous system (CNS) involvement, cyclosporine treatment, C-reactive protein (CRP), and imaging-confirmed hepatomegaly. LCA modeling demonstrated that the 2-class model exhibited the best fit (AIC?=?1,507.750, BIC?=?1,573.030, Entropy?=?0.702), stratifying patients into a low-risk group (72%) and a high-risk group (28%). Survival analysis revealed a significantly lower 30-day cumulative survival rate in the high-risk group compared to the low-risk group (51.40% vs. 88.50%, Log-rank P <?0.001). Notably, patients in the high-risk group faced a 7.42-fold increased risk of 30-day mortality ( HR =?7.4184, 95% CI: 2.6274–20.9205, P <?0.001). SHAP analysis indicated that APTT contributed most to the prediction, while RCS analysis revealed a monotonically increasing relationship between APTT and mortality risk, with elevated MCH levels demonstrating a significant protective trend.
The analytical strategy integrating the Boruta algorithm and LCA effectively identifies clinical subtypes of pediatric HLH with distinct mortality risk trajectories based on routine peripheral blood indicators.
Introduction:
Pediatric hemophagocytic lymphohistiocytosis (HLH) is an immune dysregulation syndrome triggered by genetic defects, infections, malignancies, or autoimmune diseases. Its core pathological mechanism involves dysfunction of cytotoxic T lymphocytes and natural killer (NK) cells, leading to excessive macrophage activation and secretion of large amounts of pro-inflammatory cytokines, subsequently inducing cytokine storm and systemic multiple organ dysfunction syndrome ( 1 , 2 ). With the promotion of the HLH-2004…
Read more