Highlight
- The HLH-Risk-Calculator is a novel machine learning-based tool designed to predict initial disease severity (IDS) and mortality in secondary hemophagocytic lymphohistiocytosis (sHLH) patients.
- The study analyzed 167 adult sHLH patients from multiple European centers, utilizing random forest models anchored on eight clinical and laboratory features.
- Key predictive biomarkers include serum soluble interleukin-2 receptor (sIL-2R), albumin, and platelet counts, underscoring their clinical relevance.
- The calculator offers risk predictions across several time points but requires external validation before clinical application.
Study Background
Secondary hemophagocytic lymphohistiocytosis (sHLH) is an acute, hyperinflammatory syndrome characterized by excessive immune activation, leading to multiorgan dysfunction and high mortality. Distinguishing sHLH from other inflammatory states and predicting its clinical course remain challenging due to its variable presentation and heterogeneous triggers, including infections, malignancies, and autoimmune conditions. Existing diagnostic frameworks primarily address disease identification; however, tools capable of forecasting severity and mortality at specific time points are lacking. Given the rapid and often fatal progression of sHLH, precise prognostication could guide clinical decision-making, resource allocation, and therapeutic strategies. Integrating machine learning (ML) approaches offers potential advantages by leveraging complex datasets to capture predictive patterns beyond traditional statistics.
Study Design
This retrospective multicenter cohort study included 167 adult patients diagnosed with sHLH from six European centers across three countries. The study collated comprehensive clinical, demographic, and laboratory data at baseline and during follow-up. The primary aims were twofold: to develop ML models predicting initial disease severity (defined as requiring ICU admission or death within 90 days without ICU admission) and to forecast mortality at 30, 60, 90, 180, and 365 days post-diagnosis.
Two sets of eight clinically relevant features were selected for modeling based on their biological and prognostic relevance. Random forest algorithms—a type of ensemble ML approach known for robustness to overfitting and capacity to model nonlinear relationships—were trained separately for predicting initial disease severity and for mortality at each specified time point.
Model calibration was performed to improve predictive accuracy, and the algorithms were evaluated on hold-out test sets comprising 32 patients (for IDS prediction) and 43 patients (for mortality prediction), ensuring validation against unseen data.
Key Findings
The models demonstrated strong discriminatory power and reliable overall performance across the prediction tasks. For IDS, the random forest model effectively distinguished patients at risk of severe clinical courses, with serum levels of soluble interleukin-2 receptor (sIL-2R) and albumin emerging as the most influential predictors. The inverse relationship of albumin with disease severity aligns with its role as a negative acute-phase reactant and marker of nutritional and inflammatory status.
Mortality prediction models at various time points consistently highlighted sIL-2R and platelet counts as key contributors. Elevated sIL-2R reflects heightened immune activation, whereas thrombocytopenia signals hematologic compromise, both pathophysiologically pertinent to sHLH progression and outcome.
The models’ calibration against the hold-out sets demonstrated accurate estimation of individual patient risk, underscoring the potential of integrating these ML tools into clinical workflows to assist prognostication. However, performance metrics such as area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and confidence intervals were not detailed in the abstract and remain to be reported for comprehensive appraisal.
Expert Commentary
The HLH-Risk-Calculator represents a significant innovation by applying machine learning to address a critical unmet need in managing sHLH, a disease marked by rapid deterioration and diagnostic complexity. Its reliance on readily measurable laboratory parameters lends practical feasibility. The emphasis on sIL-2R is congruent with growing evidence of its role in immune dysregulation and disease activity monitoring in sHLH.
Nevertheless, the retrospective design and moderate sample size necessitate caution in interpretation. External validation in independent and geographically diverse cohorts is indispensable to confirm reproducibility and generalizability. Furthermore, integration with clinical variables such as underlying etiologies, comorbid conditions, and therapeutic interventions might enhance predictive accuracy. Prospective studies could elucidate how risk stratification influences clinical management and outcomes.
Conclusion
The HLH-Risk-Calculator leverages machine learning to provide a novel prognostic tool for secondary hemophagocytic lymphohistiocytosis, demonstrating promising ability to predict initial severity and mortality across multiple time points. Its foundation on accessible biomarkers facilitates potential real-world application, pending rigorous external validation. The tool exemplifies the intersection of clinical hematology and artificial intelligence, offering a pathway toward personalized risk assessment and optimized care in sHLH.
Clinicians and researchers are encouraged to engage with the calculator at www.hlh-risk-calculator.com for research purposes and to contribute to data collection efforts aimed at refining its predictive algorithms.
Funding and Clinical Trials
The original study did not explicitly report funding sources or clinical trial registration in the abstract. Further details may be accessible in the full publication or supplementary materials.
References
- Ruzicka M, Stubbe HC, Fauser J, et al. The HLH-Risk-Calculator is a machine learning-based tool to predict course & mortality of secondary hemophagocytic lymphohistiocytosis. Intensive Care Med. 2026 Jun 30. PMID: 42377458.
- Janka GE, Lehmberg K. Hemophagocytic lymphohistiocytosis: pathogenesis and treatment. Hematology Am Soc Hematol Educ Program. 2013;2013:605-611.
- La Rosee P, Horne A, Hines M, et al. Recommendations for the management of hemophagocytic lymphohistiocytosis in adults. Blood. 2019;133(23):2465-2477.
- Jordan MB, Allen CE, Weitzman S, Filipovich AH, McClain KL. How I treat hemophagocytic lymphohistiocytosis. Blood. 2011;118(15):4041-4052.
