Haemophagocytic lymphohistiocytosis (HLH) is a rare, potentially fatal disorder marked by uncontrolled immune activation and systemic inflammation (1–4). Although initially recognised in paediatric populations (primary), secondary HLH can affect individuals across all age groups and is frequently triggered by infections, malignancies, autoimmune diseases, or immunodeficiencies in adults (4). Its ability to present with non-specific, sepsis-like symptoms contributes to frequent misdiagnosis and delays in appropriate treatment (5, 6). Pathophysiologically, defective granule-mediated cytotoxicity of natural-killer (NK) and CD8⁺ T cells sustains interferon-γ–driven macrophage activation and cytokine storm (4, 7). HLH is classically divided into primary and secondary disease (4): Primary HLH includes perforin-pathway defects (PRF1, UNC13D, STX11, STXBP2, etc.) together with an expanding group of inborn errors of immunity that undermine cytotoxic granule release or inflammasome regulation (4, 8, 9). Whilst primary HLH is most often diagnosed in children, secondary HLH can also occur in paediatric patients, driven by infections, malignancies, or autoimmune disease, similar to adult cases (3, 4, 10). Secondary HLH arises without a monogenic defect when external drivers push immune activation beyond a pathological threshold, with typical precipitants including acute infection, haematologic malignancy, autoimmune flare, and iatrogenic immune activation (e.g., checkpoint inhibitors and CAR-T) (2, 4).

Clinically, HLH presents with non-specific symptoms, including fever, cytopenia, hepatosplenomegaly, and multi-organ dysfunction. Elevated ferritin levels are a hallmark but non-specific finding, necessitating a high index of suspicion, particularly in critically ill patients unresponsive to standard sepsis treatments (3, 11–14).

Although HLH-2004 criteria and the HScore show reasonable diagnostic accuracy in some adult ICU cohorts (11, 12), bedside use is limited by overlap with severe systemic inflammation (e.g., septic shock) (5, 12, 15), reliance on timely bone-marrow examination, and restricted access to specialised assays (16).

The HScore estimates the probability of HLH using a weighted combination of clinical, laboratory, and cytological features—including immunosuppression, fever, cytopenias, organomegaly, ferritin, triglycerides, fibrinogen, AST, and bone marrow haemophagocytosis (11). Whilst scores above 250 indicate a >99% likelihood of HLH, performance in ICU populations is diminished by sepsis-related overlap (6, 12).

A recent multicentre validation of the HLH-2024 criteria in adults demonstrated improved diagnostic feasibility (17). However, data from strictly ICU populations remain limited. Although reports of adult HLH are accumulating (6, 12, 15, 18, 19), awareness across interdisciplinary ICUs remains limited, raising the risk of under-recognition. Detailed descriptions of presentation, disease course, and outcomes in critically ill adults are therefore scarce, underscoring the need for the present study, which also explores predictors of mortality.

A total of 43 adults with HLH were managed in the ICU (Table 1). Median age was 45 years (IQR: 33–60), and 28 (65%) were male. Median weight and body mass index (BMI) were 75 kg (IQR: 66–83) and 25.0 kg/m² (IQR: 22.0–27.8), respectively. The time from hospital admission to HLH diagnosis was 12 days (IQR 4–25). ICU length of stay was 9 days (IQR: 5–18), which was significantly shorter in non-survivors (8 days [IQR: 3–13] vs. 16 days [IQR: 7–37], p = 0.012). Overall, ICU mortality was 65.1% (28/43).

Organ support therapies were frequently required (Supplementary Table S1): IMV in 36/43 (83.7%; median 4.0 days, IQR 1.2–10.6), NIV in 22/43 (51.2%; 0.9 days, IQR 0.1–3.0), CRRT in 31/43 (72.1%; 3.3 days, IQR 0.6–9.8), and venovenous ECMO in 3/43 (7.0%; 4.6 days, IQR 1.3–5.4).

Our analysis of 43 adult ICU patients with HLH revealed a high disease burden and 65% ICU mortality. The median age was 45 years—substantially lower than in previously reported ICU cohorts (6, 12, 18)—with respiratory failure (62.8%) and sepsis or infection (41.9%) being the most common causes of admission. The severity of illness was reflected by a median SOFA score of 14 at admission and a high demand for organ support. IMV was required in 83.7% of patients, notably exceeding the 53–59% reported in prior studies (6, 15, 19), whilst RRT was required in 72.1%, comparable only to the cohort described by Knaak et al. (12) Survivors were ventilated longer, presumably reflecting early death amongst non-survivors rather than because prolonged support was beneficial. At presentation (see Table 1), a SOFA of 14 and SAPS II of 52 imply a predicted mortality ≥ 70–90%, well above general-ICU estimates for the same scores (45–55%), underscoring the rapidly progressive nature of HLH (23, 24).

In the subgroup of patients with HIV (12%), HLH was consistently triggered by opportunistic infections. All were previously known HIV cases, and no new diagnoses occurred during the study period. Antiretroviral treatment status was not systematically available. Amongst malignancy-associated HLH, mortality was 60%, consistent with prior reports (4). Interestingly, Löfstedt et al. recently described an increasing incidence of malignancy-associated HLH in their population-based study from Sweden, with rates rising from 0.026 per 100,000 adults (1997–2007) to 0.45 per 100,000 adults (2012–2018). In their cohort, malignancy-associated HLH was identified in 0.6% of haematological malignancies, with the highest proportion amongst young males (2.5%), despite better early survival, 2-year mortality remained 75% (25). In our series, all malignancy-associated HLH occurred in patients with a pre-existing malignancy under treatment. The majority had haematologic neoplasms (lymphoma, leukaemia, or myelodysplastic syndrome). Two patients had local solid tumours, whilst none had metastatic solid cancer.

In our multivariable logistic regression model (see Figure 1), a lactate concentration > 6 mmol/l was the only independent predictor of ICU death; age, sex, HScore, CCI, CRP, AST, ferritin, vasopressor use, immunosuppression, and haematologic disease were not significant. Given the model’s limited power and the heterogeneity of our cohort, these results should be interpreted with caution; however, they support the relevance of routine bedside parameters for early risk stratification.

The HLH-2004 diagnostic criteria remain widely used but were originally developed for paediatric populations (2, 3, 26). Because diagnosis requires ≥ 5 of 8 criteria (27), many ICU patients display suggestive features without crossing the threshold (6).

In initial studies, the HScore appeared robust (cutoff 169: sensitivity 93%, specificity 86%) (11), but those numbers stem from non-ICU cohorts and likely overestimate performance in critical care, where sepsis mimics many score components. A recent multicentre study in adult cohorts (17), including ICU patients, reported that lowering the HLH-2004 threshold to four items yielded the best diagnostic accuracy, whereas the revised HLH-2004 and the HScore ≥169 performed slightly less well. In our series, the median HScore was 245 (IQR 210–273), and—as expected for a tool built for diagnosis rather than prognosis—it did not differ significantly between survivors and non-survivors (median 228 vs. 249, p = 0.64; see Supplementary Table S2). We emphasise that any modification of existing cutoffs requires validation against matched ICU non-HLH controls, as sepsis and other hyperinflammatory states share many overlapping features (Table 4). To provide practical guidance, we outline a pragmatic diagnostic pathway tailored for ICU settings (Box 1).

Hyperferritinaemia, a hallmark of HLH, is frequently observed in critically ill ICU patients without HLH, often as part of the acute phase response (15, 28, 29). In a 12-year ICU cohort study, 2,623 ICU patients had ferritin ≥ 500 μg/l, but only 1.5% of them were ultimately diagnosed with HLH. Amongst those with HLH, peak ferritin concentrations were markedly higher than septic or septic-shock controls (median 31,674 vs. 1,545 and 1,448 μg/l, p < 0.001) (29). As highlighted by Liedgens et al., only 17% of patients with ferritin levels >50,000 μg/l were ultimately diagnosed with HLH, illustrating the poor specificity of even extreme hyperferritinaemia in ICU settings (11, 15). Mild-to-moderate transaminase rises are equally non-specific. Ischaemic hepatitis (“shock liver”) affects up to 13.4% of ICU admissions (30, 31), and AST/ALT elevations may also stem from drug-induced liver injury (32) heart failure (33, 34), rhabdomyolysis (31), acute or chronic liver failure (31), or parenteral nutrition-associated liver disease (PNALD) (35). Cytopenias are ubiquitous in critical illness: thrombocytopenia appears in 20–50% of patients (36, 37), and anaemia develops in nearly all long-stay ICU patients (38); leukopenia or neutropenia may follow sepsis or myelosuppressive therapy (39). Fever—a cornerstone HLH sign—was present in only 39.5% of our cases. Antipyretics, deep sedation, neuromuscular blockade, and heat-dissipating treatments (e.g., CRRT and large-volume infusions) routinely depress temperature; in a 587-patient CRRT cohort, temperature curves were identical in infected and non-infected patients (40), highlighting the reduced sensitivity of this sign.

In our cohort, bone marrow aspirates were performed in only 30 of 43 patients (69.8%), and haemophagocytosis was identified in just 15 of those (50%), reflecting limited yield and possible difficulty of sampling unstable patients (see Supplementary Table S2). Such limitations might delay diagnosis and contribute to under-recognition of HLH in the ICU setting, as haemophagocytosis may not be required for diagnosis (18).

Published data demonstrate that interferon-γ–inducible chemokines, particularly CXCL9, are significantly higher in HLH compared with sepsis, whereas IL-6 is elevated in both conditions but typically reaches higher levels in sepsis (41). In EBV-associated HLH, additional studies have shown disproportionately high IFN-γ and IL-10 with low IL-6, yielding IFN-γ/IL-6 and IL-10/IL-6 ratios that robustly differentiate EBV-HLH from sepsis (42). More recently, immune profiling has confirmed that acute EBV infection induces CXCL9/10/11 and IFN-γ–driven T-cell activation, providing a mechanistic link between EBV biology and the cytokine storm underlying HLH (43). With respect to presepsin (sCD14-ST), evidence in HLH is sparse. A small series in malignancy-associated HLH reported that presepsin levels alone were not prognostic, but a composite of presepsin and sIL-2R was associated with 90-day mortality (44). These assays were not available in our ICU, and their diagnostic role requires prospective validation before incorporation into routine panels. Because HLH shares many features with sepsis, septic shock, non-inflammatory multi-organ failure, and multisystem inflammatory syndrome (MIS), we provide a comparative differential diagnosis table (Table 4) highlighting overlapping findings and key discriminating tests relevant for ICU practise.

Future multicentre studies should combine refined clinical scores with rigorously validated biomarker panels to accelerate diagnosis, guide immunomodulatory therapy, and improve outcomes in ICU-HLH.

Our study has several limitations. First, the retrospective design may have introduced selection and information biases, particularly regarding data completeness and case identification. Clinical recognition and coding practices may also have evolved between 2008 and 2024, which could have affected case capture. In addition, conditions such as seronegative rheumatoid arthritis, immune reconstitution in HIV, or bone marrow reconstitution were likely misclassified in some cases. Information on the timing of bone marrow aspiration was not systematically recorded, and in several cases, only the pathology report date or the ICU discharge summary documented positive findings. Consequently, we could not determine a reliable interval between diagnosis and sampling. The limited diagnostic yield (haemophagocytosis in 15/30 aspirates) may therefore partly reflect delayed or post-treatment sampling.

Second, the single-centre nature limits the generalisability of our findings, as HLH management strategies, diagnostic approaches, and ICU admission criteria may vary across institutions. Third, the reliance on the clinical day of diagnosis to align temporal data. Whilst this is clinically relevant, the timing of HLH diagnosis is subject to physician discretion and available diagnostics, potentially misaligning disease onset with laboratory and clinical trends. This could affect the interpretation of disease progression and treatment responses. Fourth, pre-ICU treatments for HLH could not be assessed systematically, preventing us from fully evaluating their impact on ICU outcomes. Finally, functional assays for cytotoxicity (e.g., perforin expression) and genetic testing for inborn errors of immunity (next-generation sequencing panels, whole-exome sequencing, or whole-genome sequencing) were not systematically performed or recorded in our cohort and thus were not accessible for analysis. As these investigations are not part of routine ICU diagnostics, we cannot exclude that rare, previously unrecognised inborn error of immunity contributed to susceptibility in individual cases.

This study highlights the substantial disease burden and high mortality (65%) and organ-support requirements associated with HLH in adult ICU patients. Although nearly all cases met conventional HLH-2004 criteria and far exceeded the diagnostic HScore threshold, many of the individual variables—fever, cytopenias, transaminase elevations, hyperferritinaemia—also occur in sepsis and multi-organ failure, blurring the distinction between HLH and other hyper-inflammatory states in critical care. A lactate concentration > 6 mmol/l emerged as the sole bedside variable independently linked to ICU mortality in our exploratory model, but the HScore itself, designed for diagnosis rather than prognosis, showed no survival discrimination.

In critically ill adults, the individual components of current HLH scores frequently overlap with sepsis-related abnormalities, complicating bedside interpretation. To clarify the real-world performance of these tools in the ICU—and to determine whether adjunctive immunologic markers add value—larger, prospective cohorts that include both HLH and non-HLH patients are needed.