Real-world clinical outcomes of macrophage activation-like syndrome and hyperferritinemia in sepsis: a retrospective database analysis Authors: Konstantinos Ouranos, Mark Michael, Tilemachos Koutouratsas, Siddharth Das, Evangelia K Mylona, Fadi Shehadeh, Deepa B Gotur, Eleftherios Mylonakis Published: 2026 ABSTRACT SummaryBackgroundMacrophage activation-like syndrome (MALS) is characterized by hyperferritinemia, yet its clinical course in real-world practice and the prognostic and clinical significance of varying ferritin levels have not been fully defined. The aim of this study was to characterize the epidemiology and clinical outcomes associated with sepsis and MALS.MethodsWe performed a retrospective analysis of the TriNetX database, identifying sepsis patients with MALS and ferritin level ≥4420 ng/mL within 3 days of diagnosis; patients with sepsis and ferritin <4420 ng/mL were classified as no-MALS. Primary outcomes were 28-day and 90-day all-cause mortality, and secondary outcomes included need for endotracheal intubation, intensive care unit (ICU) admission, continuous renal replacement therapy (CRRT), and vasopressors. We performed subgroup analyses according to age, immune suppression status, and bacteremia and assessed varying ferritin cutoff values (2000, 3000, 4420, 6000, and 10,000 ng/mL) for MALS diagnosis against study outcomes. We also performed an exploratory analysis assessing survival among patients with MALS treated with anakinra. We performed propensity score matching using nearest-neighbor matching and used hazard ratios (HRs) with 95% confidence intervals (CIs) to report results.FindingsFrom January 2017 to January 2025, 10,169 patients met criteria for MALS and 214,338 were classified as no-MALS. After 1:1 propensity score matching, both cohorts included 10,168 matched patients. MALS was associated with higher 28-day (HR: 2.73) and 90-day (HR: 2.49) mortality and increased need for intubation (Day 28: HR: 2.04; Day 90: HR: 2.00), ICU admission (Day 28: HR: 1.86; Day 90: HR: 1.78), CRRT (Day 28: HR: 2.19; Day 90: HR: 2.16), and vasopressors (Day 28: HR: 1.80; Day 90: HR: 1.73). In subgroup analyses, patients aged ≥65 years, those with bacteremia, and immunocompromised patients had worse 28-day and 90-day outcomes. Mortality increased with higher ferritin cutoff values at day 28 (thresholds [ng/mL]: 2000 [HR: 2.44]; 3000 [HR: 2.57]; 6000 [HR: 2.94]; 10,000 [HR: 3.05]) and day 90 (thresholds [ng/mL]: 2000 [HR: 2.21]; 3000 [HR: 2.36]; 6000 [HR: 2.60]; 10,000 [HR: 2.67]), with similar trends for secondary outcomes. Anakinra was not associated with improved survival at day 14 or day 28.InterpretationIn patients with sepsis, MALS was associated with worse 28- and 90-day outcomes compared to patients with no-MALS. Increasing ferritin cutoff values, even below the current diagnostic threshold for MALS diagnosis, were associated with worse primary and secondary outcomes. These findings support the need for systematic ferritin measurement in patients with sepsis for earlier risk stratification and clinical decision-making.FundingNone.Keywords: Sepsis, MALS, Macrophage activation-like syndrome, Ferritin, Anakinra Research in context Evidence before this study We searched the PubMed database, from database inception to December 15, 2025, for papers published in English, using the terms “sepsis”, “septic shock”, “septicemia”, “ferritin”, “hyperferritinemia”, “macrophage activation-like syndrome”, “MALS”, “anakinra”, “endotype”. Our search yielded 1143 results. Sepsis involves heterogeneous immune dysregulation ranging from hyperinflammation to immunosuppression. A key hyperinflammatory endotype is macrophage activation-like syndrome (MALS), defined by marked macrophage activation and very high ferritin concentrations (≥4420 ng/mL). MALS is associated with excessive interleukin-1-mediated inflammation, cytokine release, severe organ dysfunction, and poor survival compared with other sepsis endotypes. Most evidence describing MALS outcomes derives from clinical trials, which consistently report higher mortality and organ dysfunction in MALS compared with no-MALS. However, real-world data describing the clinical course and outcomes of sepsis-associated MALS remain limited. In addition, although a ferritin cutoff of 4420 ng/mL is used to define MALS, the prognostic implications of lower ferritin thresholds in sepsis are not well established. Added value of this study In this large retrospective analysis of the TriNetX database, 10,169 patients met criteria for MALS and 214,338 were classified as no-MALS. After 1:1 propensity score matching, both cohorts included 10,168 patients. MALS was associated with significantly higher 28-day (hazard ratio [HR] 2.73) and 90-day (HR 2.49) mortality, as well as increased rates of intubation, intensive care unit admission, vasopressor use, and continuous renal replacement therapy. Subgroup analyses showed worse outcomes among patients aged ≥65 years, those with bacteremia, and immunocompromised individuals with MALS. Interestingly, mortality increased with higher ferritin cutoff values at day 28 and day 90, with similar trends for secondary outcomes. Implications of all the available evidence In this real-world cohort of patients with sepsis, MALS was associated with worse 28- and 90-day outcomes compared to patients with no-MALS. Higher ferritin concentrations, even below the currently proposed diagnostic cutoff, were linked to increased mortality and organ support requirements. These findings support routine ferritin measurement in sepsis to improve early risk stratification and inform clinical decision-making. Introduction Sepsis remains a major global health burden, with an estimated 48.9 million cases and 11 million related deaths worldwide each year.1 This substantial global burden reflects, in part, the biological heterogeneity of sepsis, which comprises distinct endotypes characterized by divergent immune responses and clinical trajectories.2 A key proinflammatory endotype is macrophage activation-like syndrome (MALS), defined by profound macrophage activation and markedly elevated blood ferritin concentrations (≥4420 ng/mL).3 This phenotype is associated with excessive interleukin-1 (IL-1)-mediated immune activation and widespread cytokine release that culminates in severe organ dysfunction.4 MALS accounts for approximately 5–10% of patients with sepsis and is associated with greater disease severity and markedly worse survival compared with other sepsis endotypes; this prognostic relevance highlights the need for endotype-specific therapeutic strategies.3 The impact of MALS in sepsis has been largely studied in clinical trials, with outcomes showing worse mortality and higher rates of organ dysfunction in the MALS versus no-MALS groups.5 There is a need to better define the real-world clinical course and outcomes of sepsis-associated MALS outside the context of randomized trials. Even more, although the diagnosis of MALS is established by using a ferritin cutoff value of 4420 ng/mL, it is currently unknown whether alternative cutoffs, even below the currently proposed threshold, are linked to worse outcomes in sepsis with MALS. Because ferritin is a simple and widely available biomarker, evaluating the clinical impact of hyperferritinemia and MALS may help identify high-risk patients earlier and inform risk stratification, escalation of care, and the design of immunomodulatory trials. TriNetX is a global federated research network that provides secure access to deidentified electronic health record and claims data for observational outcomes research.6 Utilizing TriNetX, we conducted a retrospective analysis to characterize the epidemiology and clinical outcomes associated with sepsis and MALS. Given the recent publication of the ImmunoSep trial7 evaluating anakinra in patients with sepsis and MALS, we also evaluated real-world outcomes associated with anakinra in this population. Methods Data source We conducted a retrospective cohort study using data from the TriNetX U.S. Collaborative Network, a clinical research platform that aggregates deidentified electronic medical records from health care organizations across the North and South America.8 The platform captures comprehensive clinical information, including patient demographics, diagnostic codes (International Classification of Disease, 10th Revision-Clinical Modification [ICD-10-CM]), procedural data (ICD-10-Procedure Coding System [PCS] and Current Procedural Terminology [CPT]), medication prescriptions (Veterans Affairs [VA] Drug Classification and RxNorm), laboratory measurements (Logical Observation Identifiers Names and Codes [LOINC]), and health care-encounter data. Ethics statement The Houston Methodist Research Institute Institutional Review Board exempted this study from review because data from TriNetX was de-identified and no informed consent was needed. The results were reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) recommendations.9 The study adhered to the Declaration of Helsinki. Ethics approval was not required for this study. Study population and exposure(s) This retrospective study included adults (≥18 years of age) with a diagnosis of sepsis based on ICD-10 codes who required hospital admission. The study period ranged from January 1, 2017, to January 1, 2025. Admitted patients with sepsis who had a blood ferritin level measured were divided in two subgroups: MALS subgroup, if they had a blood ferritin level ≥4420 ng/mL; and no-MALS subgroup, if the blood ferritin level was <4420 ng/mL. Recognizing that ICD-10 coding may not coincide precisely with the clinical onset of sepsis, we allowed for potential misalignment between diagnosis and documentation by including ferritin measurements obtained within a 3-day window before and after the date of sepsis coding. Patients with no available blood ferritin level within a 3-day window before and after diagnosis of sepsis were not included in the analysis. Please refer to Fig. 1 for more information regarding the study design, and Supplemental Table S1 for full description of inclusion and exclusion criteria. In Supplemental Table S2 we summarize the coding algorithms used to identify the study population and related subgroups. Study design. Primary and secondary study outcomes The co-primary study outcomes were 28-day and 90-day all-cause mortality between the MALS and no-MALS subgroups. Secondary endpoints included need for intensive care unit (ICU) admission, intubation and mechanical ventilation, continuous renal replacement therapy (CRRT), and vasopressor administration (defined as administration of norepinephrine, vasopressin, phenylephrine and/or angiotensin-II) by day 28 and day 90. Exploratory analysis assessing the impact of anakinra in the MALS group We performed an exploratory analysis for patients in the MALS subgroup by identifying patients that were administered anakinra 3 days before to 3 days after diagnosis of sepsis. Exclusion criteria for patients entering the anakinra analysis included anti-cytokine agent administration, diagnosis of coronavirus disease-2019 (COVID-19), and underlying comorbidities that could explain the administration of anakinra and could confound the analysis result, a month prior to sepsis diagnosis. Patients requiring anakinra use 4–365 days prior to diagnosis of sepsis were also excluded from the analysis to support a new-user design. We used an intention-to-treat framework, as patients remained in the group of their initial treatment allocation. Please refer to Supplemental Table S1 for full description of inclusion and exclusion criteria. In the MALS subgroup, we compared 14- and 28-day all-cause mortality, ICU admission, intubation, CRRT, and vasopressor use between patients receiving anakinra plus standard care versus standard care alone. Statistical analysis We present the baseline characteristics in the MALS and no-MALS subgroups by reporting frequency distributions (counts and percentages) for categorical variables and descriptive statistics (means with standard deviations [SDs]) for continuous variables. We conducted 1:1 propensity score matching using nearest-neighbor matching without replacement, applying a caliper of 0.1 SDs of the logit of the propensity score. Covariate balance was evaluated using standardized mean differences, with values below 0.1 indicating adequate balance. Matching was applied for demographics and underlying medical comorbidities based on the anticipated influence of such variables on clinical outcomes.10, 11, 12 After matching, survival probability was estimated with the Kaplan–Meier method, and treatment effects for outcomes of interest were evaluated using Cox proportional-hazards models, with results reported as hazard ratios (HRs) and 95% confidence intervals (CIs). All tests were two-sided, with a significance threshold of P < 0.05. Analyses were conducted using the TriNetX platform. To enhance the robustness of our findings, we performed subgroup analyses for the following variables: age ≥65 years, infections with Gram positive versus a Gram negative pathogen, source of infection (bacteremia or pneumonia), diagnosis of septic shock, and immune suppression status (documentation of solid tumor malignancy or leukemia/lymphoma diagnosis and/or receipt of immunosuppressive medications within 12 months prior to diagnosis of sepsis). Finally, we conducted a sensitivity analysis by excluding patients with COVID-19 infection. To account for multiple comparisons, a Bonferroni correction was applied. For each outcome in the study, nine subgroup/sensitivity analyses were performed; therefore, the significance threshold was adjusted by dividing a two-sided α of 0.05 by nine. Statistical significance after correction was defined as an unadjusted p-value <0.0056. To evaluate whether ferritin thresholds other than 4420 ng/mL were associated with differential clinical outcomes, we performed sensitivity analyses using alternative ferritin thresholds (≥2000, ≥3000, ≥6000, and ≥10,000 ng/mL) as informed by prior literature.13,14 Cohorts were redefined for each threshold and analyzed with the same propensity score matching and outcome models as the primary analysis. Effect estimates across thresholds were compared to evaluate consistency and potential dose–response relationships. Please refer to Supplemental Table S1 for additional information regarding subgroup and sensitivity analyses conducted in our study. Role of the funding source There was no funding source for this study. Results Baseline characteristics of the patients included in the analysis A total of 1,385,617 hospitalized patients with a diagnosis of sepsis were found from the TriNetX database (Fig. 2). Of these, 1,161,110 were excluded from the analysis because they did not have available blood ferritin levels within a 3-day window before and after the diagnosis of sepsis. Of the remaining 224,507 patients, 10,169 (4.5%) patients had blood ferritin levels ≥4420 ng/mL and were included in the MALS subgroup, and 214,338 (95.5%) patients had blood ferritin levels <4420 ng/mL and were classified as no-MALS. Flowchart for the selection of patients included in the analysis. Before propensity score matching, mean age (±SD) of patients in the MALS and no-MALS subgroups was 59.2 (15.3) and 61.1 (16.2) years, respectively. A total of 6081 out of 10,169 (59.8%) and 111,747 out of 214,338 (52.1%) patients in the MALS and no-MALS subgroups were male. A total of 5430 out of 10,169 (53.4%) and 135,514 out of 214,338 (63.2%) patients in the MALS and no-MALS subgroups identified as white. Regarding baseline comorbidities, diabetes mellitus, chronic obstructive pulmonary disease, chronic ischemic heart disease and chronic kidney disease were present in 4125 (40.6%), 1365 (13.4%), 3114 (30.6%) and 4180 (41.1%) of patients with MALS, whereas respective diagnoses in patients with no-MALS were documented in 92,578 (43.2%), 39,316 (18.3%), 66,242 (30.9%) and 75,854 (35.4%) patients. Solid organ tumor or hematological malignancy was reported in 4812 out of 10,169 (47.3%) and 82,124 out of 214,338 (38.3%) patients in the MALS and no-MALS subgroups (Table 1). Table 1Baseline characteristics of the MALS and no-MALS subgroups before and after propensity score matching.VariablesBefore matching, No. (%)After matching, No. (%)MALS subgroup (n = 10,169)no-MALS subgroup (n = 214,338)Standardized mean differenceMALS subgroup (n = 10,168)no-MALS subgroup (n = 10,168)Standardized mean differenceAge, mean (SD), y59.2 (15.3)61.1 (16.2)0.12259.2 (15.3)59.6 (15.6)0.025Sex Male6081 (59.8)111,747 (52.1)0.1556080 (59.8)6233 (61.3)0.031 Female4088 (40.2)102,582 (47.9)0.1554088 (40.2)3934 (38.7)0.031Race and Ethnicity White5430 (53.4)135,514 (63.2)0.25430 (53.4)5522 (54.3)0.018 Black or African American2797 (27.5)45,521 (21.2)0.1462796 (27.5)2807 (27.6)0.002 Hispanic or Latino1135 (11.2)25,434 (11.9)0.0221135 (11.2)958 (9.4)0.057 Asian699 (6.9)9715 (4.5)0.101699 (6.9)689 (6.8)0.004Comorbidities Diabetes mellitus4125 (40.6)92,578 (43.2)0.0534124 (40.6)3996 (39.3)0.026 COPD1365 (13.4)39,316 (18.3)0.1351365 (13.4)1235 (12.1)0.038 COVID-192705 (26.6)43,972 (20.5)0.1442704 (26.6)2147 (21.1)0.129 Chronic ischemic heart disease3114 (30.6)66,242 (30.9)0.0063114 (30.6)3062 (30.1)0.011 Atrial fibrillation and/or flutter2596 (25.5)50,543 (23.6)0.0452596 (25.5)2487 (24.5)0.025 Heart failure3462 (34.0)68,346 (31.9)0.0463461 (34.0)3281 (32.3)0.038 Cerebrovascular disease2244 (22.1)46,434 (21.7)0.012244 (22.1)2068 (20.3)0.042 CKD4180 (41.1)75,854 (35.4)0.1184179 (41.1)4100 (40.3)0.016 HIV diagnosis248 (2.4)3245 (1.5)0.066248 (2.4)245 (2.4)0.002 SLE287 (2.8)3672 (1.7)0.075287 (2.8)257 (2.5)0.018 Liver disease4080 (40.1)56,171 (26.2)0.2994079 (40.1)4033 (39.7)0.009 Liver cirrhosis694 (6.8)16,373 (7.6)0.031694 (6.8)598 (5.9)0.039 Neoplasms4812 (47.3)82,124 (38.3)0.1834811 (47.3)4819 (47.4)0.002 History of organ transplant884 (8.7)8201 (3.8)0.202883 (8.7)820 (8.1)0.022Continuous variables are presented as mean (±SD) and categorical variables are presented as No. (%), unless otherwise specified.CKD: chronic kidney disease; COPD: chronic obstructive pulmonary disease; COVID-19: coronavirus disease-2019; HIV: human immunodeficiency virus; MALS: macrophage activation-like syndrome; SD: standard deviation; SLE: systemic lupus erythematosus. Baseline characteristics of the MALS and no-MALS subgroups before and after propensity score matching. Continuous variables are presented as mean (±SD) and categorical variables are presented as No. (%), unless otherwise specified. CKD: chronic kidney disease; COPD: chronic obstructive pulmonary disease; COVID-19: coronavirus disease-2019; HIV: human immunodeficiency virus; MALS: macrophage activation-like syndrome; SD: standard deviation; SLE: systemic lupus erythematosus. Following propensity score matching, a total of 10,168 patients were included in each subgroup, and the MALS and no-MALS subgroups were matched with regards to demographic variables and medical comorbidities. We present demographic variables and comorbidities among different subgroups included in the analysis in Supplemental Table S3 through Supplemental Table S10. Primary and secondary outcomes All-cause mortality By day 28, 3362 patients (33.1%) in the MALS subgroup and 1456 (14.3%) in the no-MALS subgroup had died (HR: 2.73; 95% CI: 2.56–2.90) (Fig. 3). At day 90, mortality remained higher in the MALS subgroup (40.8% vs 20.3%; HR: 2.49; 95% CI: 2.36–2.62) (Fig. 4). In Fig. 5, we present the Kaplan–Meier survival curves of 28- and 90-day mortality in patients with and without MALS. 28-day outcome results in the MALS and no-MALS patient groups. 90-day outcome results in the MALS and no-MALS patient groups. Kaplan–Meier survival curves of 28- and 90-day mortality in patients with and without MALS. Kaplan–Meier survival curves comparing 28- and 90-day survival between patients with sepsis and MALS and patients with sepsis without MALS. Time (days) is shown on the x-axis, and estimated survival probability is shown on the y-axis. Hazard ratios were estimated using Cox proportional hazards models. For 28-day survival analysis, mortality was higher in the MALS subgroup (33.1% vs 14.3%; HR: 2.73; 95% CI: 2.56–2.90). For 90-day survival analysis, mortality remained higher in the MALS subgroup (40.8% vs 20.3%; HR: 2.49; 95% CI: 2.36–2.62). Abbreviations: CI: confidence interval; HR: hazard ratio; MALS: macrophage activation-like syndrome. Among patients ≥65 years of age, mortality was higher in the MALS subgroup at both Day 28 (HR: 2.63; 95% CI: [2.42–2.86]) and Day 90 (HR: 2.26; 95% CI: [2.09–2.44]). Similar associations were observed among patients with bacteremia (Day 28: HR: 2.18; 95% CI: [1.97–2.36]; Day 90: HR: 1.93; 95% CI: [1.79–2.09]), those with solid organ malignancy, leukemia, or lymphoma (Day 28: HR: 2.26; 95% CI: [2.05–2.50]; Day 90: HR: 1.96; 95% CI: [1.81–2.13]), and those with a diagnosis of septic shock (Day 28: HR: 1.86; 95% CI: [1.75–1.97]; Day 90: HR: 1.71; 95% CI: [1.63–1.81]). Need for intubation and mechanical ventilation By day 28, 3334 patients (32.8%) in MALS and 1855 (18.2%) in no-MALS required intubation and mechanical ventilation (HR: 2.04; 95% CI: 1.92–2.16). At day 90, 3535 patients (34.8%) in MALS and 2043 (20.1%) in no-MALS required intubation (HR: 2.00; 95% CI: 1.90–2.11). Subgroup analyses showed consistently higher risk of intubation and mechanical ventilation in the MALS subgroup across the aforementioned strata, as shown in Figs. 3 and 4. These findings were observed among patients 65 years of age or older (Day 28: HR: 1.81; 95% CI: [1.65–1.97]; Day 90: HR: 1.70; 95% CI: [1.54–1.87]), those with bacteremia (Day 28: HR: 1.64; 95% CI: [1.51–1.78]; Day 90: HR: 1.62; 95% CI: [1.50–1.75]), patients with solid organ malignancy, leukemia, or lymphoma (Day 28: HR: 1.64; 95% CI: [1.46–1.85]; Day 90: HR: 1.56; 95% CI: [1.40–1.74]), and those with a diagnosis of septic shock (Day 28 and Day 90: HR: 1.26; 95% CI: [1.19–1.32]). Need for ICU admission By Day 28, ICU admission was more frequent among patients in the MALS subgroup than among those in the no-MALS subgroup, affecting 3977 patients (39.1%) and 2446 patients (24.1%), respectively (HR: 1.86; 95% CI: [1.76–1.95]). This difference remained evident at Day 90, when ICU admission was more frequent in the MALS subgroup than in the no-MALS subgroup (4288 [42.2%] vs 2852 [28.0%]; HR: 1.78; 95% CI: [1.70–1.87]). Subgroup analyses showed that in the MALS group, the risk of ICU admission was greater among patients 65 years of age or older (Day 28: HR: 1.58; 95% CI: [1.46–1.70]; Day 90: HR: 1.57; 95% CI: [1.45–1.70]), those with bacteremia (Day 28: HR: 1.59; 95% CI: [1.47–1.71]; Day 90: HR: 1.58; 95% CI: [1.47–1.70]), patients with solid organ malignancy, leukemia, or lymphoma (Day 28: HR: 1.43; 95% CI: [1.30–1.57]; Day 90: HR: 1.40; 95% CI: [1.29–1.52]), and those with septic shock (Day 28: HR: 1.34; 95% CI: [1.28–1.41]; Day 90: HR: 1.35; 95% CI: [1.28–1.41]). Need for CRRT By Day 28, the need for CRRT was more common in the MALS subgroup than in the no-MALS subgroup, affecting 1607 patients (15.8%) and 814 patients (8.0%), respectively (HR: 2.19; 95% CI: [2.02–2.39]). This association persisted through Day 90, when CRRT was required in 1829 patients (18.0%) in the MALS subgroup and in 977 patients (9.6%) in the no-MALS subgroup (HR: 2.16; 95% CI: [1.99–2.33]). Subgroup analyses demonstrated higher hazard ratio for CRRT among patients 65 years of age or older (Day 28: HR: 2.19; 95% CI: [1.91–2.51]; Day 90: HR: 2.07; 95% CI: [1.81–2.37]), those with bacteremia (Day 28: HR: 1.93; 95% CI: [1.70–2.19]; Day 90: HR: 1.76; 95% CI: [1.57–1.98]), patients with solid organ malignancy, leukemia, or lymphoma (Day 28: HR: 1.86; 95% CI: [1.53–2.26]; Day 90: HR: 1.89; 95% CI: [1.58–2.27]), and those with a diagnosis of septic shock (Day 28: HR: 1.78; 95% CI: [1.63–1.94]; Day 90: HR: 1.76; 95% CI: [1.63–1.91]). Need for vasopressor administration By day 28, vasopressor administration was required more frequently in the MALS subgroup than in the no-MALS subgroup, affecting 4304 patients (42.3%) and 2799 patients (27.5%), respectively (HR: 1.80; 95% CI: [1.72–1.89]). This difference persisted at day 90, when vasopressors were administered in 4640 patients (45.6%) in the MALS subgroup and in 3236 patients (31.8%) in the no-MALS subgroup (HR: 1.73; 95% CI: [1.65–1.81]). Subgroup analyses showed consistently higher hazards of vasopressor use among patients 65 years of age or older (Day 28: HR: 1.66; 95% CI: [1.55–1.79]; Day 90: HR: 1.59; 95% CI: [1.47–1.71]), those with bacteremia (Day 28: HR: 1.69; 95% CI: [1.57–1.82]; Day 90: HR: 1.64; 95% CI: [1.53–1.75]), patients with solid organ malignancy, leukemia, or lymphoma (Day 28: HR: 1.38; 95% CI: [1.26–1.51]; Day 90: HR: 1.30; 95% CI: [1.20–1.41]), and those with septic shock (Day 28: HR: 1.22; 95% CI: [1.16–1.27]; Day 90: HR: 1.21; 95% CI: [1.16–1.27]). Sensitivity analyses with varying ferritin cutoff values for the diagnosis of MALS All-cause mortality In sensitivity analyses using alternative ferritin thresholds for MALS, 28-day all-cause mortality remained higher in MALS vs no-MALS, with progressively increasing HRs at higher cutoffs (thresholds [ng/mL]: 2000 (HR: 2.44; 95% CI: [2.35–2.54]); 3000 (HR: 2.57; 95% CI: [2.45–2.70]); 6000 (HR: 2.94; 95% CI: [2.73–3.17]); 10,000 (HR: 3.05; 95% CI: [2.75–3.39])). Similar findings were observed for 90-day mortality, with stepwise increases in HRs at higher thresholds (thresholds [ng/mL]: 2000 (HR: 2.21; 95% CI: [2.14–2.29]); 3000 (HR: 2.36; 95% CI: [2.27–2.46]); 6000 (HR: 2.60; 95% CI: [2.44–2.77]); 10,000 (HR: 2.67; 95% CI: [2.44–2.91])). In Supplemental Figure S1 and Supplemental Figure S2, we present the results of the sensitivity analyses using alternative ferritin cutoff values for the diagnosis of MALS. Need for intubation and mechanical ventilation At 28 days, the risk of intubation and mechanical ventilation was higher in MALS vs no-MALS, with increasing HRs at higher ferritin thresholds (thresholds [ng/mL]: 2000 (HR: 1.88; 95% CI: [1.81–1.95]); 3000 (HR: 1.97; 95% CI: [1.88–2.06]); 6000 (HR: 2.09; 95% CI: [1.95–2.24]); 10,000 (HR: 2.09; 95% CI: [1.90–2.30]). Findings were similar at 90 days, with stepwise increases in HRs at higher cutoffs (thresholds [ng/mL]: 2000 (HR: 1.82; 95% CI: [1.76–1.88]); 3000 (HR: 1.92; 95% CI: [1.84–2.00]); 6000 (HR: 2.02; 95% CI: [1.89–2.16]); 10,000 (HR: 2.05; 95% CI: [1.87–2.25]). Need for ICU admission At 28 days, ICU admission was more frequent in MALS vs no-MALS, with higher HRs at increasing ferritin thresholds (thresholds [ng/mL]: 2000 (HR: 1.70; 95% CI: [1.64–1.75]); 3000 (HR: 1.75; 95 % CI: [1.69–1.82]); 6000 (HR: 1.87; 95 % CI: [1.76–1.99]); 10,000 (HR: 1.87; 95% CI: [1.71–2.04]). Similar results were observed at 90 days, with stepwise increases in HRs at higher cutoffs (thresholds [ng/mL]: 2000 (HR: 1.63; 95% CI: [1.58–1.68]); 3000 (HR: 1.70; 95% CI: [1.63–1.76]); 6000 (HR: 1.77; 95% CI: [1.67–1.88]); 10,000 (HR: 1.78; 95% CI: [1.64–1.93]). Need for CRRT At 28 days, the need for CRRT was significantly greater among patients with MALS than among those without MALS, with higher HRs observed with the application of higher cutoff values (thresholds [ng/mL]: 2000 (HR: 2.17; 95% CI: [2.05–2.29]); 3000 (HR: 2.07; 95% CI: [1.94–2.21]); 6000 (HR: 2.28; 95% CI: [2.06–2.53]); 10,000 (HR: 2.64; 95% CI: [2.27–3.06]). At day 90, the need for CRRT remained higher in the MALS versus no-MALS groups, with a stepwise increase in HRs as higher cutoff thresholds were applied (thresholds [ng/mL]: 2000 (HR: 2.17; 95% CI: [2.07–2.29]); 3000 (HR: 2.06; 95% CI: [1.94–2.19]); 6000 (HR: 2.16; 95% CI: [1.97–2.38]); 10,000 (HR: 2.41; 95% CI: [2.10–2.76]). Need for vasopressor administration At 28 days, need for vasopressor administration was significantly higher among patients with MALS than among those without MALS, and the HR increased with the use of higher cutoff values (thresholds [ng/mL]: 2000 (HR: 1.67; 95% CI: [1.62–1.72]); 3000 (HR: 1.74; 95% CI: [1.67–1.80]); 6000 (HR: 1.78; 95% CI: [1.68–1.89]); 10,000 (HR: 1.91; 95% CI: [1.76–2.08]). At day 90, need for vasopressor administration was also higher in the MALS versus no-MALS groups, with a stepwise increase in the HR as higher cutoff thresholds were applied (thresholds [ng/mL]: 2000 (HR: 1.60; 95% CI: [1.56–1.65]); 3000 (HR: 1.67; 95% CI: [1.62–1.73]); 6000 (HR: 1.67; 95% CI: [1.58–1.77]); 10,000 (HR: 1.78; 95% CI: [1.65–1.93]). Sensitivity analysis in patients without COVID-19 infection In a sensitivity analysis excluding patients with COVID-19 infection (Supplemental Table S11), MALS was associated with higher mortality versus no-MALS (Day 28: HR: 2.93; 95% CI: [2.71–3.17]; Day 90: HR: 2.52; 95% CI: [2.34–2.71]). MALS was also associated with increased need for intubation and mechanical ventilation (Day 28: HR: 2.12; 95% CI: [1.97–2.28]; Day 90: HR: 2.05; 95% CI: [1.90–2.22]), ICU admission (Day 28: HR: 1.85; 95% CI: [1.74–1.97]; Day 90: HR: 1.82; 95% CI: [1.70–1.94]), CRRT initiation (Day 28: 2.24; 95% CI: [2.02–2.50]; Day 90: HR: 2.30; 95% CI: [2.06–2.57]), and vasopressor use (Day 28: HR: 1.79; 95% CI: [1.69–1.90]; Day 90: HR: 1.71; 95% CI: [1.61–1.82]). Impact of anakinra administration in the MALS subgroup In the MALS subgroup, 109 patients received anakinra within 3 days of sepsis diagnosis and 4714 received standard care alone; 5346 were excluded for prior anti-cytokine therapy, COVID-19, autoimmune disease, or earlier anakinra use. After propensity score matching, each subgroup included 104 patients (Supplemental Table S12). Our exploratory analysis showed, that, by day 14, 33 (31.7%) patients in the anakinra group vs 19 (18.3%) patients in the no-anakinra group died (HR: 1.83; 95% CI: [1.04–3.22]), with no mortality difference by baseline ferritin strata (4420–10000 ng/mL: HR: 0.91; 95% CI: [0.42–1.96]); >10,000 ng/mL: HR: 1.01; 95% CI: [0.57–1.80]). By day 28, deaths occurred in 43 (41.3%) versus 30 (28.8%) patients in the anakinra versus no-anakinra groups (HR: 1.54; 95% CI: [0.97–2.46]). Secondary outcome analyses did not reveal significant differences with regards to ICU admission, intubation, CRRT initiation or vasopressor administration between the anakinra and no-anakinra group. In Supplemental Table S13, we present the results of all the analyses for the anakinra versus no-anakinra subgroups. Bonferroni correction to address multiple comparisons in the analysis All subgroup analyses demonstrated statistically significant associations in unadjusted analyses for all five outcomes of interest (all-cause mortality, ICU admission, need for intubation, CRRT initiation, vasopressor administration). All comparisons remained statistically significant after Bonferroni correction (all unadjusted p-values <0.0056) across all outcomes under consideration (Supplemental Table S14). Thus, adjustment for multiple comparisons did not alter the interpretation of statistical significance, and the observed associations were robust across all subgroups and endpoints evaluated. Discussion In this large real-world analysis of patients with sepsis, mortality was markedly higher in the MALS group than in the no-MALS group at both 28 days (HR, 2.73) and 90 days (HR, 2.49). Patients with MALS were approximately twice as likely to require endotracheal intubation and had a significantly higher likelihood of ICU admission at 28 days (HR, 1.86) and 90 days (HR, 1.78). The risk of CRRT initiation was nearly doubled in the MALS group, and the need for vasopressor support was also significantly increased at 28 days (HR, 1.80) and 90 days (HR, 1.73), as compared with patients without MALS. In addition, subgroup analyses in patients aged ≥65 years, those with bacteremia, and immunocompromised patients showed worse 28-day and 90-day outcomes in the MALS group. Moreover, in sensitivity analyses, the use of progressively higher ferritin thresholds to define MALS was associated with stepwise increases in the risks of both primary and secondary outcomes at 28 and 90 days after sepsis diagnosis. The incidence of MALS in our sepsis cohort was 4.5%, defined by a blood ferritin level ≥4420 ng/mL. This threshold has been reported to provide 100% specificity and 100% negative predictive value for MALS classification4 and has been adopted by recent controlled trials.7,15 Prior studies using the same definition have reported MALS incidences ranging from 5.1%16 to 24.8%.15 The lower incidence observed in our study, compared with the ImmunoSep7 and PROVIDE15 trials, likely reflects that those trials focused on selected infections (pneumonia and bacteremia), whereas our cohort included diagnosis of sepsis from all sources. In our analysis, patients with MALS exhibited higher mortality compared to patients with no-MALS. Specifically, mortality was substantially higher in the MALS group than in the no-MALS group at both day 28 (33.1% vs. 14.3%) and day 90 (40.8% vs. 20.3%). Similarly, a retrospective observational study16 including adult patients with sepsis admitted to an ICU showed that mortality in patients with sepsis and MALS was twice as high than in patients with no-MALS (62.4% vs. 30.5%). The higher mortality observed in that cohort likely reflects its restriction to critically ill patients requiring ICU-level care, whereas our study encompassed patients with sepsis and MALS regardless of level of care. Outside the context of MALS, ferritin has been found to be an independent predictor of mortality in patients with sepsis. Liu et al. conducted a retrospective database analysis of 1553 patients with sepsis and found that serum ferritin (measured as a continuous variable) and serum ferritin-to-albumin ratio were independently associated with worse 28-day mortality in patients with sepsis.17 Zhang et al. conducted a retrospective study of 451 elderly patients with community-acquired pneumonia and found that ferritin was independent predictor of 28-day mortality and need for vasopressor administration.18 Collectively, these results support our findings and highlight the need for prospective studies incorporating systematic ferritin measurement to facilitate early identification of patients with sepsis who are at increased risk for decompensation. Systematic measurement of ferritin is particularly important given that approximately 80% of patients in our initially identified cohort did not have a ferritin level obtained within 3 days before or after the diagnosis of sepsis, likely reflecting current clinical practice in which the prognostic relevance of ferritin in sepsis is not routinely recognized. In subgroup analyses, patients aged ≥65 years, those with bacteremia, and immunocompromised patients had worse 28-day and 90-day outcomes in the MALS group compared with the no-MALS group. These populations are already known to have poorer outcomes in sepsis,19, 20, 21, 22 and our findings extend this knowledge by demonstrating that MALS further amplifies risk in these high-risk subgroups. Prior randomized studies in MALS were limited by small sample sizes and were not designed to evaluate the impact of MALS within clinically relevant patient subgroups. Secondary outcomes, including ICU admission, intubation, CRRT, and vasopressor initiation, were also more frequent in the MALS subgroup, highlighting the need for heightened surveillance for MALS in older, bacteremic, and immunocompromised patients. In sensitivity analyses, progressively higher ferritin thresholds were associated with stepwise increases in mortality risk, supporting a dose–response relationship between hyperferritinemia and severity of sepsis. While Kyriazopoulou et al.4 proposed a ferritin cutoff of ≥4420 ng/mL to identify a hyperinflammatory MALS phenotype with high specificity, our findings suggest that ferritin may function less as a binary diagnostic marker and more as a continuum reflecting escalating immune dysregulation. Prior studies support this concept. In the study by He et al., incremental increases in ferritin (1000–4000 ng/mL) were associated with progressively higher mortality risk,23 and in the study by Sang et al., among patients with sepsis, stepwise increases in ferritin across pre-defined quartiles (<102 to >645 ng/mL) were associated with progressively higher mortality and a longer duration of ICU stay.24 Our results extend these observations within a large real-world cohort and align with emerging models of biologically defined sepsis endotypes. Notably, ferritin elevations below the traditional MALS threshold were consistently associated with worse clinical outcomes, including increased risk of ICU admission, mechanical ventilation, CRRT, and vasopressor use. Together, these findings support a graded hyperinflammatory risk spectrum and suggest that ferritin trajectories may inform risk stratification beyond fixed diagnostic cutoffs. Conversely, markedly elevated ferritin levels were associated with substantially worse outcomes in our study, supporting the concept that extreme hyperferritinemia identifies a subset of patients with more profound hyperinflammation and higher mortality risk. This dose–response pattern suggests that ferritin functions not only as a binary diagnostic marker for MALS, but also as a continuous prognostic biomarker in sepsis. Taken in totality, these findings expand the conceptual role of ferritin in sepsis to an early risk stratification tool with implications for clinical decision-making. Modestly elevated ferritin levels may warrant closer monitoring for evolving immune dysregulation, whereas very high levels may identify patients at particularly high risk for adverse outcomes and potentially greater benefit from immunomodulatory strategies. However, increasing ferritin thresholds improve specificity at the cost of reduced sensitivity; diagnostic precision should be balanced with the need for early identification of patients who may benefit from targeted interventions. In this real-world analysis of hospitalized patients with sepsis and MALS, our exploratory analysis showed that treatment with anakinra was not associated with lower mortality at day 14 or day 28 as compared with standard care. Although early mortality was higher among patients who received anakinra, this finding most likely reflects confounding by indication, given the more severe inflammatory phenotype and higher ferritin levels in treated patients. Consistent with this interpretation, no difference in day 14 mortality was observed after stratification by ferritin level. These findings align with those of the ImmunoSep trial,7 in which anakinra did not improve survival in patients with MALS. However, ImmunoSep was a randomized trial that enrolled a highly selected population with pneumonia or bacteremia and evaluated changes in the sequential organ failure assessment (SOFA) score as the primary outcome, whereas our study leveraged real-world data from a broader population with sepsis from all sources and focused on other clinically meaningful outcomes instead of score surrogates, including mortality and major organ-support requirements. In addition, the limited number of patients with MALS who received anakinra in both studies likely constrained statistical power to detect modest treatment effects. Future studies should enroll larger cohorts of patients with sepsis and MALS to determine whether targeted therapy with anakinra or other agents25 previously evaluated in hyperinflammatory sepsis, can improve clinical outcomes. Ongoing trials, including IgM-fat26 and PALETTE,27 are evaluating the efficacy of several interventions (IgM, corticosteroids, anakinra, and tocilizumab) in patients with sepsis and are expected to provide insights into precision immunotherapy strategies in this population. This analysis based on a large real-world database can help evaluate certain clinical questions, such as the need for systematic ferritin measurement in patients with sepsis, and optimal ferritin cutoff values for MALS diagnosis. However, it has several limitations. A key limitation relates to selection bias arising from ferritin measurement practices, as approximately 80% of patients with sepsis did not have ferritin assessed within the study window, reflecting real-world practice in which ferritin testing is typically obtained in patients with suspected heightened inflammation or immune dysregulation. Consequently, the no-MALS cohort represents patients who underwent ferritin testing but did not meet criteria for MALS, rather than the general sepsis population. This enrichment of both groups for higher-risk patients likely biases effect estimates toward the null; therefore, the adverse impact of MALS compared with unselected sepsis populations may be underestimated. Nevertheless, MALS remained consistently associated with worse outcomes across multiple endpoints, suggesting that it carries a strong prognostic significance even within this higher-risk cohort. Also, although propensity-score matching was performed for major demographic characteristics and comorbid conditions associated with outcomes, residual confounding from unmeasured variables cannot be excluded. Second, owing to limitations of the TriNetX database, changes in the SOFA score, a surrogate end point used in prior immunotherapy trials in sepsis, could not be evaluated. Third, the number of patients receiving anakinra was small, which may have limited the statistical power to detect differences in mortality between the anakinra and no-anakinra subgroups. As such, we considered the impact of anakinra on study outcomes to be an exploratory analysis, and firm conclusions regarding the effectiveness of anakinra in patients with MALS cannot be reliably drawn. Fourth, despite the application of strict exclusion criteria to minimize alternative indications for anakinra use, misclassification remains possible because TriNetX does not link medication administration to a specific diagnosis and we were unable to assess the duration of anakinra therapy because of absence of granular treatment data in the TriNetX database. All-cause mortality was analyzed using standard time-to-event methods. For the secondary outcomes (ICU admission, intubation, CRRT initiation, and vasopressor use), patients were followed until the event occurred or until they were no longer at risk due to death, loss to follow-up, or the end of the study period. In these analyses, patients who died before experiencing a given outcome were treated as no longer at risk for that outcome. This approach has an important implication: death may preclude the occurrence of certain outcomes and thereby influence their observed incidence. Because formal competing-risk methods could not be applied, the findings for secondary outcomes should be interpreted with appropriate caution. Specifically, observed differences in these outcomes may partially reflect differences in survival, rather than solely differences in the underlying risk of the events themselves. In conclusion, in this large database analysis, patients with sepsis and MALS had significantly higher all-cause mortality than those without MALS. Secondary outcomes, including need for intubation, CRRT, vasopressors, and ICU admission, were also worse for MALS, particularly among patients older than 65 years, with bacteremia, or immunocompromise. This study challenges the binary approach to diagnosing MALS by ferritin cutoffs and supports a graded hyperinflammatory risk spectrum. Higher ferritin levels, even below the current diagnostic threshold for MALS, were associated with progressively worse primary and secondary outcomes, supporting the value of routine ferritin measurement for early risk stratification and clinical decision-making in sepsis. Future studies should incorporate systematic ferritin measurement alongside biologically defined sepsis endotypes to refine risk stratification, assess the prognostic value of longitudinal ferritin trends, and evaluate whether therapies targeting hyperferritinemia-associated hyperinflammatory pathways decrease mortality and organ dysfunction. Contributors Conceptualization: KO; data curation: KO, MM, and TK; formal analysis: KO, MM, and TK; funding acquisition: there was no funding source for this study; supervision: EM; writing—original draft: KO; writing—review & editing: KO, MM, TK, SD, EKM, FS, DBG and EM. KO, MM, and TK accessed and verified the underlying data. All authors read and approved the final version of the manuscript. Data sharing statement This study was conducted within the TriNetX Analytics Platform, a commercial, cloud-based environment for real-world data research. All analyses were performed using the platform's native analytical tools. Because TriNetX is a proprietary, closed-source system, the underlying code is not publicly accessible. However, investigators with authorized access to TriNetX may reproduce the analyses by applying the methods detailed in the manuscript and Supplementary Appendix. Declaration of interests The authors declare no competing interests. Acknowledgements In compliance with the International Committee of Medical Journal Editors (ICMJE) recommendations, the authors acknowledge the utilization of AI services in this work: the authors of this article used ChatGPT (GPT-5.2) by OpenAI exclusively for minor English correction and polishing in this work. All sentences revised by GPT-5.2 were reviewed and verified by the authors. No content was generated by GPT-5.2 or any other AI service. Footnotes Supplementary data related to this article can be found at https://doi.org/10.1016/j.eclinm.2026.104013. Supplementary data Supplementary Tables and Figures References Associated Data Supplementary Materials Supplementary Tables and Figures