Serial assessment of serum HMGB1 and ΔSOFA for predicting 28-day mortality in sepsis patients: a prospective cohort study in the emergency department

Abstract

Background

Prognostic assessment in sepsis involves evaluating the dynamic progression of organ dysfunction and the inflammatory response. This study compared the predictive value of serially measured serum HMGB1, a key late mediator of sepsis, with the ΔSOFA score for 28-day mortality.

Methods

This prospective cohort study enrolled 250 sepsis patients admitted to the emergency department of a tertiary hospital from January 2022 to August 2024. Serum HMGB1 levels and SOFA scores were dynamically assessed on Days 1, 4, and 7. Receiver operating characteristic (ROC) curve analysis and Kaplan-Meier survival analysis were utilized to compare their prognostic performance. The primary endpoint was 28-day all-cause mortality.

Results

A total of 232 patients (median age 71.5 years, 56.0% male) with a 28-day mortality rate of 13.8% (32/232) were included. Serum HMGB1 levels peaked on Day 4 and were significantly higher in non-survivors and septic shock patients (p < 0.05). The area under the ROC curve (AUC) for predicting 28-day mortality was 0.858 (95% CI: 0.789–0.925) for D4-HMGB1 and 0.893 (95% CI: 0.830–0.935) for D7-ΔSOFA. The AUC for D4-HMGB1 was not significantly different from that of D7-ΔSOFA (0.858 vs. 0.893, p = 0.29), but was significantly higher than that of D1-HMGB1 (0.858 vs. 0.699, p = 0.04) and D4-ΔSOFA (0.858 vs. 0.767, p = 0.01). The AUC was enhanced when HMGB1 was combined with SOFA scores, with the combination yielding higher AUC values (D7-HMGB1 + D7-SOFA: AUC = 0.898, 95% CI: 0.829–0.977, p = 0.09; D4-HMGB1 + D4-SOFA: AUC = 0.877, 95% CI: 0.792–0.928, p = 0.90) compared to the individual components at respective time points, but these increases were not statistically significant. The optimal cut-off value for D4-HMGB1 was 6.4 ng/mL. Kaplan-Meier analysis showed that patients with D4-HMGB1 ≥ 6.4 ng/mL had significantly higher mortality (log-rank test, p = 0.001). This prognostic performance remained consistent across key patient subgroups, including those with acute kidney injury, autoimmune diseases, or respiratory comorbidities.

Conclusion

Dynamic monitoring of serum HMGB1 levels in sepsis patients for 28-day mortality provides potential prognostic information. Specifically, D4-HMGB1 levels showed similar predictive performance for 28-day mortality to the D7-ΔSOFA score, highlighting its potential as an earlier warning tool in the emergency department.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12873-026-01547-2.

Background

Sepsis is defined as life-threatening organ dysfunction resulting from a dysregulated host response to infection. It represents a significant global health burden, with mortality rates ranging from 30% to 50%, as indicated by recent epidemiological studies [1, 2]. Accurate prognostic assessment is as equally crucial as early diagnosis in the management of sepsis, as it enables appropriate resource allocation and guides therapeutic interventions.

The Sequential Organ Failure Assessment (SOFA) score has been established as a key tool for diagnosing sepsis and assessing its severity, according to international guidelines [1–3]. Although the initial SOFA score provides valuable information, there is growing evidence that dynamic monitoring using ΔSOFA (the change in SOFA score over time) offers superior prognostic capability [4–7]. The evolving pattern of organ dysfunction, as captured by serial SOFA measurements, more accurately reflects disease progression and therapeutic response than a single assessment. Notably, recent research published in JAMA indicates that the SOFA-2 score provides updated definitions to describe organ dysfunction in adult patients requiring critical care and offers readily quantifiable criteria to grade the degree of dysfunction in individual organ systems [8]. This score takes into account contemporaneous changes in patient management and outcomes and has demonstrated improved prognostic utility for ICU mortality. However, this score still lacks an assessment of the host’s immune function, highlighting a critical gap between purely clinical parameters and underlying biomolecular mechanisms.

Concurrent research efforts have focused on identifying biomarkers that reflect the complex pathophysiology of sepsis. In addition to traditional markers such as C-reactive protein (CRP) and procalcitonin (PCT) [9], a variety of novel biomarkers have emerged. The soluble urokinase plasminogen activator receptor (suPAR) acts as a broad marker of immune activation; however, its standalone diagnostic power is limited due to moderate sensitivity [10]. Cell surface and soluble receptors, such as neutrophil CD64 [11], presepsin [12], and soluble TREM-1 [13], provide insights into immune cell activation. However, their utility can be limited by technical constraints or elevated levels in non-septic inflammatory conditions. Cytokines and chemokines [14], including IL-6 and MCP-1, robustly reflect the magnitude of inflammation, but their transient expression restricts their usefulness in late-phase assessment. Furthermore, biomarkers such as mid-regional pro-adrenomedullin (MR-proADM) and its active form, bio-ADM, are strongly associated with endothelial dysfunction and shock, demonstrating superior prognostic performance for mortality [15]. Similarly, proenkephalin (Penkid) has emerged as a sensitive and early indicator of sepsis-associated acute kidney injury [16]. Unlike early-response cytokines, high mobility group box 1 protein (HMGB1) displays delayed release kinetics and plays a role in sustained inflammatory responses that lead to organ injury [17, 18]. Its extracellular functions include the activation of innate immune pathways via Toll-like receptors (TLRs) and the receptor for advanced glycation end products (RAGE), the amplification of pro-inflammatory cytokine production, and the promotion of endothelial dysfunction and pyroptosis [19–21]. Clinical studies have demonstrated correlations between elevated plasma HMGB1 levels and the severity of organ dysfunction and mortality in septic shock [22, 23], highlighting organ-specific dysfunction and dysregulated immune response [24]. Notably, HMGB1 in sepsis patients exhibits a distinct time-dependent pattern, with its serum level typically peaking around Day 4 of the disease course. This dynamic characteristic suggests that monitoring HMGB1 levels, particularly the timing and magnitude of its peak, may provide additional information for assessing patient prognosis. Although the prognostic role of HMGB1 in sepsis has been explored in prior clinical [22, 23] and experimental studies [25], there remains a paucity of literature directly and chronologically comparing the value of its serial measurement at different time points with dynamic organ dysfunction assessment tools, such as the change in Sequential Organ Failure Assessment (ΔSOFA) score. D7-ΔSOFA, which can only be calculated on Day 7, is an established dynamic clinical metric for prognostic evaluation in sepsis [4–7]. However, especially in the emergency department (ED) setting where early risk stratification is crucial, whether a biomarker measured at an earlier time point can provide prognostic information comparable to the later-calculated D7-ΔSOFA has not been thoroughly investigated.

This prospective cohort study, conducted in the emergency department setting, systematically characterized the expression dynamics of serum HMGB1 levels through serial measurements in sepsis patients. The core objective is to directly compare the predictive performance of serially measured HMGB1, dynamically assessed ΔSOFA (specifically D7-ΔSOFA), and to construct biomarker-clinical score combination for 28-day mortality. The study will evaluate whether HMGB1 measured at earlier time points can provide prognostic information comparable to or potentially superior to D7-ΔSOFA, which can only be calculated later in the disease course (by Day 7), thereby exploring its potential future integration value for early clinical risk stratification.

Materials and methods

Study design and patients

This prospective observational cohort study was conducted in the emergency department (ED) of a tertiary care hospital in Shanghai, China, between January 2022 and August 2024. A total of 250 consecutive adult patients (age ≥ 18 years) meeting the Sepsis-3 diagnostic criteria [2] were initially screened for eligibility. The inclusion criteria required: (1) presentation to the ED within 72 h of symptom onset; and (2) fulfillment of Sepsis-3 criteria. Exclusion criteria comprised: (1) diagnosis of active malignancy; (2) missing essential clinical data or loss to follow-up; or (3) declined to provide informed consent. Ultimately, 232 patients completed the 28-day follow-up and were included in the final analysis (Fig. 1). Serum HMGB1 levels and SOFA scores were prospectively monitored in all enrolled patients at three predefined times: within 2 h of diagnosis on Day 1 (admission), and on the mornings of Days 4 and 7. Fasting blood samples were collected at specified times by nursing staff. The time points (Day 1, Day 4, and Day 7) were selected based on considerations of the sepsis pathophysiology: Day 1 reflects the initial state; Day 4 may capture the peak of late-phase inflammatory mediators like HMGB1; and Day 7 is commonly used for prognostic assessment with dynamic clinical scores such as ΔSOFA. The primary endpoint for prognostic evaluation was 28-day all-cause mortality.

Fig. 1.

Flowchart of the study population

Diagnostic criteria were based on the Sepsis-3 definitions jointly published by the Society of Critical Care Medicine (SCCM) and the European Society of Intensive Care Medicine (ESICM) [2]. Septic shock was identified as a subset of sepsis characterized by profound circulatory, cellular, and metabolic abnormalities, clinically defined by the requirement of vasopressors to maintain a mean arterial pressure ≥ 65 mmHg and serum lactate level > 2 mmol/L (> 18 mg/dL) despite adequate fluid resuscitation. Treatment protocols followed the 2018 Chinese Guidelines for Emergency Management of Sepsis/Septic Shock, which emphasize 1-hour bundle therapy (including broad-spectrum antibiotics, fluid resuscitation, and source control) alongside dynamic monitoring of organ function. The study protocol was approved by the Institutional Review Board of Shanghai Jiao Tong University School of Medicine Affiliated Renji Hospital (Approval No: RA-2021-487). Written informed consent, detailing the study objectives, procedures, and potential risks, was obtained from all participants or their legally authorized representatives.

Data collection

Upon enrollment, blood samples were collected from each patient for baseline analysis. Subsequent sampling for serum HMGB1 quantification was performed on Day 4 and Day 7. However, not all patients contributed samples at these time points due to attrition: on Day 4, 221 samples were available, while on Day 7, 159 samples were available. The reasons for missing samples included death (Day 4, n = 1; Day 7, n = 8), discharge (Day 4, n = 9; Day 7, n = 48), and patient refusal (Day 4, n = 1; Day 7, n = 6).

The following data were systematically recorded for each participant upon ED admission: demographic characteristics (age, gender), comorbidities, primary infection site, laboratory parameters, SOFA scores, and clinical outcomes. Laboratory assessments included complete blood count, electrolyte levels, coagulation profile, hepatic and renal function tests, C-reactive protein, procalcitonin, lactate, and cytokine levels (including IL-6, IL-8, IL-10, IL-17 A, and TNF-α). All laboratory assessments were performed according to standardized manufacturer protocols. Data on critical interventions and complications were prospectively collected, including mechanical ventilation (MV), acute kidney injury (AKI), and disseminated intravascular coagulation (DIC). AKI was diagnosed in accordance with the Kidney Disease: Improving Global Outcomes (KDIGO) guideline [26]. DIC was diagnosed according to the International Society on Thrombosis and Haemostasis (ISTH) scoring system [27].

Microbiological identification was conducted through a comprehensive diagnostic approach. Blood cultures were obtained for all patients, while cultures from other sites—including tracheal aspirate, cerebrospinal fluid (CSF), urine, and central venous catheter (CVC) tips—were collected when clinically indicated. Furthermore, pathogen detection was supplemented with advanced methodologies, including next-generation sequencing (NGS), serological immunoassays, and polymerase chain reaction (PCR) when required, to ensure comprehensive microbiological assessment.

The Sequential Organ Failure Assessment (SOFA) score was calculated for all patients on Day 1 (admission), Day 4, and Day 7. The change in SOFA score (ΔSOFA) was defined as the difference between the score on a given day and the baseline (Day 1) score, calculated as: D4-ΔSOFA = SOFA (Day 4) - SOFA (Day 1); D7-ΔSOFA = SOFA (Day 7) - SOFA (Day 1).

Sample collection and measurement of serum HMGB1

Blood samples for serum HMGB1 analysis were collected at the aforementioned time points. The sample collection procedure was standardized as follows: three milliliters of venous blood were drawn from each patient and transferred into anticoagulant-containing tubes. The samples were subsequently centrifuged at 3,000 revolutions per minute for 15 min to separate serum from cellular components. The supernatant serum was carefully aliquoted and stored at -80 °C until further analysis. Serum HMGB1 concentrations were quantified using a commercially available double-antibody sandwich enzyme-linked immunosorbent assay (ELISA) kit (Arigo, AREX Biosciences Ltd., China), strictly following the manufacturer’s instructions. To minimize measurement bias, all samples were re-coded before analysis, and researchers performing the assays were blinded to clinical information and group assignments. Each sample was tested in duplicate, and the average of the two measurements was used for statistical analysis. The lower limit of detection for the assay was 0.3125 ng/mL.

Cytokine concentrations

Serum concentrations of interleukin (IL)-6, IL-8, IL-10, IL-17a, and tumor necrosis factor-alpha (TNF-α) were quantitatively measured using Luminex multiplex cytokine assays. These assays employ distinct sets of prestained microbeads, each with a unique fluorescence signature detectable by flow cytometry. Each microbead set is coated with a specific capture antibody that binds the target analyte, forming an immunocomplex with a fluorescently labeled detection antibody. All measurements were performed using the Fluorokine^® MAP Multiplex Kit (R&D Systems, Minneapolis, MN) on a Luminex^® 100/200™ system (Luminex Corporation, Austin, TX) in strict accordance with the manufacturer’s protocols. The assay detection range for all cytokines was 18–10,000 pg/mL. All samples were analyzed in duplicate to ensure technical reproducibility. Samples exceeding the upper limit of detection were appropriately diluted and reanalyzed to obtain values within the quantitative range.

Statistical analysis

Categorical data are presented as numbers (percentages). Continuous variables are presented as mean ± standard deviation (SD) or median (interquartile range [IQR]), depending on their distribution. The normality of distribution was assessed using the Shapiro-Wilk test. Comparisons of continuous variables between groups were performed using the independent Student’s t-test or the Mann-Whitney U test, as appropriate. Categorical variables were compared using Pearson’s chi-square test or Fisher’s exact test. Receiver operating characteristic (ROC) curve analysis was conducted to evaluate the predictive performance of HMGB1 and SOFA scores for 28-day mortality. The optimal cut-off values were determined by maximizing Youden’s index. Areas under the ROC curve (AUCs), 95% confidence interval (CI), along with sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), were calculated. Differences between areas under the ROC curves (AUCs) were compared using Delong’s test. Correlations between HMGB1 levels and SOFA scores, as well as between HMGB1 and inflammatory cytokine levels, were analyzed using Spearman’s rank correlation coefficient.

Kaplan-Meier survival curves were plotted, with patients stratified by the optimal cut-off value of D4-HMGB1, and group differences were compared using the log-rank test. The prognostic performance of D4-HMGB1 was further evaluated in subgroup analyses of patients with acute kidney injury, autoimmune diseases, or respiratory comorbidities. A two-sided p-value < 0.05 was considered statistically significant for all analyses. All statistical analyses were performed using GraphPad Prism version 8.0 (GraphPad Software, San Diego, CA, USA) and SigmaPlot version 14.0 (Systat Software, San Jose, CA, USA).

Results

Demographic and clinical characteristics of patients with sepsis

A total of 232 patients with sepsis were included in this cohort study, comprising 130 males and 102 females. All patients were followed up for 28 days and were categorized by outcome into survivors (n = 200) and non-survivors (n = 32). Based on disease severity, patients were further stratified into sepsis (n = 154) and septic shock (n = 78) groups. The baseline characteristics of the patients are presented in Table 1 and Supplementary Table S1. The SOFA scores and HMGB1 levels across time points are detailed in Table 2 and Supplementary Table S2.

Table 1.

Baseline demographics and clinical characteristics according to 28-day mortality

Variable	Total (n = 232)	Survivors (n = 200)	Non-survivors (n = 32)	p value
Demographic characteristics
Gender, male, n (%)	130 (56.0)	102 (56.0)	18 (56.3)	0.979
Age, median (years)	71.5 (58.5–77.8)	72.0 (59.0-78.8)	70.0 (59.0-78.5)	0.125
BMI, mean ± SD (kg/m2)	23.4 ± 4.3	23.5 ± 4.3	22.3 ± 5.2	0.512
Comorbidities, n (%)
Diabetes mellitus	64 (27.5)	58 (29.0)	6 (18.8)	0.228
Hypertension	91 (39.2)	83 (41.5)	8 (25.0)	0.076
Coronary artery disease	36 (15.5)	28 (14.0)	8 (25.0)	0.111
Cerebrovascular disease	29 (12.5)	23 (11.5)	6 (18.8)	0.250
Autoimmune diseases	37 (15.9)	23 (11.5)	14 (43.8)	< 0.001
Pulmonary diseases	10 (4.3)	7 (3.5)	3 (9.4)	0.129
Primary site of infection, n (%)
Lung	120 (51.7)	94 (47.0)	26 (81.3)	< 0.001
Digestive tract	18 (7.7)	16 (8.0)	2 (6.3)	0.731
Urinary tract	45 (19.3)	43 (21.5)	2 (6.3)	0.051
Biliary tract	30 (12.9)	27 (13.5)	3 (9.4)	0.409
Central nervous system	4 (1.7)	4 (2.0)	0 (0)	0.184
Skin and soft tissue	7 (3.0)	7 (3.5)	0 (0)	0.366
Unknown/others	8 (3.4)	8 (4.0)	0 (0)	0.400
Pathogen(s), n (%)
Bacteria	54 (23.2)	48 (24.0)	6 (18.8)	0.514
Fungi	12 (5.2)	11 (5.5)	1 (3.1)	0.573
Virus	2 (0.9)	1 (0.5)	1 (3.1)	0.136
Mixed	15 (6.5)	10 (5.0)	5 (15.6)	< 0.001
Laboratory parameters, n (%)
Leucocytes (*109/L)	10.5 (7.3–15.2)	10.4 (7.2–15.1)	11.1 (8.0-15.2)	0.520
Platelets (*109 /L)	148.5 (91.5–211.0)	147.5 (92.8-214.8)	157.0 (81.5-197.3)	0.445
CRP (mg/L)	110.5 (57.1-179.8)	110.4 (57.9-178.5)	117.9 (46.8-189.4)	0.459
PCT (ng/mL)	3.4 (0.6–20.1)	3.8 (0.6–21.1)	2.2 (0.6–9.7)	0.208
Lactate (mmol/L)	2.3 (1.6–3.1)	2.2 (1.6-3.0)	2.5 (1.9–3.5)	0.080
ALC (cells/uL)	820.7(488.7-1247.9)	890.0(560.0-1340.0)	373.7(162.5-552.2)	0.006
IL-6 (pg/mL)	42.5 (15.9-105.5)	37.9 (15.0-93.3)	180.7 (71.4-6378.6)	0.007
IL-8 (pg/mL)	48.6 (25.7-106.7)	45.8 (24.1–95.5)	261.6 (81.5-1212.2)	0.001
IL-10 (pg/mL)	9.8 (5.4–26.8)	9.6 (5.4–21.0)	33.8 (24.2–41.5)	0.014
TNF-α (pg/mL)	2.2 (1.6–2.7)	2.2 (1.6–2.7)	2.5 (1.9–3.3)	0.163
IL-17 A (pg/mL)	2.4 (1.8–3.5)	2.4 (1.7–3.4)	3.1 (2.6–15.0)	0.030
Complications, n (%)
MV, n (%)	51 (21.9)	30 (15.0)	21 (65.6)	< 0.001
AKI, n (%)	93 (40.1)	79 (39.5)	14 (43.8)	0.649
DIC, n (%)	20 (8.6)	9 (4.5)	11 (34.4)	< 0.001

Values are presented as the mean ± standard deviation, median (interquartile range) or number (%)

BMI: body mass index; SD: standard deviation; ALC: absolute lymphocyte count; CRP: C-reactive protein; PCT: procalcitonin; SOFA: sequential organ failure assessment score; HMGB1: high-mobility group protein B1; MV: mechanical ventilation; AKI: acute kidney injury; DIC: disseminated intravascular coagulation

Table 2.

Serum HMGB1 levels, SOFA, and ΔSOFA between survivors and non-survivors

Variable	Total (n = 232)	Survivors (n = 200)	Non-survivors (n = 32)	p value
SOFA score, median
Day 1 (n = 232)	5.0 (3.0–9.0)	5.0 (3.0–8.0)	9.0 (5.0–12.0)	< 0.001
Day 4 (n = 221)	4.0 (2.0–7.0)	3.0 (1.0–7.0)	10.0 (5.0-12.5)	< 0.001
Day 7 (n = 159)	3.0 (1.0–6.0)	2.0 (1.0–5.0)	11.0 (6.0-12.5)	< 0.001
ΔSOFA, Day 4 - Day 1	-1.0 (-2.0-1.0)	-1.0 (-2.0-1.0)	1.0 (0–1.0)	< 0.001
ΔSOFA, Day 7 - Day 1	-2.0 (-3.0-0.0)	-2.0 (-4.0–1.0)	1.0 (0–2.0)	< 0.001
HMGB1 (ng/mL), median
Day 1 (n = 232)	5.0 (4.2–5.4)	5.1 (4.3–5.5)	4.2 (3.4–5.1)	0.001
Day 4 (n = 221)*	5.8 (5.4–6.34)	5.6 (5.3-6.0)	6.8 (6.2-7.0)	< 0.001
Day 7 (n = 159)#	3.5 (3.3–3.9)	3.5 (3.3–3.8)	5.1 (3.8–6.6)	0.013

Values are presented as the median (interquartile range)

^*The sample size decreased to 221 on Day 4 due to 1 death, 9 discharges, and 1 withdrawal of consent

^# The sample size decreased to 159 on Day 7 due to 9 deaths, 57 discharges, and 7 patient withdrawals

SOFA: sequential organ failure assessment score; HMGB1: high-mobility group protein B1

The baseline levels of absolute lymphocyte count (ALC) and key inflammatory cytokines (IL-6, IL-8, and IL-10) were significantly different between patient groups stratified by both 28-day mortality and disease severity (Table 1, Supplementary Table S1). As shown in Table 1, non-survivors had a significantly lower ALC than survivors (median, 373.7 vs. 890.0 cells/µL; p = 0.006). Conversely, levels of the pro-inflammatory cytokines IL-6 (median, 180.7 vs. 37.9 pg/mL; p = 0.007) and IL-8 (median, 261.6 vs. 45.8 pg/mL; p = 0.001), as well as the anti-inflammatory cytokine IL-10 (median, 33.8 vs. 9.6 pg/mL; p = 0.014), were significantly higher in non-survivors. A similar pattern of immune and inflammatory markers was observed when patients were stratified by disease severity (Supplementary Table S1).

The overall 28-day mortality rate was 13.8% (32/232). Non-survivors had a higher prevalence of autoimmune diseases (43.8% vs. 11.5%) and lung infections (81.3% vs. 47.0%) compared to survivors. Dynamic SOFA scores (measured on Days 1, 4, and 7) and HMGB1 levels (on Days 4 and 7) were significantly higher in non-survivors than in survivors. A similar pattern was observed when comparing patients with septic shock to those with sepsis. Additionally, both non-survivors and patients with septic shock were more likely to require mechanical ventilation and to develop disseminated intravascular coagulation (DIC) compared to their respective control groups (survivors and sepsis patients). However, the occurrence of acute kidney injury (AKI) did not differ significantly between the sepsis and septic shock groups, or between survivors and non-survivors.

Serum HMGB1 levels in sepsis patients

Figure 2 illustrates the dynamic changes in serum HMGB1 levels across the study period. As shown in Fig. 2A and B, serum HMGB1 levels peaked on Day 4 across all patient groups and regardless of the baseline SOFA score stratification. Non-survivors and patients with septic shock exhibited significantly higher HMGB1 levels than survivors and patients with sepsis alone, respectively, at each corresponding time point (Fig. 2A). The overall dynamic trend across all patients showed HMGB1 levels increasing to a peak on Day 4, followed by a decline (Fig. 2C).

Fig. 2.

Serial measurements of serum HMGB1 levels in patients with sepsis. A) Serum HMGB1 levels on Days 1, 4, and 7, compared between survivors and non-survivors, and between patients with sepsis and septic shock; B) Serum HMGB1 levels on Days 1, 4, and 7, stratified by the baseline (Day 1) SOFA score; C) Temporal trends of serum HMGB1 levels from Day 1 to Day 7 in survivor and non-survivor groups

Additionally, patients whose HMGB1 levels increased during the first 3 days had a higher mortality rate than those whose levels decreased during this period (Supplementary Figure S1A). This association was particularly pronounced in patients who maintained high HMGB1 levels into the subsequent 3-day period (Days 4–7). The dynamic trends of SOFA scores were broadly consistent with those of HMGB1 levels over time (Supplementary Figure S1B).

Spearman’s rank correlation analysis indicated that serum HMGB1 levels were positively associated with SOFA scores and inflammatory factors (IL-6, IL-8, IL-10, IL-17 A, TNF-α) in patients with sepsis (Supplementary Figure S2 and S3). As shown in Supplementary Figure S2A, no significant correlation was observed between HMGB1 levels and SOFA scores on Day 1 (r = -0.03, p = 0.43). A significant positive correlation was observed on Day 4 (r = 0.33, p < 0.01) (Supplementary Figure S2B), which persisted until Day 7 (r = 0.28, p < 0.01) (Supplementary Figure S2C). As illustrated in Supplementary Figure S2D and S2E, a weak but statistically significant positive correlation was observed between HMGB1 levels and ΔSOFA scores (D4-ΔSOFA: r = 0.22, p = 0.01; D7-ΔSOFA: r = 0.28, p < 0.01). D1-HMGB1 levels demonstrated only weak to negligible correlations with a panel of key pro-inflammatory and anti-inflammatory cytokines, including IL-6, IL-8, IL-10, IL-17 A, and TNF-α, and all correlation coefficients were < 0.3 (Supplementary Figure S3).

HMGB1 for the prediction of 28-day mortality in sepsis patients

Table 3; Fig. 3 present the ROC analysis of serum HMGB1 levels and SOFA scores for predicting 28-day mortality in patients with sepsis. Serum HMGB1 levels on Day 4 (D4-HMGB1; AUC = 0.858, 95% CI: 0.789–0.925, sensitivity = 76.9%, specificity = 88.1%) and the change in SOFA score by Day 7 (D7-ΔSOFA; AUC = 0.893, 95% CI: 0.830–0.935, sensitivity = 87.4%, specificity = 75.8%). The AUC for D4-HMGB1 was not significantly different from that of D7-ΔSOFA (0.858 vs. 0.893, p = 0.29), but was significantly higher than that of D1-HMGB1 (0.858 vs. 0.699, p = 0.04) and D4-ΔSOFA (0.858 vs. 0.767, p = 0.01). The AUC was enhanced when HMGB1 was combined with SOFA scores, with the combination yielding higher AUC values (D7-HMGB1 + D7-SOFA: AUC = 0.898, 95% CI: 0.829–0.977, p = 0.09; D4-HMGB1 + D4-SOFA: AUC = 0.877, 95% CI: 0.792–0.928, p = 0.90) compared to the individual components at respective time points, but these increases were not statistically significant. The AUC for the combined model of D4-HMGB1 and D4-SOFA was not significantly different from that of D7-ΔSOFA alone (0.877 vs. 0.893, p = 0.59) or the combined model of D7-HMGB1 and D7-SOFA (0.877 vs. 0.898, p = 0.52).

Table 3.

Predictive accuracy of HMGB1, ΔSOFA, and their combination at different time points for 28-day mortality in patients with sepsis

Variables	D1-HMGB1 (ng/mL)	D4-HMGB1 (ng/mL)	D7-HMGB1 (ng/mL)	D4-ΔSOFA	D7-ΔSOFA	D4-HMGB1 + D4-SOFA	D7-HMGB1 + D7-SOFA
AUC (95%CI)	0.699 (0.625–0.795)	0.858 (0.789–0.925)	0.817 (0.659–0.957)	0.758 (0.626–0.799)	0.893 (0.830–0.935)	0.874 (0.792–0.928)	0.898 (0.829–0.977)
Best Cut-off Value	4.4	6.4	4.6	-0.5	-1.5	-	-
Sensitivity (%)	64.5	76.9	70.7	80.0	87.4	84.0	66.7
Specificity (%)	78.5	88.1	98.5	69.6	75.8	84.0	98.5
NPV (%)	93.5	96.6	95.5	96.4	98.7	97.6	95.0
PPV (%)	31.7	46.5	95.5	25.3	33.9	40.4	87.5

AUC: area under the curve; 95%CI: 95% confidence interval; NPV: negative predictive value; PPV: positive predictive value

Fig. 3.

Diagnostic values of serum HMGB1 levels and SOFA score for predicting 28-day mortality in sepsis patients. ^*D4-HMGB1 vs. D4-HMGB1 + D4-SOFA (p = 0.90); ^#D7-HMGB1 vs. D7-HMGB1 + D7-SOFA (p = 0.09)

The optimal cut-off value for D4-HMGB1 was determined to be 6.4 ng/mL. Kaplan-Meier analysis demonstrated that patients with high D4-HMGB1 levels (≥ 6.4 ng/mL) had a significantly higher 28-day mortality than those with low levels (< 6.4 ng/mL; log-rank test, p = 0.001; Fig. 4). Furthermore, in subgroup analyses, this association remained significant in patients with comorbidities such as acute kidney injury, autoimmune diseases, and respiratory diseases (Supplementary Figure S4).

Fig. 4.

Values of serum HMGB1 levels on Day 4 for predicting 28-day all-cause mortality of sepsis patients

Discussion

Sepsis remains a major challenge in emergency medicine due to its high mortality and complex pathophysiology. Accurate and timely prognostic assessment is essential for optimizing clinical management. In this prospective study, we evaluated the dynamic changes of serum HMGB1 and SOFA scores in sepsis patients. Our findings demonstrate that serial monitoring of HMGB1, particularly the Day 4 (D4) concentration, provides robust predictive value for 28-day mortality, comparable to the D7-ΔSOFA score. Given the low mortality rate, the positive predictive value (PPV) is modest, while the negative predictive value (NPV) is high, suggesting that in this context, the marker might be more valuable for ruling out a high-risk outcome.

HMGB1 has long been recognized as a late-phase mediator of sepsis [28]. Unlike early cytokines such as TNF-α and IL-1β, which peak within hours, HMGB1 is released more slowly and remains elevated for a prolonged period. Our data confirmed this kinetic pattern, with serum HMGB1 levels peaking on Day 4 across the cohort. This delayed peak provides a wider therapeutic and diagnostic window, making it a potentially more reliable biomarker for monitoring disease progression than transient early-phase markers. This timing suggests HMGB1 may be involved in amplifying “second-hit” organ injury, which could potentially contribute to its different prognostic associations compared to some initial biomarkers in our study. Mechanistically, extracellular HMGB1 binds to Toll-like receptors (TLRs) and the receptor for advanced glycation end products (RAGE), thereby activating intracellular signaling pathways including p38 mitogen-activated protein kinase (MAPK), extracellular signal-regulated kinase 1/2 (ERK1/2), and NF-κB. This cascade promotes the secretion of transforming growth factor-beta 1 (TGF-β1) and platelet-derived growth factor (PDGF), exacerbating tissue damage. HMGB1 also activates the NLRP3 inflammasome, inducing gasdermin-D-mediated pyroptosis. This process, coupled with the subsequent release of pro-inflammatory cytokines that further stimulate HMGB1 secretion, establishes a persistent positive feedback loop [18, 21, 29–31].

In addition, this exploratory analysis found that the baseline levels of ALC, IL-6, IL-8, and IL-10 were associated with mortality risk and disease severity in sepsis patients. Correlation analysis showed weak but statistically significant positive associations between serum HMGB1 levels and SOFA scores, as well as with inflammatory cytokines including IL-6, IL-8, IL-10, IL-17 A, and TNF-α. These findings showed a link between HMGB1 and both organ dysfunction and immune dysregulation in sepsis. Collectively, these markers might illustrate the core feature of sepsis immunopathology—a dual state where inflammation and immunosuppression coexist [3].

The SOFA score is widely used to predict outcomes in critically ill patients; however, dynamic serial assessments, particularly the ΔSOFA, provides a more direct evaluation of prognosis than the initial score [5, 6]. In an international study [4], the rate of change in SOFA by Day 7 (D7-ΔSOFA) relative to the admission SOFA was an effective predictor of 28-day mortality (AUC 0.84, sensitivity 78%, specificity 80%). Mortality increased approximately 15-fold in patients with a reduction in admission SOFA of less than 25% by Day 7. Our results concur with previous studies in which D7-ΔSOFA demonstrated comparable prognostic performance in predicting 28-day mortality (AUC 0.89, sensitivity 87%, specificity 76%). This confirms that D7-ΔSOFA can identify high-risk patients and holds significant clinical value for prognostic assessment in sepsis. A review by Grooth et al. [7] similarly concluded that ΔSOFA, rather than a fixed-day SOFA, demonstrates a stable and persistent association with sepsis mortality, strongly recommending its prioritization as a surrogate endpoint in RCTs. However, the requirement to wait until Day 7 to determine D7-ΔSOFA represents a significant delay. Data from this observational study showed that serum HMGB1 levels on Day 4 had similar predictive performance for 28-day mortality to D7-ΔSOFA, but were measured earlier. Thus, our findings suggest that measuring HMGB1 on Day 4 could potentially provide prognostic information at an earlier time point. Recent research published in JAMA [8] introduced the SOFA-2 score, a modification of the original SOFA that includes new variables and revised thresholds to better characterize organ dysfunction and its associated mortality (SOFA-2 AUC, 0.79; 95% CI, 0.76–0.81 vs. SOFA-1 AUC, 0.77; 95% CI, 0.74–0.81). Sequential SOFA-2 assessments from ICU Day 1 to Day 7 maintained predictive validity. However, the SOFA-2 score did not incorporate immune dysfunction due to insufficient data and a lack of content validity. In contrast, this study explored, within an observational framework, the potential of combining HMGB1, an immune-related indicator, with the SOFA score, showing comparable prognostic performance.

In contrast to procalcitonin and C-reactive protein, which aid in early diagnosis but lack prognostic specificity beyond 72 h, the delayed peak of HMGB1 offers a unique temporal window for monitoring treatment response. Notably, its prognostic power appears to surpass that of lactate (AUC typically 0.63–0.73), as it reflects a direct link to cellular injury rather than hypoperfusion alone [12]. The Day 4 peak might also coincide with the transition of sepsis from hyperinflammation to immunosuppression. HMGB1 exacerbates organ injury through multiple mechanisms: it sustains NF-κB activation and induces pyroptosis, impairs mitochondrial function, and promotes endothelial glycocalyx shedding, thereby worsening capillary leak. This explains why non-survivors maintained higher HMGB1 levels through Day 7 compared to survivors, reflecting persistent tissue damage.

In this study, 15.9% of patients with sepsis had underlying autoimmune diseases. Despite this potential source of variability, the predictive accuracy of the Day 4 HMGB1 peak was retained. We also performed a stratified analysis by grouping patients based on the presence or absence of acute kidney injury (AKI) and then separately examining the association between HMGB1 levels and outcomes. With only 32 total death events, the number of death events in each subgroup is likely to be in the single digits. Kaplan-Meier analyses or statistical tests based on such limited data tend to produce unstable and potentially misleading results (whether false-positive or false-negative). Therefore, all subgroup analyses remain exploratory. We aim to preliminarily describe the behavior patterns of HMGB1 across different clinical contexts, thereby pointing out directions for future exploration in larger-scale studies. Although HMGB1 clearance may be influenced by renal function, our analysis demonstrated that HMGB1 may be a potential predictor of 28-day mortality even in the subgroup of patients with sepsis and AKI.

Our study has some limitations that must be taken into consideration. First, as an observational study, it is subject to potential residual confounding, and thus, we can only establish a statistical association with mortality rather than causality. Second, the fixed timing on Day 1, Day 4, and Day 7, rather than a continuous dynamic detection, might miss individual peak variation and affect the precision of assessing serum HMGB1, and the prognostic performance of ΔHMGB1 (the change in HMGB1 levels over time) was not evaluated. Additionally, elevated HMGB1 levels are not specific to infection and can occur in other conditions such as chronic inflammatory disorders [32], tissue damage [33], and neoplasia [34], while renal function was assessed using a crude binary measure; these factors may affect the precision of interpretation. Third, HMGB1 measurement is not yet standardized, is time-consuming and costly, and remains unsuitable for routine emergency department use at present. Its future clinical integration would require overcoming several challenges: prospective interventional studies to validate its utility, the development of rapid assays suitable for the ED environment, and cost-effectiveness analyses. Finally, the single-center design limits the generalizability of our findings, and external validation is required to confirm the accuracy of these results.

Conclusion

This prospective study demonstrates that serum HMGB1 levels, characterized by a distinct peak on Day 4, accurately predict 28-day mortality in sepsis patients, with performance comparable to the D7-ΔSOFA score. Our findings suggest that D4-HMGB1 could serve as a potential early warning tool. However, its future clinical integration would require overcoming several challenges: prospective interventional studies to validate its utility, the development of rapid assays suitable for the emergency department environment, and cost-effectiveness analyses.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

HMGB1

High-mobility group protein B1

SOFA

Sequential organ failure assessment score

ROC

Receiver operating characteristic

AUC

Area under the curve

CI

Confidence Interval

IL

Interleukin

TNF-α

Tumor Necrosis Factor-alpha

IL-17A

Interleukin-17α

ICU

Intensive Care Unit

JAMA

Journal of the American Medical Association

CRP

C-reactive protein

PCT

Procalcitonin

suPAR

Soluble urokinase plasminogen activator receptor

CD64

Fc gamma receptor I

TREM-1

Triggering Receptor Expressed on Myeloid cells-1

MCP-1

Monocyte Chemoattractant Protein-1

MR-proADM

Mid-regional pro-adrenomedullin

bio-ADM

Biologically Active Adrenomedullin

TLRs

Toll-like receptors

RAGE

Receptor for advanced glycation end products

ED

Emergency department

SCCM

Society of Critical Care Medicine

ESICM

European Society of Intensive Care Medicine

MV

Mechanical ventilation

AKI

Acute kidney injury

DIC

Disseminated intravascular coagulation

KDIGO

Kidney Disease: Improving Global Outcomes

CSF

Cerebrospinal fluid

CVC

Central venous catheter

PCR

Polymerase Chain Reaction

NGS

Next-generation sequencing

ELISA

Enzyme-linked immunosorbent assay

SD

Standard deviation

IQR

Interquartile range

PPV

Positive predictive value

NPV

Negative predictive value

BMI

Body Mass Index

ALC

Absolute Lymphocyte Count

Lac

Lactic acid

LPS

Lipopolysaccharide

DAMP

Damage-associated molecular pattern

MAPK

Mitogen-activated protein kinase

ERK1/2

Extracellular signal-regulated kinase 1/2

NLRP3

NOD-, LRR- and pyrin domain-containing protein 3

TGF-β1

Transforming growth factor-beta 1

PDGF

Platelet-derived growth factor

RCT

Randomized Controlled Trial

NF-κB

Nuclear factor kappa B

Author contributions

SQ and YJ contributed to methodology, statistical analysis, manuscript drafting, and revision; HS and FC participated in data collection and assembly, statistical analysis; CY participated in study design and manuscript review; QM and ZW contributed to the study conception, design, and coordination, and manuscript review. All authors read and approved the final manuscript.

Funding

The study was funded by the Shanghai Municipal Health Commission, China (202140459, 202240018).

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Ethical approval was obtained from the Institutional Review Board of Shanghai Jiao Tong University School of Medicine Affiliated Renji Hospital (Approval No: RA-2021-487). Also, informed consent was obtained from all participants. This study was carried out in accordance with the principles of the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.