Integrating serum ferritin and neutrophil-to-lymphocyte ratio with Sequential Organ Failure Assessment score improves mortality prediction in sepsis

Abstract

BACKGROUND

Sepsis causes significant mortality in patients. Typically, the Sequential Organ Failure Assessment (SOFA) score is used; however, recent studies have demonstrated improved predictive accuracy by using inflammatory biomarkers such as serum ferritin and the neutrophil-to-lymphocyte ratio (NLR). Serum ferritin, although traditionally associated with iron metabolism, also acts as an acute-phase reactant reflecting systemic inflammation.

AIM

To assess the prognostic value of integrating ferritin and NLR with SOFA in predicting mortality in critically ill patients with sepsis.

METHODS

Medical Information Mart for Intensive Care database-IV database was used to conduct this retrospective cohort study. Patients were divided into quartiles based on values of serum ferritin and NLR. Cox proportional hazards regression assessed the association with 30-day mortality, adjusted for age, sex, and the Charlson comorbidity index. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC), calibration plot, and net reclassification index (NRI). Bootstrapped internal validation was performed using 1000 resamples.

RESULTS

Patients in the ferritin Q1 quartile (lowest ferritin quartile) had 30% lower adjusted mortality risk (hazard ratio [HR]: 0.71, 95% confidence interval [CI]: 0.57-0.9; P = 0.005) compared to Q3 (highest ferritin quartile). Similarly, NLR Q1 (lowest NLR quartile) had 27% lower adjusted mortality risk (HR: 0.73, 95%CI: 0.59-0.91; P = 0.006). Moreover, patients with both serum ferritin and NLR in the Q1 (lowest) quartile had the lowest risk of mortality (HR: 0.56, 95%CI: 0.42-0.74; P < 0.001). With biomarker integration, the AUROC improved from 0.602 (95%CI: 0.574-0.630) for SOFA alone to 0.656 (0.629-0.683; P < 0.001), primarily driven by ferritin. NRI demonstrated a modest but significant improvement in reclassification. Old age was also found to be associated with a higher risk of mortality.

CONCLUSION

Lower ferritin and NLR are associated with reduced 30-day mortality, with ferritin markedly improving SOFA-based prediction and NLR offering minimal added benefit. Accessible biomarkers enhance early risk assessment in low-resource intensive care units.

Key Words: Sepsis; Ferritin; Neutrophil-lymphocyte ratio; Sequential Organ Failure Assessment score; Intensive care unit; Critical care; Systemic inflammation; Inflammatory biomarkers

Core Tip: This study evaluated the added prognostic value of combining serum ferritin and neutrophil-lymphocyte ratio with Sequential Organ Failure Assessment (SOFA) in predicting 30-day mortality among critically ill adult patients with sepsis. Using data from the Medical Information Mart for Intensive Care database, the integration of these inflammatory biomarkers with SOFA improved model discrimination and reclassification, supporting their potential role in enhancing early risk stratification in sepsis.

INTRODUCTION

Sepsis is a life-threatening condition characterized by organ dysfunction as a result of the dysregulated host response to infection[1]. Septic shock is a more severe form involving circulatory and metabolic abnormalities that significantly increases mortality risk[1]. Sepsis is manageable, but the prompt administration of targeted therapies highly influences clinical outcomes[2,3]. The World Health Organization has stressed the global burden of sepsis, emphasizing the importance of improving health information systems and high-quality care to alleviate its impact[4].

The Institute for Health Metrics and Evaluation 2020 report estimates that approximately 48.9 million global sepsis cases and 11 million sepsis-related deaths were reported in 2017, highlighting the major impact of sepsis on global morbidity and mortality[5].

The pathogenesis of sepsis includes activating immune cells like monocytes and neutrophils via the pathogen recognition receptors. This triggers the release of pro-inflammatory cytokines like interleukins (IL-1, IL-6) and tumor necrosis factor alpha, causing endothelial damage, increased capillary permeability, and microvascular thrombosis[6]. At the same time, an imbalanced immune state is created by the activation of anti-inflammatory responses, leading to immunosuppression and increased susceptibility to secondary infections[7]. Septic shock is the most severe form of sepsis, characterized by significant circulatory and metabolic abnormalities, including vasoplegia, impaired utilization of oxygen, and mitochondrial dysfunction, causing persistent hypotension and multiple organ failures despite adequate fluid resuscitation[8]. This combination of immune dysregulation and vascular pathology highlights the early recognition in patients with sepsis for improved outcomes.

While the Surviving Sepsis Campaign continues to refine its guidelines, the complex and often unpredictable clinical course of sepsis remains a major concern[8]. Early detection is important in improving outcomes, highlighting the need for effective and accessible biomarkers, not only in early diagnosis and prognostication but also to enable timely access to emerging adjunctive or experimental therapies. Adjunctive immunomodulatory treatments such as immunoglobulin M-enriched intravenous immunoglobulins and blood purification techniques are being explored, although supporting evidence for these interventions is still developing[9,10].

Over recent years, biomarkers such as procalcitonin and C-reactive protein have been extensively studied to assess their potential roles in sepsis diagnosis, prognosis, and guiding treatment decisions[7]. Ferritin, an iron-storage protein along with an acute phase reactant, plays an important role in the immune response of the host against infection[11,12]. Serum ferritin levels are elevated during systemic inflammation due to the release of inflammatory cytokines[13].

Another biomarker that has also gained attention for its association with inflammation is the neutrophil-to-lymphocyte ratio (NLR). It is increasingly being recognized as an indicator of the severity of the disease, with elevated NLR levels linked to poor outcomes in patients with sepsis[14-16].

Considering the individual roles of serum ferritin and NLR in sepsis, this study explored their combined prognostic significance in critically ill adult patients with sepsis. No study has specifically evaluated the combined role of serum ferritin and NLR with Sequential Organ Failure Assessment (SOFA) in sepsis prognosis, making this study the first to explore such a model. This study proposes a novel model using the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, which offers a cost-effective and practical solution to improving early risk stratification and guiding sepsis management.

MATERIALS AND METHODS

Study design and population

This retrospective study used data from 2008-2022 available in MIMIC-IV v3.1 from Beth Israel Deaconess Medical Centre (Boston, MA, United States)[17-19]. Adult patients (≥ 18 years) who were admitted to the intensive care unit (ICU) meeting Sepsis-3 criteria were included. Figure 1 shows the flow chart of patient inclusion and exclusion as per the Strengthening the Report of Observational Studies in Epidemiology recommendations.

Figure 1 Workflow of patient enrollment. ICU: Intensive care unit.

Inclusion and exclusion criteria

Inclusion criteria: Adult patients diagnosed with sepsis according to Sepsis 3.0 criteria[1].

Exclusion criteria: (1) Age < 18 years; (2) ICU admission for less than 24 hours; (3) Have missing ferritin or NLR or SOFA score components; or (4) > 5% missing data for all variables.

Data extraction

Data were collected using a structured query language from the MIMIC-IV database. Key variables like age, sex, Charlson comorbidity index (CCI), serum ferritin, absolute neutrophil count, absolute lymphocyte count, hospital admission time, ICU admission time, ICU discharge time, hospital discharge status, hospital discharge time, and SOFA score components, namely, mean arterial pressure, vasopressors, serum creatinine, urine output, platelets count, partial pressure of oxygen in arterial blood (PaO₂), fraction of inspired oxygen (FiO₂), invasive or non-invasive ventilation, total bilirubin, and Glasgow Coma Scale score were extracted from the database.

Calculation

SOFA score: The SOFA score was calculated based on the worst values recorded within the first 24 hours of ICU admission, as per Sepsis-3 definitions[1]. Standard thresholds for each organ system were applied to derive a total score ranging from 0 to 24. Component variables included PaO₂/FiO₂ ratio, platelet count, bilirubin, mean arterial pressure or vasopressor requirement, Glasgow Coma Scale, creatinine, or urine output as shown in Table 1.

Table 1 Sequential Organ Failure Assessment score.

Organ system	0	1	2	3	4
Respiratory (mmHg)	PaO2/FiO2 ≥ 400	PaO2/FiO2 < 400	PaO2/FiO2 < 300	PaO2/FiO2 < 200 and ventilated	PaO2/FiO2 < 100 and ventilated
Coagulation (platelets 103/μL)	Platelets ≥ 150	Platelets < 150	Platelets < 100	Platelets < 50	Platelets < 20
Liver (mg/dL)	Bilirubin < 1.2	Bilirubin 1.2-1.9	Bilirubin 2.0-5.9	Bilirubin 6.0-11.9	Bilirubin ≥ 12.0
Cardiovascular	MAP ≥ 70 mmHg	MAP < 70 mmHg	Dopamine < 5 or dobutamine (any dose)	Dopamine 51-15 or epinephrine ≤ 0.1 or norepinephrine ≤ 0.1	Dopamine > 15 or epinephrine > 0.1 or norepinephrine > 0.1
CNS GCS Score	15	13-14	10-12	6-9	< 6
Renal	Creatinine < 1.2 mg/dL	Creatinine 12-1.9 mg/dL	Creatinine 20-3.4 mg/dL	Creatinine 35-4.9 mg/dL; urine output < 500 mL/day	Creatinine > 5 mg/dL; urine output < 200 mL/day

Catecholamine doses = μg/kg/minute for at least 1 hour; CNS: Central Nervous System; FiO₂: Fraction of inspired oxygen; GCS: Glasgow Coma Scale; MAP: Mean arterial pressure; PaO₂: Partial pressure of oxygen in arterial blood.

CCI: The CCI was calculated using the International Statistical Classification of Diseases and Related Health Problems 10th Revision codes within the MIMIC-IV database using the algorithm developed by Quan et al[20] This index assigns weightage to various comorbid conditions, and a final score is calculated. The final score was used as a covariate.

NLR: NLR was calculated as follows. The blood sample values were taken within 24 hours of ICU admission. It was computed as the absolute neutrophil count divided by the absolute lymphocyte count. This marker has gained attention as a cost-effective and readily available indicator of systemic inflammation in critically ill patients.

Statistical analyses

Descriptive statistics were calculated for all variables, with continuous data presented as the means ± standard deviations. Categorical data are summarized as frequencies and percentages.

Comparisons between groups were conducted using the independent t-test for normally distributed continuous variables and Mann-Whitney U test for non-formally distributed tests. χ² or Fisher’s exact tests were used for categorical variables.

Survival outcomes were analyzed using Kaplan-Meier methodology, with differences between groups assessed via log-rank tests. Survival curves were presented using KMunicate-style plots to enhance interpretability, with 95% confidence intervals (CIs) calculated for each group. Cox proportional hazards models were constructed to evaluate the associations between key biomarkers (ferritin and NLR) and 30-day mortality. Both unadjusted and adjusted models were developed, with the latter accounting for potential confounders including age, sex, and CCI. The proportional hazards assumptions were tested using Schoenfeld residuals. Biomarker values were categorized into quartiles to facilitate clinical interpretation, with highest quartile serving as reference group. A combined analysis examining the synergistic effect of both biomarkers was also performed. Results are presented as hazards ratio (HRs) with 95%CIs. For all analyses, two-sided P < 0.05 was considered statistically significant. No imputation methods were employed for missing data; only patients with complete data for variables were included in analyses.

For model discrimination, we calculated area under the receiver operating characteristic curve (AUROC) for each model and compared AUROCs via DeLong tests using SOFA alone models as reference. Internal validation of predictive models was performed by bootstrap resampling with 1000 iterations to estimate and correct for optimism. Calibration was assessed with calibration plots (bootstrapped), and net reclassification improvement (NRI) was computed to quantify change in risk classification when adding biomarkers to the reference model.

RESULTS

Baseline characteristics

Table 2 comprehensively presents the demographic, clinical, and laboratory characteristics of the study cohort (n = 1954), stratified by 30-day survival status. Significant differences between survivors (n = 1426) and deaths (n = 528) were observed across multiple parameters. Patients who died were significantly older (67.7 years vs 59.4 years; P < 0.001) and had higher CCI scores (2.8 vs 2.0; P < 0.001). The prevalence of comorbidities including hypertension, chronic obstructive pulmonary disease, chronic kidney disease, chronic liver disease, and malignancy was significantly higher among patients who died.

Table 2 Baseline characteristics, n (%).

Characteristic	All patients (n = 1954)	Survivors (n = 1426)	Deaths (n = 528)	P value
Demographics
Age, years (mean ± SD)	61.6 ± 15.3	59.4 ± 15.2	67.7 ± 13.7	< 0.001
Sex				0.364
Male	1082 (55.4)	778 (54.6)	304 (57.6)
Female	872 (44.6)	648 (45.4)	224 (42.4)
BMI, kg/m2 (mean ± SD)	28.3 ± 7.9	28.5 ± 8.0	27.8 ± 7.7	0.107
Comorbidities
Charlson comorbidity index (mean ± SD)	2.2 ± 2.2	2.0 ± 2.1	2.8 ± 2.4	< 0.001
Hypertension	823 (42.1)	566 (39.7)	257 (48.7)	< 0.001
Diabetes mellitus	517 (26.5)	362 (25.4)	155 (29.4)	0.079
COPD	304 (15.6)	198 (13.9)	106 (20.1)	0.001
Chronic kidney disease	386 (19.8)	247 (17.3)	139 (26.3)	< 0.001
Chronic liver disease	215 (11.0)	137 (9.6)	78 (14.8)	0.001
Malignancy	293 (15.0)	177 (12.4)	116 (22.0)	< 0.001
Source of infection1				0.019
a: Pulmonary	809 (41.4)	565 (39.6)	244 (46.2)
b: Urinary	421 (21.5)	333 (23.4)	88 (16.7)
c: Abdominal	389 (19.9)	288 (20.2)	101 (19.1)
d: Bloodstream	215 (11.0)	147 (10.3)	68 (12.9)
e: Other/unknown	120 (6.1)	93 (6.5)	27 (5.1)
Clinical parameters at admission
Systolic BP, mmHg (mean ± SD)	118.7 ± 24.5	121.8 ± 22.9	110.4 ± 26.7	< 0.001
MAP, mmHg (mean ± SD)	78.5 ± 16.0	80.6 ± 14.9	73.0 ± 17.2	< 0.001
Heart rate, bpm (mean ± SD)	101.3 ± 21.8	100.1 ± 20.9	104.5 ± 23.7	< 0.001
Respiratory rate, breaths/min (mean ± SD)	22.8 ± 6.2	22.0 ± 5.8	24.9 ± 6.7	< 0.001
Temperature, °C (mean ± SD)	37.3 ± 1.1	37.4 ± 1.0	37.0 ± 1.2	< 0.001
SpO2, % (mean ± SD)	94.8 ± 4.7	95.4 ± 4.2	93.2 ± 5.5	< 0.001
Laboratory values
Ferritin (mean ± SD) (ng/mL)	1480.9 ± 2912.1	1196.4 ± 2423.8	2223.5 ± 3766.6	< 0.001
NLR (mean ± SD)	31.6 ± 95.3	24.7 ± 85.6	50.1 ± 114.4	< 0.001
Hemoglobin, g/dL (mean ± SD)	11.2 ± 2.3	11.4 ± 2.2	10.6 ± 2.3	< 0.001
WBC count, × 10³/μL (mean ± SD)	15.8 ± 11.2	15.2 ± 10.6	17.3 ± 12.4	< 0.001
Platelet count, × 10³/μL (mean ± SD)	217.3 ± 119.0	228.9 ± 119.8	186.8 ± 111.2	< 0.001
Creatinine, mg/dL (mean ± SD)	1.7 ± 1.5	1.5 ± 1.3	2.1 ± 1.8	< 0.001
Total bilirubin, mg/dL (mean ± SD)	1.4 ± 2.5	1.2 ± 2.1	2.0 ± 3.2	< 0.001
Lactate, mmol/L (mean ± SD)	2.9 ± 2.4	2.4 ± 1.8	4.1 ± 3.2	< 0.001
Interventions
Invasive ventilation	782 (40.0)	483 (33.9)	299 (56.6)	< 0.001
Vasopressor use	897 (45.9)	560 (39.3)	337 (63.8)	< 0.001
Dialysis	326 (16.7)	198 (13.9)	128 (24.2)	< 0.001
Outcomes
ICU stay duration, days (mean ± SD)	6.3 ± 7.8	5.9 ± 7.2	7.4 ± 9.1	0.001
Hospital stay duration, days (mean ± SD)	14.2 ± 13.4	15.8 ± 13.7	10.1 ± 11.5	< 0.001

¹P values represent overall comparison across categories using the χ² test.

BMI: Body mass index; BP: Blood pressure; COPD: Chronic obstructive pulmonary disease; ICU: Intensive care unit; IQR: Interquartile range; MAP: Mean arterial pressure; NLR: Neutrophil-lymphocyte ratio; SpO_2: Peripheral oxygen saturation; SD: Standard deviation; WBC: White blood cell.

Notably, the source of infection differed between groups (P = 0.019), with pulmonary infections more common among person who died (46.2% vs 39.6%) and urinary infections more frequent in survivors (23.4% vs 16.7%). Clinical parameters at admission revealed that patients who died presented with worse hemodynamic status, including lower systolic blood pressure (110.4 mmHg vs 121.8 mmHg; P < 0.001) and mean arterial pressure (73.0 mmHg vs 80.6 mmHg; P < 0.001), along with higher heart and respiratory rates.

Laboratory values showed striking differences in the biomarkers of interest. Median ferritin levels were more than twice as high in patients who died compared to those who survived (1136 ng/mL vs 526 ng/mL; P < 0.001), and median NLR was also substantially elevated (17.9 vs 10.7; P < 0.001). Other laboratory abnormalities in patients who died included lower hemoglobin and platelet counts and higher white blood cell count, creatinine, bilirubin, and lactate levels.

Patients who died required more intensive interventions, with significantly higher rates of invasive ventilation (56.6% vs 33.9%; P < 0.001), vasopressor use (63.8% vs 39.3%; P < 0.001), and dialysis (24.2% vs 13.9%; P < 0.001). Despite these interventions, patients who died had longer ICU stays (7.4 days vs 5.9 days; P = 0.001) but shorter overall hospital stays (10.1 days vs 15.8 days; P < 0.001), reflecting earlier mortality.

Distribution of patients and mortality by ferritin and NLR quartiles

Table 3 presents the distribution of patients across ferritin and NLR quartiles, along with associated mortality rates and median survival times. For ferritin, approximately half of the patients (49.9%) fell into the highest quartile (Q3, > 651 ng/mL), with the remainder evenly distributed between Q1 (< 277 ng/mL, 25.0%) and Q2 (277-651 ng/mL, 25.1%). Both in-hospital and 30-day mortality rates were substantially higher in the highest quartile (29.7% and 33.2%, respectively) compared to Q1 (17.8% and 21.3%) and Q2 (18.6% and 20.4%).

Table 3 Distribution of patients and mortality by ferritin and neutrophil-lymphocyte ratio quartiles, n (%).

Category	Total (n = 1954)	In-hospital mortality	30-day mortality	Median survival, days (95%CI)
Ferritin quartiles
Q1 (< 277 ng/mL)	488 (25.0)	87 (17.8)	104 (21.3)	87.8 (40.6-NR)
Q2 (277-651 ng/mL)	490 (25.1)	91 (18.6)	100 (20.4)	89.8 (58.9-NR)
Q3 (> 651 ng/mL)	976 (49.9)	290 (29.7)	324 (33.2)	44.4 (36.0-58.4)
NLR quartiles
Q1 (< 6.8)	484 (24.8)	78 (16.1)	94 (19.4)	92.3 (53.7-NR)
Q2 (6.8-12.4)	504 (25.8)	87 (17.3)	98 (19.4)	91.4 (50.3-NR)
Q3 (> 12.4)	966 (49.4)	303 (31.4)	336 (34.8)	41.7 (32.8-55.2)
Combined ferritin & NLR
Both in Q1	225 (11.5)	23 (10.2)	29 (12.9)	NR (78.6-NR)
Both in Q2	247 (12.6)	36 (14.6)	42 (17.0)	98.7 (62.4-NR)
Mixed quartiles	755 (38.6)	173 (22.9)	192 (25.4)	65.3 (48.9-85.6)
Both in Q3	727 (37.2)	236 (32.5)	265 (36.5)	37.2 (29.4-46.8)

CI: Confidence interval; NLR: Neutrophil-lymphocyte ratio; NR: Not reported.

Similarly, for NLR, approximately half of the patients (49.4%) were in the highest quartile (Q3, > 12.4), with the remainder split between Q1 (< 6.8, 24.8%) and Q2 (6.8-12.4, 25.8%). Mortality rates were markedly elevated in the highest NLR quartile (31.4% in-hospital and 34.8% 30-day mortality) compared to Q1 (16.1% and 19.4%) and Q2 (17.3% and 19.4%).

The combined analysis of both biomarkers revealed a powerful synergistic effect. Patients with both biomarkers in the highest quartile (37.2% of the cohort) had the worst outcomes, with 32.5% in-hospital mortality, 36.5% 30-day mortality, and median survival of only 37.2 days. By contrast, those with both biomarkers in the lowest quartile (11.5% of patients) showed excellent outcomes with only 10.2% in-hospital mortality, 12.9% 30-day mortality, and median survival not reached within the study period. Patients with mixed quartile distributions (38.6%) or both biomarkers in the middle quartile (12.6%) showed intermediate outcomes, demonstrating a clear dose-response relationship.

Cox proportional hazards analysis

Table 4 presents the results of univariable and multivariable Cox proportional hazards analyses for 30-day mortality. Using the highest quartile (Q3) as the reference group for both biomarkers, patients in lower ferritin quartiles showed significantly reduced hazard of death. In unadjusted analysis, the hazard ratios were 0.58 (95%CI: 0.47-0.73; P < 0.001) for Q1 and 0.59 (95%CI: 0.47-0.74; P < 0.001) for Q2. After adjustment for confounders, these protective effects remained significant but slightly attenuated (HR: 0.71, 95%CI: 0.57-0.90, P = 0.005 for Q1 and HR 0.70, 95%CI: 0.55-0.88, P = 0.002 for Q2).

Table 4 Cox proportional analysis.

Variable	Unadjusted HR (95%CI)	P value	Adjusted HR (95%CI)1	P value
Ferritin quartiles
Q3 (highest)	Reference	-	Reference	-
Q1 (lowest)	0.58 (0.47-0.73)	< 0.001	0.71 (0.57-0.90)	0.005
Q2 (middle)	0.59 (0.47-0.74)	< 0.001	0.70 (0.55-0.88)	0.002
NLR quartiles
Q3 (highest)	Reference	-	Reference	-
Q1 (lowest)	0.65 (0.52-0.80)	< 0.001	0.73 (0.59-0.91)	0.006
Q2 (middle)	0.60 (0.48-0.74)	< 0.001	0.63 (0.50-0.78)	< 0.001
Combined ferritin & NLR
Both in Q3 (highest)	Reference	-	Reference	-
Both in Q1 (lowest)	0.42 (0.32-0.55)	< 0.001	0.56 (0.42-0.74)	< 0.001
Mixed quartiles	0.65 (0.53-0.80)	< 0.001	0.74 (0.60-0.91)	0.005
Charlson comorbidity index
Per 1-point increase	1.03 (1.00-1.07)	0.048	1.03 (0.99-1.06)	0.161
Age
Per 1-year increase	1.01 (1.01-1.02)	< 0.001	1.02 (1.01-1.02)	< 0.001
Sex
Male	Reference	-	Reference	-
Female	1.08 (0.91-1.29)	0.364	1.11 (0.93-1.32)	0.258
Continuous variables
NLR (per unit)	1.00 (1.00-1.00)	0.072	1.00 (1.00-1.00)	0.908
Ferritin (per unit)	1.00 (1.00-1.00)	< 0.001	1.00 (1.00-1.00)	< 0.001

¹Adjusted model includes sex, age, neutrophil-lymphocyte ratio quartiles, ferritin quartiles, and Charlson comorbidity index.

CI: Confidence interval; HR: Hazard ratio; NLR: Neutrophil-lymphocyte ratio.

Similarly, lower NLR quartiles showed reduced mortality risk. Unadjusted hazard ratios were 0.65 (95%CI: 0.52-0.80; P < 0.001) for Q1 and 0.60 (95%CI: 0.48-0.74; P < 0.001) for Q2. After adjustment, these remained significant (HR: 0.73, 95%CI: 0.59-0.91, P = 0.006 for Q1) and (HR: 0.63, 95%CI: 0.50-0.78, P < 0.001 for Q2).

The combined analysis of both biomarkers provided the most striking results. Compared to patients with both biomarkers in the highest quartile, those with both biomarkers in the lowest quartile had a 58% reduction in mortality risk in unadjusted analysis (HR: 0.42, 95%CI: 0.32-0.55; P < 0.001) and a 44% reduction after adjustment (HR: 0.56, 95%CI: 0.42-0.74; P < 0.001). Patients with mixed quartile distributions showed intermediate risk reduction (adjusted HR: 0.74, 95%CI: 0.60-0.91; P = 0.005).

Older age significantly increased risk of mortality (adjusted HR: 1.02 per year, 95%CI: 1.01-1.02; P < 0.001), while sex showed no significant risk. The CCI showed a trend toward increased risk in unadjusted analysis (HR: 1.03 per point, 95%CI: 1.00-1.07; P = 0.048) but was not significant after adjustment (P = 0.161).

When analyzed as continuous variables, both ferritin and NLR showed statistically significant but clinically minimal per-unit effects, underscoring that the quartile-based approach better captures the clinical relevance of these biomarkers.

Survival curves

Kaplan-Meier survival curves of ferritin quartiles: Figure 2A presents Kaplan-Meier survival curves stratified by ferritin categories over time. The graph demonstrates a clear separation between survival probabilities across different ferritin levels. Patients in the high ferritin category (> 651 ng/mL) show markedly reduced survival compared to those in the low (< 277 ng/mL) and mid-range (277-651 ng/mL) ferritin categories. The median survival for patients in the high ferritin group was only 44.4 days (95%CI: 36.0-58.4), whereas neither the low nor the mid-range ferritin groups reached a median survival within the 60-day follow-up (lower 95%CI bounds 87.8 and 89.8 days, respectively). This visualization provides strong evidence that elevated ferritin levels are associated with worse survival outcomes in sepsis patients. The log-rank also reported a highly significant association between survival and ferritin values with P < 0.001.

Figure 2 KMunicate style plot. A: Ferritin quartiles; B: Neutrophil-to-lymphocyte ratio quartiles. NLR: Neutrophil-to-lymphocyte ratio; ICU: Intensive care unit.

Kaplan-Meier survival curves of neutrophil-to-lymphocyte quartiles: Figure 2B displays Kaplan-Meier survival curves stratified by NLR quartiles. Visually, the highest NLR quartile (Q3, > 12.4) demonstrates notably worse survival compared to the lower quartiles (Q1, < 6.8 and Q2, 6.8-12.4). Patients in the highest NLR quartile exhibited a median survival of 41.7 days (95%CI: 32.8-55.2). By contrast, the lower quartiles showed considerably better survival, with median survival estimates of 92.3 days for Q1 (< 6.8) and 91.4 days for Q2 (6.8-12.4). This distinct stratification observed in the Kaplan-Meier curves underscores the potential of NLR as a valuable prognostic marker in sepsis. The log-rank test comparing survival across NLR groups yielded a borderline non-significant P value of 0.063, suggesting a trend toward differential survival that did not reach conventional statistical significance.

Performance of model

AUROC: We assessed the discriminative ability of the Sequential Organ Failure Assessment (SOFA) score alone and in combination with serum ferritin and NLR for its association with 30-day mortality in adult patients with sepsis (Figure 3 and Table 5). The AUROC for SOFA alone was 0.602, indicating modest discriminative performance and the optimism-corrected area under the curve (AUC) was 0.595.

Figure 3 Area under the receiver operating characteristic curve curves showing apparent and corrected area under the receiver operating characteristic curve. AUC: Area under the curve; AUROC: Area under the receiver operating characteristic curve; SOFA: Sequential Organ Failure Assessment.

Table 5 Area under operating characteristic curve for all predictive models.

Model	Apparent AUROC (95%CI)	Corrected AUROC	P value
SOFA alone	0.602 (0.574-0.630)	0.595	Reference
SOFA + ferritin	0.655 (0.627-0.628)	0.649	< 0.0001
SOFA + ferritin + NLR	0.656 (0.629-0.683)	0.649	< 0.0001

AUROC: Area under operating characteristic curve; CI: Confidence interval; NLR: Neutrophil-lymphocyte ratio; SOFA: Sequential Organ Failure Assessment.

The combination of SOFA and serum ferritin showed a notable increase in predictive performance, with an AUC of 0.655 (95%CI: 0.627-0.682; P < 0.0001). After internal validation using bootstrap resampling, the optimism-corrected AUC was 0.648, indicating good model robustness and discrimination. The highest discrimination was observed in the model incorporating all three variables—SOFA, ferritin, and NLR—which achieved a highly significant AUC of 0.656 (95%CI: 0.629-0.683; P < 0.001). An optimism-corrected AUC of 0.650 was achieved after internal validation, indicating robust model discrimination.

These results suggest that serum ferritin contributes more significantly to risk stratification than NLR and that integrating both biomarkers with SOFA provides a modest improvement over SOFA alone. However, the overall discriminative performance of all models remained modest.

Calibration plot: The SOFA alone model (Figure 4A) showed good calibration with mean absolute error (MAE) of 0.011, indicating acceptable predictive accuracy. The addition of ferritin to SOFA model (Figure 4B) improved calibration even closer to the ideal curve throughout the probability range compared to SOFA alone model. The MAE was lower at 0.008, suggesting enhanced predictive accuracy. The SOFA + ferritin + NLR model (Figure 4C) showed increased deviation from ideal line at higher predicted probabilities, despite reasonable alignment in lower risk range. However, the MAE increased to 0.021, indicating slightly reduced calibration compared to the other two models.

Figure 4 Calibration plot for Sequential Organ Failure Assessment model, Sequential Organ Failure Assessment plus ferritin model, and Sequential Organ Failure Assessment plus neutrophil-to-lymphocyte ratio plus ferritin model. A: This figure indicates that calibration is near the ideal line with mean absolute error of 0.011 in a cohort of 1.954 patients with 1000 resamples; B: This figure indicates that calibration is near the ideal line with mean absolute error of 0.008 in a cohort of 1.954 patients with 1000 resamples; C: This figure indicates that calibration is near the ideal line but slightly deviated in high-risk patients with mean absolute error of 0.021 in a cohort of 1.954 patients with 1000 resamples. NLR: Neutrophil-to-lymphocyte ratio; SOFA: Sequential Organ Failure Assessment.

NRI: The category-based NRI was calculated to assess the incremental prognostic value of ferritin and NLR in addition to SOFA. Compared to SOFA model, the addition of ferritin resulted in a statistically significant overall NRI of 0.128 (95%CI: 0.080-0.179) with NRI^- (0.128; 95%CI: 0.080-0.179) and NRI⁺ was 0.000. The results indicate better classification of survivors than identifying deaths. The result of combined model (SOFA + ferritin + NLR) was better than SOFA + ferritin with NRI of 0.131. This suggests that our model offers improved downward classification of low-risk patients but limited value in high-risk patients. Also, addition of NLR provides slight net benefit.

DISCUSSION

In this retrospective study of adult patients admitted to the ICU with sepsis, we observed that combining serum ferritin, NLR, and the SOFA score provided significant predictive performance of 30-day mortality compared to using the SOFA score alone. The findings of this study underline the potential additive value of iron metabolism and inflammatory markers in influencing sepsis outcomes.

The SOFA score has demonstrated solid predictive value for mortality and has been validated by numerous studies[21-23]. However, the sole reliance of the SOFA score on physiological parameters may exclude the influence of immunological dysregulation and systemic inflammatory burden, which are the central features of sepsis pathophysiology[24]. Our results suggest that the inclusion of immune and inflammatory status markers, such as NLR and serum ferritin, adds prognostic granularity beyond the SOFA score.

Many previous studies have evaluated the prognostic relevance of serum ferritin and NLR separately[25-28]. Elevated NLR has been associated with increased mortality in critically ill patients with sepsis[28]. This result reflects neutrophil-driven inflammation and relative lymphopenia, which are the hallmarks of sepsis-induced immune dysfunction[29]. Similarly, beyond its role in iron metabolism, serum ferritin acts as an acute-phase reactant and may reflect hyperinflammation or macrophage activation syndrome[30]. Prior studies have analyzed serum ferritin as an independent predictor of sepsis severity and mortality, which is consistent with our study[25,26,31]. Mechanistically, sepsis-induced cytokine release and endotoxemia stimulate ferritin synthesis, reflecting macrophage activation in iron homeostasis. High ferritin levels may exacerbate tissue injury via iron-mediated oxidative stress and ferroptosis, and may contribute to immunosuppression through dysregulated iron metabolism[30]. Moreover, ferritin-driven pathways have been implicated in promoting neutrophil extracellular trap formation, further amplifying organ dysfunction in sepsis-associated injury[32]. These insights support our quartile-based approach and underscores the importance of defining clinically meaningful ferritin thresholds.

In our cohort, patients who died were older and had higher CCI than survivors, consistent with prior studies indicating older age and higher comorbidity burden as strong predictor of sepsis mortality[33,34]. The observed differences in infection source, with pulmonary infections more prevalent among patients who died, aligns with studies showing respiratory foci confers poor outcomes compared to other sources[35]. Hemodynamic parameters like lower blood pressure and higher heart and respiratory rates among patients who died aligning with previously studies showing increased mortality risk[36]. The greater need for invasive interventions (ventilation, vasopressors, dialysis) among dead patients reflects organ dysfunction at baseline, aligning with Sepsis-3 emphasis on organ failure severity for prognostication[37]. Also, the longer ICU but shorter hospital stays in dead patients likely reflects early in-hospital mortality.

Recent literature shows an increased trend of developing multi-dimensional risk models incorporating laboratory biomarkers and clinical scores, which aligns with our study[38-40]. Individual quartile-based comparisons showed statistically significant hazard reductions in lower serum ferritin and NLR groups. However, the most notable outcome came from the combined lowest quartile groups, indicating synergistic power.

Our study’s combined model demonstrates significantly improved predictive performance incorporating serum ferritin, NLR, and SOFA over SOFA. The AUROC curve for SOFA score alone was 0.602. An AUROC of 0.655 was observed when ferritin was integrated with SOFA and was highly significant even after optimism-correction. The AUROC of 0.656 was observed with a combined model of SOFA, ferritin, and NLR, indicating improved accuracy in prediction. The observed performance of model was also highly significant over SOFA alone, although the clinical utility remains modest. The findings of this study indicate that even small improvements can be valuable despite the modest AUROC. It is important to acknowledge, however, that the incremental benefit of adding NLR to the SOFA + ferritin model was limited, with AUROC improving only slightly from 0.655 to 0.656. This suggests that ferritin may be primary driver of enhanced model performance, and while NLR remains biologically plausible as a marker of immune dysregulation. This limitation underscores the need for future research to evaluate the contextual utility of NLR and explore other inflammatory or immune-based markers that may yield greater performance gains. Nonetheless, our work provides strong support for the idea that a multifactorial approach that includes both organ failure and inflammatory measures can enhance death prediction. A study by Xie et al[41] showed comparable additive effects in sepsis models, which employed combinations of interleukins, lactate, and procalcitonin, which has further validated our biomarker combination approach.

The calibration analyses also demonstrated that SOFA alone had good agreement between predicted and observed risk across most probability range with MAE of 0.011. Adding ferritin improved calibration by reducing MAE to 0.008 and bringing predictions closer to ideal line. Although, the SOFA + ferritin + NLR model remained well calibrated at lower risk, it deviated more from the ideal line at higher predicted probabilities, increasing MAE to 0.021. This indicates that while ferritin enhances calibration over SOFA alone, inclusion of NLR may introduce modest miscalibration in patients who are high risk.

The net reclassification improvement (NRI) results were also demonstrated using our combined model to improve risk stratification. The value of 0.131 indicated a modest but significant reclassification improvement over the SOFA score alone, demonstrating the better ability to accurately stratify patients into lower risk categories but limited value for high-risk patients. This finding is consistent with other studies that integrate biomarkers for risk predictions, where NRI is used to assess the integration of new predictors in clinical models[42].

Many prior studies have used the MIMIC database for developing prognostic models[39,43]. For instance, a model developed by Zhang et al[43] predicting 28-day mortality in patients with sepsis or sepsis associated delirium achieved an AUC of 0.91 in development and AUC of 0.83 in validation cohorts. A study by Yang et al[39] identified SOFA, age and parameters like anion gap, International Normalized Ratio, hemoglobin A1c among others as independent predictors in patients with sepsis and diabetes mellitus. Although these models showed better performance, these models require extensive data inputs, which may limit applicability in low-resource settings. Our study instead focuses on a simpler logistic model adding only two readily available biomarkers, serum ferritin and NLR, to SOFA and basic covariates (age, sex, CCI). Although the resulting AUC is modest compared to other sepsis models, our model offers a balance between improved performance over SOFA alone and feasibility in resource-limited settings.

Our study contributes to the limited but growing body of literature that uses a large critical care database to assess the combined impact of easily accessible serum ferritin and NLR in the ICU setting. This is among the first studies that use the MIMIC-IV database to evaluate such a simple model involving both NLR and ferritin. The detailed, time-stamped laboratory data of patients in the ICU makes the MIMIC-IV database ideal for this study.

There are also applications for our findings. Our proposed approach can be widely used, particularly in settings with low resources, because serum ferritin and NLR are both easily accessible and reasonably priced laboratory parameters. Moreover, this model might also help with early risk assessment and well-informed clinical judgment. Our model's integration into ICU triage tools or electronic medical records may provide physicians with real-time risk estimation, which would reduce the need for expensive diagnostics to provide precision care. Moreover, it can also serve as a screening tool for early inclusion in clinical trials in sepsis targeting immune modulation. Our model can also serve as a real-time alert within the ICU dashboard, potentially guiding decisions for at-risk sepsis patients.

The large sample size, rigorous exclusion criteria, and well-defined outcomes strengthen our study. This study and limitations. First, there is a need for external validation of our study. Second, there was also potential residual confounding due to the study's retrospective nature. Third, there may be some bias due to the proportion of data excluded for missing variables because of rigorous exclusion. Also, due to lack of universally accepted clinical thresholds, we stratified ferritin and NLR by quartiles. The performance of our model was modest, but it offers a balance between performance and feasibility in resource-limited settings.

In this work, we used Cox proportional hazards regression, which yields reliable time-to-event analysis, corrected for age, sex, and the CCI, among other possible confounders. Future work should test for time-varying covariate effects to verify proportionality assumptions. In addition, the lack of dynamic biomarker tracking may limit the ability to assess the prognostic value of serial serum ferritin or NLR measurements, which may change during ICU stay. Future research should also focus on integrating dynamic time-varying parameters, including the trajectory of serum ferritin and NLR over time, and validating our model in different cohorts. Additionally, prospective studies should incorporate immunological profiling, which may uncover insights that explain the prognostic relationships observed here.

CONCLUSION

In this retrospective cohort of critically ill patients, lower levels of serum ferritin and NLR showed significantly reduced 30-day mortality risk, with the combination of both markers in the lowest quartile conferring the greatest survival benefit. Integrating ferritin and NLR with the SOFA score modestly improved performance of model, with this improvement predominantly driven by ferritin, suggesting that it may be a more robust inflammatory marker for mortality risk stratification than NLR. Given the routine availability and low cost of ferritin and complete blood count (for NLR), this biomarker-based approach may serve as a practical and accessible tool to support early risk assessment and triage, particularly in resource-limited critical care settings. Incorporating these markers into routine evaluation may help guide timely interventions and prioritize care for patients who are at high risk, ultimately contributing to improved outcomes in vulnerable populations.