Introduction

Sepsis is described as a severe infectious disease characterized by an abnormal inflammatory response of the body to an infection, leading to tissue damage and organ dysfunction [1]. Despite advancements in sepsis research in recent years, it remains a significant cause of death among hospitalized patients, imposing a tremendous economic burden on patients' families [2]. Recent studies have reported alarming mortality rates for sepsis in the United States, with the annual rate among hospitalized patients as high as 26.7%, and reaching a staggering 41.9% in the intensive care unit [3]. Therefore, early detection and treatment of sepsis are crucial for improving patients' prognosis. Currently, we use scoring tools such as qSOFA, SOFA, and APACHE to assess the severity of sepsis in patients [4]. However, these scores require the collection of multiple indicators, making their clinical application less convenient. Consequently, there is an urgent need to identify a simple, fast, and cost-effective predictive indicator to evaluate the severity of sepsis.

Ferritin is a widely present protein in plasma, cells, and tissue organs [5]. Studies indicate its crucial role in infection and inflammation processes [6, 7]. As a result, serum ferritin level is often considered a significant indicator for infectious and inflammatory diseases. An increase in ferritin levels suggests the presence of inflammation or the severity of a disease. Moreover, research has found associations between abnormal serum ferritin levels and conditions, such as digestive disorders [8], Liver disease [9, 10] and COVID-19 [11]. However, it is worth noting that ferritin levels are influenced by various factors, including inflammation, liver function, and nutritional status. Therefore, relying solely on ferritin levels may not provide reliable results. On the other hand, serum albumin, a negative acute-phase reactant protein, is closely linked to the degree of inflammation, disease prognosis, and mortality rates [12]. Current studies demonstrate a strong correlation between serum albumin and mortality and prognosis in conditions, such as sepsis [13], tumors [14], and kidney diseases [15]. It is important to consider that patients' nutritional status can also impact albumin levels. Therefore, using albumin levels alone to predict disease prognosis may have certain limitations.

Currently, there have been studies indicating the association between ferritin/albumin and the prognosis of COVID-19 patients [16]. Nevertheless, the relationship between ferritin/albumin levels and mortality rates in patients with sepsis is still not clear. Therefore, we assessed the relationship between ferritin/albumin and the 28-day all-cause mortality rate in patients with sepsis based on the MIMIC-IV database. In accordance with the STROBE guidelines, we present the following article.

Methods

Database introduction

The data for this study was obtained from the MIMIC-IV (v2.0) database, a publicly available database for medical research which contains inpatient data from Beth Israel Deaconess Medical Center in Boston, USA, from 2008 to 2019. MIMIC-IV 2.0 contains various types of patient information, including demographic data, physiological characteristics, diagnoses, surgeries, medication prescriptions, laboratory test results, and vital signs. To gain access to and make effective use of the database for this study, Zhengting Liu, the corresponding author, successfully completed the Collaborative Institutional Training Initiative (CITI) program, specifically passing exams on "Conflicts of Interest" and "Data or Specimen Only Research" (ID: 48255890). Consequently, the research team is now authorized and eligible to access and extract the necessary data from the database for their study.

Population selection criteria

According to the MIMIC-IV database, a total of 73,181 patients were admitted to the ICU between 2008 and 2019, of which 50,920 were first-time hospitalizations and first-time ICU stays. Among them, 24,390 met the definition of Sepsis-3, which is defined as an increase of ≥ 2 points in Sequential Organ Failure Assessment (SOFA) score plus documented or suspected infection. After a thorough screening process, patients who fail to meet the following criteria will be excluded: (1) patients who are under the age of 18; (2) patients who have had less than 24 h of stay in the ICU; (3) patients with missing death time data; and (4) patients who are not recorded for serum ferritin and albumin within 24 h of hospitalization. This study ultimately included a cohort of 1553 patients (Fig. 1).

Fig. 1
figure 1

Flowchart of study patients

Full size image

Data extraction

The FAR is the primary variable under investigation. To reduce the influence of subsequent treatments on serum ferritin and albumin, we opted to measure these factors for the first time after patients were admitted. Furthermore, we identified and controlled for several other potential confounding factors, including demographic variables (age and gender), vital signs (heart rate, respiratory rate, systolic and diastolic blood pressure, mean arterial pressure, temperature, and oxygen saturation), laboratory indicators (albumin, ferritin, anion gap, bicarbonate, calcium, chloride, sodium, potassium, hematocrit, hemoglobin, red blood cells, erythrocyte distribution width, glucose, blood urea nitrogen, creatinine, platelets, and white blood cells), comorbidities (myocardial infarction, congestive heart failure, chronic pulmonary disease, diabetes, renal disease, malignant cancer, severe liver disease, and metastatic solid tumor), and sequential organ failure assessment (SOFA). Data extraction was performed using structured query language (SQL) queries run on PostgresSQL software (v13.7.1) and Navicat Premium software (version 15) 0.2.4 Grouping and endpoint events.

Grouping and endpoint events

In this study, patients were stratified into two groups: a 28-day survival group and a 28-day death group, based on their mortality status within the specified timeframe. The primary endpoint of the study was defined as mortality from any cause within 28 days.

Management of missing data

To minimize bias, any variable with missing data greater than 15% (such as aspartate aminotransferase, lactate dehydrogenase, lactate, C-reactive protein, and coagulation indicators) was excluded from the study. For variables with less than 5% missing data, including mean arterial pressure, heart rate, oxygen saturation, temperature, calcium ion, hematocrit, hemoglobin, and red blood cells, the missing values were replaced with the mean value of the variable. Platelets and white blood cells had less than 5% missing data and were replaced with the median value of the variable.

Statistical analysis

In this study, normally distributed continuous variables were presented as mean ± standard deviation, while non-normally distributed continuous variables were represented using the median or interquartile range. Categorical variables, on the other hand, were expressed as percentages. To analyze the baseline characteristics, t tests or one-way analysis of variance (ANOVA) were used for continuous variables, while Chi-square tests were employed for categorical variables. Univariate COX regression analysis was conducted to identify potential risk factors for 28-day mortality in sepsis patients, specifically considering variables with a multivariate COX regression analysis P value of less than 0.001. Moreover, ROC curve analysis was carried out to evaluate and compare the prognostic performance of serum ferritin, serum albumin, the ferritin/albumin ratio (FAR), and Sequential Organ Failure Assessment (SOFA) score in predicting 28-day mortality among sepsis patients. Sensitivity and specificity measures were computed for each indicator. The Youden index was used to determine the optimal cutoff value for FAR, and patients were separated into high and low FAR groups for survival rate comparisons using Kaplan–Meier survival analysis. Finally, subgroup analysis was performed to determine if an interaction effect of FAR existed in different subgroups. Statistical analyses were conducted using R 4.1.1 software and Freestat software version 1.8. The significance level was set at P < 0.05 for all tests, using a two-tailed approach.

Results

Baseline patient demographics

Table 1 displays the baseline characteristics of sepsis patients in both the 28-day survival and non-survival groups. This study included a total of 1553 patients who met the inclusion and exclusion criteria, consisting of 687 females and 866 males, with an average age of 66.8 ± 15.1 years. The 28-day mortality rate for sepsis patients was 37.34%. Non-survivors had significantly lower hospital length of stay, SBP, DBP, MAP, and SpO2, and higher RR and SOFA scores than survivors (P < 0.05). In addition, laboratory indices such as FAR, ferritin, AG, RDW, BUN, CREA, and WBC were higher in the non-survival group, whereas ALB, bicarbonate, chloride, and PLT were lower (P < 0.05). No statistically significant differences were found between the two groups for the remaining covariates (P > 0.05). Among the sepsis patients, 28.3%, 20.7%, 13.1%, and 8.8% had nephropathy, severe liver disease, tumor, and diabetes mellitus, respectively.

Table 1 Baseline characteristics between survivors and non-survivors
Full size table

The LAR is an independent risk factor for 28-day mortality in patients with sepsis

Our univariate Cox regression analysis revealed a significant association between FAR (log2) and 28-day mortality in sepsis patients (HR: 1.17, 95% CI 1.13–1.22, P < 0.001), as shown in Table 2. Subsequently, we included covariates with P < 0.001 from Table 2 in our multivariate Cox regression analysis to examine the independent effect of FAR on 28-day mortality in sepsis patients. We constructed four different multivariate Cox regression models, and the hazard ratios (HR) and 95% confidence intervals (CI) are presented in Table 3.

Table 2 Univariate Cox analysis of risk factors for death within 28 days in patients
Full size table
Table 3 Multivariate Cox regression analysis of risk factors for death within 28 days in patients
Full size table

In the minimally adjusted model (Model I), we observed a persistent significant association between FAR and 28-day mortality in sepsis patients (HR: 1.18, 95% CI 1.13–1.23, P < 0.001). Even after full adjustment for covariates (Model III), FAR was still an independent risk factor. To further assess the robustness of our findings, we converted the continuous variable FAR to a dichotomous variable based on the FAR threshold derived from the ROC curve in Fig. 2. Using a cutoff point of FAR < 364.2215 as the reference baseline, we obtained similar conclusions, indicating that FAR is an independent risk factor for 28-day death in sepsis patients.

Fig. 2
figure 2

ROC curve analysis

Full size image

ROC curve analysis and survival curve analysis

We performed an ROC curve analysis on four indicators, namely, FAR, ferritin, albumin, and SOFA score, to predict the 28-day mortality rate in sepsis patients. Table 4 shows the data from Fig. 2. The results revealed that FAR had a larger AUC (61.0141%) than ferritin (60.4778%), albumin (55.5618%), and even the SOFA score (56.9719%), indicating a significant predictive advantage of FAR.

Table 4 Information of ROC curves in Fig. 2
Full size table

Furthermore, we have determined the optimal threshold for FAR to be 364.2215, enabling the categorization of sepsis patients into high FAR group (FAR ≥ 364.2215) and low FAR group (FAR < 364.2215). Subsequently, as shown in Fig. 3, survival curves were constructed to compare the survival rates between the two groups of patients, revealing a significantly higher mortality rate in the high FAR group compared to the low FAR group (P < 0.001).

Fig. 3
figure 3

Kaplan–Meier survival analysis curves for 28-day mortality in patients with sepsis

Full size image

Subgroup analysis

The forest plot (Fig. 4) demonstrated that there was no observed interaction between FAR and any of the subgroups when stratified by variables, such as age, gender, diabetes, congestive heart failure, liver disease, kidney disease, chronic pulmonary disease, and malignant cancer (P > 0.05). These findings strongly suggest that, regardless of these demographic and clinical factors, FAR remains an independent risk factor for 28-day mortality in sepsis patients.

Fig. 4
figure 4

Subgroup analysis forest plot depicting the relationship between 28-day mortality rate and FAR in sepsis patients

Full size image

Discussion

The findings of this retrospective study strongly indicate that FAR is a significant independent factor associated with a 28-day all-cause mortality in sepsis patients and serves as a reliable predictor for predicting 28-day mortality in sepsis patients. The analysis of the ROC curve demonstrates that FAR exhibits a higher AUC value (61.0141%) compared to ferritin (60.4778%), albumin (55.5618%), and SOFA score (56.9719%). Moreover, the comprehensive analysis, including multifactor COX regression and Kaplan–Meier survival analysis, reveals a positive correlation between FAR and the 28-day mortality rate among sepsis patients. Notably, as FAR increases, the risk of 28-day mortality decreases. The findings are further reinforced by the subgroup analysis, ensuring the robustness and validity of our results.

In recent years, an increasing number of researchers have found that inflammatory biomarkers, such as ALB [17], ferritin [6], Albumin/Fibrinogen [18, 19], hematocrit [20], platelets [21], Red blood cell distribution width [22], systemic immune-inflammation index [23], Prognostic nutrition index [24], and urea nitrogen/albumin [25], etc. FAR holds potential as a prognostic indicator for predicting the outcome of sepsis patients. FAR has been widely employed as a novel predictive factor for COVID-19 [16], exhibiting excellent prognostic accuracy for mortality rates. However, studies specifically exploring the application of FAR in predicting the prognosis of sepsis patients have yet to be reported.

Ferritin plays a vital role in iron storage and regulation, transportation of iron ions, immune modulation, and antioxidation within the body. Our research findings strongly support the close association between elevated serum ferritin levels and poor prognosis in sepsis patients, consistent with the study conducted by Liudang He et al. [6]. In addition, the study by Yi-Peng Fang et al. [26] revealed a significant correlation between high serum ferritin levels and increased in-hospital mortality rate in sepsis patients, establishing it as an independent prognostic indicator for predicting mortality in septic patients. However, it is important to consider that factors, such as inflammation, liver and kidney function, as well as dietary and nutritional status, can also influence serum ferritin levels. Thus, relying solely on ferritin as a prognostic marker may have limitations and lack reliability to predict patient outcomes.

The pathogenesis of sepsis is closely tied to the inflammatory response, which triggers vasodilation, thrombus formation, cell apoptosis, and tissue damage. Serum albumin, being a negative acute-phase reactant protein, is correlated with the severity of infection [27, 28]. During sepsis, the inflammatory response induced by infection influences the synthesis and degradation of albumin, resulting in reduced levels of albumin in the plasma and causing hypoalbuminemia [29, 30]. Recent research indicates that hypoalbuminemia in sepsis patients is not primarily caused by suppressed liver synthesis, but rather by enhanced systemic clearance [31]. Studies indicate that serum albumin is an independent prognostic risk factor for sepsis patients [13, 17]. Nevertheless, it is important to consider that serum albumin levels can be influenced by various factors, including nutritional status, hepatic and renal function, chronic inflammatory conditions, and other underlying medical conditions of the patients. Therefore, this study aims to evaluate prognostic risks in sepsis patients more comprehensively by analyzing the relationship between the ferritin-to-albumin ratio and the 28-day mortality rate, thereby mitigating the influence of individual factors on sepsis prognosis.

Although there have been numerous studies investigating the relationship between ferritin and albumin and various diseases, the association between the ferritin-to-albumin ratio (FAR) and sepsis remains unexplored. However, research indicates that elevated ferritin levels and decreased albumin levels serve as important markers for mortality and prognosis in sepsis patients. Based on this information, we propose that FAR may hold greater significance in the context of sepsis compared to individual measurements of ferritin and albumin. In our study, we discovered a positive correlation between FAR and the 28-day mortality rate among sepsis patients, suggesting that higher FAR levels are associated with an increased risk of death in this population. These findings align with the conclusions drawn by Öztürk Taşkin et al. regarding the prognostic value of FAR in predicting mortality in COVID-19 patients [16]. Furthermore, ROC curves demonstrated that FAR exhibits good predictive capabilities with regard to the 28-day mortality rate among sepsis patients.

Our study has several advantages. First, the data used in our study is obtained from the MIMIC-IV database, ensuring the authenticity and reliability of the data. Second, this is a retrospective study, where we employed various statistical methods such as multiple-factor Cox regression analysis and subgroup analysis to control for potential confounding factors in the study. In addition, we selected the levels of ferritin and albumin measured for the first time after patient admission, excluding the interference of subsequent treatments, thereby enhancing the accuracy of the results. Finally, our study is the first to explore the relationship between FAR and the 28-day mortality rate in sepsis patients. Moreover, the convenient and cost-effective acquisition of serum ferritin and albumin levels further enhances the feasibility of this approach.

Certainly, Our study does have certain limitations. First, One significant limitation of our study is that we solely relied on the initial measurement of FAR after admission, failing to continuously track its fluctuations throughout the entire duration of sepsis. As a result, we were unable to assess the potential impact of dynamic changes in FAR on the prognosis of sepsis patients. Second, the data used in our study was obtained from the MIMIC-IV public database. As a retrospective study, despite our best efforts to control for confounding factors, it is still challenging to completely avoid the influence of unknown confounders. Finally, the measurement of laboratory indicators was not strictly standardized, and different hospitals may use different instruments, leading to measurement deviations in the results.

Conclusion

FAR, which has the potential to serve as an indicator for predicting the prognosis of sepsis patients, shows a correlation with the 28-day mortality rate. A higher FAR value is associated with an elevated risk of mortality within 28 days among sepsis patients. However, to comprehensively validate the relationship between FAR and sepsis prognosis, a large-scale study involving multiple medical centers is necessary.