Introduction

Sepsis is a potentially fatal clinical syndrome triggered by various infectious syndromes and characterized by a dysregulated systemic inflammatory response. Despite substantial progress in medical interventions and an improved comprehension of its underlying pathophysiological mechanisms, sepsis remains a leading cause of intensive care unit (ICU) admissions. The persistent high incidence of sepsis-related ICU admissions highlights the urgent need for continued research to reduce its clinical burden [1,2,3]. Septic shock and sepsis rapidly advance inflammatory states marked by immunosuppression [4]. Lymphocytes, a critical subset of white blood cells (WBC), comprise 20–40% of the total WBC count [5]. Sepsis-induced lymphocyte apoptosis, associated with adverse outcomes, involves increased cell death and impaired lymphocyte proliferation, significantly reducing circulating lymphocyte percentages [6, 7]. Sepsis triggers the release of pro-apoptotic factors such as glucocorticoids, reactive oxygen species, and pro-inflammatory cytokines, which promote lymphocyte apoptosis [8]. The depletion of lymphocytes, mainly T, B, and natural killer cells, compromises innate and adaptive immune responses. [9, 10]. Sepsis-associated lymphopenia makes patients susceptible to secondary infections, often resulting in prolonged hospital stays and increased mortality rates [11, 12]. Recent studies have explored the relationship between lymphocyte counts and patient outcomes. Kazan et al. found that among COVID-19 patients admitted to the ICU, those with higher lymphocyte levels had better survival rates than individuals with decreased lymphocyte counts, while lower lymphocyte counts on the fifth day of hospitalization correlated with increased mortality [13]. Despite advances in sepsis research, our understanding of the relationship between lymphocyte counts and mortality risk in critically ill sepsis patients remains incomplete [14,15,16]. Furthermore, lymphocytosis is recognized as an indicator of poor clinical outcomes under specific conditions [17].

We hypothesized that a non-linear relationship exists between lymphocyte counts and hospital mortality risk in critically ill sepsis patients. To investigate this, we analyzed the association between lymphocyte counts and all-cause hospital mortality in sepsis patients using data from two large intensive care databases: the Medical Information Mart for Intensive Care IV 2.0 (MIMIC-IV 2.0) and the eICU Collaborative Research Database 2.0 (eICU-CRD 2.0). We aimed to provide a comprehensive understanding of the prognostic significance of lymphocyte counts in sepsis. This research could inform clinical decision-making and improve risk stratification strategies in critical care settings.

Methods

Data source

This retrospective study utilized data from two comprehensive, publicly accessible critical care databases: the Medical Information Mart for Intensive Care IV 2.0 (MIMIC-IV 2.0) and the eICU Collaborative Research Database 2.0 (eICU-CRD 2.0).

The MIMIC-IV 2.0 is a publicly accessible database containing adult (≥ 18 years) medical data from Beth Israel Deaconess Medical Center in Boston, Massachusetts. It covers admissions to medical, surgical, coronary, and cardiac surgery intensive care units from 2008 to 2019. The database provides extensive information, including demographic data, vital sign measurements, comprehensive laboratory test results, detailed medication records, procedural information, ICD coding, and hospital length of stay [18].

The eICU-CRD 2.0, conducted by the Philips eICU Research Institute, is a multi-center database representing over 200 hospitals across the United States from 2014 to 2015. This telehealth archive focuses exclusively on adult ICU patients and includes continuous and intermittent vital signs, laboratory measurements, pharmaceutical records, detailed care plan information, admission diagnoses, and treatment information [19].

Participants

This study enrolled sepsis participants aged 18 or older from the MIMIC-IV 2.0 and eICU-CRD 2.0 databases. Inclusion criteria were as follows:1) Patients with confirmed infections and a Sequential Organ Failure Assessment (SOFA) score of 2 or greater, meeting the Sepsis-3.0 diagnostic criteria, were eligible for inclusion [4]. 2) Complete peripheral blood counts recorded within twenty-four hours of ICU admission.

Exclusion criteria were as follows: 1) ICU stay of less than twenty-four hours; 2) presence of immunodeficiency virus (HIV), rheumatism, malignancies, metastatic neoplasms, or hematological disorders including aplastic anemia; 3) the absence of lymphocyte data on the day of admission; 4) Only the first ICU admission was considered in cases of recurrent hospitalizations.

Data extraction

Clinical data were extracted using SQL queries from the MIMIC-IV 2.0 and eICU-CRD 2.0 databases. The extracted data, primarily from the first 24 h of ICU admission, included: 1) Laboratory assessments: Platelets, monocytes, neutrophil count, lymphocytes, WBC, blood glucose, hemoglobin, albumin, blood urea nitrogen (BUN), creatinine, sodium, potassium, alanine aminotransferase (ALT), aspartate aminotransferase (AST), high-density lipoprotein, low-density lipoprotein, total cholesterol, and triglycerides. 2) Demographic details and vital signs: sex, age, heart rate, systolic and diastolic, temperature, blood pressure, and SOFA score. 3) Blood gas analysis: Lactate, PH, PaO2fio2ratio, Fio2, Paco2, Pao2. 4) ICU-related data: Length of ICU stay and survival status at discharge. 5) Comorbidities and Treatments: congestive heart failure, myocardial infarction, renal disease, chronic pulmonary or liver disease, mechanical ventilatory support, and renal replacement therapy. Average values were employed for variables documented multiple times in the first twenty-four hours post-ICU admission.

Data extraction was performed by our research team member Guyu Zhang, who has extensive experience working with these databases and obtained approval certification number (55,849,941). Before data extraction, all team members underwent training on the structure and content of the MIMIC-IV 2.0 and eICU-CRD 2.0 databases.

Outcome measures

The primary endpoint was all-cause hospital mortality. Secondary outcomes included 60-day and 90-day all-cause hospital mortality, allowing for assessment of both short-term and intermediate-term mortality risks associated with lymphocyte levels.

Statistical analysis

Clinical data were extracted using SQL queries. Continuous variables were presented as mean ± standard deviation for normally distributed data or median and interquartile range for non-normally distributed data. Continuous variables were analyzed using either the Student's t-test for normally distributed data or the Mann–Whitney U test for non-normally distributed data. We employed the Chi-square test or Fisher's exact test as appropriate for categorical variables, presented as proportions.

To explore the independent prognostic significance of lymphocytes on hospital mortality, we performed univariable and multivariable analyses using Cox proportional hazard models facilitated by the 'survival' package in R. Results were reported as hazard ratios (HR) with 95% confidence intervals (CI). Spearman correlation analyses using the R package ‘ggpur’ were performed to prevent overfitting due to multicollinearity. Clinically relevant and prognosis-associated variables such as WBC, gender, age, albumin, creatinine, potassium, temperature, first-day sofa score, and respiratory rate were included in the restricted cubic spline (RCS) model to investigate the associations between lymphocytes and hospital mortality by R package 'rms'. Patients were categorized into quartiles based on lymphocyte counts. Kaplan–Meier survival analysis was conducted to evaluate the incidence rate of hospital mortality across lymphocyte level groups, with group discrepancies assessed by log-rank tests using the R package 'Survminer'. Analyses were further stratified according to diabetes, congestive heart failure, renal disease, liver disease, cerebrovascular disease, invasive ventilation, renal replacement therapy, and chronic pulmonary disease to verify the robustness of the prognostic effect of lymphocytes on hospital mortality. Interactions between lymphocytes and stratification variables were examined using likelihood ratio tests.

Variables with more than 30% missing data, such as lactate, PaO2/FiO2 ratio, pH, FiO2, PaCO2, PaO2, high-density lipoprotein, low-density lipoprotein, total cholesterol, and triglycerides, were excluded from the analysis (Figure S1). The remaining 33 predictor variables, identified at ICU admission, were included. Missing values among these selected variables were imputed using predictive mean matching (PMM) through the 'mice' package. Data analysis was performed with R software, version 4.1.3 (Beijing, China), and statistical significance was established at a two-tailed P-value threshold of < 0.05 for all analyses conducted in the study.

Fig. 1
figure 1

A flow chart illustrating the regulatory model of patient enrollment and analysis workflow

Full size image

Results

Demographics and clinical characteristics

Based on the inclusion criteria, our study included 37,054 sepsis patients (Fig. 1). Table 1 reveals that in the survival group, levels of white blood cells (WBC), neutrophils, monocytes, platelet count, blood urea nitrogen (BUN), creatinine, glucose, potassium, alanine aminotransferase (ALT), aspartate aminotransferase (AST), heart rate, respiratory rate, and age were significantly lower compared to those in the non-survival group. Conversely, levels of lymphocytes, albumin, calcium, and hemoglobin were higher in the survival group. Comorbidities were less prevalent in the survival group than in the non-survival group. Specifically, we observed lower rates of congestive heart failure (24.3% vs. 29.8%, P < 0.001), chronic pulmonary disease (22.5% vs. 24.1%, P < 0.008), myocardial infarction (10.3% vs. 13.5%, P < 0.001), renal diseases (19.9% vs. 25.3%, P < 0.001), liver disease (7.49% vs. 15.7%, P < 0.001), and cerebrovascular disease (12.6% vs 16.3%, P < 0.001). The non-survival group had a higher need for invasive ventilation (55.1% vs 74.6%, P < 0.001) and renal treatment (19.9% vs 25.3%, P < 0.001). Patients were categorized into four groups based on their lymphocyte count during the first twenty-four hours of admission, using quartile for classification (Q1: 0.05–0.81; Q2: 0.81–1.3; Q3: 1.3–2; Q4: 2–8). As shown in Table 2, the third quartile (Q3) group exhibited the highest survival rate (87.6%, P < 0.001). Furthermore, the incidence rate of myocardial infarction was not statistically significant across different groups. A similar pattern was observed in patients with septic shock (TableS1).

Table 1 Baseline characteristics of sepsis patients in MIMIC-IV and eICU-CRD
Full size table
Table 2 Baseline characteristics of sepsis patients according to lymphocytes quartile
Full size table

Lymphocyte counts and hospital mortality risk

A univariable Cox regression analysis was conducted to identify variables influencing hospital mortality. As shown in Table 3, WBC, neutrophils, monocytes, gender, age, albumin, creatinine, potassium, temperature, respiratory rate, and the first-day sofa score were significantly associated with hospital mortality. Conversely, ALT, AST, heart rate, blood pressure components, platelets, glucose, and sodium were found to have limited clinical significance,with hazard ratios close to 1. We conducted a correlation analysis to address potential collinearity among the remaining continuous variables. Figure 2A demonstrates strong correlations among WBC, neutrophils, and monocytes. Based on these findings, WBC was selected as the adjusted variable. A multivariable Cox regression analysis was then utilized to assess the independent prognostic significance of lymphocytes (Table 3). After adjusting for WBC, gender, age, albumin, creatinine, potassium, temperature, respiratory rate, and the first-day sofa score, the RCS model revealed a U-shaped relationship between lymphocytes and hospital mortality (Fig. 2B). The optimal lymphocyte level associated with the lowest in-hospital mortality was identified as 1.85 × 109/L, within the Q3 interval. Both lower and higher lymphocyte counts were significantly associated with increased hospital mortality rates. This pattern remained consistent in patients with and without septic shock (Fig. 3).

Table 3 Cox regression analysis of the variables
Full size table
Fig. 2
figure 2

Analysis of the correlation between clinical variables (A) and Restricted cubic spline: lymphocyte count and hospital mortality risk (B)

Full size image
Fig. 3
figure 3

Restricted cubic spline: lymphocyte count and hospital mortality risk in patients with septic shock (A) and without septic shock (B)

Full size image

Survival outcomes and subgroup analyses

The Kaplan–Meier survival analysis curve was employed to illustrate the hospital mortality rates at 60 and 90 days, revealing significant differences across various groups. The second (Q2) and third (Q3) quartile groups demonstrated longer median survival time of 50.5 and 57.6 days, respectively, compared to the first quartile (Q1) group (P < 0.05). In contrast, the Q1 and fourth quartile (Q4) groups exhibit similar median survival times of 40.5 and 41 days, respectively, with no significant difference (P > 0.05) (Fig. 4A). This pattern remained consistent for 60-day hospital mortality (Fig. 4B) and septic patients (FigureS2). As shown in Fig. 5, the subgroup analysis further elucidated the relationship between lymphocyte counts and hospital mortality across various patient subgroups. A U-shaped relationship was observed in patients receiving invasive ventilation therapy and those with congestive heart failure, diabetes, or renal disease(Fig. 5A-D), indicating that low and high lymphocyte counts were associated with increased mortality in these groups. In contrast, no significant relationship between lymphocyte counts and hospital mortality was found in patients with cerebrovascular or chronic pulmonary disease (Fig. 5G-H). Patients with liver disease and undergoing renal replacement therapy exhibited a linear correlation between lymphocyte counts and hospital mortality (P for overall = 0.021, P for nonliner = 0.158, P for overall = 0.025, P for nonliner = 0.759) (Fig. 5E-F). Notably, significant interactions were observed in the third (Q3) quartile groups of patients with liver disease (P for interaction < 0.01) and undergoing renal replacement therapy (P for interaction < 0.01), suggesting the relationship between lymphocyte counts and mortality may differ in these specific patients (Fig. 6).

Fig. 4
figure 4

Kaplan–Meier survival analysis curves for all-cause mortality in sepsis patients. Footnote lymphocyte count quartiles: Q1 (0.05–0.81), Q2 (0.81–1.3), Q3 (1.3–2), Q4 (2–8). Kaplan–Meier curves showing cumulative probability of all-cause mortality according to groups at ninety days (a), and sixty days (b)

Full size image
Fig. 5
figure 5

Restricted cubic spline curve for different subgroups among sepsis patients (A) Invasive ventilation therapy (B) Congestive heart failure (C) Diabetes (D) Renal disease (E) Liver disease (F) Renal replacement therapy (G) Cerebral disease (H) Chronic pulmonary disease

Full size image
Fig. 6
figure 6

Forest plot for the primary outcome in different subgroup. HR, hazard ratio; CI, confidence interval

Full size image

For patients with septic shock, a similar pattern was observed among those receiving invasive ventilation therapy and those with congestive heart failure or diabetes. However, no significant relationship was found between lymphocyte counts and hospital mortality in patients with cerebrovascular disease, chronic pulmonary disease, liver, or renal replacement therapy (Figure S3-4).

Discussion

Sepsis is a life-threatening condition that triggers systemic organ dysfunction and an exaggerated immune response to infection, leading to metabolic disturbances and severe immunosuppression [20]. Lymphocytes, crucial for adaptive and innate immune systems, are essential for an effective defense against sepsis [21]. Studies by Cillion's [22] and Ceccato's [23] have demonstrated that lymphocytopenia can independently predict ICU mortality. The underlying pathophysiological mechanism may involve early-stage impairment of B lymphocytes during sepsis onset, resulting in decreased IgM levels of natural antibodies against bacterial infection [24]. Furthermore, T lymphocytes, when continuously exposed to antigens, upregulate the expression of checkpoint inhibitors like CTLA-4, PD-1, LAG-3, and TIM-3, which weakens the host's capacity to mount a robust cellular immune response to subsequent infections [25, 26]. Clinical observational studies have shown a strong correlation between sepsis-induced lymphopenia and poor prognosis [27,28,29]. Persistent lymphopenia on the fourth day after sepsis diagnosis has emerged as a reliable biomarker for predicting 28-day and 1-year survival [30]. Lower lymphocyte counts and sustained lymphopenia are associated with elevated risks of ICU-acquired infections and 28-day mortality [31]. Our findings reveal a U-shaped relationship between lymphocyte count and hospital mortality risk in sepsis patients. The potential mechanisms behind this relationship may involve a delicate balance between the protective and harmful effects of lymphocytes in sepsis. Low lymphocyte counts likely signify impaired immune function and an inability to effectively respond to infections, while high counts could suggest uncontrolled inflammation causing organ damage [32, 33]. While lymphocytopenia has been considered a predictor of poor outcomes in sepsis, our study suggests that lymphocytosis also warrants equal attention. Recognizing both extremes in lymphocyte counts as indicators of poor prognosis represents a significant advancement in the field, pointing toward a more balanced approach to evaluating immune status beyond the simplistic association of low lymphocyte counts with higher mortality risk. The subgroup analyses reveal a linear association between higher lymphocyte counts and improved prognosis in sepsis patients with liver disease or requiring renal replacement therapy, highlighting the importance of robust cell-mediated immunity for survival in these cohorts [34]. In contrast, no significant association was found between lymphocyte counts and outcomes in patients with chronic pulmonary disease or cerebrovascular disease, suggesting disease-specific differences in the prognostic value of lymphopenia that merit further investigation.

Importantly, our study also explored the relationship between lymphocyte counts and outcomes in septic shock patients. Subgroup analysis revealed that septic shock patients exhibited similar patterns to those observed in the overall sepsis cohort. Specifically, we found a U-shaped association between lymphocyte counts and hospital mortality in septic shock patients with congestive heart failure, diabetes, renal disease, and receiving invasive ventilation therapy (Figure S3-4). This finding suggests that the complex interplay between lymphocyte levels and mortality risk persists even in the more severe septic shock. However, it's noteworthy that no significant relationship was found between lymphocyte counts and hospital mortality in septic shock patients undergoing renal replacement therapy. This divergence from the overall sepsis cohort could indicate that in septic shock, the impact of extreme organ dysfunction might outweigh the prognostic value of lymphocyte counts in specific subgroups. Alternatively, It may suggest that the immune dysregulation in septic shock is more complex than in sepsis conditions.

Our study underscores the role of lymphocytes as a potent biomarker for risk stratification and prognosis in sepsis and septic shock. The complex interplay between alterations in lymphocyte distribution, functionality, and inflammatory response underlines the need for personalized management approaches tailored to specific sepsis phenotypes. These findings have significant clinical implications, indicating that regular monitoring of lymphocyte counts could be a crucial aspect of sepsis management. Such monitoring can inform therapeutic decisions and enhance patient risk assessment, potentially improving outcomes.

This study's retrospective design and reliance on observational databases inherently introduce potential confounding factors. While our findings underscore the significance of lymphocytes in influencing mortality rates in sepsis patients, these results should be interpreted with caution and verified through prospective trials that meticulously control pertinent clinical variables. Our analysis is constrained by the limitations inherent in the MIMIC-IV 2.0 and eICU-CRD 2.0 databases. Notably, the inadequate documentation of specific critical data, such as inflammatory biomarkers and the dynamic changes in lymphocyte counts over time, precludes their comprehensive analysis. It is essential to acknowledge that, given the observational nature of our study, the association identified between lymphocyte counts and mortality does not imply causation. Future research is needed through prospective studies or randomized controlled trials to elucidate this relationship. Subsequent research should evaluate the implications of sustained lymphopenia during the ICU rather than focusing solely on initial lymphocyte counts. This approach could provide a more comprehensive understanding of the dynamic role of lymphocytes in sepsis progression and outcomes.

Conclusion

Our findings demonstrate a U-shaped association between lymphocyte counts and hospital mortality risk in sepsis patients, with the lowest risk observed at moderately elevated levels around 1.85 × 109/L. Both lymphocyte deficiency and extreme lymphocytosis are associated with a poorer prognosis, underscoring the delicate balance between impaired immunity and hyperinflammation in influencing outcomes. This finding suggests that an optimal range of lymphocyte counts may be crucial for better patient outcomes.