Association between serum bicarbonate and 28-day mortality in septic shock A cohort study

Authors: Dan Niu, Boling Li, Huihui Bai

Abstract

Septic shock, a high-mortality subtype of sepsis, is associated with significantly elevated mortality rates compared to sepsis without shock. Although metabolic acidosis (reflected by serum bicarbonate levels) is a common complication in septic shock, robust evidence regarding the prognostic value of initial serum bicarbonate specifically in intensive care unit (ICU)-admitted septic shock patients remains limited. This study therefore aimed to evaluate the association between initial serum bicarbonate and 28-day mortality in a large cohort of septic shock patients. This retrospective cohort study utilized data from the Medical Information Mart for Intensive Care IV (v2.2) database, which contains de-identified records of 73,181 adult ICU admissions (2008–2019). Adult patients with septic shock were identified using international classification of diseases -9 (785.52) and international classification of diseases-10 (R6521, T8112XA) codes (n = 7216). Inclusion required an available first recorded serum bicarbonate level during the index ICU admission; patients with missing bicarbonate or mortality data were excluded. The primary exposure was serum bicarbonate, and the primary outcome was 28-day all-cause mortality. Multivariable logistic regression was used to assess the association, expressed as odds ratios (ORs) with 95% confidence intervals (CIs). Nonlinearity was explored using generalized additive models. Among 5287 ICU patients (median age 67.7 years), 1662 (31.4%) died within 28 days. Generalized additive modeling revealed a significant U-shaped association between serum bicarbonate and 28-day mortality (edf = 2.96, χ² = 62.33, P < .001). Below 20 mEq/L, each 1 mEq/L increase was associated with a significantly reduced 28-day mortality risk (OR = 0.91, 95% CI 0.88–0.93, P < .0001). Above 27 mEq/L, each 1 mEq/L increase was associated with a significantly increased mortality risk (OR = 1.11, 95% CI 1.02–1.20, P = .0185). In patients with septic shock, the association between serum bicarbonate and 28-day mortality risk followed a U-shaped curve. Both lower and higher serum bicarbonate levels were associated with an increased risk of 28-day mortality. These findings highlight the need for personalized acid-base management strategies.

1. Introduction

Sepsis, defined as life-threatening organ dysfunction caused by a dysregulated host response to infection, poses a substantial global health challenge.^[1]^ Septic shock represents a distinct and more severe clinical phenotype within the sepsis spectrum. It is characterized by persistent hypotension requiring vasopressors to maintain a mean arterial pressure ≥65 mm Hg and a serum lactate level >2 mmol/L despite adequate fluid resuscitation, along with profound underlying circulatory, cellular, and metabolic abnormalities.^[1,2]^ This condition is associated with a significantly elevated mortality risk compared to sepsis without shock.^[2,3]^ Epidemiological data consistently underscore the high mortality burden; a recent meta-analysis across high-income settings reported mean 30-day and 90-day mortality rates of 34.7% and 38.5%, respectively, for septic shock-significantly exceeding corresponding rates of 24.4% and 32.2% for sepsis without shock.^[3]^ Consequently, the identification of objective, early prognostic indicators is paramount for effective risk stratification and for guiding therapeutic interventions in septic shock management.^[4,5]^ Accurate early identification of high-risk patients remains a critical challenge in optimizing resource allocation and improving clinical outcomes.

Current prognostic assessment in sepsis relies heavily on composite clinical scoring systems (e.g., sequential organ failure assessment [SOFA], Acute Physiology and Chronic Health Evaluation II) and indicators of organ dysfunction.^[6,7]^ Although valuable, these tools have inherent limitations, including calculation complexity and a potential delay in reflecting acute physiological deterioration. Consequently, there is significant interest in identifying readily accessible biochemical markers to augment early risk stratification. Acid-base disturbances, particularly metabolic acidosis, are frequent complications in septic shock, arising from tissue hypoperfusion, anaerobic metabolism, and impaired acid excretion.^[8,9]^ Pathophysiologically, metabolic acidosis is defined by a primary reduction in serum bicarbonate concentration, which triggers compensatory hypocapnia and ultimately leads to a decrease in arterial pH.^[10]^ Thus, serum bicarbonate serves as a central biochemical parameter that reflects both the presence and severity of metabolic acidemia. Evidence supporting its prognostic value includes a large cohort study of patients with cardiogenic shock, which demonstrated that baseline serum bicarbonate independently predicted short-term intensive care unit (ICU) mortality.^[11]^ Furthermore, a 2009 ICU cohort study observed significantly more pronounced metabolic acidosis in non-survivors versus survivors of septic shock and severe sepsis.^[12]^ Given its role as a direct indicator of acidotic burden, initial serum bicarbonate concentration emerges as a biologically plausible prognostic indicator in septic shock. However, robust evidence specifically regarding the predictive utility of serum bicarbonate levels in septic shock patients admitted to the ICU remains scarce. Therefore, this study aimed to investigate the relationship between serum bicarbonate levels and 28-day mortality in patients with septic shock using a large-scale database.

2. Materials and methods

2.1. Data source

This retrospective observational cohort study utilized data from the Medical Information Mart for Intensive Care IV version 2.2 (MIMIC-IV v2.2) database. The MIMIC-IV v2.2 is a publicly available, single-center database comprising de-identified clinical data of 73,181 adult patients admitted to the intensive care unit (ICU) at Beth Israel Deaconess Medical Center in Boston between 2008 and 2019. To gain access to the database, the authors completed the required online training course and passed the corresponding examinations (Record ID: 58,868,345). The creation and release of the MIMIC-IV database were approved by the Institutional Review Boards of both Beth Israel Deaconess Medical Center and the Massachusetts Institute of Technology. Informed consent was waived as all patient health information had been anonymized.

2.2. Patients

The MIMIC-IV database contained medical records for 299,712 adult patients. Among the 73,181 adult patients admitted to the ICUs, 7216 were identified as having septic shock based on international classification of diseases (ICD) codes. Case identification relied on the following septic shock-related ICD-9 code 785.52 and ICD-10 codes R65.21 and T81.12XA. The outcome variable was defined as all-cause mortality within 28 days following ICU admission, and the exposure variable was serum bicarbonate, treated as a continuous variable. Serum bicarbonate levels measured at ICU admission were extracted for analysis. Patients with missing or unreported serum bicarbonate values were excluded from the study. In cases of multiple ICU admissions during the study period, only data from the first admission were included.

2.3. Variables and outcome measures

Structured Query Language was used to extract the data; Structured Query Language cords were obtained from https://github.com/MIT-LCP/mimic-code/. The following variables were measured or calculated in this sex (male/female), age, weight, comorbidity, Charlson comorbidity index, vital signs at admission (heart rate, respiratory rate, systolic blood pressure, diastolic blood pressure, and mean arterial pressure), laboratory indices (white blood cells, hemoglobin, platelet count, urea nitrogen, serum creatinine, potassium, sodium, chloride, phosphate, magnesium, glucose, bicarbonate, and anion gap), mechanical ventilation (MV) and continuous renal replacement therapy (CRRT), and the use of hydrocortisone, vasopressors, intravenous immunoglobulin therapy (IVIG), sodium bicarbonate, antibiotics (carbapenem, cephalosporins, penicillin, and vancomycin). If a variable was measured multiple times during the study, the first record was used for analysis. Severity at admission was measured by the SOFA score and simplified acute physiology score (SAPS II). These 2 scores were estimated for all patients within 24 hours of ICU admission. The authors’ clinical experience and research literature are the main sources of information for the selection of these covariates.^[11,13]^

2.4. Missing data description

Patients with missing data on exposure and outcome variables were excluded from this study. Because <5% (0–4.2%) of the covariates in this study were missing, there was no need for multiple interpolations to fill in the gaps.

2.5. Statistical analysis

Data on categorical variables are presented as percentages, and data on continuous variables are expressed as mean (SD) or interquartile range. As this was a cohort study, the exposure variables were divided into 4 quartiles, and the distribution of the baseline characteristics of patients differed across quartiles. One-way analysis of variance (normal distribution), the Kruskal–Wallis test (non-normal distribution), and the Chi-square test (categorical variables) were performed to evaluate statistically significant differences between the groups. Logistic regression models were used to examine the relationship between serum bicarbonate levels and 28-day mortality in patients with septic shock. The results are presented as odds ratios (ORs) with its 95% confidence intervals (95% CIs). Covariates were included as potential confounding variables in the final models if they changed the effects of serum bicarbonate level on 28-day mortality by >10% or were significantly associated with 28-day mortality. The following covariates were adjusted gender, age, weight, urea nitrogen, serum creatinine, hypertension, diabetes, coronary heart disease (CHD), chronic obstructive pulmonary disease (COPD), chronic kidney disease (CKD), chronic liver disease (CLD), end-stage renal disease (ESRD), CRRT, vasopressors, hydrocortisone, IVIG, MV, carbapenems, cephalosporin, penicillin, and vancomycin. We used a generalized additive model to investigate the the relationship between serum bicarbonate levels and 28-day mortality in patients with septic shock (Fig. 2). The inflection point value was calculated using a recursive algorithm, and a two-piece-wise linear model was used to calculate the OR at a 95% confidence interval on either side of the inflection point (Table 3). Furthermore, subgroup analysis was performed to determine whether there were differences between subgroups in the prediction of clinical outcomes based on serum bicarbonate levels. Interactions between subgroups were examined using the log-likelihood ratio test. Statistical significance was defined as a two-tailed P-value of smaller than .05. All data processing steps and statistical tests were performed using EmpowerStats (www.empowerstats.com; X&Y Solutions Inc., Boston ) and the statistical software package R (The R Foundation for Statistical Computing, Vienna, Austria; http://www.r-project.org; version 3.4.3).

3. Results

3.1. Description of the patient screening process

This study initially enrolled 299,712 patients; 226,531 individuals were excluded because they were not admitted to the ICU, 65,965 patients were excluded because they did not meet the diagnostic criteria for septic shock, 1862 patients were excluded because they were not admitted for the first time, and 67 patients were excluded because they had missing serum bicarbonate information. Ultimately, 5287 patients were included in the final data analysis. The flowchart provides detailed information (Fig. 1).

3.2. Baseline characteristics of patients

The mean age of the 5287 patients was 67.70 ± 15.48 years, and 2885 (54.57%) were men. Table 1 summarizes baseline characteristics of these septic shock patients stratified by serum bicarbonate quartiles. Significant interquartile differences (Q1: 3.00–16.00 mEq/L, n = 1113; Q4: 23.00–44.00 mEq/L, n = 1630) were observed for most parameters (P < .05). Patients in Q1 exhibited greater disease severity, evidenced by higher SOFA (9.95 ± 3.96 vs 6.85 ± 3.59; P < .001) and SAPS II scores (56.0 ± 15.9 vs 42.7 ± 14.1; P < .001), despite being younger (66.2 ± 15.9 years vs 68.6 ± 14.9 years; P = .001). Metabolic derangements in Q1 included higher anion gap (21.6 ± 6.8 vs 14.1 ± 3.5, P < .001), creatinine (median 2.00 mg/dL vs 1.20 mg/dL, P < .001), and potassium (4.44 ± 1.02 mEq/L vs 4.10 ± 0.77 mEq/L, P < .001). Interventions were more frequent in Q1: CRRT (22.6% vs 10.7%, P < .001), mechanical ventilation (59.8% vs 51.6%, P < .001), and sodium bicarbonate use (45.5% vs 18.4%, P < .001). Crucially, Q1 patients had significantly higher 28-day mortality (43.0% vs 28.9%; P < .001), contrasting with the overall cohort mortality of 31.4% (1662/5287). No significant differences existed in hemoglobin (10.16 ± 2.31 g/dL vs 10.07 ± 2.07 g/dL; P = .600) or Charlson comorbidity index (6.56 ± 3.03 vs 6.72 ± 2.84; P = .087).

3.3. Relationship between serum bicarbonate and 28-day mortality in patients with septic shock

The results of univariate logistic regression analysis for 28-day mortality are shown in Table 2. Critical interventions significantly increased CRRT (OR = 3.19, 95% CI: 2.71–3.75, P < .0001), MV (OR 2.53, 95% CI: 2.24–2.86, P < .0001), and vasopressor use (OR = 1.88, 95% CI: 1.60–2.21, P < .0001). Disease severity scores per-unit rise predicted SOFA (OR = 1.20, 95% CI: 1.18–1.22), SAPS II (OR = 1.06, 95% CI: 1.06–1.07), and Charlson comorbidity index (OR = 1.16, 95% CI: 1.13–1.18, all P < .0001). Major comorbidities included CLD (OR = 2.01, 95% CI: 1.71–2.37, P < .0001) and ESRD (OR = 1.55, 95% CI: 1.25–1.92, P < .0001). Protective factors were longer hospitalization (OR = 0.93, 95% CI: 0.92–0.94, P < .0001) and cephalosporin therapy (OR = 0.75, 95% CI: 0.67–0.85, P < .0001). Hypertension demonstrated reduced mortality risk (OR = 0.86, 95% CI: 0.76–0.97, P = .0174). Gender, diabetes, hemodynamic parameters, and intravenous immunoglobulin showed no significant association.

Different covariate adjustment strategies were used to elucidate the relationship between serum bicarbonate levels and 28-day mortality in patients with septic shock. The nonadjusted and adjusted models are presented in Table 3. In the non-adjusted model, the risk of 28-day mortality in patients with septic shock was reduced by 4% for every unit increase in serum bicarbonate (OR = 0.96, 95% CI: 0.95–0.97, P < .0001). After controlling for gender, age, and weight, the risk of 28-day mortality decreased by 5% (OR = 0.95, 95% CI: 0.94–0.96, P < .0001). In the adjusted-II model (gender, age, weight, urea nitrogen, serum creatinine levels, hypertension, diabetes, CHD, COPD, CKD, CLD, ESRD, CRRT, vasopressors, hydrocortisone, IVIG, MV, and taking carbapenems, cephalosporin, penicillin, and vancomycin were adjusted), the risk of 28-day mortality decreased by 3% (OR = 0.97, 95% CI: 0.97–0.98, P < .0001). Sensitivity analysis was performed using quartile-based categorical variables and P-values were calculated for the trend tests. However, inconsistent results were obtained when comparing serum bicarbonate levels as continuous or categorical variables (P for trend < .0001) (Table 3). The unequal variances in OR values between the different serum bicarbonate groups suggested a nonlinear relationship between serum bicarbonate levels and 28-day mortality.

We observed a nonlinear dose–response relationship between serum bicarbonate levels and 28-day mortality in patients with septic shock (Fig. 2 and Table 4). Inflection points were determined using a two-stage linear model and recursive techniques. The results showed that the inflection points were 20 and 27 mEq/L. When serum bicarbonate levels were <20 mEq/L, each unit increase in serum bicarbonate was associated with a significantly reduced risk of 28-day mortality (adjusted OR = 0.91, 95% CI: 0.88–0.93, P < .0001). However, within the intermediate range (20–27 mEq/L), the association was nonsignificant (adjusted OR = 1.03, 95% CI: 0.99–1.07, P = .172). Conversely, when serum bicarbonate levels were above 27 mEq/L, each unit increase in bicarbonate was associated with a significantly increased risk of 28-day mortality (adjusted OR = 1.11, 95% CI: 1.02–1.20, P = .0185) (Table 4). All analyses were adjusted for gender, age, weight, urea nitrogen, serum creatinine, hypertension, diabetes, coronary heart disease, chronic obstructive pulmonary disease, chronic renal disease, chronic liver disease, end-stage renal disease, and treatments including CRRT, vasopressors, hydrocortisone, intravenous immunoglobulin, mechanical ventilation, carbapenems, cephalosporin, penicillin, and vancomycin. The generalized additive model identified a statistically significant nonlinear (U-shaped) relationship between serum bicarbonate and 28-day mortality (edf = 2.96, χ² = 62.33, P < .001; Fig. 2). Given this nonlinearity, a two-piece-wise linear regression model provided a significantly better fit than a standard logistic regression model (log-likelihood ratio test P < .001; Table 4). We therefore present the piecewise model as the optimal characterization of this association.

3.4. Subgroup analyses

Table 5 presents the results of subgroup analyses of the relationship between serum bicarbonate levels and 28-day mortality, which were performed according to age, gender, SOFA and SAPS II scores, systolic blood pressure, pulse O2 saturation, CKD, COPD, diabetes, CHD, and ESRD.

The results demonstrated that following careful adjustments, the relationship between serum bicarbonate levels and 28-day mortality was stable in all subgroups. A U-shaped relationship between serum bicarbonate level and 28-day mortality was observed in all patients, except those with a SAPS II score of <47 (P = .0352 for the interaction term).

4. Discussion

This retrospective cohort study of 5287 septic shock patients from the MIMIC-IV database identified an independent, U-shaped association between serum bicarbonate levels and 28-day mortality. Following covariate adjustment and sensitivity analyses, the results demonstrated that both low and high serum bicarbonate levels were associated with an increased risk of death. Specifically, levels below 20 mEq/L were linked to higher mortality (risk decreased by 9% per unit increase; OR: 0.91, 95% CI: 0.88–0.93, P < .0001), while levels above 27 mEq/L were also associated with higher mortality (risk increased by 11% per unit increase; OR: 1.11, 95% CI: 1.02–1.20, P = .0185).

The range associated with the most favorable outcomes was 20 to 27 mEq/L, which is notably wider than the conventional reference range for healthy individuals (22–26 mEq/L).^[14]^ This discrepancy likely reflects the unique pathophysiology of septic shock, characterized by complex acid-base disturbances and compensatory mechanisms.^[9]^ It is essential to clarify that these thresholds are derived from mortality risk stratification within a specific septic shock population and should not be construed as universal diagnostic criteria or direct therapeutic targets. Important limitations include the retrospective design, population specificity, and the inherent constraints of observational data in establishing causality. Thus, these findings do not supersede current clinical guidelines such as the Surviving Sepsis Campaign,^[15]^ which emphasize treating the underlying cause of shock and comprehensive acid–base evaluation (incorporating pH, hemodynamics, and clinical context) over correction of isolated bicarbonate values.

The therapeutic use of sodium bicarbonate in metabolic acidosis remains contentious. Certain studies indicate potential benefits in specific subgroups, particularly patients with acute kidney injury (AKI) and severe acidosis (pH < 7.2).^[16,17]^ The BICAR-ICU (sodium bicarbonate therapy for patients with severe metabolic acidaemia in the intensive care unit) trial, for example, demonstrated that bicarbonate administration reduced mortality and the need for renal replacement therapy in patients with AKI stages 2 to 3 and profound acidosis.^[18]^ Similarly, a target trial emulation by Blank et al reported a modest mortality reduction with bicarbonate therapy in patients with metabolic acidosis, especially those with AKI or severely low pH.^[17]^ In contrast, other investigations, including that by Zhang et al, found no overall mortality benefit in septic patients with metabolic acidosis, though a subgroup with AKI and pH < 7.2 showed improved survival.^[16]^ These divergent outcomes underscore the complexity of bicarbonate therapy and highlight the necessity for individualized treatment decisions that consider risks such as metabolic alkalosis, hypernatremia, and hypocalcemia.^[18]^

The U-shaped association observed in our study is consistent with findings in other clinical populations. Both low and high bicarbonate levels have been linked to increased mortality in generally healthy elderly adults,^[19]^ patients with CKD,^[20,21]^ and individuals undergoing hemodialysis.^[22]^ This consistent pattern across diverse settings suggests that extremes of bicarbonate concentration, irrespective of origin, are associated with adverse outcomes.

The mechanisms contributing to poor outcomes at bicarbonate extremes are multifactorial. Severe acidemia can impair cardiac contractility, alter ventilatory responses, and promote anaerobic metabolism.^[23,24]^ Conversely, alkalemia may induce ionized hypocalcemia, increase carbon dioxide partial pressure, and predispose to complications such as arrhythmias and impaired oxygen delivery.^[23]^ These pathophysiological consequences underline the critical importance of maintaining acid–base homeostasis in critically ill patients.

Despite considerable advances in the management of septic shock (including refined protocols for antibiotics,^[25]^ vasopressors,^[26]^ fluid resuscitation,^[27]^ and novel therapies^[28,29]^) mortality rates remain substantial.^[3]^ Acid–base disorders are highly prevalent in ICU settings and often arise from sepsis, renal impairment, hepatic dysfunction, or iatrogenic interventions.^[23]^ Notably, not only metabolic acidosis but also elevated bicarbonate levels may signal underlying pathology and correlate with increased mortality.^[13]^

Our results align with those reported by Libório et al,^[13]^ who also identified a U-shaped relationship between bicarbonate levels and mortality in critically ill patients using the MIMIC-II database, albeit with slightly different optimal thresholds (24 and 31 mEq/L). A fundamental methodological difference lies in the exposure Libório et al analyzed the maximum bicarbonate level during the ICU stay, which may be influenced by compensatory responses or treatments, whereas we used the first measurement within 24 hours of admission. While this initial value reflects the acid–base status at a critical clinical juncture and may be valuable for early risk stratification, the acid–base profile in septic shock is highly dynamic, influenced by resuscitation, organ support (e.g., CRRT), and the evolving disease process. Nevertheless, neither a single initial nor a maximum value fully captures the dynamic evolution of acid–base balance in septic shock, which likely plays an important role in prognosis and therapeutic response.^[30]^ Future studies should consider incorporating serial bicarbonate measurements or trend analyses to more comprehensively assess its relationship with outcomes and to explore whether “bicarbonate trajectories” offer superior predictive value compared to a single measurement.

This study has several strengths. First, the large sample size yielded reliable results and a better understanding of the relationship between serum bicarbonate levels and 28-day mortality in patients with septic shock. Second, this study employed algorithms that elucidate nonlinearity to better reflect the true relationship between serum bicarbonate levels and the 28-day mortality in these patients. Third, stratification variables were used to further explore this relationship, and a U-shaped relationship was observed between serum bicarbonate levels and 28-day mortality in patients with a SAPS II score >47.

This study has several limitations. First, its retrospective and single-center design introduces the potential for unmeasured confounding and may limit the generalizability of our findings to other ICU populations due to potential institutional biases and limited patient diversity. Second, although all-cause mortality is a robust and reliably captured endpoint in the MIMIC-IV database, we lacked data on cause-specific mortality. Third, our analysis relied on a single, static measurement of serum bicarbonate (the first value within 24 hours of ICU admission). This snapshot cannot capture the dynamic evolution of acid–base status in response to resuscitation and ongoing critical illness, which may be vital for prognosis. Future research incorporating repeated measures or trajectory modeling could provide deeper pathophysiological insights and enhance the clinical utility of bicarbonate for outcome prediction. Fourth, we lacked important data on inflammatory markers (e.g., C-reactive protein) and nutritional assessments (e.g., albumin levels), which may confound the relationship between bicarbonate and mortality. Fifth, specifically, the lack of measured serum lactate levels at admission limits our ability to fully characterize the initial severity of metabolic derangement in septic shock. Finally, our findings demonstrate an association but not causation, and the observational design does not allow for an assessment of bicarbonate’s predictive performance for mortality.

5. Conclusion

Our findings demonstrate a significant U-shaped association between serum bicarbonate levels and 28-day mortality risk in patients with septic shock. Both low and high bicarbonate concentrations were independently associated with increased mortality, highlighting the critical need for personalized acid-base management strategies. These results support the development of biomarker-directed clinical trials to optimize outcomes in this high-risk population.

Acknowledgments

The authors thank all the researchers who created and managed the MIMIC-IV database.

Author contributions

Conceptualization: Boling Li.

**Data ** Dan Niu, Huihui Bai.

Methodology: Dan Niu, Boling Li, Huihui Bai.

Resources: Dan Niu.

Software: Dan Niu, Huihui Bai.

Supervision: Dan Niu, Boling Li.

**Writing – original ** Dan Niu.

**Writing – review & ** Dan Niu, Boling Li.