Serum Bicarbonate: Reconsidering the Importance of a Neglected Biomarker in Predicting Clinical Outcomes in Sepsis

Background Despite being an important pathophysiological component, information on the predictive value of serum bicarbonate level in sepsis is limited. Study design and method This is a single-centered retrospective study involving 4176 patients admitted to the medical ICU (MICU) with a diagnosis of sepsis. Patients were divided into two groups based on the presence or absence of chronic kidney disease (CKD) on admission: CKD and non-CKD, respectively. Each group was then divided into three sub-groups based on serum bicarbonate level at presentation (in mEq/l)- low (<22), normal (22-28), and high (>28). We compared the clinical outcomes between the sub-groups in each group, with in-hospital mortality as the primary endpoint. Secondary endpoints included vasopressor-free days, ventilator-free days, ICU-free days, and hospital-free days. Result In both the CKD and non-CKD groups, low serum bicarbonate was associated with significantly increased in-hospital mortality. There was no difference in the mortality between the sub-groups with normal and high serum bicarbonate. When adjusted for other known predictors of mortality, the association of low serum bicarbonate with increased in-hospital mortality was statistically significant only in the patient group with a Sequential Organ Failure Assessment (SOFA) score of ≥9. Additionally, the SOFA score had a better predictive value for in-hospital mortality, ICU-free days, and ventilator-free days when the serum bicarbonate level was <22. Interpretation Serum bicarbonate is a good predictor of clinical outcomes in sepsis and can be used along with other markers of sepsis to predict clinical outcomes.


Introduction
Sepsis is a life-threatening syndrome of organ dysfunction caused by infection and dysregulated host response to the infection [1,2]. It is the leading cause of death worldwide, with an estimated incidence of 31.5 million cases of sepsis, 19.4 million cases of severe sepsis, and 5.3 million deaths per year [3]. Markers of sepsis that predict morbidity and mortality, such as Acute Physiology and Chronic Health Evaluation (APACHE) score, Sequential Organ Failure Assessment (SOFA) score, Simplified Acute Physiology Score (SAPS) score, serum lactate level, and mean arterial pressure, have been extensively studied [4][5][6][7][8][9]. Despite being an important pathophysiological component of sepsis, information on serum bicarbonate level and its predictive value in sepsis are limited. Hemodynamic instability in critically ill patients with sepsis can induce multiple organ compromise, including renal dysfunction and tissue hypoperfusion, both of which are widely believed to cause decreased serum bicarbonate levels. Recent studies have suggested that serum bicarbonate can neither be a surrogate for lactate nor can it accurately predict the lactate level in patients with sepsis [10][11][12]. To the best of our knowledge, ours is the first large study to look at the relation between serum bicarbonate and clinical outcomes in patients with sepsis. We hypothesize that serum bicarbonate level at presentation is a predictor of morbidity and short-term mortality in patients with sepsis admitted to the medical intensive care unit (MICU).

Materials And Methods
After approval by the University of Kentucky Institutional Review Board (approval number 47751), we performed a single-center retrospective research study on patients admitted to the MICU of the University of Kentucky Medical Center, Lexington, Kentucky, United States, with a diagnosis of sepsis, severe sepsis, and septic shock from January 2012 to December 2017. We used International Classification of Diseases (ICD) 9 and 10 codes to identify these patients. The exclusion criteria were age < 18 years, end-stage renal disease requiring dialysis, or missing serum bicarbonate level at presentation defined as within four hours of admission to MICU. A total of 5293 patients met the inclusion criteria, of which 1117 patients were excluded. The final cohort consisted of 4176 subjects. The primary outcome was in-hospital mortality, defined as death before hospital discharge. Secondary outcomes included hospital-free days, ICU-free days, ventilator-free days, and pressor-free days, all calculated at 28 days. Patients with a diagnosis of severe sepsis were included in the study because we also used ICD 9 for the identification of patients.
Stata Statistical Software: Release 13, 2013 (StataCorp LP, College Station, Texas, United States) was used for statistical analysis. We first evaluated the unadjusted predictive value of serum bicarbonate in sepsis. Then, we adjusted the outcomes for co-morbidities and other known predictors of clinical outcomes in sepsis.
For the unadjusted predictive value of serum bicarbonate, the study cohort was first divided into three groups based on serum bicarbonate level at presentation: serum bicarbonate level <22, serum bicarbonate level 22-28, and serum bicarbonate level >28. In-hospital mortality was then compared among these groups. The study cohort was then divided into two major groups based on the presence or absence of chronic kidney disease (CKD) on admission: CKD and non-CKD groups, respectively. Each group was then divided into three subgroups based on serum bicarbonate at presentation: low bicarbonate (<22 mEq/l), normal bicarbonate (22-28 mEq/l), and high bicarbonate (>28 mEq/l). The mortality in the three sub-groups within each group was then compared using a Chi-square test.
To calculate the adjusted predictive value of serum bicarbonate, the study cohort was divided into three groups: group 1 (bicarbonate level <22 mEq/L), group 2 (bicarbonate level 22-28 mEq/L), and group 3 (bicarbonate level >28 mEq/L). The clinical outcomes in groups 1 and 3 were then compared to those in group 2. We used a multiple regression model adjusted for age, gender, SOFA score, and co-morbidities (CKD, chronic obstructive pulmonary disease (COPD), liver cirrhosis). Based on the distribution shown in plots ( Figure 1 and Figure 2), we conducted Zero-Inflated Poisson regression models to look at the hospitalfree days and ICU-free days as clinical outcomes.

FIGURE 2: ICU-free days distribution plot
Based on the distribution shown in plots ( Figure 3 and Figure 4), we first transformed the outcome variables by dividing the ventilator-free days and pressor-free days by 28. We then conducted Zero-One Inflated Beta regression models to look at the impacts of bicarbonate on these clinical outcomes.

Results
Patient characteristics are given in Table 1.  In the final cohort of 4176 patients, irrespective of the presence or absence of CKD ( Table 2), in-hospital mortality was significantly higher in the patients with low serum bicarbonate compared to normal or high serum bicarbonate (p-value<0.0001). Additionally, noted in the low bicarbonate subgroup of both the CKD and non-CKD groups were significantly high SOFA score, low mean arterial pressure, low temperature, low pH, low albumin, high lactate, low estimated glomerular filtration rate (eGFR), and high anion gap. (Table 1 ).  When the final cohort was divided into CKD and non-CKD groups ( Table 3), the low bicarbonate sub-group had higher in-hospital mortality compared to the normal bicarbonate sub-group in both the CKD (p=0.013) and non-CKD (p<0.001) groups. In comparing the low bicarbonate sub-group to the high bicarbonate subgroup, this difference was statistically significant only in the non-CKD group (p<0.001). For secondary outcomes, in both the CKD and non-CKD groups, more patients with low serum bicarbonate levels required vasopressors and these patients had significantly fewer pressor-free days compared to patients with normal or high bicarbonate levels (p-value<0.001). Regarding other secondary outcomes, only patients in the low bicarbonate sub-group of the non-CKD group had significantly fewer hospital-free days, ICU-free days, and ventilator-free days (p-value<0.001), and no statistically significant difference was seen in the three subgroups with CKD.

CKD: Chronic Kidney Disease
Using a multivariate logistic regression model adjusted for age, gender, CKD, COPD, liver cirrhosis, and SOFA score, the primary outcome was compared between group 1 (bicarbonate level <22 mEq/L), group 2 (bicarbonate level 22-28 mEq/L), and group 3 (bicarbonate level >28 mEq/L). As we found a significant interaction between the SOFA score and low bicarbonate, this interaction term was then introduced into the regression model. We found that when adjusted for these variables, the association of low serum bicarbonate with increased in-hospital mortality was seen only when the SOFA score was ≥ 9. When bicarbonate was higher than 28, the mortality rate was higher but not statistically significant ( Table 4).

Mortality
Hospital-free days ICU-free days Ventilator-free days Pressor-free days  In addition, the interaction term between the low bicarbonate group and the SOFA score was significantly positive, suggesting that the impact of the SOFA score on mortality was higher among patients in group 1.
The odds ratio of the SOFA score for the normal bicarbonate group was 1.236 ( Table 5), suggesting a 23.6 % increase in the odds of mortality for every unit increase in the SOFA score. The odds ratio of SOFA score for the low bicarbonate group was 1.300 (exp (0.212+0.05) ( Table 4), suggesting a 30.0% increase in the odds of mortality for every unit increase in the SOFA score.  The estimation coefficients of the Poisson regression model for hospital-free days are listed in Table 4. The interaction between group 1 and SOFA score was not significant and was not included for further analysis. For the non-zero portion, Group 1 had significantly higher hospital-free days, while group 3 had significantly lower hospital-free days. The incident rate ratio for group 1 was 1.039 and that for group 3 was 0.921.

Mortality
The estimation coefficients of the Poisson regression model for ICU-free days are listed in Table 4. For the non-zero portion, patients in group 3 (bicarbonate > 28) had significantly fewer ICU-free days (p<0.01). The interaction term of bicarbonate group 1 and SOFA score was significantly negative, suggesting that the difference in ICU-free days between the low bicarbonate group and the normal bicarbonate group would depend on patients' SOFA score. When the SOFA score was ≥ 11 (0.141/0.013; Table 4), patients in group 1 had fewer ICU-free days. This also suggested that the impact of SOFA score on the ICU-free days was more significant on patients in group 1 compared to group 2. For group 2, for every unit increase in the SOFA score, there was a 1.8 % decrease in ICU-free days ( Table 4). For group 1, for every unit increase in the SOFA score, there was a 3.1 % decrease in ICU-free days.
The estimation coefficients of the Beta regression model portion for ventilator-free days are listed in Table  4. The interaction term of bicarbonate group 1 and SOFA score was negative suggesting that the difference in ventilator-free days between group 1 and group 2 was dependent on the SOFA score, although the difference was not statistically significant (p=0.052). When the SOFA score was ≥ 12 (0.298/0.025; Table 4), patients in group 1 had fewer transformed ventilator-free days suggesting that the impact of SOFA score on the ventilator-free days was larger on patients in group 1 compared to group 2. For patients in group 2, with every unit increase in SOFA score, there was a 4.5 % decrease in the average transformed ventilator-free days. For patients in group 1, with every unit increase in SOFA score, there was a 6.9 % decrease in the average transformed ventilator-free days.
The estimation coefficients of the Beta regression model portion for pressor-free days are listed in Table 4.
The interaction between group 1 and SOFA score was not significant and was not included for further analysis. We found that there was no significant difference in pressor-free days between the groups.

Discussion
Sepsis and its complications remain the leading cause of morbidity and mortality worldwide [3,13]. Metabolic acidosis [14,15], elevated lactate [16][17][18][19][20], and acute kidney injury [21][22][23] are known predictors of worse clinical outcomes in patients with sepsis. While these factors have been extensively investigated for their role in sepsis, the predictive value of serum bicarbonate in sepsis remains unexplored. Recent studies have shown that serum bicarbonate and lactate do not correlate, suggesting that serum bicarbonate cannot be a surrogate for lactate and vice versa [10][11][12]24]. Moreover, with a significant proportion of hospitals in rural areas operating without access to rapid lactate levels [25,26], knowing the predictive value of serum bicarbonate in sepsis will always help in risk stratification.
In this retrospective analysis including 4176 patients, we investigated the predictive value of serum bicarbonate in sepsis. Our study suggests that serum bicarbonate level at presentation can predict inhospital mortality as well as the ICU-free days, ventilator-free days, and hospital-free days. In our unadjusted analysis, serum bicarbonate level < 22 mEq/l had a significant association with elevated inhospital mortality; however, when controlled for confounders, the difference in the in-hospital mortality was not statistically significant except for when the SOFA score was ≥9. When the SOFA score was ≥9, the mortality rate among patients with serum bicarbonate levels < 22meq/l was significantly higher. Similarly, patients with low serum bicarbonate (< 22 mEq/l) had significantly fewer ICU-free days when the SOFA score was ≥11 and significantly fewer ventilator-free days when the SOFA score was ≥12. In contrast, patients with low serum bicarbonate had significantly more hospital-free days while the patients with high serum bicarbonate had fewer hospital-free days compared to patients with normal serum bicarbonate. It was not clear from this study, however, why the patients with low serum bicarbonate had higher hospital-free days.
An interesting finding of our study was the interaction between SOFA score and serum bicarbonate at presentation suggesting that the predictive value of SOFA score for in-hospital mortality, ICU-free days, ventilator-free days, and hospital-free days was higher among patients with low serum bicarbonate. This brings up a valid question if we should incorporate serum bicarbonate levels into the SOFA score to predict clinical outcomes.
Increasing attention has been dedicated to the utilization of sodium bicarbonate in the treatment of sepsis. Current guidelines [13] recommend against its use based on two small cohort studies [27,28]. However, a few recent studies suggested improved mortality with the administration of sodium bicarbonate in a subset of patients with acute kidney injury [29,30] based on which, a weak recommendation was made in favor of sodium bicarbonate therapy in a subset of patients with septic shock, severe metabolic acidosis (pH ≤ 7.2), and acute kidney injury (AKI) (Acute Kidney Injury Network (AKIN) score 2 or 3). Of note, patients in this study were included if their enrollment SOFA score exceeded 4. Based on our study, low serum bicarbonate had a worse clinical outcome only when the SOFA score was ≥9. With these findings, we propose that the clinical outcomes might be better if the patient population is appropriately chosen with a SOFA score of ≥9 for further studies involving the administration of bicarbonate in patients with sepsis.
Given the retrospective nature of our study, multiple limitations exist. The limitations of our study include the fact that (a) it was a single-center study, (b) the study had a retrospective design, and (c) the reason behind the high bicarbonate level (>28 mEq/L) was not investigated.

Conclusions
Given the widespread availability of serum bicarbonate levels in health care facilities around the globe, we believe that the utilization of serum bicarbonate levels at presentation along with other markers of sepsis can help in the prognostication of these patients with sepsis. Also noted in our study is that the predictive value of the SOFA score was dependent on the serum bicarbonate level at presentation. Based on this result, we believe that incorporating serum bicarbonate level in the SOFA score might increase the predictive value of the SOFA score. We propose that further prospective studies be performed to evaluate the utilization of serum bicarbonate for these purposes.

Additional Information Disclosures
Human subjects: Consent was obtained or waived by all participants in this study. University of Kentucky Institutional Review Board issued approval 47751. Animal subjects: All authors have confirmed that this study did not involve animal subjects or tissue. Conflicts of interest: In compliance with the ICMJE uniform disclosure form, all authors declare the following: Payment/services info: All authors have declared that no financial support was received from any organization for the submitted work. Financial relationships: All authors have declared that they have no financial relationships at present or within the previous three years with any organizations that might have an interest in the submitted work. Other relationships: All authors have declared that there are no other relationships or activities that could appear to have influenced the submitted work.