Metabolic acidosis (arterial pH<7.3) is observed in about 8% of intensive care unit (ICU) admissions1 and is associated with particularly high mortality (up to 57%).1,2 The etiology of severe metabolic acidosis typically includes tissue hypoxemia, shock, diabetic ketoacidosis, hepatic and renal failure, and intoxications.2 Metabolic acidosis has numerous deleterious consequences on various physiological systems, including enzyme-/protein functionality, tissue metabolism, and increased production of nitric oxide leading to vasodilatation,3 which may further amplify organ dysfunction. Furthermore, acidosis is associated with reduced adrenoceptor numbers on cellular surfaces,3 depression of myocardial function including contractility and relaxation,3,4 cardiac arrhythmia,3 and a shift of the oxyhemoglobin dissociation curve.3

Previous data show that low serum bicarbonate levels predict mortality at ICU admission in e.g. cardiogenic shock patients.5,6 Intravenous sodium bicarbonate can be applied to rapidly correct metabolic acidosis and reports indicate that about two thirds of North American critical care specialists prescribe sodium bicarbonate for this purpose.7 However, it remains uncertain whether metabolic acidosis should be corrected using sodium bicarbonate (or other buffers).8,9 Importantly, there are concerns that sodium bicarbonate therapy could lead to sodium and/or fluid overload, increased lactate and/or carbon dioxide production, and/or decreased ionized serum calcium.10 Thus, sodium bicarbonate might theoretically worsen the outcome of affected patients.11

Current evidence on whether sodium bicarbonate impacts on mortality in the critically ill is controversially discussed, with sparse data from mostly retrospective studies12–14 available.

As most of the available studies are performed in acidotic patients with sepsis13,15 or hyperlactatemia,12,16 we thus embarked to perform a retrospective analysis with the primary objective to investigate whether infusion of sodium-bicarbonate impacts on all-cause mortality in critically ill patients with all-cause metabolic acidosis including but not restricted to septic and lactic acidosis.

A total of 18,754 admissions of ICU patients were screened in the study interval. After removal of duplicates and patients with incomplete data sets, 16,966 data sets remained. Of these, 749 patients received sodium bicarbonate (SB group) during ICU stay. We performed a 1:3 pairing of patients with regard to age, gender and APACHE II score at ICU admission to narrow further data search on patients with comparable disease severity. For respective n=2165 patients, further data collection was performed. While exploring the data, we removed n=69 very extreme outliers with aberrant values (eventually data entry mistakes) in ICU length of stay and very high vasopressor support in addition to patients without acidosis (pH>7.3, n=1125). 441 (SB) and 530 (noSB) patients remained in the final “crude‿ dataset. A Consort flow chart is given (Fig. 1).

Figure 1.

Flowchart of patient selection.

(0.53MB).

Patient characteristics of this “crude‿ data set are given in Table 1. Median dose of sodium bicarbonate 8.4% given in the SB-group was 100ml (interquartile range [IQR] 100–242ml). Patients received SB after a median time of 5.14h following ICU admission (IQR 14.64h). The SB-group had lower APACHE II scores than the noSB-group (median 30; IQR 11 vs. 32; IQR 19; p<0.001). Further, statistically significant differences were observed between groups regarding pH (SB median 7.17; IQR 0.15] vs. noSB 7.23; IQR0.09; p<0.01), arterial HCO3− (SB median 12.4mmol/l; IQR 8.11] vs. noSB 17.7mmol/l; IQR 5.28; p<0.001) and additional acidosis-related indices (Table 1). Patients in the SB group differed significantly from nonSB patients in regard to admission diagnosis such as sepsis, metabolic disorders, intoxications, need for vasoactives and blood products, and need for RRT (all p<0.05). ICU mortality was higher in the sodium-bicarbonate group (SB 46% vs noSB 31%; p<.001) (Table 1). Median time to death in non-survivors was 7.66 days (IQR 279.3days) in the noSB-Group vs. 3.32 days in the SB-group (IQR 37.53days) (p<.001).

The ORs for the ATT with regard to (all-cause) ICU mortality and (all-cause) in-hospital excess mortality (mortality beyond ICU mortality) in the “crude‿ dataset adjusted for SB treatment status, sex, age, BMI and APACHE scores were 1.56 (95% confidence interval [CI] 1.17–2.07; p=.002) resp. 0.67 (95%CI 0.43–1.03; p=.07). The ATT OR for excess mortality at 30 days, between day 31 and 364 and over one year are 0.52 (95%CI 0.32–0.84; p=.007), 1.31 (95%CI 0.74–2.31; p=.357) and 0.72 (95%CI 0.47–1.09; p=.121) respectively (Supplement Table 1).

Propensity score analysis

We constructed the PS model that fitted to 971 observations and explained 34% of variability in treatment prescription. While dropping 615 observations (described in Suppl. Table 2), we achieved a sound matching between the treatment and control groups (Suppl. Fig. 2.).

Using propensity scores, we obtained a dataset with 348 patients equally split between the control and sodium bicarbonate groups. Patient characteristics are given (Fig. 1, Table 2). Median cumulative dose of sodium bicarbonate was 100ml (IQR 100, 200) applied after a median of 8.77h following ICU admission (IQR 2.56, 23.51). ICU-mortality, in-hospital excess mortality and excess mortality after hospital discharge as well as median time-to-death did not differ significantly (p for all >.05) (Table 2). Median hospital LOS was longer in the SB group than in the noSB group (9 days; IQR 17.2 days vs. 6 days; IQR16days; p=.01)

Table 2.

Patient demographics, ICU treatment, and follow-up data of propensity score matched patient population (n=348).

Variable	n(no SB)=174	n(SB)=174	p
Gender	Female: 55 (32%)	Female: 50 (29%)	.64
BMI (kg/m2)	25.5 [6.25]	26.83 [7.91]	.2
Age (years)	68 [16]	68 [17]	.94
APACHE II	32 [8]	28 [11]	<.001
DG Cardiovascular	68 (39%)	55 (32%)	.18
DG Respiratory	18 (10%)	33 (19%)	.03
DG Gastrointestinal	27 (16%)	26 (15%)	1
DG Neurological	15 (9%)	10 (6%)	.41
DG Trauma incl. TBI	9 (5%)	5 (3%)	.41
DG Metabolic & Intox	10 (6%)	13 (7%)	.67
DG Hemic disease	3 (2%)	3 (2%)	1
DG Renal	4 (2%)	2 (1%)	.68
DG Sepsis	15 (9%)	18 (10%)	.71
apH	7.23 [0.08]	7.23 [0.09]	1
aHCO3–(mmol/L)	15.1 [3.4]	15.25 [3.7]	1
aLactate (mmol/L)	1 [0.8]	1 [1.0]	.45
apCO2(mmHg)	28 [6.1]	28.1 [6]	.99
BE	−11 [−4.1]	−11.25 [−4]	.17
Hc	0.27 [0.07]	0.26 [0.07]	.98
Na (mmol/L)	134 [4.2]	134 [6]	.98
K (mmol/L)	3.6 [0.7]	3.6 [0.8]	.98
CreaS (mmol/L)	118 [84.5]	125 [131]	<.001
Adrenaline (y/n)	76 (44%)	85 (49%)	.39
Adrenaline CD (μg)	2675.07 [11,215.1]	5956.06 [11,098.4]	.13
Noradrenaline (y/n)	109 (63%)	135 (78%)	<.001
Noradrenaline CD (μg)	2471.82 [6512.9]	3330.52 [13,183.4]	.09
Ventilation (y/n)	126 (72%)	111 (64%)	.11
Blood Products (y/n)	91 (52%)	111 (64%)	.04
RRT (y/n)	44 (25%)	43 (25%)	1
Length of ICU stay (days)	2.65 [4.2]	2.87 [5.1]	.54
Length of hospital stay (days)	6 [16]	9 [17.2]	.01
ICU mortality	55 (32%)	72 (41%)	.07
InHospital excess mortality	22 (13%)	18 (10%)	.61
30d excess mortality	20 (11%)	11 (6%)	.13
Excess mortality d31–365	12 (7%)	15 (9%)	.69
Excess mortality at last follow upa	26 (15%)	23 (13%)	.76
Time to death (days)	5.22 [274.9]	6.25 [47.3]	.91

Median values [interquartile ranges] or counts (percentages) are given. ‘p’ refers to Kolmogorov–Smirnov test for continuous variables, or to chi-squared test for categorical variables (for the null hypothesis of equal categorical frequencies between two groups). Laboratory data are at admission to ICU. DG, Diagnosis Group is referring to the APACHE IV-classification of diagnosis ad admission with a separate sepsis group (all groups include operative and non-operative diagnoses); Cardiovasc., Cardiovascular; Gastroint., Gastrointestinal; Haemat., Hematological; NoSB, non sodium bicarbonate recipients; SB, sodium-bicarbonate recipients; APACHE II, Acute Physiology and Chronic Health Evaluation II Score; Hc, hematocrit; Na, sodium; K, potassium; CreaS, serum creatinine; TBI, traumatic brain injury; CD, cumulative dose; RRT, renal replacement therapy; ICU, intensive care unit.

A p value < 0.05 was considered statistically significant.

a

Last follow up: September 2019. Please note that excess mortality is given for in-hospital mortality and mortality at d30, d31–365 and at last follow up (mortality is excess mortality plus ICU mortality).

In the matched data set, we observed no association of sodium bicarbonate with ICU-mortality or death outcomes neither in the ATE OR nor in the ATT OR (Table 3).

Table 3.

Odds ratios (OR) for treatment effects for mortality in the matched dataset.

	ATE (matched sample), n=348	p-Value	ATT (matched sample), n=348	p-Value
ICU mortality	1.08 (0.991, 1.17)	.08	1.09 (0.987, 1.2)	.09
In Hospital excess mortality	0.982 (0.924, 1.04)	.55	0.984 (0.92, 1.05)	.62
30d excess mortality	0.96 (0.91, 1.01)	.13	0.958 (0.903, 1.02)	.16
Excess mortality d31–365	1.02 (0.968, 1.08)	.44	1 (0.94, 1.06)	.98
Excess mortality at last follow-upa	0.994 (0.935, 1.06)	.82	0.993 (0.928, 1.06)	.83

Reported are the odds ratios (OR) for the treatment effects for mortality at various stages in presence of sodium bicarbonate treatment. Estimated by logit logistic regressions, regressing mortality outcome on SB treatment status, sex, age, BMI and APACHE.

a

Last follow up was in September 2019. ATE, Average Treatment Effect; ATT, Average Treatment effect in Treated. Please note that excess mortality is given for in-hospital mortality and mortality at d30, d31–365 and at last follow up (mortality is excess mortality plus ICU mortality).

As for the majority of patients (97.5%), ICU length of stay was below 20 days, we estimated crude Kaplan–Meier survival curves up to 20 days (i.e. if death did not occur after 20 days from admission, we reported it as no death event during 20 days). The Kaplan–Meier estimate of the “crude‿ data showed that the patients who received the sodium bicarbonate treatment had considerably lower survival probability (log-rank test: p=.005; Fig. 2), while using the matched data, there was no significant difference (stratified log-rank test p=.53). Applying the ATE and ATT weights in the “crude‿ sample (n=971) did not result in a difference in probability of survival (stratified log-rank test p=.91 and .98, respectively) (Fig. 2). Estimating Kaplan–Meier survival curves for the whole follow-up time showed no different results (Suppl. Fig. 3).

Figure 2.

Kaplan–Meier survival curves (20 days) obtained using different propensity score methods. In the top-left panel crude Kaplan–Meier survival curves for treated and untreated subjects in the “crude‿ sample (n=971; log-rank test: p=0.005). In the top-right panel Kaplan–Meier survival curves for treated and untreated subjects in the propensity score matched sample (n=348 stratified log-rank test p=0.53). In the bottom-left and bottom-right panels survival curves in the sample weighted using the ATE weights (n=971) and the sample weighted using the ATT weights (n=971) are reported (adjusted log-rank test with p-values 0.91 and 0.98). ATE, Average Treatment Effect; ATT, Average Treatment effect in Treated.

(0.36MB).

Cox regression modeling did not reveal a significant difference in survival at 20 days (Hazard ratio [HR] (multivariable, matched sample, n=348: 1.16; 95%CI 0.86–1.56; p=.33) (Table 4).

Table 4.

Hazard ratios estimated with Cox regressions.

Predictor for 20 day-survival	HR (univariable, “crude‿ sample n=971)	HR (multivariable, “crude‿ sample n=971)	HR (multivariable, matched sample n=348)	HR (multivariable, full sample with ATE weights, n=971)	HR (multivariable, full sample with ATT weights, n=971)
Sodium bicarbonate treatment	1.30 (1.08–1.55, p=.005)	1.08 (0.87–1.34, p=.50)	1.16 (0.86–1.56, p=.33)	1.07 (0.93–1.22, p=.35)	1.05 (0.87–1.27, p=.58)
Propensity score	1.17 (1.10–1.23, p<.001)	1.15 (1.07–1.23, p<.001)	–	1.14 (1.09–1.19, p<.001)	1.18 (1.09–1.28, p<.001)
APACHE II	1.03 (1.02–1.05, p<.001)	1.03 (1.02–1.05, p<.001)	1.03 (1.00–1.05, p=.05)	1.02 (1.01–1.03, p<.001)	1.02 (1.01–1.04, p=.002)

Hazard ratios (HR with 95% confidence intervals) of predictor variables are given. ATE, Average Treatment Effect; ATT, Average Treatment effect in Treated.

A p value < 0.05 was considered statistically significant.

In this monocentric propensity score matched analysis, we observed no influence of sodium bicarbonate infusion on ICU-mortality in critically ill patients with metabolic acidosis.

In critically ill patients, it appears that sodium bicarbonate infusion is mostly prescribed as symptomatic measure aiming to influence effects of severe metabolic acidosis (e.g. on the cardiovascular system) until a more causal therapy (e.g. improvement of tissue oxygenation, source control strategies, and/or antibiotics) becomes effective. Currently, there is uncertainty whether SB infusion would impact on patient survival beneficially or whether it would e.g. induce sodium and/or fluid overload, increase paCO2 production, and would potentially be detrimental.10 Previous studies supported this assumption and showed that SB administration in metabolic acidosis may actually increase mortality.16 In contrast, the BICAR-ICU-trial12 and a retrospective study of Zhang et al.13 showed no impact on mortality when sodium bicarbonate was administered to critically ill patients with metabolic acidosis. In line with the aforementioned studies, we did not observe an increased mortality after sodium bicarbonate treatment. In contrast to the BICAR-study12 and the study of Zhang,13 however, we likely studied a rather broad population of ICU patients with metabolic acidosis as we included not only patients with acidosis and hyperlactatemia12 and/or septic patients.13

In addition, and in accordance with previous data,12,13 our study did not observe mortality benefits in patients with sodium bicarbonate infusion. However, although speculative, e.g. negative SB-induced effects could theoretically be counterweighted by beneficial effects such as improved cardiac contractility and/or vascular responsiveness to catecholamine therapy,3 without effects on the mortality endpoints. Further, effects (whether beneficial or not) could theoretically be short- rather than long-lived.

Importantly, about 50% or more patients who received sodium bicarbonate did not have a “severe‿ metabolic acidosis with pH≤7.21 (Tables 1 and 3). However, when compared to other investigations, our results (median pH of 7.17 in “crude‿ and 7.23 in matched groups) may be considered rather comparable to other retrospectives studies (e.g. Kim et al.: mean pH 7.244±0.168)16; Zhang et al.: minimum mean pH 7.16±0.1)13). Moreover, the median arterial pCO2 in the matched groups was 28mmHg. Therefore, the observed pH at admission may be respiratory compensated and true pH in fact considerably lower.

Further, one might assume that SB patients would have had a higher risk of death. However, in our subset of ICU patients, SB treated patients had a slightly, but statistically significant, lower baseline APACHE II score, whereas ICU mortality tended to be increased. This stands in contrast to previous data13,14 but may theoretically underline that metabolic acidosis is additionally associated with a higher mortality.

A number of important additional limitations of our analysis deserve discussion. First, limitations arise from the retrospective, single-center, and explorative design of this study and all respective inherent limitations apply that are driven by study design. In this retrospective study, even if mortality analyses were adjusted for typical potential confounders, our data may theoretically be subject to some degree of unmeasured confounding. Additionally, by definition, the matched sample analysis required extensive exclusion of patients of the overall sample, which may per se introduce (e.g. selection) bias. Moreover, this may point to heterogeneity of the total cohort and may be one reason why respective patients were particularly challenging to match. By deliberately curtailing our ICU data set to a subset of patients and by removing some outliers that aimed for a comparable subset of ICU patients, it may theoretically be possible that some degree of selection bias was introduced. Second, we only considered intravenous sodium bicarbonate formulations of 8.4% for analysis and might have missed other preparations, including e.g. chronic oral bicarbonate use in few patients with chronic kidney disease. Third, mortality may not be an optimal endpoint to examine effects of an early single intervention in a critically ill population, which may also underline our interest in early outcomes of our cohort (i.e. 20 days of ICU stay). It appears that endpoints related to therapy-induced effects (e.g. increase in pH) might be particular interesting in subsequent investigations. Fourth, our retrospective study design made it impossible to identify the medical reason for the (independent treating) physician to prescribe sodium bicarbonate and for the timing of the prescription. We can thus not exclude with certainty that SB was also given for aetiologies of metabolic acidosis with a better prognosis (e.g. diabetic ketoacidosis, intoxications) or other reasons than for metabolic acidosis only, e.g. in cases of additional rhabdomyolysis. Although the current analysis may be one of the largest investigations available, it appeared that the sample size was too limited to conclude back on effects of SB dose and/or exact timing as well as to conclude on subgroups of critically ill patients (e.g. acute kidney injury [AKI] or sepsis patients, patient's post-surgical interventions). Thus, the presented analysis may theoretically be “underpowered‿. Fifth, we did not evaluate other important covariates such as interventions during the ICU stay (e.g. emergency hemodialysis) as well as respective urgency nor the impact of fluid intake other than sodium bicarbonate including fluid balance. Sixth, the presented mortality data reflects all-cause mortality and we are unable to conclude back on specific acidosis and/or treatment-related adverse effects nor on different causes of death. Seventh, as our retrospective analysis made it impossible to conclude back on acid base analysis after sodium bicarbonate infusion we cannot describe any influence of sodium bicarbonate on the pH and the bicarbonate level after treatment. Moreover, we refrained from estimating dose-effects of the sodium bicarbonate given.

In this propensity score matched analysis, intravenous administration of sodium bicarbonate did not appear to effect on mortality in ICU patients with metabolic acidosis. Additional prospective controlled clinical investigations seem required to further determine potential effects of sodium bicarbonate infusion in subgroups of patients with metabolic acidosis, including potential high-risk groups, such as critically ill patients with AKI.

JW: Designed the study, performed data collection and assessment, wrote the first draft, coordinated the input of all authors, supervised the study and revised the manuscript for important intellectual content.

BH: Designed the study, performed data collection and assessment, wrote the first draft, coordinated the input of all authors, and revised the manuscript for important intellectual content.

LC: Contributed to data interpretation, revised the manuscript for important intellectual content.

II: Performed all statistical analyses, revised the manuscript for important intellectual content.

CAP: Co-designed the study, contributed to data interpretation, supervised the study, and revised the manuscript for important intellectual content.

JCS: Co-designed the study, contributed to data interpretation, supervised the study and revised the manuscript for important intellectual content.

All authors approved the final version of the manuscript.

Full departmental disclosure: CAP, LC, and JCS report grants from Orion Pharma, Abbott Nutrition International, B. Braun Medical AG, CSEM AG, Edwards Lifesciences Services GmbH, Kenta Biotech Ltd, Maquet Critical Care AB, Omnicare Clinical Research AG, Nestle, Pierre Fabre Pharma AG, Pfizer, Bard Medica S.A., Abbott AG, Anandic Medical Systems, Pan Gas AG Healthcare, Bracco, Hamilton Medical AG, Fresenius Kabi, Getinge Group Maquet AG, Dräger AG, Teleflex Medical GmbH, Glaxo Smith Kline, Merck Sharp and Dohme AG, Eli Lilly and Company, Baxter, Astellas, Astra Zeneca, CSL Behring, Novartis, Covidien, Hemotune, Phagenesis, Philips Medical, Prolong Pharmaceuticals and Nycomed outside the submitted work. The money went into departmental funds. No personal financial gain applied.

II is affiliated with CTU Bern, University of Bern, which has a staff policy of not accepting honoraria or consultancy fees. However, CTU Bern is involved in design, conduct, or analysis of clinical studies funded by not-for-profit and for-profit organizations. In particular, pharmaceutical and medical device companies provide direct funding to some of these studies. For an up-to-date list of CTU Bern's conflicts of interest see http://www.ctu.unibe.ch/research/declaration_of_interest/index_eng.html

The additional authors declare that they have no competing interests.