Background
Derangements in acid–base balance are frequently observed in critically ill patients at the intensive care unit (ICU) and present in various patterns [
1‐
4]. Severe acid–base disorders, especially metabolic acidosis, have been associated with increased mortality [
5,
6]. As a consequence, acid–base status in critically ill patients with various disease entities has been extensively studied.
Yet, only a few studies assessed the impact of underlying chronic liver disease on acid–base equilibrium in critical illness [
7,
8]. While a balance of offsetting acidifying and alkalinizing metabolic acid–base disorders with a resulting equilibrated acid–base status has been described in stable cirrhosis [
9], severe derangements with resulting net acidosis owing to hyperchloremic, dilutional and lactic acidosis were observed when cirrhosis was accompanied by critical illness [
7,
8]. Acute liver failure (ALF) is characterized by a different acid–base pattern with dramatically increased lactate levels [
10]. The acidifying effect of this increase in lactate was neutralized by hypoalbuminemia in non-paracetamol-induced ALF [
11].
Despite advantages in intensive care medicine, which have led to an improved outcome over the last decade [
12], mortality in cirrhotic patients admitted to ICU is still high [
13‐
15]. Measurement and knowledge of specific acid–base patterns and their implications in critically ill patients with liver cirrhosis may help to improve patient management, especially in the ICU setting [
16]. However, to our knowledge, the acid–base profile of critically ill cirrhotic patients with acute-on-chronic liver disease (ACLF) has not been compared to critically ill patients without acute or chronic liver disease. Most information on the acid–base status of critically ill patients with cirrhosis was obtained by comparing these patients with healthy controls [
8]. Yet, part of metabolic disturbances in critically ill patients with liver cirrhosis may be attributable to critical illness per se, rather than to the presence of chronic liver disease.
The aim of this study was to assess acid–base patterns of critically ill patients with liver cirrhosis and ACLF, respectively, in comparison with critically ill patients without acute or chronic liver disease.
Methods
Patients
All patients admitted to 3 medical ICUs at the Medical University of Vienna between July 2012 and August 2014 were screened for inclusion in the study. For the present study, only patients who had arterial blood samples drawn within 4 h after ICU admission were eligible for inclusion. Patients with acute liver injury in the absence of chronic liver disease were excluded. One hundred and seventy-eight patients with liver cirrhosis were identified as eligible for inclusion. The control group of 178 critically ill patients without acute or chronic liver disease was selected by propensity score matching (PSM).
On admission, Simplified Acute Physiology Score II (SAPS II) [
17], SOFA [
18], infections and organ dysfunctions were documented.
All patients were screened for the presence of acute kidney injury (AKI) defined by urine output and serum creatinine according to the Kidney Disease: Improving Global Outcomes (KDIGO) Clinical Practice Guidelines for Acute Kidney Injury [
19].
The presence of liver cirrhosis was defined by a combination of characteristic clinical (ascites, caput medusae, spider angiomata, etc.), laboratory and radiological findings (typical morphological changes of the liver, sings of portal hypertension, etc., in ultrasonography or computed tomography scanning), or via histology, if available. ACLF was identified and graded according to recommendations of the chronic liver failure (CLIF) consortium of the European Association for the Study of the Liver (EASL) [
20]. CLIF-SOFA score [
20] and CLIF-C ACLF score [
21] were calculated. Septic shock was defined according to the recommendations of the Surviving Sepsis Campaign [
22].
Twenty-eight-day mortality and 1-year mortality were assessed on site or by contacting the patient or the attending physician, respectively.
This study is based on a post hoc analysis of prospectively collected data [
23]. The Ethics Committee of the Medical University of Vienna waived the need for informed consent due to the observational character of this study.
Sampling and blood analysis
On admission, arterial blood samples were collected from arterial or femoral artery and parameters for the assessment of acid–base status were instantly measured.
pH, partial pressure of carbon dioxide (PaCO2), ionized calcium (Ca2+) and lactate were measured with a blood gas analyzer (ABL 725; Radiometer, Copenhagen, Denmark). Samples of separated plasma were analyzed for concentrations of sodium (Na+), potassium (K+), chloride (Cl−), magnesium (Mg2+), inorganic phosphate (Pi), albumin (Alb), plasma creatinine, blood urea nitrogen (BUN), aspartate aminotransferase (AST) and alanine aminotransferase (ALT) by a fully automated analyzer (Hitachi 917; Roche Diagnostics GmbH, Mannheim, Germany). Na+ and Cl− were measured using ion-selective electrodes. Lactate was measured with an amperometric electrode.
Acid–base analysis
Arterial concentration of bicarbonate (HCO3
−) was calculated from measured pH and PaCO
2 values according to the Henderson–Hasselbalch equation [
24,
25]. Base excess (BE) was calculated according to the formulae by Siggaard-Andersen [
24‐
26].
Quantitative physical–chemical analysis was performed using Stewart’s biophysical methods [
27], modified by Figge and colleagues [
28].
Apparent strong ion difference (SIDa) was calculated:
$$\begin{aligned} {\text{SIDa}} &= {\text{Na}}^{ + } + {\text{K}}^{ + } + 2 \times {\text{Mg}}^{2 + } + 2 \times {\text{Ca}}^{2 + } - {\text{Cl}}^{ - } - {\text{lactate}} \\ & \left( {{\text{SIDa}}\;{\text{in}}\;{\text{mEq/l;}}\;{\text{all}}\;{\text{concentrations}}\;{\text{in}}\;{\text{mmol/l}}} \right) \\ \end{aligned}$$
Effective strong ion difference (SIDe) was calculated in order to account for the role of weak acids [
29]:
$$\begin{aligned} {\text{SIDe}} & = 1000 \times 2.46 \times 10^{ - 11} \times \frac{{{\text{PaCO}}_{2} }}{{10^{{ - {\text{pH}}}} }} + {\text{Alb}} \times \left( {0.123 \times {\text{pH}} - 0.631} \right) + {\text{Pi}} \times \left( {0.309 \times {\text{pH}} - 0.469} \right) \\ & \left( {{\text{SIDe}}\;{\text{in}}\;{\text{mEq/l;}}\;{\text{PaCO}}_{ 2} \;{\text{in}}\;{\text{mmHg,}}\;{\text{Alb}}\;{\text{in}}\;{\text{g/l}}\;{\text{and}}\;{\text{Pi}}\;{\text{in}}\;{\text{mmol/l}}} \right) \\ \end{aligned}$$
The effect of unmeasured charges was quantified by the strong ion gap (SIG) [
30]:
$$\begin{aligned} {\text{SIG}} = {\text{SIDa}} - {\text{SIDe}} \hfill \\ \left( {{\text{all}}\;{\text{parameters}}\;{\text{in}}\;{\text{mEq/l}}} \right) \hfill \\ \end{aligned}$$
Based on the concept that BE can be altered by plasma dilution/concentration reflected by sodium concentration (BE
Na), changes of chloride (BE
Cl), albumin (BE
Alb), lactate (BE
Lac) and unmeasured anions (BE
UMA), the respective components contributing to BE were calculated according to Gilfix et al. [
31]. The detailed formulae for the BE subcomponents are shown in “
Appendix.”
Thus, total BE is calculated by the sum of the BE subcomponents:
$${\text{BE}} = {\text{BE}}_{\text{Na}} + {\text{BE}}_{\text{Cl}} + {\text{BE}}_{\text{Alb}} + {\text{BE}}_{\text{Lac}} + {\text{BE}}_{\text{UMA}}$$
Reference values were obtained from a historical cohort of healthy volunteers, as published elsewhere [
8]. Acidemia and alkalemia were defined by pH < 7.36 and > 7.44, respectively. HCO3
−< 22 and > 26 mmol/l, respectively, defined metabolic acidosis and alkalosis [
2]. Respiratory acidosis and alkalosis were identified by PaCO
2 > 45 and < 35 mmHg, respectively. BE
Na < − 5 and > 5 mmol/l defined dilutional acidosis and alkalosis, respectively. Hyperchloremic acidosis and hypochloremic alkalosis were defined by BE
Cl < − 5 and > 5 mmol/l, respectively. BE
Alb > 5 mmol/l identified hypoalbuminemic alkalosis. Lactic acidosis was defined by BE
Lac < − 1.1 mmol/l (calculated BE
Lac for lactate at the upper limit of normal) and metabolic acidosis owing to unmeasured anions by BE
UMA < − 5 mmol/l.
Statistical analysis
Data are presented as median and interquartile range (25–75% IQR), if not otherwise specified. PSM was used to minimize the confounding effect of severity of disease on acid–base status when comparing cirrhosis to non-cirrhosis patients. One-to-one PSM (1:1) was done by cirrhosis versus non-cirrhosis based on the following variables: SOFA score, need for mechanical ventilation and the presence of AKI. IBM SPSS 22 (with SPSS Python essentials and FUZZY extension command) was used for PSM. McNemar test was used for the comparison of binary and Wilcoxon’s signed-rank test for the comparison of metric variables between cirrhosis and matched controls. Nonparametric one-way ANOVA (Kruskal–Wallis test) with Dunn’s post hoc analysis was performed to assess differences in acid–base parameters between matched controls, cirrhosis patients without ACLF and ACLF patients. Within each group, comparisons were made using Chi-squared test or Mann–Whitney U test, as appropriate. Spearman’s rank correlation was used to assess correlations between metric variables. A receiver operating curve (ROC) analysis was performed, and the area under the ROC curve (AUROC) was calculated to evaluate the prognostic value of different metric variables. Impact of acid–base disorders on mortality was assessed using Cox regression. A p value < 0.05 is considered statistically significant. Statistical analysis was conducted using IBM SPSS Statistics version 22.
Discussion
Disturbances in acid–base equilibrium are common in critical illness [
16]. In this study, we demonstrate that critically ill patients with cirrhosis and ACLF, respectively, differentiate considerably from patients without hepatic impairment in terms of acid–base balance.
In accordance with earlier reports, we observed in our cohort a marked hyperchloremic acidosis with coexisting hypoalbuminemic alkalosis [
8,
9,
11]. This phenomenon, however, was not limited to patients with cirrhosis and should therefore not be considered an exclusive acid–base pattern of liver disease. Instead, this seems to be a characteristic pattern of critical illness per se [
3]. Yet, hypoalbuminemia and resulting alkalosis were most pronounced in patients with ACLF. However, the main distinguishing metabolic acid–base characteristic between critically ill patients with and without cirrhosis was a marked metabolic acidosis attributable to an increased lactate (and unmeasured anions). In cirrhosis, coexisting respiratory alkalosis partly compensated for metabolic acidosis, thereby resulting in almost normal pH values. However, respiratory alkalosis failed to compensate for net metabolic acidosis in patients with ACLF.
Increased lactate levels in critically ill patients can result from both increased production (e.g., tissue malperfusion, impaired cellular oxygen metabolism during sepsis, hypermetabolic states) and reduced lactate clearance (e.g., loss of functioning hepatocytes in acute hepatic injury or chronic liver disease) [
32‐
34]. The liver not only is a crucial player in the disposal of lactate, but may also become a net producer of lactate, especially during hepatic parenchymal hypoxia. Although lactic acidosis has been described in the literature in critical ill patients with cirrhosis [
7,
8], this is the first study investigating the association of metabolic disturbances with ACLF compared to a matched cohort of critically ill patients without liver disease. Indeed, the extent of lactic acidosis was directly associated with ACLF grade. Accordingly, lactic acidosis was present in almost 80% of all patients with ACLF grade III. Moreover, lactate levels were correlated with INR and bilirubin, thereby suggesting that lactate levels are directly related to liver function. Vasopressor support and severity of disease (as reflected by SOFA score) were also significantly associated with increased lactate levels. In sum, our data suggest that a combination of hepatic impairment and tissue hypoxia may contribute to lactic acidosis in critically ill patients with liver cirrhosis.
Great effort has been put in revealing the nature of unmeasured anions in critical illness [
2,
35‐
38]. Still, source and clinical implications of unmeasured anions are incompletely understood [
39,
40]. Recently, it was shown in a large cohort of critically ill patients that increased concentrations of unmeasured anions were independently associated with increased mortality [
41]. Citrate, acetate, fumarate,
α-ketoglutarate and urate have been identified as potential candidates contributing to acidosis associated with high SIG in hemorrhagic shock [
36]. Apart from states of shock, renal failure has been linked to increased levels of unmeasured anions in several studies [
8,
42,
43]. As compared to non-ACLF cirrhosis patients, the presence of ACLF was associated with an increase in unmeasured anions, as reflected by BE
UMA and SIG. Both variables were strongly associated with acute kidney injury. Patients with liver cirrhosis are especially susceptible to renal failure [
44‐
47], and renal impairment constitutes a central criterion for ACLF [
20]. In sum, our findings indicate that impairment of renal function, rather than “hepatic failure,” may be responsible for the increase in levels of unmeasured anions observed in patients with ACLF.
In the present study, metabolic acidosis and acidemia, respectively, were associated with increased 28-day mortality in liver cirrhosis. Accordingly, 28-day mortality rate was 91% in cirrhosis patients with arterial pH values < 7.2 and 86% in those with arterial HCO
3− values < 15 mmol/l. Lactic acidosis and acidosis attributable to unmeasured anions were identified as main contributors to acid–base imbalance in critically ill patients with liver cirrhosis. Earlier studies have challenged the prognostic value of unmeasured anions or lactate in critically ill patients [
40]. Yet, the relationship between lactate levels, unmeasured anions and mortality and poor outcome has been described multiply in the literature [
7,
8,
32,
33,
48], and lactate levels have recently been suggested as a parameter, indicating severity of disease in patients with chronic liver disease [
49]. In our critically ill cirrhosis patients, we observed a dramatic independent impact of both lactate and BE
UMA on 28-day mortality. Thus, acid–base status in critically ill patients with cirrhosis and ACLF, respectively, is an early and independent predictor of outcome (Fig.
2). By contrast, acid–base status was of poor prognostic value in our propensity score-matched controls. This may be attributable to the fact that our control patients were matched to critically ill cirrhosis patients, thereby resulting in the exclusion of less severely ill non-cirrhosis patients with better acid–base profiles and lower mortality rates.
This study has strengths and limitations. First, this is a post hoc analysis; however, our study comprises structured acid–base analyses from a large cohort of critically ill patients stratified according to the presence of liver cirrhosis. Second, this study was performed in patients admitted to the ICU. Thus, our findings may not entirely reflect acid–base status of cirrhotic patients treated at normal wards. However, our study also incorporates cirrhosis patients without ACLF and patients of all ACLF categories. Third, there are pros and cons of propensity score matching. In this study, we have decided to use propensity score-matched controls in order to minimize the confounding effect of severity of disease on acid–base balance. Although we were able to achieve good comparability, inherent differences between cirrhotic and non-cirrhotic patients affecting acid base balance cannot be entirely abolished by matching procedures. Moreover, the loss of heterogeneity (by selection of the most severely ill patients) hampers survival analyses in the control group. Fourth, residual confounding is, as always, a matter of concern and cannot be entirely excluded. Future studies should confirm these results and focus on therapeutic implications for patients with liver disease at the ICU.
Authors’ contributions
AD, TH, BS and VF participated in conception and design of the study. KRo, KRu, RB, CZ, PS and GH contributed to acquisition and interpretation of data. AD, TH and VF performed the statistical analysis. AD and TH drafted the manuscript. GCF, MT, BS and VF critically read and revised the manuscript for important intellectual content. All authors read and approved the final manuscript.