More than 20 yr ago, Stewart published a new theory of acid–base equilibrium, proposing three independent variables, the strong ion difference (SID), the total weak acids and the carbon dioxide tension (PCO2), as determinants of the hydrogen ion concentration [H+] [1,2].
SID is the sum of strong ions, which are sodium (Na+), potassium (K+), chloride (Cl−) and also, at physiological pH, lactate. In normal conditions, SID is +40 mEq L−1 . The total weak acids are proteins (mainly albumin) and phosphate.
According to Stewart's theory, six different acid–base abnormalities can be identified: (1) low SID acidosis, (2) high PaCO2 acidosis, (3) high total weak acid acidosis, (4) high SID alkalosis, (5) low PaCO2 alkalosis and (6) low total weak acids alkalosis. According to the law of electroneutrality, the sum of SID and anions from weak acids and CO2 should equal zero. However, this is not the case in certain pathological conditions, in which unmeasured anions and cations are present. For this reason, Kellum and colleagues  introduced the strong ion gap (SIG), the difference between SID and the sum of anions derived from CO2, phosphate and albumin. In normal settings, SIG is presumed to be zero. In critically ill patients, SIG can be greater than zero, indicating an excess of unmeasured anions (acidosis), or, rarely, less than zero, indicating an excess of unmeasured cations (alkalosis).
In the last decade, several experimental [5-8] and clinical investigations have been conducted which compare Stewart's and traditional theory in the interpretation of acid–base conditions, especially in critically ill patients [9-14]. The results have shown that Stewart's theory allows for the identification and quantification of complex acid–base aberration, where standard theory demonstrates normal physiology. Based on Stewart's studies, a third approach to the evaluation of acid–base disturbances has been proposed by Fencl and Leith [10,15] and Gilfix and colleagues , in which equations to correct the base excess (BE) for changes in sodium, chloride and albumin have been introduced.
The purposes of this prospective study are threefold. First, to investigate the occurrence of non-respiratory (metabolic) acid–base disturbances in critically ill patients and their trend over the first 3 ICU days; second, whether application of Stewart's theory offers advantages over the traditional theory in the diagnosis of acid–base metabolic disturbances; and third, whether variables derived from Stewart's and Fencl's methods are more strongly predictive of mortality than the traditional theory.
A prospective cohort study involving 136 patients admitted to the Institute of Anaesthesiology and Intensive Care, University of Brescia, Italy, between March 2000 and October 2001 was performed. The population consisted of surgical, medical and trauma patients requiring intensive care. Patients admitted in the ICU for monitoring or for elective surgery were excluded. The data collection for this study was approved by the Institutional Ethics Committee, which waived the need for informed consent.
Blood samples were collected starting from the first day of admission, between 7.00a.m. and 8.00a.m., from an arterial line and were analysed for arterial blood gases, electrolytes and proteins. PaCO2, pH, bicarbonate, BE and standard base-excess (SBE) were measured with an ammeter method (ABL 520; Radiometer, Copenhagen, Denmark). Na+, K+ and Cl− were measured with an indirect method (Dimension RXL; GMI, Inc., Ramsey, MN, USA); phosphorous, calcium and magnesium with a colorimetric reaction (Hitachi 717; GMI, Inc. Ramsey, MN, USA); proteins with an electrophoretic method (Alfa Biotec CTE 8000, Milan, Italy); and lactate with an enzymatic method (Dimension RXL).
Anion gap and Stewart's variables were calculated based on the following formulae :
A−, the dissociated fraction of weak acids, was calculated as
The components of BE were obtained from formulas in accord with Fencl and colleagues  and Gilfix and colleagues :
BE ≤ −3 mEq L−1, SID ≤ 38 mEq L−1, A− ≥ 18 mEq L−1 or SIG ≥ 3 mEq L−1 were considered indicative of metabolic acidosis; while BE ≥ 3 mEq L−1, SID ≥ 42 mEq L−1, A− ≤ 12 mEq L−1 or SIG ≤ −3 mEq L−1 were considered indicative of metabolic alkalosis. Lactate was deemed increased if found to be greater than 2.2 mEq L−1.
We used Ringer's Lactate as the main crystalloid for fluid resuscitation and we generally used normal saline for neurological patients. With respect to colloid solutions we used neither polygeline nor 4% albumin, but only starch.
In each patient, the acid–base status was analysed using both the traditional and Stewart's approaches as well as Fencl's method. Acid–base abnormalities were diagnosed by traditional and Stewart's method, while Fencl's variables were used solely for logistic regression.
We expressed discrete variables as counts (percentage) or median (range) and continuous variables as mean ± SD, unless otherwise stated. Categorical data were compared using the χ2-test or Fisher's exact test, if appropriate. Differences in the acid–base variables during the first 3 ICU days were analysed using analysis of variance. In order to analyse the independent predictors of the 28-day mortality, a multivariable stepwise forward logistic regression model was used, using the variables statistically different in the bivariate analysis, as well as those considered clinically relevant. Explanatory variables were entered in the model with a 0.15 level of significance . The area under the receiver operating characteristic (AUROC) curve was then calculated to assess the overall performance of the logistic regression model . Finally, in order to identify the cut-off value of independent covariates that best predicted the 28-day mortality, we used the point on the receiver operating characteristic (ROC) curve corresponding to the best compromise between sensitivity and specificity. We considered P < 0.05 to be statistically significant. The data were analysed with STATA 8.0 (Stata Corporation, College Station, TX, USA).
In all, 136 patients were enrolled and a total of 380 samples analysed. The mean number of samples per patient was 2 (range 1–6). Patient characteristics, SAPS II, SOFA and admission diagnoses are presented in Table 1. Mean ± SD of traditional, Stewart's and Fencl's variables are reported in Table 2. The occurrence of metabolic acidosis was 92.9% and 15.0% using Stewart's method and the traditional method, respectively (χ2 = 151; P < 0.0001). Unmeasured anions were the major cause of metabolic acidosis (Table 3). Blood lactate >2.2 mEq L−1 was found in 62 (16%) samples.
The occurrence of metabolic alkalosis was 93.4% and 18.7% using Stewart's method and the traditional method (χ2 = 386; P < 0.0001), respectively. Reduced weak acid (hypoalbuminaemia) was the major cause of metabolic alkalosis (Table 3).
Interestingly, in 245 (64.5%) samples a diagnosis of mixed metabolic alkalosis and acidosis was achieved accorded with Stewart's theory, while a normal acid–base status was found using the traditional method.
Metabolic acidosis and metabolic alkalosis are usually seen as mutually exclusive in the traditional theory, except for rare specific situations (e.g. in the context of delta anion gap in relationship with a disproportionate change in bicarbonate concentration). However, with Stewart's analysis, simultaneous metabolic alkalosis and acidosis are frequently observed. The presence of two acid–base alterations of opposite sign explains the apparent normality described by the traditional theory, and can be illustrated by two examples. In both of these examples, analysis based on Stewart's theory proved superior in describing the clinical circumstances than that obtained by application of the traditional theory.
A 77-yr-old male was admitted with septic shock requiring massive doses of norepinephrine and dobutamine in addition to volume replacement in order to maintain an arterial pressure of 70/40 mmHg. He also required tracheal intubation and mechanical ventilation with tidal volume 8.0 mL kg−1 and FiO2 0.40. At 3-h after ICU admission, analysis of the acid–base status with the traditional method showed a normal situation (pH 7.42; PaCO2 36.9 mmHg; BE −0.7). Analysis with Stewart's method revealed a metabolic acidosis caused by decreased SID (33.9 mEq L−1) combined with a metabolic alkalosis caused by deficit of weak acids (8.1 mEq L−1).
A 37-yr-old male was admitted to the ICU with severe malaria. He developed sepsis and multiple organ failure (respiratory, renal and coagulation systems) requiring mechanical ventilation (tidal volume 6.0 mL kg−1, FiO2 0.45), antibiotics, high-dose diuretics and fresh frozen plasma. Analysis of acid–base status with the traditional method showed mild respiratory acidosis (pH 7.30, PaCO2 49 mmHg) with normal metabolic conditions (BE −2.2; plasma bicarbonate 22.4 mEq L−1). Analysis with Stewart's method revealed a complex acid–base disturbance with severe metabolic acidosis due to unmeasured anions (SIG 17.7 mEq L−1) combined with metabolic alkalosis caused by increased SID (51.6 mEq L−1) and decreased weak acids (9.07 mEq L−1).
BE and SIG had strikingly different time trends during the first 3 ICU days (Fig. 1): the BE remained within the normal range with a non-significant tendency towards increased values (alkalosis); conversely, the SIG was strongly acidic on day 1 and became progressively more acidic during the following days. A− continued to decrease (indicating progressive metabolic alkalosis) during the first 3 ICU days (data not shown). Quantitatively, the increase in SIG (day 1: −7.41 mEq L−1; day 2: −7.81 mEq L−1; day 3: −7.92 mEq L−1; F = 2.85; P = 0.0493) was greater than the reduction in A− (day 1: 10.61; day 2: 10.16; day 3: 9.78 mEq L−1; F = 7.16; P = 0.0009).
Bivariate analysis of the acid–base variables (traditional method: AG, BE and lactate; Stewart's method: SIG, SID and A−; Fencl's method: BEcl, BEua, BEalb and BEfw) are shown in Table 4.
SIG (adjusted odds ratio 1.25; 95% CI 1.05–1.49; P = 0.013) and lactate (adjusted odds ratio 1.40; 95% CI 1.11–1.77; P = 0.004) were the only independent predictors of the 28-day mortality in the multivariate logistic regression. The AUROC of this model was 0.6943. The cut-off value of SIG was 9 mEq L−1 (sensitivity = 0.70; specificity = 0.49) and 1.6 mEq L−1 for lactate (sensitivity = 0.74, specificity = 0.40). Their respective adjusted odds ratios were 2.38 (95% CI 1.47–3.86; P = 0.000) and 1.90 (95% CI 1.15–3.12; P = 0.012).
In this study, based on a mixed population of critically ill medical, surgical and trauma patients, we observed frequent acid–base alterations despite normal pH and BE. Metabolic acidosis due to unmeasured anions and metabolic alkalosis due to reduced weak acids (hypoalbuminaemia) were the most common alterations, and they were frequently simultaneously present, explaining the normal pH. Notably, metabolic acidosis and, to a lesser degree, metabolic alkalosis increased significantly over the first 3 ICU days. Finally, unmeasured anions and lactic acid were independent predictors of mortality.
We interpret these results as indicating that metabolic acidosis was the primary disturbance in the early phase of critical illness and metabolic alkalosis was a secondary event to offset the acidic state. Different considerations could support our theory. Firstly, acidosis is the direct consequence of cellular energy failure, a condition that is typically observed during microcirculatory insufficiency, sepsis or multiple organ dysfunction and failure [19-21]. Secondly, acidosis is also a direct consequence of hypercatabolic states , a common situation in the early phase of critical illness. Finally, metabolic acidosis, variably described as increased blood lactate, reduced BE or increased SIG , has been demonstrated to be correlated with a poor outcome in a variety of patient subsets (trauma, sepsis and respiratory failure). This result is also confirmed by our data.
The present study demonstrates that the acidic state in the early phase of critical illness not only persists after resuscitation but is also progressive despite intensive care and it is due to unmeasured anions. These unmeasured anions are probably ketones, sulphate, formate, protein dissociation products and intermediates of the energy metabolism accumulated as a consequence of the critical condition. These unmeasured anions are also present in polygeline used as an infusion solution in critically ill patients . In addition, starches can provoke acid via reduction of SID . Massive infusions of polygelines may increase SIG, erroneously indicating an iatrogenic acid–base disturbance as a pathological problem due to critical illness. However, this complication is irrelevant for our study because the patients in our cohort have been treated with starch but not polygelines. Moreover, should the increase in SIG be due to polygelines infused before ICU admission, we would anticipate an initial increase in SIG within the first day after admission and a subsequent decrease, a set of conditions we have not observed in our patients. Although the nature and origin of unmeasured anions in critical illness remain unknown (except for lactate and ketoanions), the present study strongly suggests that unmeasured anions are an end product of critical illness rather than iatrogenic. In addition, the present study demonstrates that unmeasured anions, which are the effect and not the cause of underlying critical illness, are correlated with the severity of illness. Our conclusion is similar to that of Balasubramanyan and colleagues , who studied a paediatric population and that of Kaplan and Kellum , who studied trauma patients with vascular lesions. We studied a more general adult population of critically ill medical and surgical patients.
Several authors have demonstrated that Stewart's method is more useful than the traditional method for diagnosing complex acid–base alterations in various clinical situations. Hypoalbuminaemia, which is frequent in critically ill patients, is a recognized cause of metabolic alkalosis in Stewart's method, but it is not considered in the traditional method. Furthermore, the correct definition of BE assumes that patients have normal water content, normal blood electrolyte concentration and normal serum albumin, although these variables are often altered in critically ill patients. Our study confirms that unmeasured anions, which are considered only in Stewart's theory, are a further factor explaining the differences between the two acid–base theories. Moreover, we demonstrate that, contrary to the prevailing view, the simultaneous presence of metabolic acidosis and metabolic alkalosis is common, despite an apparent normal BE .
A limitation of our study was that the AUROC curve of the final logistic model was 0.69, where 0.8 is considered the cut-off value below which the discriminatory power of diagnostic tests becomes suboptimal . This result is similar to that obtained by Cusack and colleagues , who found the AUROC curves for pH, standard BE and APACHE II score to be 0.72, 0.71 and 0.76, respectively. We did not consider SAPS II or APACHE II scores in our model, since we wanted to compare the acid–base variables of the three methods (traditional, Stewart's and Fencl's methods), rather than making a general prognostic model, which would have required a substantial greater number of patients to take account of inter- and intra-patient variability. Nonetheless, the multivariable logistic model was able to demonstrate that SIG was an independent predictor of mortality, thereby excluding the iatrogenic hypothesis.
We conclude that metabolic acidosis by unmeasured anions is a clinically relevant phenomenon which is correlated with mortality. Progressive metabolic acidosis may be ongoing in the early phase of critical illness despite the absence of acidaemia.
We thank Dr Guido Bertolini, MD Laboratory of Clinical Epidemiology, Mario Negri Institute, Ranica, Italy, for his methodological support and Nathan Astrof for the revision of this manuscript.
1. Stewart PA. Independent and dependent variables of acid–base control. Respir Physiol
2. Stewart PA. Modern quantitative acid–base chemistry. Can J Physiol Pharmacol
3. Ronco C, Bellomo R. Crit care Nephrol
. Dordrecht, Boston: Kluwer Academic, 1998.
4. Kellum JA, Kramer DJ, Pinsky MR. Strong ion gap: a methodology for exploring unexplained anions
. J Crit Care
5. Alfaro V, Peinado VI, Palacios L. Factors influencing acid–base status during acute severe hypothermia in unanesthetized rats. Respir Physiol
6. Russell KE, Hansen BD, Stevens JB. Strong ion difference approach to acid–base imbalances with clinical applications to dogs and cats. Vet Clin North Am Small Anim Pract
7. Constable PD. A simplified strong ion model for acid–base equilibria: application to horse plasma. J Appl Physiol
8. Pesquero J, Alfaro V, Palacios L. Acid–base analysis during experimental anemia in rats. Can J Physiol Pharmacol
9. Balasubramanyan N, Havens PL, Hoffman GM. Unmeasured anions
identified by the Fencl–Stewart method predict mortality
better than base excess, anion gap, and lactate in patients in the pediatric intensive care unit. Crit Care Med
10. Fencl V, Jabor A, Kazda A, Figge J. Diagnosis of metabolic acid–base disturbances in critically ill patients. Am J Respir Crit Care Med
11. Durward A, Skellett S, Mayer A, Taylor D, Tibby SM, Murdoch IA. The value of the chloride: sodium ratio in differentiating the aetiology of metabolic acidosis
. Intensive Care Med
12. Cusack RJ, Rhodes A, Lochhead P et al.
The strong ion gap does not have prognostic value in critically ill patients in a mixed medical/surgical adult ICU. Intensive Care Med
13. Rocktaeschel J, Morimatsu H, Uchino S, Bellomo R. Unmeasured anions
in critically ill patients: can they predict mortality
? Crit Care Med
14. Kaplan LJ, Kellum JA. Initial pH, base deficit, lactate, anion gap, strong ion difference, and strong ion gap predict outcome
from major vascular injury. Crit Care Med
15. Fencl V, Leith DE. Stewart's quantitative acid–base chemistry: applications in biology and medicine. Respir Physiol
16. Gilfix BM, Bique M, Magder S. A physical chemical approach to the analysis of acid–base balance in the clinical setting. J Crit Care
17. Hosmer DW, Lemeshow S. Model-building strategies and methods for logistic regression. In: Hosmer DW, Lemeshow S, eds. Applied Logistic Regression
. New York: John Wiley & Sons, Inc., 1989: 82–134.
18. Bewick V, Cheek L, Ball J. Statistics review 13: receiver operating characteristic curves. Crit Care
19. Levraut J, Grimaud D. Treatment of metabolic acidosis
. Curr Opin Crit Care
20. De Backer D. Lactic acidosis. Intensive Care Med
21. Fink MP, Evans TW. Mechanisms of organ dysfunction in critical illness: report from a Round Table Conference held in Brussels. Intensive Care Med
22. Shills M. Modern Nutrition in Health and Disease
. Philadelphia: Lea & Fabiger, 1988.
23. Hayhoe M, Bellomo R, Liu G, McNicol L, Buxton B. The aetiology and pathogenesis of cardiopulmonary bypass-associated metabolic acidosis
using polygeline pump prime. Intensive Care Med
24. Waters JH, Bernstein CA. Dilutional acidosis following hetastarch or albumin in healthy volunteers. Anesthesiology
25. Kellum JA. Diagnosis and treatment of acid–base disorders. In: Shoemaker WC, Ayers SM, Grenvik A, Holbrook PR, eds. Textbook of Critical Care (TCC)
. Philadelphia: WB Saunders, 1999: 839–853.