Original Contribution

Reliability of anion gap calculated from data obtained using a Blood gas analyzer: is the probability of error predictable?

Norio Otani MD^a,?, Sachiko Ohde EdM^b,

Toshiaki Mochizuki MD^a, Shinichi Ishimatsu MD, PhD^a

^aDepartment of Emergency and Critical Care Medicine, St. Luke‘s International Hospital, Tokyo 104-8560, Japan

^bCenter for Clinical Epidemiology, St. Luke‘s Life Science Institute, Tokyo 104-8560, Japan

Received 2 January 2009; revised 3 February 2009; accepted 4 February 2009

Abstract

Background: anion gap is a useful index for assessing the clinical condition of critically ill patients especially in intoxication. Recently, AG can be obtained easily using a blood gas analyzer (BGA); however, its reliability requires validation.

Methods: We enrolled patients who simultaneously underwent blood gas analysis and blood test in the central hospital laboratory and patients who visited the emergency department of our hospital from January 1, 2004, to December 31, 2007.

The deviation of AG calculated using the BGA and that calculated by the central hospital laboratory were extracted. From the data obtained using the BGA, the independent risk factor causing a significant error in AG was statistically analyzed.

Results: A total of 2922 patients were enrolled, of which 339 were defined as the significant error group. Male sex, abnormal HCO -, abnormal lactate, abnormal K, abnormal Cl, and abnormal Na were

3

the independent risk factors producing the significant error. The results indicate that regardless of whether the original electrolyte data of the patients are abnormal, when the electrolyte measurement results obtained using the BGA are abnormal, the calculated AG might show a significant error. In addition, the fact that lactate was determined as a risk factor indicates that AG might be more useful in patients who have intoxication than in those under an unstable state in terms of vital signs.

Conclusion: When risk factors are present, the medical condition of a patient should be reevaluated by comparing results without heavily relying on the AG obtained by a BGA.

(C) 2010

Introduction

Anion gap (AG) is a useful index for measuring negative or positive ions to assess the clinical condition of critically ill patients in the acute phase [1], especially in

* Corresponding author. Tel.: +81 3 3541 5151; fax: +81 3 5550 7066.

E-mail address: [email protected] (N. Otani).

lactic acidosis, ketoacidosis, renal failure, and certain intoxication [2-4].

With considerable progress in the development of a blood gas analyzer (BGA), electrolyte measurement has become readily possible, making the rapid calculation of AG feasible.

However, AG calculated from the results measured by a central hospital laboratory may not be in agreement with AG calculated from the measurement results obtained

0735-6757/$ – see front matter (C) 2010 doi:10.1016/j.ajem.2009.02.011

using a BGA [5], and this discrepancy remains unclear to physicians.

We experienced some cases of the salicylates intoxication that AG elevation was not found in the AG obtained by BGA results of a measurement, although AG elevation was found in the AG, which central laboratory results of a measurement provided. The salicylates intoxication is famous as the poisoning that an AG elevation is a characteristic and which needs to perform specific treatment. We thought that this AG error might be a risk to retard a start of the treatment in a poisoning practice.

In clinical practice involving critically ill patients, the rapid acquisition of clinical data using the BGA is very useful for enabling swift clinical decision making [6,7]. To efficiently use the measurement results, it is necessary to determine a factor affecting the precision of the obtained results. The present question is which indepen- dent risk factor from the clinical data obtained using the BGA can estimate the possibility that the calculated AG is incorrect.

Methods and materials

We conducted a retrospective cohort study by using data collected in the normal course of patients care. Prior ethical approval was obtained from Research Ethics Committee of St. Luke’s International Hospital, which waived the need to obtain informed consent. The data we analyzed in this study were collected only for this study, and the data were not a part of a previously published data set.

The subjects enrolled were patients who simultaneously underwent blood gas analysis and blood test in the central hospital laboratory and those who visited the emergency department of St. Luke’s International Hospital, Tokyo, Japan, from January 1, 2004, to December 31, 2007. “Simultaneously” here means that patients had a history of acceptance of both specimens for analysis within 30 minutes. The emergency department of the hospital serves 38 000 patients annually of different severities without bias.

We obtained laboratory examination data of the patients from their medical records retrospectively and analyzed them. We excluded patients with data deficiency of the constitution factor in the AG calculating formula and those less than 18 years old.

The traditional AG was calculated as AG = [Na] + [K] –

[HCO -] – [Cl]. The concentration of potassium in the blood is usually relatively low compared with that of sodium, chloride, and bicarbonate. Many clinicians omit this variable when calculating AG, and thus, the formula is generally given as AG = [Na] – [HCO -] – [Cl] [1,8]. It has been reported that various corrections for AG are necessary [5,9]; however, the correction factor cannot be determined at the time of clinical examination for correction of blood gases. Because this study is based on the standpoint of a

3

3

clinician, the formula for calculating AG should be AG = [Na] – [HCO -] – [Cl].

We used the ABL-725 (Radiometer, Copenhagen, Den- mark) for the BGA and use the Bio Majesty JCA-BM1650 (Japan Electron Optics Laboratory, Tokyo, Japan) for the central hospital laboratory. Both laboratory systems were adequately calibrated periodically. All elements in the formula for calculating AG were measured with an ion- selective electrode method in both measuring systems.

3

Here, we defined AG calculated from the results obtained in the central hospital laboratory as labAG and defined AG calculated from the results obtained using the BGA as gasAG. We hypothesized that labAG was the correct AG and therefore analyzed differences between labAG and gasAG as deltaAG (?AG; ?AG = gasAG – labAG). For the analysis of differences between labAG and gasAG, we conducted statistical analysis using the absolute value of ?AG (|?AG|).

We initially considered by preliminary analysis that a 90% quantile in the |?AG| distribution was equivalent to |?AG| = 10, and we defined higher than |?AG| = 10 as the significant error group (Fig. 1). If an obtained blood test result was within the reference value range, we defined it as a normal value; if it was outside, we defined it as an abnormal value and performed statistical analysis.

A multivariable adjusted logistic regression model was constructed to evaluate the adjusted associations of each of the laboratory examination item factors with AG discre- pancy. The model was also adjusted for demographic data. Adjusted odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for both models. The Statistical Package for Social Sciences software version 15.0J (SPSS Inc., Chicago, IL) was used for all statistical analyses.

Results

In the case of the BGA, results were obtained within several minutes after placement of blood in the analyzer. In

Fig. 1 Outline of distribution of |?AG|. The left and right limits of the boxes indicate the 25th and 75th percentiles, and the bar within the boxes indicates the median value. The bars extending from the boxes indicate the range of the percentiles, from the 10th to the 90th. The maximal value of |?AG| was 58 and the minimum was 0. ?AG = gasAG – labAG.

Table 1 Attributes of all subjects in this study and the subgroup of |?AG| >=10

All patients |?AG| >=10

No. of patient 2912 339

Age (mean +- SD) 63.5 +- 21.1 61.5 +- 18.8

Sex (male:female) 1717 (59.0%): 236 (69.6%):

1195 (41.0%) 103 (30.4%)

the case of analysis by the central hospital laboratory, approximately 1 hour after specimen acceptance was needed	Base excess SaO2	213 136	63.4 40.5	1341 1126	52.2 43.8	b.05 .253
until results were sent.	Hco –	190	56.0	1037	40.3	b.05
Overall, there were 2922 patients examined as subjects.	Lac	273	80.5	1679	65.3	b.05
Of these, the significant error group (|?AG| >=10) consisted	K	190	56.0	1255	48.8	.012
of 339 patients (11.6%). The outlines of both groups are shown in Tables 1 and 2. For the index showing the general disease background of the patients, the distribution of the hospitalization of patients per department is shown in Table 2. The disease background of the extracted error group showed no marked changes relative to the whole population. The results of the univariate analysis for 339 patients in the significant error group are shown in Table 3 and those of the multivariable analysis in Table 4.	Cl 159 46.9 834 32.4 b.05 Na 267 78.8 1128 43.8 b.05 Glucose 290 85.5 1974 76.7 b.05 All subjects were divided into those with |?AG| >=10 and those with |?AG| b10. The examined categories were patients’ sex and the presence or absence of an abnormal value in the BGA’s examination item. The ?2 test in the nominal scale was used for data analysis, and the t test was used for the analysis of the sample independent in the continuous variable. Both statistical analyses considered P b .05 as significant. The results, which we could not obtain using a measurement error, were analyzed for missing values.

Male sex (OR, 1.64; 95% CI, 1.27-2.12 ), abnormal

Table 3 Results of a univariate analysis

|?AG| >=10 (n = 339)

|?AG| b10 (n = 2573)

P

Sex, male Laboratory

examination item

n

236

N

% n

69.6 1481

% N

%

57.6

%

b.05

3

HCO – value (OR, 1.41; 95% CI, 1.04-1.90), abnormal

3

lactate value (OR, 2.06; 95% CI, 1.52-2.80), abnormal K

value (OR, 1.43; 95% CI, 1.12-1.82), abnormal Cl (OR,

1.47; 95% CI, 1.15-1.87), and abnormal Na value (OR,

5.04; 95% CI, 3.80-6.69) were found to be associated with the significance error of |?AG|.

Discussion

Table 2 Admission departments of all patients enrolled as subjects and patients with |?AG| N10

All patients |?AG| >=10

Anion gap is a useful index representing the concentra- tion of all unmeasured anions in the plasma and is valuable

Table 4 Results of multivariable analysis

pH	134	39.5	1363	53.0	b.05
PCO2	211	62.2	1314	51.1	b.05
PO2	272	80.2	2055	79.9	.943

Admission department	No. of patient	%	Admission department	No. of patient	%
Critical care	786	27.0	Critical care	108	31.9
medicine			medicine
Cardiology	404	13.9	Cardiology	47	13.9
Respiratory	352	12.1	Gastroenterology	40	11.8
medicine
Gastroenterology	337	11.6	Respiratory	35	10.3
			medicine
Neurosurgery	224	7.7	General surgery	26	7.7
Neurology	208	7.1	Neurosurgery	23	6.8
General surgery	160	5.5	Neurology	16	4.7
Infectious	138	4.7	Infectious	13	3.8
medicine			medicine
Nephrology	62	2.1	Endocrinology	11	3.2
Endocrinology	57	2.0	Nephrology	5	1.5
Cardiovascular	44	1.5	Cardiovascular	4	1.2
surgery			surgery
Others	140	4.8	Others	11	3.2
Total	2,912		Total	339

OR 95% CI P

Sex, male 1.64 1.27-2.12 b.05 Laboratory examination item pH 1.25 0.96-1.63 .1 PCO2 0.96 0.73-1.27 .78 PO2 1.06 0.77-1.44 .73 Base excess 1.12 0.84-1.51 .44 SaO2 0.92 0.71-1.18 .5 HCO – 1.41 1.04-1.90 b.05 3 Lac 2.06 1.52-2.80 b.05 K 1.43 1.12-1.82 b.05 Cl 1.47 1.15-1.87 b.05 Na 5.04 3.80-6.69 b.05 Glucose 1.21 0.86-1.70 .28

Data were analyzed using logistic-regression analysis, and Pb .05 was considered significant.

in clinical practice for assessing the condition of a critically ill patient [10].

In general, an elevated AG indicates lactic acidosis (type A), uremia, sepsis, rhabdomyolysis, ketoacidosis, toxic ingestions, and metabolic alkalosis with volume depression. On the other hand, a decreased AG indicates increased levels of an unmeasured cation (eg, in cases of hypocalce- mia, hypomagnesemia, and elevated IgG), decreased levels of an unmeasured anion (eg, in cases of hypoalbuminemia), and drug usage (eg, bromide, iodide, lithium) [11].

In clinical practice involving poisoning patients in particular, it is very important to know AG elevation. MUDPILES is a mnemonic that has been used for many years to remember important intoxication-induced AG elevation.

MUDPILES is an age-old tool that has been scrutinized in recent years and describes several well-known etiologies of metabolic acidosis: methanol and met-formin, uremia, diabetic ketoacidosis, paraldehyde and phen-formin, iso- niazid and iron, lactic acid, ethanol and Ethylene glycol, and salicylates.

However, other etiologies have not been clearly identified to date. Extending the mnemonic to MUDPILECATS as suggested by many would thus include cyanide, acetyl salicylic acid, alcoholic ketoacidosis, toluene, starvation ketoacidosis, and solvents as causes of an increased AG in metabolic acidosis [12].

Utility in the emergency department

In an emergency department, a BGA provides the results of a blood test immediately. The use of the main electrolytes in the blood as measurement items in the BGA is very useful for physicians in the emergency department in terms of making quick decisions [6,7]. However, previous studies have raised questions regarding the reliability of the results obtained using a BGA [3,13].

In a recent well-formulated study, it has been suggested that the AG calculated using elements from the results obtained using a BGA shows a high probability of error [5].

In the emergency department, patients present various clinical conditions, and the overall environment is chaotic. Therefore, from the viewpoints of the clinician responsible for drawing blood, the sampling method, sampling quantity, and the hemal type (artery or vein), obtaining uniform specimens is difficult.

The rationale for conducting this study was to analyze the advantages of the benefits derived from the rapid acquisition of measurement results using a BGA in an emergency department environment and to evaluate the reliability of the obtained results by comparing with measurement results from the central hospital laboratory. Specifically, this study aimed to determine the trustworthi- ness of gasAG calculated from the results obtained using the BGA.

Data definition and measurement method

As for the BGA, examination is performed just after drawing blood. But the blood test in the central hospital laboratory needs time for a process to submit. Therefore, in this study, we have estrangement at submission time and the Blood drawing time as for the blood test in the central hospital laboratory. On the basis of the fact, in this study, if submission time of both laboratory examination line was “within 30 minutes,” we defined it as approximately simultaneous drawing blood.

This research was not able to standardize hemal modality (arterial blood or venous blood). In addition, we were not able to unify these, although the drawing blood site was Peripheral vessels. However, in the results of this study, pH, pCO2, and PO2 in blood did not present a significant correlation in AG error; the point that these were not unified may become a bias.

We measured plasma ingredients after hemal plasma skimming of blood specimens using laboratory procedures in the central hospital laboratory, whereas we measured whole blood specimens using the BGA directly. This was the basic difference of the 2 laboratory examination techniques performed, which may have generated the discrepancies in the results. Such variation has been reported previously [5,9].

Ideally, we must confirm the reproducibility of results obtained using the BGA by using specimens showing an abnormal value to make the results more reliable. We should also measure the same types of specimen (whole blood or plasma) using both laboratory systems simultaneously. Because this study is only a retrospective observational study, the above-mentioned concerns must be addressed in future studies.

Risk factor

Na, Cl, and HCO3 were identified as the constitution factors in the calculation formula of AG; however, the fact that these constitution factors present an abnormal value was taken as an independent risk factor to cause an AG error.

At the time point when measurement results are obtained using the BGA, it remains unknown whether a patient does have electrolyte abnormality, or it may be possible that the results contain errors.

Regardless of whether the original electrolyte data of the patients were abnormal values, we can reliably state that when the results of electrolyte measurement (particularly for Na and Cl) obtained using the BGA are abnormal values, the calculated AG (gasAG) shows a significant difference from labAG. This suggests that gasAG is strongly doubtful when abnormal values as a calculating formula component are used in AG calculation.

Moreover, abnormal values of K and lactate and the male sex were identified as risk factors. These are considered as

factors unrelated to the component of the formula for calculating AG.

We usually consider a condition as elevated AG when the results of blood gas analysis indicate acidosis, particularly metabolic acidosis [14]. However, the elevation of lactate concentration was an independent risk factor for producing an error in the results. In other words, the reliability of gasAG is low in hypoxemia, peripheral circulatory failure, and the condition of a patient in shock.

Lactic acidosis is known as a major cause of AG increase [11,14]. Thus, in this sense, gasAG should not be used for lactic acidosis evaluation. Because the apparatus we used in this study could measure lactate concentration directly, we were able to analyze it as a covariate.

In patients showing no increase in lactate concentration, gasAG is considered reliable. Lactate concentration report- edly increases in patients with circulatory failure or under anoxic condition. Therefore, gasAG should be carefully used for estimating the clinical condition of patients in the shock state. In other words, it is more appropriate to consider the differentiation of acute intoxication, which is characterized by an elevated AG.

Regarding abnormality in K given as a risk factor, we suspect the presence of renal dysfunction constituting the disease background. However, this could not be elucidated in this study and requires further investigation.

The male sex should be given attention because it was considered a risk factor. Here, we can only assume the probability that the disease background that men play as a confounder is a true risk factor, as well as the probability that a hormonal factor actually affected the laboratory examina- tion system. However, the study design could not confirm this assumption, and further study is recommended.

Specificity of the subject group

The group of patients who became the subjects in this study included those who simultaneously underwent blood gas analysis and blood test at the central hospital laboratory. It appears that there is bias in modality and severity of the disease when we consider the clinical condition of the patients who need to be administered blood gas analysis and a blood test at the same time. Therefore, many test results that markedly deviated from the normal range were possibly included in the analysis. It is therefore difficult to generalize the results in all patients considering this background. However, the results can be considered meaningful if their application is limited to emergency and seriously ill patients.

When discussing having limited the subjects to critically ill patients, it seems that when absolute value of gasAG was fairly high, this AG error would not be as relevant. However, as for the value of gasAG, maximum was 40.1, and minimum was -53. On the other hand, the maximal value

of the |?AG| was 58. Based on these facts, AG error should always be considered in clinical decision making.

Because this research was conducted using a single instrument of the single institution, we recommend that a similar study be conducted by other institutions for comparison and to verify the results.

In conclusion, from data obtained using a BGA, the independent risk factors that can estimate the possibility that the calculated AG is incorrect include the following:

• • 1. AG calculating formula constitution factors (i.e., Na, Cl and HCO₃) showing an abnormal value
    2. K and lactate showing an abnormal value
    3. Male sex.

When these risk factors are present, confirmatory measurement results from the central hospital laboratory should be obtained for the evaluation of the medical condition of a patient, without relying heavily on the measurement results obtained using the BGA.

References

1. Oster JR, Perez GO, Materson BJ. Use of the anion gap in clinical medicine. South Med J 1988;81:229-37.
2. Lim S. Metabolic acidosis. Acta Med Indones 2007;39:145-50.
3. Hovda KE, Hunderi OH, Rudberg N, et al. Anion and osmolal gaps in the diagnosis of methanol poisoning: clinical study in 28 patients. Intensive Care Med 2004;30:1842-6.
4. Brenner BE, Simon RR. Management of salicylate intoxication. Drugs

1982;24:335-40.

1. Morimatsu H, Rocktaschel J, Bellomo R, et al. Comparison of point- of-care versus central laboratory measurement of electrolyte concen- trations on calculations of the anion gap and the strong ion difference. Anesthesiology 2003;98:1077-84.
2. Leventhal RI, Goldman JM. Immediate plasma Potassium levels in treating diabetic ketoacidosis. Arch Intern Med 1987;147:1501-2.
3. Strickland RA, Hill TR, Zaloga GP. Bedside analysis of arterial blood gases and electrolytes during and after cardiac surgery. J Clin Anesth 1989;1:248-52.
4. Kraut JA, Madias NE. Serum anion gap: its uses and limitations in clinical medicine. Clin J Am Soc Nephrol: CJASN 2007;2:162-74.
5. Story DA, Morimatsu H, Egi M, et al. The effect of albumin concentration on plasma sodium and chloride measurements in critically ill patients. Anesth Analg 2007;104:893.
6. Berkman M, Ufberg J, Nathanson LA, et al. Anion gap as a screening tool for elevated lactate in patients with an increased risk of developing sepsis in the emergency department. J Emerg Med 2008.
7. Mokhlesi B, Leiken JB, Murray P, et al. Adult toxicology in critical care: part I: general approach to the Intoxicated patient. Chest 2003; 123:577-92.
8. Klauer KM. Life beyond MUDPILES. Air Med J 2002;21:37-41.
9. King R, Campbell A. Performance of the radiometer OSM3 and ABL505 blood gas analysers for determination of sodium, potassium and haemoglobin concentrations. Anaesthesia 2000;55:65-9.
10. Morris CG, Low J. Metabolic acidosis in the critically ill: part 2. Causes and treatment. Anaesthesia 2008;63:396-411.