CLIN.CHEM.36/12, 2093-2096 (1990)

CLINICAL CHEMISTRY, Vol. 36, No. 12, 1990 2093

Discordance between Measured and Calculated Total Carbon DioxideJ. P. J. Ungerer,1 M. J. Ungerer,2 and W. J. H. Vermaak’

Recent studies on the agreement and correlation betweenmeasured and calculated total CO2 (TCO2) have yieldedconflicting results. Pre-analytical variation could have beenpartially responsible. While keeping such variables at anabsolute minimum, we found excellent correlation (r = 0.98)in 88 samples, with only a small variation in agreementbetween measured and calculated TCO2 values (SD = 1.1mmol/L), which could be a function of variation in apparentpK(pK’). A subsequent evaluation of 913 consecutive sam-ples, routinely analyzed, yielded similar results. These re-sults suggest that some of the discrepancies reported in theliterature could be ascribable to differences in sample typesand sample handling. Rigid control of pre-analytical proce-dures is therefore a prerequisite in studies on this topic. Thetwo methods were found to agree over a wide range ofvalues, such that either of them could be used to evaluateclinical acid-base status accurately.

Addftlonal Keyphrases: blood gases . acid-base statussample handling . variation, source of

In the assessment of acid-base status, total carbon diox-ide can be measured directly (measTCO2) or calculated(calcTCO2) from measured pH and pCO2 by way of theHenderson-Hasselbalch equation:

pH = pK’ + log [HCOI/(S - pco)]

The calcTCO2 is equal to [HCO3 + (S Pco2)]. pK’ repre-sents the apparent first dissociation constant of carbonicacid and S the solubility coefficient for CO2 gas in plasma.

Despite numerous papers on the subject of measTCO2and calcTCO2, controversy still exists with respect to theaccuracy of each method and the degree with which theycorrelate (1-12). The constants used in the Henderson-Hasselbalch equation, especially the pK’, are not constantin blood samples but vary because of the effect of changes inionic strength, pH, temperature, and the concentrations of(e.g.) proteins, electrolytes, and urea. However, no consen-sus exists as to the degree of variability, especially inrelation to its use in clinical medicine (13-20). Any varia-tion in pK’ would subsequently lead to inaccuratecalcTCO2. Some authors argue that the inconstancy of pK’diminishes the accuracy of calcTCO2 to the extent that theuse of measTCO2 is warranted (1-6); others disagree (6-12). Contributing to this controversy is the fact that resultsfound in studies comparing measTCO2 and calcTCO2 differ(1-12).

By scrutinizing the various publications on this topic, wecame to the conclusion that much of the disparity of resultsamong the studies could be partly explained by factors that

Departments of 1 Chemical Pathology and 2Anaesthesiology,Faculty of Medicine, University of Pretoria, P0 Box 2034, 0001Pretoria, South Africa.

Received March 14, 1990; accepted October 5, 1990.

are not directly method related. Studies differed in manyrespects, such as analytical methods used, the type ofsamples analyzed, sample treatment, and whether or not acorrection was made to eliminate possible systemic biasbetween the two methods (1-12). For example, a fewauthors (1, 7) comparing arterial calcTCO2 with venousmeasTCO2 have, in some instances, not mentioned how thevenous samples were handled before analysis. In contrast,others used both methods on identical samples (2,9, 13).

In this study we assessed the agreement betweenmeasTCO2 and calcTCO2 methods, while minimizing thepre-analytical variables that could affect the results. Weassessed the effect of specimen handling and comparedblood-gas analyses in plasma and whole-blood specimens.MeasTCO2 and calcTCO2 were also obtained for >900consecutive samples, to assess the agreement in diverseclinical situations in the routine setting.

Materials and MethodspH and Pco2 were measured, and the TCO2 calculated

from these values, with an ABL 3 blood-gas analyzer(Radiometer, Copenhagen, Denmark). The measTCO2 wasdetermined with an Astra 8 analyzer (Beckman, Fullerton,CA). In this direct measurement method, the sample isacidified and the released CO2 is measured with an ion-selective electrode. Precision was calculated from internalquality-control measurements, by use of Qualicheck (anaqueous-based control material from Radiometer) with theABL 3 analyzer and Decision control serum (Beckman)with the Astra 8 analyzer.

We selected 88 specimens from those received at ouremergency laboratory, to include a wide range of acid-basevalues. Heparinized arterial blood samples were collectedanaerobically. An aliquot was removed and blood-gas anal-ysis was done immediately upon receipt of the sample. Theremainder of the sample was transferred to capped tubesand centrifuged (5 mm at room temperature). Blood-gasanalysis and measTCO2 were performed simultaneously onthe plasma. All analyses were completed within 30 mm ofthe sample’s receipt.

The effect of exposure to air was investigated in 31 ofthese samples. After determining the measTCO2, we di-vided the remaining plasma from each sample into twoportions. One portion was drawn into a plastic syringe andsealed, the other was left in uncapped tubes. After 4 h atroom temperature (-23 #{176}C),each sample portion was ana-lyzed for measTCO2.

To evaluate the agreement between measTCO2 andcalcTCO2 in the routine clinical setting, we comparedresults from 913 consecutive samples sent to our emer-gency laboratory for blood-gas analysis. The samples,anaerobically treated arterial blood, were analyzed forblood gases. Immediately afterwards the blood was centri-fuged in uncapped tubes and TCO2 was measured in theserum, usually within 30 mm. The samples originatedmostly from various intensive-care units and the accidentdepartment of our hospital.

r

0.9810.9780.990

Blood calcTCO2 and plasma measTCO2

08

E6

0gtO 4

Limits of agreementObservedExpected

2SD 3.28

2AD__1.4_#{149} #{149} . #{149} mean 1.75-

-. 2SD -3.28

CC

aI 0

C,’0I.-0tO -4Ii,

a

- 280 1.84

#{149}1 -

.. .. I #{149}%-- #{149}#{149}#{149} mean 1.04

2_-j.8$- OSfl -2RC

BloodTCO2,mmol/L

Pco, CICTCO2,pH mmHg mmol/L pH mmHg CaIc Mess

Mean 7.39 35.64 22.55 7.47 28.12 21.84 20.80

SD 0.12 11.49 7.83 0.13 8.42 7.47 7.39

Results

2094 CLINICAL CHEMISTRY, Vol. 36, No. 12, 1990

The blood-gas analysis and measTCO2 results obtainedfrom the 88 samples are summarized in Table 1. The pHvalues ranged from 6.94to 7.61andthePco1 from 14.9 to72.7 mmHg. Table 2 presents the results of the linear-regression analysis between the various measurements onthe different specimens.

The agreement between measTCO2 and calcTCO2 wasassessed according to the methods proposed by Bland andAltman (21) and is illustrated with bias plots (Figure 1).The analytical precision studies, performed with quality-control samples, yielded an SD of 0.6 mmoIJL for calcTCO2and 0.7 mmol/L for measTCO2. The expected and observedlimits of agreement (mean difference ± 2 SD) are shown.The expected limits, ± 1.82 mmolfL, are based on thecombined imprecision for the two methods, calculated ac-cording to the following formula:

Expected SD = \/(SD measTCO2)2 + (SD calcTCO2)2

Figure 2 illustrates the effect on the measTCO2 in the 31samples exposed to air over a 4-h period. The averagedecrease in measTCO2 was 2.95 mmol/L (SD = 1.40,maximum decrease = 5.8 mmoIJL). No change inmeasTCO2 was observed for the sample portions not ex-posed to air.

The pH and Pco1 values in the group of 913 consecutivesamples varied from 6.83 to 7.78 and 11 to 113 mmllg,respectively. To illustrate the diversity of samples, wenoted 21 samples with pH <7.2. Linear-regression analysisof the results gave the following equation: calcTCO2 =

1.0774(measTCO2) + 2.7844 (r = 0.942, SEE = 1.7571).Before plotting the variation in differences (Figure 3), wecorrected the measTCO2 according to the linear-regressionformula, so as to remove any systematic bias between thetwo methods. A similar procedure was followed by Masterset al. (9) The SD of the differences between measTCO2 andcalcTCO2 was 1.76 mmol/L. Without the correction for biasthis SD was 1.79 mmol/L. To assess its variability, we alsocalculated the pK’ in each individual case, according to theequation pK’ = pH - log[(measTCO2 - Pco - 0.0306)!(Pco2 0.0306)]. The frequency distribution of pk’ is illus-trated in Figure 4. The mean pK’ was 6.096 (SD = 0.041,CV = 0.67%).

DiscussionAccording to the regression analysis (Table 1), blood and

plasma calcTCO2 values were well correlated (r = 0.99).The plasma calcTCO2 correlated better with measTCO2than did blood calcTCO2, but in both instances the corre-lation was near perfect. The mean calcTCO2 was also 0.71mmollL lower in plasma than in whole blood, even thoughwe used anaerobic conditions.

The effect of exposure to air was clearly demonstrated,with the measTCO2 decreasing by as much as 5.8 mmolJL

Table 1. Acid-Base Resultsfor 88 SpecimensPlasma

Table2. LInear-RegressionAnalysisTCO2ResultsIntercept SEE

y x Slope mmol/L

Plasma calcTCO2 MeasTCO2 0.992 1.206 1.448Blood calcTCO2 MeasTCO2 1.037 0.978 1.631Plasma calcTCO2 Blood calcTCO2 0.944 0.538 1.041

tO 20 30 40 60 60

( (b-caIcTCO2l + Eo-calcTCO2])/2 mmol/ I

Plasma caIcTCO2 and plasma meas TCO2o Limits ol agreement

Obeerved‘ Expected

I,0

2SD 2.66

0 tO 20 30 40 50 SO

((p-calcTCO2J + (p-meaaTCO2l)12 mmol/ I

Fig. 1. Bland and Altman (21) plot of mean observed betweencalcTCO2 and measTCO2 vs the difference between the methods:(top) blood, (bottom) plasmaObserved limitsof agreement are based on ±2 SD of the mean difference(1 SD = 1.64 mmol/L blood, 1.43 mmoVL plasma). Expected limits ofagreementare based on ±2 SD of the predicted combinedanalytical impre-cisionof the methods (1 SD = 0.92 mmol/L)

(Figure 2). In contrast, plasma samples could be storedanaerobically for long periods without adversely affectingacid-base results. Meticulous care pertaining to samplehandling and time lapse before measurement is critical foraccurate results. We did not evaluate the effect of storagetime on acid-base measurements in whole blood.

Both blood and plasma calcTCO2 show good agreementwith measTCO2 (Figure 1), with the SD of the differencebeing 1.64 and 1.43 mmol/L, respectively. The positivemean difference indicates a calibration disequilibrium be-tween measTCO2 and calcTCO2. MeasTCO2 could be amore nearly accurate substitute for calcTCO2 if resultswere corrected for this difference.

We used the one-tailed statistical test of Levin (22) to testwhether the various standard deviations for mean differ-ences (Figure 1) were significantly different from each

toB

6

-2

I .rc

10 20

Limits of agreement 2SD)OhservedExpected

where observed SD represents the observed SD of thedifference between plasma calcTCO2 and measTCO2 whenthe pre-analytical variables were being minimized. Thecalculated value for the method-related SD was 1.1mmol/L. Thus the difference between measTCO2 andcalcTCO2 attributable to differences inherent to the meth-ods should be <2.2 mmol/L in 95% of samples analyzed.

If we apply the same formula to the results obtained byO’Leary and Langton (1), the method-related SD would be3.2 mmol/L, not 1.1 mmol/L. The reason for a highermethod-related SD may be related to variables not relatedto the methods per se. Sample treatment may have playedan important role. Furthermore, another contributing fac-tor could be that venous measTCO2 was compared witharterial calcTCO2. Arterial and venous acid-base analytescan differ markedly in certain clinical conditions (23-25),the difference in calculated bicarbonate being about 2mmol/L on average (26). Venous samples could also containfluids given intravenously, which would also affect

30 measTCO2. The difference between measTCO2 andcalcTCO2 should not therefore be ascribed solely to pK’changes, nor should the measTCO2 be used to calculatepK’, if conditions are not standardized and identical plasmasamples are not used.

In the group of 88 patients, which included patients withserious acid-base disorders, no obvious outliers were foundthat would have serious implications on patient treatment.In certain studies, in which the difference betweenmeasTCO2 and calcTCO2 was unacceptably large, the au-thors based their views on large discrepancies found inindividual cases (2, 3). This prompted us to investigate alarge number of consecutive acid-base determinations per-formed by our laboratory. The subjects included patientswith a wide variety of disorders, some with severe acid-base and electrolyte disturbances.

The correlation for this group of 913 samples was excel-lent (r = 0.942). The “method-related” SD, or SD notexplained by analytical variation, of the differences be-tween measTCO2 and calcTCO2, calculated as above, was1.51 mmol/L. The results from the studies of the group of 88indicate that this variance would be even smaller if thecalcTCO2 had been determined in plasma instead of whole

p0 blood and if the samples had been centrifuged anaerobi-cally. In calculating the pK’ from measTCO2, pH, and Pco2,the SD of the pK’ values was only 0.041. This variationincludes the combined analytical variation of the threemeasurements, and one can assume that the true variationin pK’ is even less. However, we did encounter a fewobvious “outliers,” i.e., samples with large differences be-tween calcTCO2 and measTCO2 (a-f in Figure 3). Theseoutliers resulted in abnormal values for calculated pK’,correspondingly marked a-f in Figure 4. Serial sampleswere analyzed in the three patients in whom outliers a, d,

5.7 5.8 5.9 6.0 6.1 6.2 6.3

CLINICAL CHEMISTRY, Vol. 36, No. 12, 1990 2095

Fig. 2. MeasTCO2in 33 samples before and after 4 h of exposure toair at room temperature

(,eIxTCO2 + ,,,#{149}xtTCO2)/ 2 ,,weol/L

Fig. 3. Bland and Altman (21) plot of mean observed plasmacalcTCO2and measTCO2vs the difference between the methods, in913 consecutive samplesa-f, outliers”, as referred to in the text. Observed SD of the difference = 1.76mmol/L; expected SD of the difference = 0.92 mmoVL

30

25

26

21

22

20

II

IS

#{149} Ia

o 12

I0

Fig. 4. Frequency distribution of pK’, as calculated from the 913consecutive samples analyzedThe outliers’ marked a-f correspondto thosein Fig. 3, (see text)

other.The SD of the difference between plasma calcTCO2and measTCO2 was less than between blood calcTCO2 andmeasTCO2 (1.43 vs 1.64 mmol/L), but not significantly so (F= 1.3153, P >0.05). The expected SD of the differencebetween calcTCO2 and measTCO2 attributable to corn-

bined analytical variation was 0.91 mmolJL. This wassignificantly less than 1.43 mniol/L, the SD of the differencebetween plasma calcTCO2 and measTCO2 (F = 2.4694, P =

<0.001). Because the plasma calcTCO2 and measTCO2were determined simultaneously on the same sample, thisgreater than expected variation was probably caused bymethod-related factors, e.g., changes in pK’, whichwouldaffect calcTCO2. This “method-related” variation can bequantified by the formula

Method-relatedSD = \/observed SD)2 - (expected SD)2

and e occurred; the results are summarized in Table 3. ThemeasTCO2 in samples a, d, and e does not appear to fit thepattern observed in the respective serial determinations; ittherefore seems likely that the discrepant measTCO2 val-ues were caused by a random laboratory error. The otherthree outliers were from an isolated analysis done onlyonce on a patient, and cannot be explained by the availabledata. pK’ differences may have been the cause but in lightof the large number of specimens examined, as well as thethree obvious errors of measTCO2, these “outliers” weremore likely the result of random laboratory or analyticalerrors in one or more measurements. This is also the viewof Gennari in explaining widely discrepant results (12).

In evaluating the results of these comparison studieswith measTCO2 and calcTCO2, it is also worth noting thework of Lustgarten et al. (27), who studied the correlationbetween various contemporary methods for determinationof measTCO2. They examined the correlation between thevarious measTCO2 methods by linear-regression analysisand obtained the following results: rand SEE ranged from0.975 to 0.969 and from 0.932 to 1.962, respectively. Theseresults actually vary more than ours (Table 2). Further-more, a few “outliers” among the 60 patients they exam-ined are evident in their linear-regression figures. Forexample, one method of measTCO2 gave a value of -=9mmol/L corresponding to a value of 3 mmolJL by anothermethod. Thus, differences between measTCO2 andcalcTCO2 cannot summarily be ascribed to variation inpK’, because measTCO2 may also be the culprit.

In conclusion, differences between calcTCO2 andmeasTCO2 may be caused by pre-analytical variation,analytical imprecision, or fluctuations in the constants(e.g., pK’) used in the Henderson-Hasselbalch equation. Itis, therefore, imperative to minimize the pre-analyticalvariables and to account for analytical imprecision in allstudies in which calcTCO2 and measTCO2 are compared, orin which the constancy of pK’ is assessed. By adhering tothese prerequisites, we found the variation in pK’ so smalland the agreement between calculated and measured TCO2good enough that clinical misinterpretation was unlikelyin the vast majority of cases. Both of these methods seem to

Table 3. Serial Results for Three Patients HavingDiscrepanciesbetween CalcTCO2and MeasTCO2

calcTCO2 measTCO2

pH7.287.357.42

calcTCO2 -

measTCO2Pco

mmHg19.317.123.6

mmol/L

9.5 12.2

9.8 26.116 18.1

SampleBeforeaAfter

Befored

After

Before8

After

-2.7-16.3

-2.1

7.48 30.2 23.57.42 38.0. 25.57.42 39.2 26.2

21.918.025.0

7.45 35.47.49 35.27.45 39.1

25.7 26.127.9 18.628.1 29.6

Samples a, d, and e correspond to the outliers identified in Figs. 3 and 4.‘Before’ and “after” refer to the samples (fromthe same patient) immediatelypreceding and following these outliers.

2096 CLINICAL CHEMISTRY, Vol. 36, No. 12, 1990

be valid and can be used interchangeably with reasonable

confidence.

References1. O’Leary TD, Langton SR. Calculated bicarbonate or total car-bon dioxide? Clin Chem 1989;35:1697-700.2. Kost GJ, Trent JKT, SaeedD. Indications for measurements oftotal carbon dioxide in arterial blood. Clin Chem 1988;34:1650-2.3. Rosan RC, Enlander D, Ellis J. Unpredictable error in calcu-lated bicarbonate homeostasis during pediatric intensive care: thedelusion of fixedpK’.Clin Chem 1983;29:69-73.4. Natelson S, Nobel D. Effect of the variation of pK’ of theHenderson-Hasselbalch equation on values obtained for total CO2calculated from Pco2 and pH values [Letter]. Clin Chem1977;23:767-9.5. Hood I, Campbell EJM. Is pK OK? N Engl J Med1982;306:864-5.6. flear CTG, Robei-ts SW, Hayes S, Stoddart JC, Covington AK.pK’ and bicarbonate concentration in plasma. Clin Chem1987;33:13-20.7. Mohler JG, Mohler PA, Pallivathugal RG. Failure of the serumCO2 determined by automation to estimate the plasma bicarbon-ate. Scand J Clin Lab Invest 1987;47(Suppl 188):61-7.8. Dear PRF, Henderson MJ. Assessing metabolic acidosis [Let-ter]. Clin Chem 1989;35:338.9. Masters P, Blackburn MEC, Henderson MJ, Barrett JFR, DearPRF. Determination of plasma bicarbonate of neonates in inten-sive care. Clin Chem 1988;34:1483-5.10. Karlowicz MG, Simmons MA, Brusilow SW, Jones MD. Car-bonic acid dissociation constant (pK1) in critically ill newborns.Pediatr Res 1984;18:1287-9.11. Ryan DH, Holt J. Is pK OK? [Letter]. N Engl J Med1982;306:863-4.12. Gennari FJ. Is pK OK? [Letter]. N Engl J Med 1982;307:683.13. De Raedt M, Vandenbergh E, Van de Woestijne KP. Directand indirect determination of partial pressure of CO2 in thearterial blood of patients with respiratoryinsufficiency. Clin Sci1968;35:347-52.14. Tibi L, Bhattacharuya SS, FlearCTG. Variability in pK1 ofhuman plasma. Clin Chim Acts 1982;121:15-31.15. Austin WH, Ferrante V, Anderson C. Evaluation of wholeblood pK’ in the acutely ill patient. J Lab Clin Med 1968;72: 129-35.

16. Trenchard D, Noble MIM, Gus A. Serum carbonic acid pK’1abnormalities in patients with acid-base disturbances. Clin Sci1967;32:189-200.17. Sinclair MJ, Hart BA, Pope HM, Campbell EJM. The use ofthe Henderson-Hasselbalch equation in routine medical practice.Clin Chim Acts 1968;19:63-9.18. Siggaard Andersen 0. The first dissociation exponent of car-bonic acid as a function of pH. Scand J Clin Lab Invest1962;14:587-97.19. Austin WH, Ferrante VV, Ritchie RF. Effect of abnormalplasma constituents on the pK’ of whole blood. Am J Clin Pathol1989;51:799-801.20. Severinghaus JW, Stupfel M, Bradley AF. Variations of serumcarbonic acid pK’ with pH and temperature. J AppI Physiol1956;9:197-200.21. Bland JM, Altman DG. Statistical methods for assessingagreement between two methods of clinical measurement. Lancet1986;i:307-10.

plC’ 22. Levin RI. Chi-square and analysis of variance. In: Levin RI.5.99 Statistics for management, 4th ed. Englewood Cliffs, NJ: Prentice-5 66 Hall, 1987:440-94.6flA 23. Adroque HJ, Rashad MN, Germ AB, Yacoub J, Machas NE.

Assessing acid-base status in circulatory failure: differences be-1 6 6 13 tween arterial and central venous blood. N EngI J Med

1989;320:1312-6.7.5 6.26 24. Weil MH, Rackow EC, Trevino R, Grundler W, Falk JL,1.2 6.12 Griffel MI. Difference in acid-base state between venous and

arterial blood during cardiopulmonary resuscitation. N Engi J0.4 6.09 Med 1986;315:153-6.9.3 6.28 25. Kost GJ. Venous total carbondioxide: correlation with arterial

-1.5 6.07 bicarbonate and clinical usefulness. Lab Med 1986;17:74.26. Relman AS. “Bloodgases”:arterial or venous? [Editorial]. NEngl J Med 1986;315:188-9.27. Lustgarten JA, Creno RJ, Byrd CG, Wenk RE. Evaluation ofcontemporary methods for serum CO2. Clin Chem 1976;22:374-8.