Agreement and Correlation of Arterial and Venous Blood Gas Analysis in a Diverse Population

Study design, setting, population, and protocol

This was a retrospective, observational study conducted at a 4-hospital health system, including a Level 1 trauma center with a total of 750 beds. All ABG and VBG in the same patient that were collected at the same time between January 1, 2006, and May 1, 2017 were included in the study. Per hospital and laboratory protocol, all blood gas samples are sent “stat” (transferred immediately without batching or delay) to the laboratory, all blood gas analyses are performed within 15 min of sample collection time, and all blood gas analyses are processed as soon as possible after arrival in the laboratory. Arterial and venous blood gases were obtained per unit protocol but were not directly observed given the retrospective nature of the study. There was no differentiation between VBGs sampled peripherally versus centrally. All patients of any clinical status were included in this study, as patient de-identified material was analyzed. The health system’s Institutional Review Board (IRB) deemed that as no identifiable patient information was collected, IRB review was not required and a non-reviewable letter was submitted and accepted.

Measurements and data analysis

The data collected from the ABG and VBG were analyzed using Bland-Altman plots and Pearson correlations. Bland-Altman analysis and plots were used to evaluate and visualize the degree of agreement between arterial and venous pH, pCO₂, and HCO 3 − . ¹³ The mean difference (MD) (ABG – VBG), 95% limits of agreement (LOA), and associated 95% confidence interval (CI) were computed. For the purposes of this study, we compared the 95% LOA with a predefined clinical limit width of −0.1 to 0.1 for pH and −2.4 to 2.4 mm Hg for pCO₂ based on previously published acceptable clinical limits of laboratory analytes.^14,15 These clinical limits were derived from intra-individual and inter-individual biological variations. Pearson correlations were used to describe the strength of the association between arterial and venous pH, pCO₂, and HCO 3 − . Finally, sensitivity and specificity were analyzed for specific cutoffs of VBG values. Acidemia was defined as pH <7.35 ¹⁶ and hypercarbia as pCO₂ >45 mm Hg. ¹⁷ A sensitivity of greater than 80%^8,9,18,19 was considered appropriate for the use of VBG to identify an abnormal ABG result. All analyses were conducted using SAS 9.3 (SAS Institute, Cary, NC, USA). Level of significance was set at P = .05.

Pearson correlation versus Bland-Altman plot analysis

Pearson correlations have been used in previous studies when comparing VBG with ABG as a surrogate for agreement. However, this statistical method measures the strength of a relation between 2 variables and estimates the linear correlation rather than agreement.^13,20 Pearson correlations are also highly dependent on the range or variation of the variables. ²¹ Thus, Kelly et al ¹⁸ indicated that Pearson correlations are not the clinically relevant measure and instead stated that the accepted method for assessing the agreement between ABG and VBG is the Bland-Altman plot analysis.

Altman and Bland ²² have previously described an alternative method to Pearson correlations when comparing 2 tests with continuous variables. This statistical method quantifies the bias (estimated by the MD) and assesses the agreement regarding the average and individual readings between 2 tests. ¹³ However, previous surveys^20,23–25 have shown that the method described by Bland and Altman ²⁶ is not properly reported or interpreted. Most recently, Chhapola et al ²⁰ showed that only 74% of studies compared the LOA with a priori defined clinical criteria and only 6% reported CIs of LOA. As the 95% LOA is defined as ±1.96 SD of the bias or MD, ²⁶ concluding that 2 tests are in agreement if 95% of the differences lie within the lower and upper LOA is not appropriate. ²⁰ Instead, the width of the LOA should be defined prior to the study and then compared with the obtained 95% LOA.^13,20,26 Bland and Altman^13,26 suggested defining the limits based on clinical judgment. However, there are published acceptable clinical limits of laboratory analytes ¹⁴ that we have used in our study to define our thresholds.

pH

Compared with our predefined threshold, the wide 95% LOA for pH precludes the replacement of ABG by VBG in the general population. This is particularly so for acidemia (pH <7.35) but is also true for pH ⩾7.35. Clinically, these fairly wide LOA in acidemia limit the usefulness of VBG pH to inform a clinician of the arterial acid-base status and may falsely suggest a mildly acidemic arterial pH or even normal arterial pH when the arterial pH may in fact reveal significant acidemia.

The 95% LOA for pH in our study is comparable with those with a diverse population but is wider than studies with specific populations.^{3,4,7,9–12,19,27} However, none of these studies reported 95% CI of the upper and lower LOA, and many of the studies did not set a threshold a priori to define what would be clinically acceptable. Only 2 of the studies used ±0.05 units as their threshold based on the consensus of 15 attending physicians (trauma surgery, emergency medicine, pulmonary/critical care, anesthesiology) ⁴ and based on the reported laboratory analytical error. ⁸ If this threshold for pH is used, most of the studies reviewed would not be able to recommend replacing ABG by VBG. Even if the threshold is wider, such as in our study, depending on the 95% CI of the upper and lower LOA of these studies, replacement of ABG by VBG would probably also not be recommended.

Our MD for pH is lower than the expected difference of 0.05 to 0.06 especially for pH >7.35 but is consistent with previous studies which also found lower MDs.^11,27–29

Studies evaluating the sensitivity of VBG pH for ABG pH are rare. McKeever et al ⁹ had suggested using a venous pH of 7.34 as a cutoff for managing acute exacerbations of chronic obstructive pulmonary disease (COPD) as this pH had a sensitivity of 88.9% and a specificity of 95.6% in identifying an arterial pH ⩾7.35 in their study. We evaluated the sensitivity of VBG to identify ABG acidemia in the general population because we felt that a screening tool should be used to identify possible disease. In our study, the sensitivity of a VBG pH <7.35 identifying an ABG pH <7.35 was 85.0% with a negative predictive value of 81.1% (see Table 3). Thus, based on our predefined threshold, VBG would be an acceptable screening tool for identifying acidemia in the general population. It goes without saying that a VBG should not replace clinical judgment—if the patient is clinically deteriorating or clinical suspicion for acidemia is high, an ABG should be obtained regardless of the VBG result. Moreover, we reiterate that a VBG pH should not be used if exact values are needed given the wide LOA, especially at pH <7.35.

pCO₂

Venous blood gas pCO₂ has poor agreement for ABG pCO₂ as seen by the wide 95% LOA which became wider when we evaluated hypercarbia (pCO₂ >45). These findings are similar to previous studies including 2 meta-analyses^28,29 which showed poor agreement between ABG and VBG for pCO₂. Although there was some overlap regarding the studies that were included in the meta-analyses, we agree that the wide LOA precludes the replacement of measurement of arterial pCO₂ with venous pCO₂, especially when an accurate measurement of pCO₂ is necessary. It should also be noted that similar to our study, the MD for many of the studies included in these meta-analyses was much lower than the expected 5 to 6 mm Hg.^28,29

More recently, Zeserson et al ¹¹ found a Pearson correlation of 0.93 for pCO₂ and an MD of 4.8 mm Hg with an LOA of 8.05 to −17.82 mm Hg in their study which included an undifferentiated critically ill patient population from the ED and the medical and surgical ICU. Although they suggest that venous and arterial pCO₂ correlation allowed for replacement of ABG by VBG for bedside clinical decisions, we cannot agree. Their study emphasized correlation rather than agreement, and they did not include predefined clinically acceptable thresholds for the LOA nor 95% CI for the lower and upper LOA. Our study showed that even for pCO₂ ⩽45, the 95% LOA was too wide to allow VBG pCO₂ to replace ABG pCO₂. Moreover, comparing their obtained 95% LOA with our predefined clinically acceptable threshold shows that there is actually poor agreement between VBG and ABG for pCO₂. This suggests that for pCO₂, arterial measurement is warranted to guide clinical decision making.

Previous studies^3,8,18,19,30 have used venous pCO₂ as a screen for potential hypercarbia. These studies found 100% sensitivity of VBG in identifying arterial hypercarbia using different venous cutoffs of 45 mm Hg (patients with COPD and acute respiratory disease)^8,18,19 and 50 mm Hg (mechanically ventilated trauma patients). ³ Ibrahim et al ³⁰ found that a venous cutoff of pCO₂ ⩾30 mm Hg was 100% sensitive in screening for arterial pCO₂ >45 mm Hg, whereas a venous cutoff of pCO₂ >45 mm Hg resulted in a sensitivity of 79%. In our study, which included a diverse population, using a venous cutoff of pCO₂ >45 mm Hg as a screen was 67.6% sensitive for an arterial pCO₂ >45 mm Hg with an 82.8% negative predictive value (see Table 3). Using a venous cutoff of pCO₂ >30 mm Hg was 99.8% sensitive for an arterial pCO₂ >45 mm Hg (see Table 3), which may be used to rule out hypercarbia. These findings are similar to Ibrahim et al, as our studies included a diverse population. This indicates that a VBG cannot accurately replace an ABG and is unsuitable as a screening tool due to a sensitivity that is less than 80% if used to evaluate the general population for hypercarbia.

HCO₃₋

The 95% LOA was modestly broad for arterial and venous H C O 3 − . Correlation was excellent as expected because it is calculated based on pH and pCO₂ which both showed good correlation when all values were included. Our MD is low compared with the expected physiological MD of 2 to 3 mEq/L. However, several studies included in the systematic review by Bloom et al ²⁸ as well as the study by Middleton et al ³¹ had similar low MDs.

Limitations

We did not have documentation of whether a VBG was obtained via centrally or peripherally, which could have an effect on the variability of the VBG values obtained. However, 2 previous studies showed that there was no significant difference between a central and peripheral VBG for pH,^12,27 pCO₂, ¹² and HCO 3 − . ¹² We have reported time for results but do not strictly have time for sample collection. Hospital and laboratory policies as outlined in the “Methods” section minimize and standardize the time between sample collection and results reporting, but a limitation of the data collected is that sample collection time was not reported.

Although ABG and VBG samples were collected per unit protocol, some of the samples may not have been collected in an anaerobic fashion, specifically using gas syringes. Introduction of air to either the ABG or VBG sample would affect the pCO₂ content and pH causing statistically significant differences compared with samples that are not air-contaminated as was shown in a previous study. ³² However, Gambino ³³ had showed that collecting VBGs in a heparinized vacutainer versus blood gas syringes did not show significant differences in pH or pCO₂. Similarly, Fleisher and Schwartz ³⁴ showed that for ABGs, there was no significant difference for pCO₂ between samples obtained from heparinized vacutainers versus blood gas syringes and only a small increase of 0.01 units for pH that was not felt to be clinically significant.

The retrospective de-identified approach precluded subpopulation analysis in our study, and some patients may have had more than 1 paired ABG-VBG samples. In addition, we were unable to correlate blood gas results with specific clinical statuses. Previous literature have noted the effects of temperature status ³⁵ including fever, ³⁶ hematocrit/hemoglobin concentration and metabolic rate, ³⁷ cardiac output and oxygenation status, ³⁸ and the use of drugs, particularly carbonic anhydrase inhibitors ³⁹ on blood gases, but a critical review of these effects is beyond the scope of this article. Furthermore, the aim of our study was to evaluate whether a VBG could replace ABG or be used as a screening tool for acidemia and hypercarbia for the general population regardless of their clinical status.