Foreword
This post is about a research project I did as a pulmonary critical care fellow in 2011. To understand it, you need to know a bit of the story behind it.
I had some ideas for converting VBG values into ABG values. To investigate further, I requested post-publication data from several researchers who had published studies comparing ABG vs. VBG data. Many generously shared their data with me.
Based on this data, I developed formulae for conversion of VBG values into ABG values. This work was initially submitted to Critical Care Medicine, where it was rejected. No major flaws were found in the analysis, but it was deemed to beirrelevant (one reviewer wrote “This meta-analysis describes well a way to calculate ABG from VBG. However, whether this information is useful in clinical practice is debatable”)(1).
The manuscript was revised a bit and submitted to a second journal. One of the reviewers selected by the journal was an investigator who had provided me with the highest quality data in the paper. The investigator (whom I will call Dr. No) mayhave been concerned that the manuscript wouldcompete with his own work. He withdrew permission to use his data.
Without Dr. No's data, the manuscript was not publishable. I reached out to some additional investigators who had recently published data, and obtained one fresh dataset. However, I couldn't find anything that matched Dr. No's data (he had painstakingly measured ABG and VBG values in a nearly simultaneous fashion, yielding surprisingly precise results).
Years passed. It has weighed on me that I failed to publish these results, which I continue to believe are valid and potentially useful. Now that I am a blogger, I canpresent this research in my blog. I have redacted data from Dr. No (including his or her identity) to avoid any potential ethical or personal conflict.
Theory
The initial concept is simple, albeit perhaps over-simplified. Imagine blood flowing from the radial artery to a vein in the hand. Tissues in the hand extract oxygen and generate carbon dioxide (in a ratio equal to the respiratory quotient). If we assume that most patient's hands have a similar respiratory quotient, then the change in CO2 between arterial and venous gas should be proportional to the change in oxygen content (where k1 is an empirically derived constant):
The next question is what effect this change in the CO2 will have on the pH. The relationship between pH and CO2 is complex, based partially on the Henderson-Hasselbach equation. However, the first-order approximation of any curve is a straight line. Thus, the change in pH may be approximated as proportional to the change in carbon dioxide:
This creates the saturation model, which may be summarized as follows (where k1 and k2 are empirically derived constants):
This provides a way to estimate ABG values based on a combination of VBG values plus simultaneous pulse oximetry. Unfortunately, most VBGs aren't obtained with a simultaneous pulse oximetry. It would be nice to have a way to convert a VBG directly into an ABG, without having to know the arterial oxygen saturation. This can be done with the use of a third assumption.
The difference in oxygen saturation is the difference between the venous and arterial oxygen saturation (shown below). The venous oxygen saturation varies quite a bit, between roughly 10%-95%. Meanwhile, among hospitalized patients, the arterial oxygen saturation is maintained in a tight range (typically between 88-100%). Therefore, the vast majority of the variation in the difference in oxygen saturation comes from variations in venous oxygen saturation. The variation in arterial oxygen saturation is so low that it may be approximated as zero (by setting the patient's arterial oxygen saturation equal to the average oxygen saturation of the entire patient population).
This approximation allows us to create the simplified saturation model, which is capable of estimating ABG values directly based on VBG values:
This model isn't perfect. However, it's probably better than the most common method used in the literature, which is to relate arterial and venous parameters to each other directly using linear regression equations (where c1-c4 are constants):
Source data
Sixteen studies were identified from the literature search as relevant for consideration. Of these, three studies involving 314 patients were analyzed. Thirteen studies were rejected for analysis for the following reasons: the corresponding author did not respond to our request to analyze their data (7), the corresponding author was unable to locate the data (4), the data did not contain oxygen saturation values (1), and the data was internally inconsistent with reported bicarbonate values that differed substantially from those calculated using the Henderson-Hasselbach equation (1). Since one of these three studies contained a control group, this study was designated as having two patient groups and, therefore, a total of four patient groups were analyzed. Data from two patients in two different studies were censored (in one case because pCO2 was immeasurably high, and in another case because the venous oxygen saturation was >25% higher than the arterial oxygen saturation).
Characteristics of source data are shown here (Ak 2006, Ibrahim 2011, O'Connor 2011):
Validation of the basic assumptions of the saturation model
The saturation model predicts the existence of two linear relationships which should exist in any dataset. These datasets support the existence of a universal and linear relationship (the red lines in each set of figuresbelow have matching slopes):
Comparison of that saturation model vs. direct correlational model
That looks nice, but we need to be a bit more precise. Let's start by analyzing these datasets using the direct correlational model (below), which is the conventional way of looking at these datasets.
If we calculate the constants involved in these equations from different datasets, the numbers are all over the place (table below). Thus, this strategy is unable to yield a universally applicable equation which can relate arterial and venous blood gas values. Similar variation is notable when evaluating the published literature regarding ABG vs. VBG comparison, explaining why these equations haven't gained clinical acceptance.
Now, let's analyze this data using the saturation model (below). The constants obtained from each dataset are consistent with each other. This implies that it may be possible to use the saturation model to create a universally applicable equation to convert VBG values into ABG values.
Final results
The saturation model and the simplified saturation model had the same performance in converting from VBG into ABG values. The final equations derived to convert from VBG to ABG values are as follows:
Accuracy?
Here is where things fall apart without Dr. No's data. In analyzing these different datasets, the primary driver of the accuracy isn't the model itself, but rather the accuracy of the underlying data (e.g. time interval between VBG and ABG, processing of ABG and VBG specimens, etc.).
One problem inherent in nearly all studies comparing VBG to ABG values is that all error is blamed on VBG-ABG differences, ignoring the following:
- Error involved in sampling of arterial and venous blood (e.g. gas bubbles).
- Error involved in delay to analysis.
- Changes in blood gas values over time (may fluctuate rapidly).
Overall, it is easy to over-estimate the error involved in extrapolating from VBG to ABG samples (based on the above sources). However, it is difficult to under-estimate this error across an entire dataset. Thus, the data set suggesting the lowest error is closest to the true error involved in extrapolation from VBG to ABG data (2).
Dr. No's data was the most precise (possibly because it required specific time intervals between ABG and VBG samples). This data suggests that a simplified saturation model may predict ABG values with a precision that could be adequate for clinical use. Of note, it is debatable precisely how much error in an ABG measurement is acceptable (e.g., is a 95% confidence interval of +/- 0.03 pH units and +/- 5 mm pCO2 accurate enough?). I would argue that important management decisions shouldn't be based on subtle differences in ABG or VBG values.
ABG values are generally taken as the gold-standard for pH assessment. However, it must be noted that clinically stable patients have random fluctuations in pH and pCO2 with a standard deviations of 0.015-0.02 and 1.5-3 mm, respectively (4). This again emphasizes that small differences in ABG values aren't clinically relevant. For example, one of the classic errors in ABG interpretation is over-interpreting random variation in sequential ABG measurements.
Validation with another dataset
One flaw in the above analysis is that datasets were used to generate coefficients in the regression equation, and then the regression equation was tested on these same datasets. This creates the possibility for circular logic.
Subsequent to failed publication attempts as described above, I requested data from investigators who had more recently published papers. I was kindly provided with one dataset from Dr. Geraldine McMahon from her publication (McCanny 2012).
The accuracy of three methods for interpreting VBG values were tested using this data:
- Estimating the ABG value as equal to the VBG value (as is often done in clinical practice).
- Conversion of VBG values into ABG values using a method published by LeMoel 2013.
- Conversion of VBG values into ABG values using the simplified saturation model with coefficients derived above.
Below are the results. The simplified saturation model provided the best prediction of ABG values. This improved accuracy substantially, compared to assuming that ABG values are roughly equal to VBG values:
Clinical bottom line
These equations aren't currently ready for clinical use (they require further validation). However, a basic physiologic truth has been illustrated here: the differences between arterial and venous blood gas are strongly relatedto differences between arterial and venous oxygen saturation.
This implies that the accuracy of a VBG can be estimated by looking at the oxygen saturation of the venous blood gas:
- If the venous oxygen saturation is high, little metabolism occurred in the tissue, so the VBG should be very close to the ABG.
- If the venous oxygen saturation is low, then substantial metabolism has occurred, so the VBG may not match up well with the ABG.
Based on some of the numbers above, this is a rough scheme that may be used to evaluate VBGs:
In practice, the oxygen saturation of VBGs is often quite high (e.g. >80%), suggesting that the VBG is extremely close to the ABG. If the VBG oxygen saturation is low, the following techniques might be used to obtain a VBG with a higher oxygen saturation:
- Minimize the duration oftourniquet application (e.g., if the patient has a venous catheter that allows blood to be withdrawn, slowly pull blood off the venous catheter without using a tourniquet).
- Don't let blood sit out at room temperature for a prolonged time (either process it immediately or place it on ice).
Limitations & Methodology
This analysis has numerous limitations, most notably some of the most precise data has been redacted. Another important limitation is that it was performed solely on studies investigating peripheral venous blood samples. Whether or not this analysis holds true for central venous specimens is unknown.
For additional details regarding methodology, attached is a copy of themanuscript from 2012.
More recent studies
Since performing this analysis, a few studies have come out suggesting that VBG values are closer to ABG values than was generally believed (e.g. Zeserson 2016). This comes as little surprise. Much of “error” in prior studies comparing VBG and ABG values was likely due to extraneous sources (e.g. sample processing, random variation in blood gas values over time, etc.).
Overall I continue to believe that VBG values are usually fine for clinical decision making. For example, if you're making major decisions based on whether the pH is 7.27 or 7.30 or 7.32, then you probably need to re-consider your medical decision-making process (3). Although the medical literature is replete with textbooks and guidelines using arbitrary ABG cutoffs, there is scant prospective evidence validating hard ABG cutoffs to guide therapy.
- The difference between ABG and VBG values depends on the amount of cellular respiration that occurs in the tissues in between.
- Oxygen saturation in the venous blood gas may be used to estimate how close VBG values are to ABG values.
- Simple formulae utilizing venous oxygen saturation may improve our ability to predict ABG values based on VBG values.
Furtherinformation:
- Original manuscript from 2012 ishere.
Notes
- There are significant differences between specialties with regards to the opinion towards using VBG data. Emergency physicians seem to best understand the utility of VBGs (because they are constantly dealing with sick, undifferentiated patients who mostly don't have arterial catheters). Currently there seems to be greater interest in VBGs in the critical care community, as we are moving away from placing A-lines and towards using end-tidal CO2 to monitor patients.
- This does assume that there is a roughly stable magnitude of error across different clinical situations.
- More on the use of ABG values in clinical decision-making in upcoming posts.
- References: Umenda 2008, Sasse 1994, Thorson 1983, Hess 1992.