I was once called to admit a patient with idiopathic pulmonary fibrosis to the ICU because of severe hypoxemia. On exam, I was surprised to find a patient who was quite comfortable, saturating 100% on BiPAP with an FiO2 of 100%.
It turned out that the patient had previously had an oxygen saturation of 90% on four liters nasal cannula (her home oxygen prescription). An ABG had been performed, revealing a PaO2 of 54 mm. The medicine team was alarmed by this low PaO2, so they titrated up her oxygen while monitoring serial ABGs. They were unable to increase her PaO2 very much using a nasal cannula, so eventually she was placed on BiPAP.
My recommendation was to stop obtaining ABGs and to titrate the oxygen based on pulse oximetry. She was weaned back to four liters and didn’t require ICU transfer.
This case may seem silly, but it highlights some common issues. In a patient with an adequate pulse oximetry waveform, what is the best way to monitor oxygenation? What if the ABG and pulse oximetry seem to disagree?
Ten reasons why pulse oximetry is generally the best way to monitor oxygenation.
For a patient with a good pulse oximetry waveform, pulse oximetry has numerous advantages compared to ABG monitoring:
#1. Pulse oximetry is a better measurement of oxygen delivery to the tissues.
PaO2, the oxygen tension in arterial blood, is the best way to determine how well the lungs are working. However, oxygen saturation is a better measurement of the systemic oxygen delivery to the tissues (DO2)(7):
DO2 = 13.4(cardiac output)(hemoglobin)(oxygen saturation)
An ABG is painful for the wrist and the wallet. The total cost of drawing and analyzing an ABG is nearly $200 (1). In contrast, pulse oximetry is noninvasive, painless, and free (2).
#3. One ABG begets another ABG…
If you’re using the ABG as a therapeutic monitoring tool, then one ABG is rarely enough. The first ABG will often reveal some abnormality, leading to a minor change in the ventilator and repeat ABG. But the second ABG isn’t quite right, so you tweak the ventilator some more and get another ABG. Before you know it, you’ve ordered a cascade of ABGs and minor ventilator adjustments (with no real benefit to the patient).
Occasionally, an arterial catheter might even be placed for the purpose of measuring frequent ABGs. This is generally a terrible idea. The availability of an easy source of arterial blood encourages frequent ABGs and other labs as well. For example, one study found that the presence of an arterial catheter correlated with a four-fold greater volume of phlebotomy (Tarpey 1990).
Infrequently, ABGs will actually represent a venous or mixed sample. The oxygen extraction of the hand isn’t very high, so the level of oxygen in these venous samples may be only slightly lower than arterial blood. Thus, it may not be obvious that the sample was venous.
#5. Point-of-care ABG analyzers don’t actually measure oxygen saturation
Many hospitals utilize point-of-care ABG analyzers. These devices typically measure PaO2 and subsequently use this to calculate the oxygen saturation (assuming a normal PaO2 vs. oxygen saturation curve). For patients with abnormal hemoglobin dissociation curves, this calculated saturation will be wrong.
#6. ABG measurement may delay critical decisions.
Measuring an ABG may delay interventions, such as ICU transfer or intubation. Sometimes it seems like physicians check an ABG when they don’t know what to do, in order to buy some time to think. At other times, it may seem like the ICU team requests an ABG to delay accepting a patient. Occasionally, physicians may feel obligated to check an ABG before calling for help, to exercise due diligence. Regardless, the practice of delaying treatment to obtain an ABG is usually unnecessary, particularly when oxygenation is concerned (3).
#7. PaO2 values are frequently misinterpreted.
We are constantly exposed to oxygen saturation values, leading to the development of a good sense about what they mean. Meanwhile, we are exposed to PaO2 values far less often, so we may struggle to interpret them.
The most common error is panicking about a low PaO2 value. PaO2 values are always much lower than oxygen saturation values. This is simply a reflection of the oxygen saturation curve (figure above). For example, a saturation of 88% correlates to a PaO2 of ~55mm. We’re generally comfortable with a saturation of 88%, but a PaO2 of 55mm may cause concern. The lower number is scarier.
This cognitive bias is often seen when ABGs are obtained in patients on mechanical ventilation. For a patient with mild hypoxemia, the PaO2 value will often be surprisingly low. This may lead to unnecessary increases in FiO2 and PEEP, delaying extubation.
#8. Checking the A-a gradient is over-utilized and potentially misleading.
The A-a gradient is the difference in oxygen tension between arterial blood and alveolar gas. Medical school courses love this. However, trying to use the ABG to diagnose the etiology of respiratory failure works poorly in real life:
- A normal A-a gradient doesn’t exclude pulmonary embolism (Stein 1995).
- ABG analysis usually fails to distinguish between different causes of respiratory failure (H&P, ultrasonography, and radiology are more accurate).
I sometimes see practitioners measure the A-a gradient of a critically ill patient who is requiring moderate to high levels of supplemental oxygen (e.g. >3 liters/min). Measuring this is pointless, because such patients will invariably have an elevated A-a gradient (if the patient had a normal A-a gradient, then they would require at most a low amount of supplemental oxygen)(4).
#9. A single ABG only measures a snapshot in time.
We’ve all been called to evaluate a patient for low oxygen saturation. Often, the saturation will bounce back rapidly on its own. Thus, we are constantly paying attention to oxygenation trends and averaging the oxygen saturation over time. Nurses are often keenly aware of this (“yeah, he desats whenever he starts coughing, don’t worry he’ll be fine in a minute”).
If we obtain an ABG, this sort of trending and averaging is impossible. We have access to only one point in time. It is impossible to know whether the oxygen saturation was transiently low, or if it was continuously low. The usual assumption is that the ABG reflects the patient’s ongoing condition (for example, if the patient was hypoxemic 15 minutes ago, then they must still be hypoxemic now). This assumption is frequently wrong.
#10. Changes in PaO2 are widely misinterpreted.
Let’s imagine that we obtain two ABGs to determine if there has been any change in oxygenation after initiating BiPAP. After starting BiPAP, the PaO2 decreases from 56 mm down to 49 mm. The oxygenation is worsening, so this indicates that we must intubate the patient.
No, it doesn’t. Please step away from the laryngoscope. Mallat 2015 compared back-to-back ABGs drawn via arterial catheters in 129 ICU patients to determine the repeatability of this test. There are large differences between these nearly simultaneous PaO2 values. The 95% confidence interval in comparing two PaO2 values was +/- 9mm:
This is consistent with previous studies (5). Therefore, PaO2 differences in sequential ABGs will often reflect merely random variation. To be >95% confident that a difference doesn’t represent random noise, it should be >9mm. However, to be >95% confident that the a clinically significant change occurred (let’s say, >10mm change), the measured difference must be even higher, perhaps >19mm.
Changes in pulse oximetry are more likely to be meaningful, because the provider will typically look at numerous values over time. For example, if a patient had been consistently saturating in the high 90’s and is now consistently saturating in the 80s, we can be fairly confident that there has been a change (rather than representing two data points, this represents dozens of data points).
When is ABG useful to investigate oxygenation?
There are some situations when it may be helpful to use an ABG to investigate oxygenation.
#1. Pulse oximetry waveform is unreliable.
The most common situation where ABG is needed to test oxygenation is when pulse oximetry cannot provide a reliable waveform. For example, some patients have non-pulsatile blood flow from a ventricular assist device (VAD) or ECMO. Poor perfusion may lead to an erratic waveform.
#2. Diagnosis of methemoglobinemia.
Methemoglobinemia will artificially lower the patient’s measured oxygen saturation, usually producing saturations between 85-90%. This will typically cause providers to give more oxygen, which doesn’t improve the measured oxygen saturation. In this context, mismatch between the low saturation versus the PaO2 (which will often be elevated) suggests a diagnosis of methemoglobinemia (6).
#3. Calculation of the PaO2/FiO2 ratio to guide a specific therapeutic decision.
The PaO2/FiO2 ratio is often used as an index of severity of hypoxemia among patients who are intubated. Most evidence on proning in ARDS was performed using the PaO2/FiO2 ratio, including cutoffs determining which patients benefit. Thus, knowing the PaO2/FiO2 ratio may be helpful if you are contemplating whether to prone a patient.
Why measuring PaO2 is generally unhelpful.
More information isn’t always better. One example of this can be borrowed from a post by Rory Speigel about the use of brain natriuretic peptide (BNP) for diagnosing heart failure. BNP has very good test characteristics for diagnosing heart failure. However, in clinical practice it doesn’t seem to help much. Why not?
The answer is that clinicians are very good at diagnosing heart failure without BNP (8). Thus, it’s unlikely that the BNP will improve our performance much. BNP could also point us in the wrong direction. It adds more information, but not necessarily superior information to what we already have.
Adding a PaO2 to an oxygen saturation is similar. Oxygen saturation alone is an excellent measurement of the patient’s oxygenation. For most patients, it is unclear what PaO2 adds above and beyond the oxygen saturation. The added information may be more likely to mislead than to inform.
For patients with an adequate oximetry waveform, pulse oximetry is usually superior to ABG for measuring oxygenation. The top ten reasons for this are:
- Saturation is a more direct measurement of tissue oxygen delivery than PaO2.
- ABG is a painful and expensive test.
- One ABG typically leads to a cascade of ABGs, multiplying costs and blood loss.
- ABGs may be contaminated with venous blood.
- Point-of-care ABG monitors calculate oxygen saturation, rather than measuring it directly.
- Obtaining an ABG may delay management.
- PaO2 values are easily misunderstood.
- Measuring the A-a gradient is over-utilized and potentially misleading.
- ABG only measures oxygenation at a single time point.
- Changes in PaO2 are widely misinterpreted.
Related: developing a sensible approach to gas exchange physiology
- ABG unhelpful in diagnosis of cardiopulmonary failure
- ABG/VBG not helpful in DKA
- Understanding oxygen delivery in ARDS
- Oxygen-ICU trial: 100% isn't an A+
- The Case of the Dubious Squire (Rory Spiegel discusses BNP in heart failure)
- Costs obviously vary based on location, perhaps within a range of roughly $100-$200. Costs include materials themselves, procedure costs, and analytic costs.
- All critically ill patients are monitored with pulse oximetry, so there is no added cost required to obtain this.
- I used to think that getting an ABG before starting BiPAP or intubation would be essential, to know what the patient’s “baseline” was. However, the patient is in a state of flux when acutely ill, so there is no real “baseline” during this period.
- The only scenario in which calculating the A-a gradient could be useful is in a patient with mild occult hypoventilation and normal lungs. In this case, the ABG will reveal a normal A-a gradient with elevated PaCO2, proving that hypoventilation is the cause of hypoxemia. If the hypoxemia is purely due to hypoventilation, then it should be easily overcome by increasing the FiO2 slightly (e.g. using 1-2 liters of oxygen). Thus, the use of ABG to determine A-a gradient may occasionally be helpful, but only in selected cases of very mild
- Thorson 1983 and Sasse 1994.
- Of course, most patients with saturation in the 80s don’t require an ABG to exclude methemoglobinemia. Investigation for methemoglobinemia may be justified when the patient has been exposed to drugs which may cause methemoglobinemia, and when there are other signs of methemoglobinemia (e.g. disproportional cyanosis).
- Of course, oxygen saturation is only part of DO2. It's possible to have a dangerously low DO2 despite having 100% oxygen saturation (e.g. if cardiac output and/or hemoglobin are low). Nonetheless, oxygen saturation is more closely related to DO2 than PaO2 is.
- Particularly with the use of lung ultrasonography. Prior to the availability of bedside cardiac and lung ultrasonography, I think BNP might have had more of a role.
Latest posts by Josh Farkas (see all)
- IBCC Chapter & Cast:GI Hemorrhage - September 19, 2018
- IBCC chapter & cast:Troponin elevation in non-cardiac critical illness - September 12, 2018
- PulmCrit- Sepsis Research Rant: More fake news, less real news - September 10, 2018