Transcutaneous CO2 monitors (TCOM) have been a breakthrough technology for neonatal medicine. I can remember the days without them when we faced the unsavory choice between poking the baby for a blood gas or flying blind. In the last few years, however, I have seen them used more & more – occasionally in circumstances where I think they can do more harm than good. While my anger may be madness, yet there is method in it. I will try to explain that method.
This is just my analysis of published data with a smattering of clinical experience and opinion. It comes with no guarantees and definitely should not be interpreted as medical advice for any specific scenario.
TCOM Basics
A TCOM is essentially a tiny heater and a pH probe. Laid on the patient’s skin, the heat dilates blood vessels and increases skin permeability to CO2. That CO2 predictably equilibrates with carbonic acid and so can be estimated by the pH probe. (cf a nice textbook review and a pretty good free summary) Although first described way back in 1960 by Severinghaus, I’ve seen availability & use in NICUs take off since 2010. Back in 1969, TCOMs were shown to have good linear correlation with blood gas PaCO2 in the range of 20 to 74 mm Hg, which is great…and also part of the problem.
TCOMs for Intubated Babies are Amazing!
In a mechanically ventilated baby, that is, one whose ability to show respiratory distress is quite limited by the endotracheal tube, TCOMs are a wonderful technology. They still have their limitations, but while those are interesting & important in their own way, let us waste no more space on them here.
TCOMs for Babies on Non-Invasive Respiratory Support are Tricky!
Using a TCOM in a baby on CPAP or NIPPV runs into some thorny issues of interpretation. The tl;dr:
Because TCOMs are a) fairly sensitive but poorly specific, b) worse with increasing CO2 readings and c) biased to overestimate CO2, a high reading in a baby without respiratory distress is very likely a false positive.
The statistical reasoning behind this is nearly identical to that of sepsis evaluations. As Bayes’ theorem reminds us, using a non-specific test with a low pre-test probability generates more false than true positives.
Data on TCOMs
Perhaps unsurprisingly, most people seem to get less exercised about this topic than I do, so data are not abundant. The best dataset I’ve found so far comes from Lambert 2018 with 1219 pairs of TCOM readings with arterial or capillary blood gases. First, the good news. As alluded in older data above, the correlation overall was pretty good – a Pearson correlation coefficient of ~0.8. Perfect correlation would be 1.0, so that’s not too shabby. As you might expect, correlation was a bit better for arterial gases. Certainly capillary gases have their own issues – a topic perhaps for a future post – but in the absence of a better source, let’s just lump them in with the ABGs as our gold standard.
We’ll start with a basic scatterplot and linear regression:
At first glance, it looks pretty good! I’ve coerced the y-intercept to be 0, so the slope, 1.08, implies we expect the TCOM to be 1.08 times the PaCO2. Pretty close to 1.00 which would be a perfect measure. Great, right?
Well, look at the same graph when I plot not the regression line but the slope = 1 that would represent a hypothetical perfect TCOM:
Suddenly the first challenge jumps off the screen:
The TCOM almost always overestimates the PaCO2, often by a lot.
In stats speak, the TCOM is a biased estimator. The TCOM overestimated the PaCO2 77% of the time. Over 20% of the time, it was at least 10 mm Hg over.
This may not seem so bad. And it may be as much feature as bug. One can imagine being in the TCOM manufacturer’s shoes – if errors are inevitable, “conservative” overestimates limit their risk & liability. (I can’t find the paper now, but this phenomenon was shown with O2 sat probes vs ABGs in poor med student volunteers some years ago. The sat systematically underestimated PaO2 particularly at lower levels.)
But the picture gets worse when we zoom in on the most relevant part of the data – where CO2 levels look high:
When the PaCO2 is > 55, the TCOM performs notably worse with a correlation coefficient down to 0.61. That’s a bummer because that’s when we really need it. In a slight consolation, it’s so noisy out here that only 66% of the TCOM readings are overestimates. And so we have our second challenge:
The TCOM is least reliable when it’s most needed.
…certainly when it’s theoretically most needed for babies on non-invasive ventilation. On a mechanical vent, those reliable TCOM readings in the 40s can be quite helpful. They’re opportunities to wean! But on CPAP, it’s a different story. If the baby looks distressed, no one is going to delay further action because of a reassuring TCOM. The argument is made that the TCOM might pick up clinically silent CO2 retention, particularly in babies getting high FiO2. Unfortunately, the TCOM is a very unreliable way to assess this. And of course…Bayes’ theorem is the best way to understand that.
Bayes’ Theorem & TCOMs
I’ll spare you the background here as it’s covered in my earlier post and explained better than I ever could in a video by the excellent 3Blue1Brown. The TCOM is a diagnostic test, and so like any other, the Reverend Thomas Bayes taught us how to interpret its results. We start with the sensitivity and specificity of the test.
For the sake of argument and to be somewhat generous to the TCOM, let’s say the clinically relevant sensitivity only requires the TCOM to be > 55 when the PaCO2 is > 60. As we’d expect from the bias we saw above, it’s unusual for the TCOM to be normal when the CO2 is high – 95% sensitivity by this definition. Of course, the specificity is much worse. Counting a true positive as one with PaCO2 > 65 when the TCOM > 70, we get a mere 60% specificity or a 40% false positive rate.
Of course the final, essential and oft neglected ingredient for Bayesian inference is the pre-test probability – that is, how likely do we believe hypercapnia to be before we run any test. I’ll share my guesses on this below, but here will just show the range of possibilities. Doing the Bayesian math, P(A|B) = [P(B|A) * P(A)] / P(B), yields this plot with our pre-test probability on the X-axis and the post-test probability on the y-axis:
The most notable thing about the Bayesian graph is how close to a straight line it is. That is, the TCOM doesn’t (or at least shouldn’t) change our beliefs about the baby’s CO2 very much. If you already think a baby has a high CO2, the TCOM might help confirm that. Given its good sensitivity, if you already think a baby is doing well, a normal TCOM is pretty reassuring. But if a baby seems well clinically, seeing an elevated TCOM really shouldn’t shift our impression much.
Another way to think about this is to look at the ratio of false : true positives as a function of the pre-test probability:
Doing tests at low pre-test probabilities just nets you oodles of false positives with little to show for it. In this case, even at substantial levels of clinical suspicion (high pre-test probability), we should expect a handful of false positives for every true case of hypercapnia2. (This conclusion and the hyperbolic curve are not unique to the TCOM. Every medical test has this property. They just have different levels of probability at which the false positive rate skyrockets.)
Just, Like, My Opinion, Man
As the graphs above show, the pre-test probability is the crucial factor in interpreting any test, TCOM included. This is where I see a big difference between babies with and without mechanical ventilation.
With an endotracheal tube in place, babies give us very few clinical clues to their respiratory status. Furthermore, by virtue of having needed intubation in the first place, something is wrong with their respiratory system, and often those wrong things can change rapidly. Thus, at the bedside, we are often know so little about their respiratory status as to be justly equivocal. In Bayesian terms, our pre-test probability of hypercapnia is roughly 50%, and so, following the graphs above, acting on the TCOM nets a fair amount of information and a low false positive rate.
But now imagine we have an end-tidal CO2 monitor (ETCO2) along with our TCOM, and the ETCO2 has a reliable looking tracing. We now have an additional piece of information to inform our pre-test probability. If the ETCO2 is 60 and the TCOM 80, I would be willing to bet the gas CO2 is around 70 and may merit further intervention. But if the ETCO2 is 45, our pre-test (and here I mean before looking at the TCOM) probability of hypercapnia is low. If then the TCOM says 70, the analysis above shows we have very good reason to suspect the TCOM is a false positive. Getting a gas may still be perfectly reasonable, but we shouldn’t be surprised to find a CO2 of 50.
I think the situation for a baby receiving non-invasive ventilatory support is much like the baby with a reassuring ETCO2.
First, just as the intubated baby must have something wrong with their breathing, the non-intubated baby is at least marginally healthier. Certainly some such are tenuous, but the mere fact of not being intubated reduces my pre-test probability of hypercapnia. (The same reasoning works in reverse – if a baby looks distressed on CPAP but the TCOM is normal, I’m not going to trust the TCOM despite its sensitivity being much better than its specificity.) It’s hard to put numbers on this. One way to think about it… Say I’m weaning support on an intubated baby and am mulling extubation. If I poll the RT, RN and other providers who know the baby on the chances of a successful extubation and get a probability guess below 1/3, I’m leaving the tube in. In the absence of some clinical indication to the contrary, I’d call 1/3 about the highest pre-test probability in a non-intubated baby.
Second, the baby’s clinical appearance is probably much more reliable and reassuring than any ETCO2 monitor. When a baby is not intubated and looks to be breathing comfortably, my pre-test probability of hypercapnia is quite low. I would estimate a pre-test probability far below 1/3. For the sake of argument here, I’ll just say it’s < 10%. At that level, the graphs above show how little the TCOM adds. Even a TCOM reading > 70 only gets me to at most a 20% chance of hypercapnia on a gas. At that level, I would usually watch clinically rather than keep the TCOM on.3 In my experience, in those situations, the TCOM is monitoring us providers more than the baby. We get anxious periodically and feel compelled to check gases while the baby is contentedly growing on CPAP.
Thus, while I think there is a possible window where a TCOM can be useful in a non-intubated baby, it’s a narrow one.
Micropreemies
In 2005, Aliwalas et al looked at TCOM characteristics in sub-28-weekers. Perhaps unsurprisingly, their results were even more cautious than those above with “moderate agreement” between TCOM and blood gas. Despite some theory that thin micropreemie skin may improve TCOM measurement, they found correlation coefficients ranged 0.45-0.73 and so warned in conclusion “noninvasive monitoring methods, as used in the study, cannot be substituted for PaCO2 analyses in preterm infants”. A little harsh perhaps when arterial access is difficult and respiratory status tenuous but a fair interpretation of their results.
Footnotes
- NICU providers, myself included, tend to discuss historical blood gas & TCOM results with their correlation. In doing so, we’re implying that we think a TCOM has a roughly fixed correlation with the PaCO2 – at least for a given baby over some uncertain time horizon. That is, we think the correlation has autocorrelation through time. These data have made me think that’s not the case in general. These data suggest the TCOM : PaCO2 correlation is just a random variable biased such that it’s expected value is 5-10 mm Hg above the PaCO2 but without any other major determining factor. The actual deltas we see in practice have exactly this distribution from what I can tell (admittedly without having made any attempt to measure it). Thus, I don’t think recording the TCOM-gas delta or getting a gas just to estimate the delta is helpful in most situations.
- Of course, this circles back to the problem of defining a true case of hypercapnia in practice. When the screening test is the TCOM and the next step to gold standard is a capillary gas, this can get messy.
- In addition to pain and risks of drawing a gas, I also worry about false positives from the gas. I’ve definitely seen cases where a gas was drawn that wasn’t necessary and found an elevated, presumably spurious potassium or a weird base deficit. Then we felt compelled to draw another lab sample with more pain, infection risk and another chance to find a false positive. This is especially tricky when we have to choose between an arterial stick and a venous or capillary gas. The latter are easier to get but much less reliable. Indeed, the exact same Bayesian argument above applies. In a non-intubated baby who looks clinically well, the probability of a falsely elevated CO2 on a venous or cap gas is pretty high. Indeed, it’s high enough that the joint probability of both the TCOM and the gas being false positives is not at all negligible. This could leave us in an awkward spot clinically.
