Accuracy of Transcutaneous Carbon Dioxide Levels in Comparison to Arterial Carbon Dioxide Levels in Critically Ill Children

Discussion

We analyzed the accuracy of P_tcCO2 monitoring in a large diverse population of critically ill children with an extensive range of P_aCO2 values, finding that P_tcCO2 provides a clinically useful estimate of P_aCO2 (± 5 mm Hg) in most clinical conditions, including many that are associated with low cardiac output or increased subcutaneous tissue. We found that subjects with cyanotic heart disease are more likely to have P_tcCO2 values ≥ 5 mm Hg higher than P_aCO2. The specific monitor used was also important in our analysis because one of the 4 monitors we used was associated with a P_tcCO2 value ≥ 5 mm Hg higher than P_aCO2. In secondary analyses we demonstrated that the P_tcCO2 and P_aCO2 difference can be used to improve the prediction ability of subsequent P_tcCO2 measurements.

Our findings are similar to previous studies in newborns, adults, and children, demonstrating a small bias and poor precision with wide limits of agreement between P_aCO2 and P_tcCO2.^4–10 Urbano et al⁵ found similar results in 11 critically ill children monitored with the SenTec monitor, a bias of −2.1 mm Hg with 95% limits of agreement of ± 10.6 mm Hg. This previous study used a range of ± 7.5 mm Hg as acceptable and found that 81.2% of measurements fell within this range.⁵ This was similar to the 83.2% of measurements in our study where the P_tcCO2 was within ± 7.5 mm Hg of the P_aCO2.

Measurement of P_tcCO2 requires warming of the skin to hyperperfuse capillaries and facilitate diffusion of CO₂ through the skin. Children with cyanotic heart disease, particularly in the perioperative period, often have low cardiac output with decreased skin perfusion. Decreased skin perfusion causes reduced removal of CO₂ from the skin through the blood, so it is not surprising that P_tcCO2 levels ≥ 5 mm Hg higher than P_aCO2 were associated with children with cyanotic heart disease.¹ We did not find a significant association between other markers of poor cardiac output such as high lactate level, VIS, or acyanotic heart disease and a P_tcCO2 ≥ 5 mm Hg higher than P_aCO2. Lactate can be elevated for reasons other than poor perfusion, which may have led to nonsignificance. In previous studies, doses of epinephrine as high as 0.3 μg/kg/min were required to observe a diminished correlation between P_aCO2 and P_tcCO2.¹¹ We did not see this association in our data, although in our study only 5 subjects were on doses of vasoconstrictors this high. It is possible that many of the children in our study with acyanotic heart disease did not have low cardiac output at the time of measurement, and only the children with cyanotic heart disease had compromised cardiac output to the degree necessary to affect skin perfusion.

In 10.9% of the measurements the P_tcCO2 was ≥ 5 mm Hg lower than P_aCO2. For some of the measurements this was in the situation of a very high P_aCO2, as demonstrated by the significant association between a P_aCO2 ≥ 55 mm Hg and a P_tcCO2 ≥ 5 mm Hg lower than P_aCO2 in the multivariable analysis. The primary theoretical reason for this type of discrepancy in other measurements would be problems with either probe placement or technical drift of the measurement. Although all respiratory therapists were trained in appropriate probe placement, it was not required that they check probe placement prior to recording data. The accuracy of P_tcCO2 measurements may improve with more frequent monitoring of probe placement or calibration, although this would also make the monitors more cumbersome to use in a clinical setting.

An accurate, noninvasive, and continuous method to estimate P_aCO2 is desirable for optimal patient care in most critically ill children. Although noninvasive continuous oxygenation monitoring is universally clinically accepted with pulse oximetry, no such method is universally accepted for CO₂ monitoring. Assessment of both oxygenation and ventilation are necessary to determine the respiratory status of a patient. High P_aCO2 increases blood flow to the brain and decreases blood flow to the lungs. It is imperative in situations such as increased intracranial pressure or pulmonary hypertension that clinicians closely monitor P_aCO2 levels as a reflection of changing pH. While P_tcCO2 monitoring does not have the precision to replace P_aCO2 measurement routinely, our data suggest that in many children it may be a useful adjunct continuous P_CO2 monitoring method.

Most mechanically ventilated children are monitored with end-tidal capnography because it provides breath-to-breath P_CO2 values, confirming appropriate endotracheal tube placement; and for some children it provides an acceptable continuous estimate of P_aCO2. However, P_ETCO2 may be inaccurate and highly variable in periods of incomplete exhalation when dynamic hyperinflation is present.¹² Furthermore, critically ill children (particularly those with significant lung or cardiac disease) often have elevated alveolar dead space (ie, alveoli that are ventilated without perfusion).^13–15 This results in a P_ETCO2 that is lower than P_aCO2. In small infants, some clinicians argue against using P_ETCO2 monitoring at all because the monitor adds airway dead space to the ventilator circuit. Furthermore, clinicians are more commonly choosing noninvasive modes of respiratory support such as high-flow humidified nasal cannula or bi-level positive airway pressure. Children on noninvasive respiratory support often have changing cardiorespiratory pathophysiology and are a population in which close monitoring of respiratory status is imperative for the detection of clinical deterioration and timely intervention. Continuous monitoring of CO₂ levels with P_tcCO2 has the potential to address many of the known problems with P_ETCO2 monitoring in some children.

In mechanically ventilated children with ARDS, concerns regarding ventilator-induced lung injury have led to guidelines that recommend permissive hypercapnia.¹⁶ Currently, blood gases are the primary method for ventilation assessment in children with ARDS, and ventilator changes occur infrequently.¹⁷ To prevent periods of overventilation, and thus abide by permissive hypercapnia, more frequent and accurate measurements of CO₂ to prompt ventilator changes are necessary. The limits of agreement we found for P_tcCO2 are likely sufficient for this purpose. Moreover, in critically ill children with acute hypoxemic respiratory failure, elevated dead space has been associated with increased mortality.^13,14 Using P_tcCO2-based prediction modeling of P_aCO2 for dead space calculation could be feasible for noninvasive monitoring of dead space for prognostic purposes in some children.

Although some may disagree on how close a P_tcCO2 measurement should be to a P_aCO2 measurement for clinical use, values ± 5 mm Hg are close to the acceptable error of measurement within point-of-care devices for P_aCO2 measurements.¹⁸ There are limits to accuracy for all measurement devices. Some of the devices that intensivists rely on heavily to be accurate often have surprisingly wide limits of agreement. For example, while pulse oximetry is considered accurate by most clinicians, research has demonstrated that, when pulse oximetry oxygen saturation values are below normal, the limits of agreement can be quite wide and are comparable to the range we found for P_tcCO2 in this study.^19,20

Our study had several limitations. We chose to operate the V-Sign Sensor 2 at 42°C to limit the risk of skin blistering or burning (no patients experienced either in our study). However, a higher monitoring temperature may have improved the accuracy of P_tcCO2 measurements.²¹ Our analysis was limited by the sample size, which may have led to a lack of power to detect some associations. For example, it is possible that, in a larger sample size, age < 2 y old may have retained an independent association in the multivariable model for P_tcCO2 ≥ 5 mm Hg higher than P_aCO2. Furthermore, we did not measure cardiac output or subcutaneous tissue in the subjects, relying instead on surrogate measures. It is possible that measured cardiac output or subcutaneous tissue would have been more strongly associated with accuracy. We did not perform blood gases for the purposes of the study, therefore there was a variable time from monitor set up and calibration to blood gas. However, the time from calibration was not associated with accuracy in our analysis. The subjects who were selected for our study all had an arterial line in place for blood gas monitoring. In general, children with an arterial line have a higher severity of illness. It is possible that P_tcCO2 monitoring would perform differently in the larger population of critically ill children without an arterial line, although we would anticipate P_tcCO2 to perform better in this population due to the lower severity of illness, not worse. We limited our study to the SenTec P_tcCO2 monitor. It is possible that other monitors perform with higher or lower accuracy.