Accuracy and Interpretation of Transcutaneous Carbon Dioxide Monitoring in Critically Ill Children

Abstract

OBJECTIVES:

Transcutaneous carbon dioxide (Tcco₂) monitoring can noninvasively assess ventilation by estimating carbon dioxide (CO₂) levels in the blood. We aimed to evaluate the accuracy of Tcco₂ monitoring in critically ill children by comparing it to the partial pressure of arterial carbon dioxide (Paco₂). In addition, we sought to determine the variation between Tcco₂ and Paco₂ acceptable to clinicians to modify patient care and to determine which patient-level factors may affect the accuracy of Tcco₂ measurements.

DESIGN:

Retrospective observational cohort study.

SETTING:

Single, quaternary care PICU from July 1, 2012, to August 1, 2020.

PATIENTS:

Included participants were admitted to the PICU and received noninvasive ventilation support (i.e., continuous or bilevel positive airway pressure), conventional mechanical ventilation, or high-frequency oscillatory or percussive ventilation with Tcco₂ measurements obtained within 15 minutes of Paco₂ measurement.

INTERVENTIONS:

None.

MEASUREMENTS AND MAIN RESULTS:

Three thousand four hundred seven paired arterial blood gas and Tcco₂ measurements were obtained from 264 patients. Bland-Altman analysis revealed a bias of –4.4 mm Hg (95% CI, –27 to 18.3 mm Hg) for Tcco₂ levels against Paco₂ levels on the first measurement pair for each patient, which fell within the acceptable range of ±5 mm Hg stated by surveyed clinicians, albeit with wide limits of agreement. The sensitivity and specificity of Tcco₂ to diagnose hypercarbia were 93% and 71%, respectively. Vasoactive-Infusion Score (VIS), age, and self-identified Black/African American race confounded the relationship between Tcco₂ with Paco₂ but percent fluid overload, weight-for-age, probe location, and severity of illness were not significantly associated with Tcco₂ accuracy.

CONCLUSIONS:

Tcco₂ monitoring may be a useful adjunct to monitor ventilation in children with respiratory failure, but providers must be aware of the limitations to its accuracy.

RESEARCH IN CONTEXT.

• Transcutaneous co₂ (Tcco₂) monitoring in pediatric respiratory failure can be particularly useful during noninvasive or high-frequency ventilation, but clinical use is not standardized, and there is conflicting data on factors that affect accuracy.

• Previous studies are limited by small sample sizes, narrow inclusion criteria, and a lack of data regarding provider preferences and practices.

• We analyzed a large cohort of pediatric patients and surveys from clinical providers to determine the acceptable and actual variation between Tcco₂ and Paco₂ and which patient-level factors affect accuracy.

AT THE BEDSIDE.

• Tcco₂ has variable accuracy, which is affected by the patient’s self-identified race and degree of vasoactive use.

• Tcco₂ tends to overestimate Paco₂ and falsely reassuring Tcco₂ values in the setting of hypercarbia are relatively rare.

• Tcco₂ monitoring can be a useful adjunct to monitor ventilation with consideration of limitations.

Devices such as end-tidal co₂ (Etco₂) monitors and transcutaneous co₂ (Tcco₂) monitors can noninvasively estimate the partial pressure of carbon dioxide (Paco₂) in critical illness. These devices can be advantageous by providing continuous monitoring and not requiring access to an arterial blood sample, but have limitations to their use and accuracy (1–3). Although Etco₂ is standard of care in operating rooms and frequently used in PICUs, use is limited in both noninvasive respiratory support and high-frequency oscillatory or percussive ventilation (HFV). Additionally, accuracy of Etco₂ is affected by increased dead space in younger patients and in those with parenchymal lung disease (4–9).

Tcco₂ monitoring first became available in the early 1980s and sensors have been updated over time (10). The technology uses the “arterialization” of the capillaries through warming of the sensor and induced local hyperemia, allowing detection of CO₂ at the skin that correlates with Paco₂. As both elevated temperature and production of CO₂ by epidermal cells leads to increased tissue CO₂, Tcco₂ monitoring devices use a correction or calibration factor based on temperature (10, 11). While an advantage of Tcco₂ measurement is that it is unaffected by dead space or mode of ventilation, it is possible that factors that affect cardiac output, tissue perfusion, and subcutaneous tissue may affect measurement (12, 13). Skin pigmentation, gender, obesity, probe location, fluid balance, and renal failure have not been shown to affect Tcco₂ accuracy, but studies have had small sample sizes and have been performed in older children and adults (9, 13, 14). Some studies have shown a significant decrease in accuracy of Tcco₂ at elevated Paco₂ levels (9, 11). Finally, there is conflicting data on how Tcco₂ accuracy is affected by decreased perfusion and vasopressor use (9, 12, 15, 16).

Our study aimed to evaluate the accuracy of Tcco₂ monitoring in critically ill children based on acceptable differences between Tcco₂ and Paco₂ as determined by clinicians and to determine which patient-level factors may affect the accuracy of Tcco₂ measurements.

MATERIALS AND METHODS

This study was conducted in two parts: a survey of clinicians and a retrospective observational cohort study of critically ill children at a quaternary, freestanding children’s hospital in an urban setting. Approval for this study was obtained from the institutional review board (IRB) at Ann and Robert H. Lurie Children’s Hospital of Chicago (IRB2021-4308 titled “Provider Survey Regarding TCM Monitoring” approved March 10, 2021; IRB 2020-3844 titled “Factors Affecting the Reliability of Transcutaneous Monitoring (TCM) of Carbon Dioxide in a Pediatric Intensive Care Unit” approved November 26, 2021). Procedures were followed in accordance with the ethical standards of the responsible institutional committee on human experimentation and with the Helsinki Declaration of 1975. The retrospective observational cohort study analyzed data from patients admitted to the PICU between July 1, 2012, and August 1, 2020, who received noninvasive ventilation (NIV) support (i.e., continuous or bilevel positive airway pressure), conventional mechanical ventilation (CMV), or HFV with Tcco₂ measurements obtained within 15 minutes of a Paco₂ measurement. Participants who had both Tcco₂ and Paco₂ measurements documented and were receiving respiratory support were included. If a patient had two recorded Tcco₂ measurements within the time window, the Tcco₂ that was closest to and preceding the Paco₂ measurement was used to avoid a Tcco₂ reflecting a ventilator change made after observing a low or high Paco₂. The Sentec Digital Monitoring System (SDMS) pH based Tcco₂ device was used in our hospital during the study period. Per protocol, the probe was heated to 41°C for neonates and 42°C for all patients over term plus 1 year per manufacturer recommendations. Of note, the upper limit of measurement of Paco₂ in the point-of-care blood gas devices resulted in very high Paco₂ levels truncated at approximately 100 mm Hg. The survey was conducted to determine the variation between Tcco₂ and Paco₂ acceptable to clinicians and determine the Tcco₂ values that are likely to lead clinicians to take clinical actions. This survey consisted of an online survey with clinical scenarios and multiple-choice questions distributed to all pediatric intensive care physicians, nurse practitioners, hospitalists, and residents at Ann & Robert H. Lurie Children’s Hospital (Supplemental Fig. 1, http://links.lww.com/PCC/C529).

Data were extracted from the electronic health records. Measurements that occurred outside the 15-minute window were excluded and were not imputed. Other variables were assumed to not be present when missing (e.g., a patient with no vasoactive support recorded was assumed to have a Vasoactive-Infusion Score [VIS] of 0). R, Version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria) was used for analysis. The correlation of Tcco₂ and Paco₂ was assessed using the Pearson’s correlation coefficient (PCC), the concordance correlation coefficient (CCC), and the repeated measures correlation coefficient (RMC) accounting for the patients. The agreement between Tcco₂ and Paco₂ categories (i.e., low, normal, or high based on the clinician survey responses) was assessed by comparing disagreement in clinically relevant categories (e.g., normal Tcco₂ with high Paco₂), as well as using the weighted Cohen’s Kappa coefficient. The agreement of Tcco₂ levels with the reference Paco₂ was assessed with Bland-Altman bias analysis using the first measurement pair obtained for each patient to avoid bias of multiple measurements in one patient. To determine which patient-level factors had a significant effect on the relationship between Tcco₂ and Paco₂, a series of mixed-effect linear regression models were conducted, with Paco₂ as the dependent variable, Tcco₂ and the patient-level factors as the fixed effects, and the patients as the random effect to account for repeated measures. All pairs of Tcco₂ and Paco₂ measurements were included in the multivariable mixed-effects model. The patient-level factors of interest included: VIS, age, location of the Tcco₂ probe, percent fluid overload, weight-for-age, race, and severity of illness based on the Pediatric Risk of Mortality (PRISM) III score. All dynamic patient-level factors (type of respiratory support, VIS, percent fluid overload) reflect values at the time of Paco₂ and Tcco₂ measurement. PICU admission weight was used for the analysis. For patients without a documented PICU admission weight (< 2% of patients), a weight within the same admission or within 7 days before admission was used.

RESULTS

Demographics

A total of 3407 paired Paco₂ and Tcco₂ measurements were obtained from 264 patients. Median age of participants was 28 months (interquartile range [IQR], 9–100 mo). Most participants (47%) received HFV support, followed by NIV (30.3%) and CMV (22.7%). The demographics of the patient cohort are summarized in Table 1. A survey to assess practices and comfort with the use of Tcco₂ monitoring was distributed to healthcare providers at various levels of training and experience. There were a total of 107 survey responses with a 45% response rate. The survey data showed that 59% of respondents felt the Tcco₂ value must be within 5 mm Hg of Paco₂ to make clinical decisions. Under the assumption that Tcco₂ is within 5 mm Hg of Paco₂, the majority of respondents would either obtain a Paco₂ or titrate respiratory support for a Tcco₂ less than 30 mm Hg or greater than 49 mm Hg and perform no action if Tcco₂ falls within the range of 30–49 mm Hg. Survey contents and results are summarized in Supplemental Figures 1 and 2 (http://links.lww.com/PCC/C529).

TABLE 1.

Study Population Demographics and Clinical Characteristics

Demographics	n
Total participants	264
Median age (mo) (IQR)	27.80 (9.28–99.85)
Race/ethnicity, n (%)
White	91 (34.5)
Black/African American	64 (24.2)
Hispanic/Latino	85 (32.2)
Asian	13 (4.9)
Other	11 (4.2)
Sex, n (%)
Male	143 (54.2)
Female	121 (45.8)
Median weight (kg) (IQR)	11.9 (7.7–23)
Respiratory support mode, n (%)
Noninvasive ventilation	80 (30.3)
Conventional mechanical ventilation	60 (22.7)
High-frequency oscillatory ventilation/high-frequency percussive ventilation	124 (47.0)
Vasoactive infusion requirement, n (%)	79 (29.9)
Pediatric Risk of Mortality III score, median (IQR)	11.00 (6.00–18.00)
Length of stay, d, median (IQR)	22.00 (12.38–40.99)
Mortality while hospitalized, n (%)	36 (13.6)

IQR = interquartile range.

Agreement of Tcco₂ and Paco₂ Measurements

The median time difference between measurement pairs of Tcco₂ and Paco₂ was 6 minutes (IQR, 3–10 min). Correlation between Tcco₂ and Paco₂ for all pairs was high (PCC = 0.82, CCC = 0.81, and RMC = 0.80) (Fig. 1). The Bland-Altman plot for Tcco₂ levels against the Paco₂ gold standard on the first measurement pair for each patient is shown in Figure 2. Mean bias between Tcco₂ and Paco₂ was –4.4 mm Hg with 95% limits of agreement ranging from –27 to 18.3. The Bland-Altman analysis was similar for those receiving NIV, CMV, and HFV support (Fig. 2).

Figure 1.

Correlation between transcutaneous carbon dioxide (Tcco₂) and partial pressure of carbon dioxide (Paco₂). The correlation between Paco₂ and Tcco₂ for all measurements of patients requiring high-frequency oscillatory ventilation/high-frequency percussive ventilation, conventional mechanical ventilation, and noninvasive ventilation is high with Pearson’s correlation coefficient (PCC) = 0.82, concordance correlation coefficient (CCC) = 0.81, and repeated measures correlation coefficient (RMC) = 0.80. The increased number of samples plotted with a Paco₂ above 100 mm Hg is due to limitations of laboratory measurement (i.e., maximum level in the blood gas device is > 110 mm Hg) and may actually be higher.

Figure 2.

Bland-Altman plot for the first transcutaneous carbon dioxide (Tcco₂)– partial pressure of carbon dioxide (Paco₂) pairs. The Bland-Altman plot is based on the first measurement per patient to avoid the bias introduced by repeat measurements in the Bland-Altman plot analysis. The mean bias between Paco₂ and Tcco₂ was –4.4 mm Hg with 95% limits of agreement, shown by the dashed lines, ranging from –27 to 18.3. There is no significant difference in the distribution among those on high-frequency oscillatory ventilation/high-frequency percussive ventilation (HFOV/HFPV), conventional mechanical ventilation (CMV), and noninvasive ventilation (NIV).

Analysis of agreement between sample pairs in clinically relevant categories based on the survey results are shown in Table 2. Similar to the Bland-Altman bias analysis, these results show that Tcco₂ tends to overestimate Paco₂ and the agreement across the ordinal categories have a weighted Cohen’s Kappa = 0.40 (95% CI, 0.31–0.50). A normal or low Tcco₂ level less than 50 mm Hg rarely corresponds to a Paco₂ greater than or equal to 50 mm Hg (146/1151 [12.7%] of the sample pairs with a Tcco₂ < 50 mm Hg). An elevated Tcco₂ greater than 50 mm Hg was most often associated with an elevated Paco₂ greater than or equal to 50 mm Hg (1852/2256 [82.1%] of the sample pairs with a Tcco₂ ≥ 50 mm Hg) (Table 2). The sensitivity and specificity of Tcco₂ to diagnose hypercarbia was 93% (1852/1998 tests with a Paco₂ ≥ 50 mm Hg) and 71% (1005/1409 tests with a Paco₂ < 50 mm Hg), respectively, suggesting that Tcco₂ rarely misses hypercarbia.

TABLE 2.

Transcutaneous Carbon Dioxide Versus Paco₂ Agreement at Low, Normal, and High Paco₂ Levels

CO2 Range	Paco2 < 30 mm Hg, n = 93, n (%)	Paco2 30–50 mm Hg, n = 1316, n (%)	Paco2 > 50 mm Hg, n = 1998, n (%)
Tcco2 < 30 mm Hg (n = 36)	21 (22.6)	14 (1.1)	1 (0.1)
Tcco2 30–50 mm Hg (n = 1115)	62 (66.7)	908 (69)	145 (7.2)
Tcco2 > 50 mm Hg (n = 2256)	10 (10.8)	394 (30)	1852 (92.7)

n = number of sample pairs within range, Paco₂ = partial pressure of carbon dioxide, Tcco₂ = transcutaneous carbon dioxide.

Ranges for both Paco₂ and Tcco₂ are as follows: low < 30 mm Hg, normal 30–50 mm Hg, and high > 50 mm Hg.

Percentages are calculated from a total (n) for each of the following: low Tcco₂, normal Tcco₂, and high Tcco_2.

Weighted Cohen’s Kappa = 0.40 (95% CI, 0.31–0.50).

Tcco₂ Measurement Confounding Variables

Multivariable mixed-effect model shows that VIS, age, and self-identified Black or African American race confounded the relationship between Tcco₂ with Paco₂ (Table 3). Type of respiratory support, percent fluid overload, weight-for-age, probe location, and severity of illness based on PRISM III score were not significantly associated with Tcco₂ accuracy.

TABLE 3.

Multivariable Mixed-Effect Model

Variable	Coefficient (95% CI) in Bivariable Mixed-Effect Analysis (Paco2 ~ Tcco2 + x]	Coefficient (95% CI) in Multivariable Mixed-Effect Analysis (Paco2 ~ Tcco2 + Age [x] + Black/African American [yr] + VIS [z])
Tcco2	Variesa	0.79 (0.77–0.80)
Age, mo	0.01 (0.002–0.02)b	0.02 (0.01–0.03)
Black/African American	–2 (–3.8 to –0.3)b	–1.9 (–3.5 to –0.24)
Pediatric Risk of Mortality III score	–0.07 (–1.4 to 0.01)
Weight-for-age	–0.04 (–0.31 to 0.23)
VIS	–0.18 (–0.22 to –0.13)b	–0.18 (–0.23 to –0.13)
Cumulative fluid overload	0.01 (–0.01 to 0.03)
Probe location
Chest	Reference
Abdomen	0.81 (–0.55 to 2.2)
Back	1.15 (–0.69 to 2.98)
Thigh	1.1 (–0.02 to 2.2)
Other	0.05 (–91 to 1)
Respiratory support
Conventional mechanical ventilation	Reference
High-frequency oscillatory or percussive ventilation	–0.57 (–1.5 to 0.37)
Noninvasive ventilation	–0.70 (–2.6 to 1.2)

Paco₂ = partial pressure of carbon dioxide, Tcco₂ = transcutaneous carbon dioxide, VIS = Vasoactive-Infusion Score.

^aTcco₂ coefficient varies across bivariable analyses. The Tcco₂ coefficient in univariate mixed-effect analysis was 0.78 (95% CI, 0.76–0.80), implying that Tcco₂ is highly correlated with Paco₂ but tends to overestimate it.

^bA coefficient’s 95% CI that does not cross zero is considered statistically significant.

The intercept for the multivariable mixed-effect model was 8.5 (7.1–10). As an example, a 24-mo-old Black/African American child on vasoactive support with a VIS of 20 and a Tcco₂ of 55 mm Hg would be expected to have a Paco₂ of: 8.5 + 55 × 0.79 + 24 × 0.02 + 1 × –1.9 + 20 × –0.18 = 47 mm Hg.

DISCUSSION

Our data show that among a large cohort of pediatric participants with varying levels of respiratory support, Tcco₂ monitoring rarely misses clinically significant hypercarbia, although significant discrepancies can occur between Tcco₂ measurement and Paco₂. The majority of clinicians surveyed would make clinical decisions based on a Tcco₂ value within 5 mm Hg of Paco₂. We found that Tcco₂ levels generally fell within this range, but it tends to overestimate the Paco₂ and has relatively wide limits of agreement. Furthermore, we found that the degree of vasoactive support, age, and self-identified Black or African American race confounded the accuracy of Tcco₂. With improved understanding of the limitations to accuracy, Tcco₂ monitoring can be better used as a noninvasive, continuous estimate of Paco₂ in the critical care setting.

Tcco₂ monitoring is used in a variety of clinical scenarios in the critical care setting. For patients requiring noninvasive respiratory support, which limits the use of Etco₂ monitoring, Tcco₂ monitoring has been shown to reliably trend Paco₂ in adults and help direct clinical changes (6, 17, 18). In our cohort, Tcco₂ accuracy was similar in participants receiving noninvasive respiratory support and participants receiving invasive respiratory support. HFV modes, which also limit the ability to use Etco₂ (19–21), represented a significant proportion of this cohort and showed similar accuracy to those receiving other forms of ventilation. Although invasive arterial sampling is necessary for interpretation of acid-base status and is the gold standard for accurate CO₂ monitoring, continuous noninvasive monitoring such as Tcco₂ can guide the frequency of blood gas measurement and alert clinicians to unexpected changes in ventilation (9, 22).

In this cohort, Tcco₂ appears to generally overestimate Paco₂ with significant effects on the relationship between Tcco₂ and Paco₂ associated with the use of vasoactive drugs, age, and self-identified Black or African American race. In our dataset, higher VIS resulted in a lower predicted Paco₂ based on Tcco₂ value. This effect is small and a prior pediatric study comparing Tcco₂ to Paco₂ did not find an association between VIS and Tcco₂ value 5 mm Hg or more above or below Paco₂ (12). Clinicians should be aware of the potential association between vasopressor use and accuracy of Tcco₂. Older age is associated with a slightly higher predicted Paco₂ based on Tcco₂ than younger children. Our findings did not find an independent association between probe location and the relationship between Tcco₂ and Paco₂. A recent meta-analysis suggests that the earlobe is the best site for Tcco₂ placement, although this was not specific to the pediatric population and earlobe placement was not used in this study (23). Several recent studies focused on pulse oximetry have noted inaccuracy in patients with darker skin tones (24–26). In our cohort, self-identified Black or African American race was associated with decreased Tcco₂ as compared with Paco₂. Although the clinical effect here is small, this further supports the need for increased investigation and regulation of medical technology for influence of racial bias in the development, calibration, and validation of devices. Type of respiratory support, fluid overload, weight, and severity of illness did not have a significant association with accuracy of Tcco₂ in this study.

The survey results suggest that most clinicians would make clinical decisions based on a Tcco₂ within 5 mm Hg of arterial blood gas measurement and pursue further clinical action if a measurement outside of the range of 30–49 mm Hg was obtained. Guidelines published in 2012 by the American Association for Respiratory Care for transcutaneous monitoring reference an acceptable clinical range of agreement of Tcco₂ within 7.5 mm Hg of Paco₂ (14, 27). There are no current pediatric specific guidelines for Tcco₂ monitoring. Survey data may help inform updated guidelines from both manufacturers and medical organizations as well as providers at institutions not yet using Tcco₂.

Our study has several strengths and limitations. This study analyzes a large cohort of patients with varying degrees of respiratory support at a large academic center and includes survey results that provide some context to better understand the importance of Tcco₂ accuracy from the clinician’s standpoint. The survey was limited to a single institution with a significant number of survey respondents in medical training at the institution, potentially resulting in similar clinical approaches and a limited longitudinal exposure to CO₂ monitoring. A larger scale survey across both academic and community institutions would better evaluate the spectrum of clinical interpretation of Tcco₂. Other limitations include the retrospective nature of our analysis, such that paired measurements were not always recorded at the exact same time and it is possible that the level of CO₂ fluctuated between measurements. The patient cohort was also limited to those for whom a Tcco₂ measurement was obtained. It is possible that more frequent Paco₂ measurements were obtained in children in more dynamic situations, such as worsening respiratory status, and thus our analyses may underestimate the agreement between Tcco₂ and Paco₂ in stable conditions. Additionally, our study includes the use of a single sensor type at a single institution. Prospective studies evaluating multiple sensor types and sensor placement locations including ear lobe probes across multiple centers could better assess accuracy and investigate optimal use of Tcco₂ (16). Prior studies in neonates have shown decreased blood gas frequency with Tcco₂ use (22). This could be assessed in further studies as a possible reduction in invasive procedures and cost.

CONCLUSIONS

Tcco₂ monitoring may be a useful adjunct to monitor ventilation in children with respiratory failure within clinically relevant ranges. Although accuracy is variable, Tcco₂ monitoring rarely provides a falsely reassuring measurement in the setting of clinically significant hypercarbia. Further, Tcco₂ tends to slightly overestimate Paco₂ and the relationship between Tcco₂ and Paco₂ is confounded by the use of vasoactive drugs, self-identified Black or African American race, and age. Clinicians must be aware of these limitations to accuracy and confirm results with blood gas measurement as clinically indicated.

ACKNOWLEDGMENTS

We thank Jennifer Arzu, MPH, for early statistical support.