1 Introduction

The monitoring of carbon dioxide (CO2) level is essential for diagnosis and therapeutic guidance in mechanically ventilated and/or tracheostomized subjects [1]. Subjects with parenchymal or non-parenchymal lung disease with invasive ventilation must be monitored to assess alveolar ventilation and also to predict the need for mechanical ventilation (MV) [1]. The current gold standard method for the measurement of partial pressure of carbon dioxide (PCO2) is intermittent arterial blood gas (ABG) analysis. In addition to being a time-consuming invasive method, ABG does not provide continuous monitoring and measures arterial PCO2 (PaCO2) with arterial puncture which may be associated with increased blood loss, potentially-permanent vessel damage and catheter associated complications. Also it does not provide real-time measurement of PCO2; delaying response time in critically ill patients [2]. However, although CO2 monitoring has several non-invasive measurement techniques, PaCO2 analysis remains as the gold standard method. With today's technology, it is not possible for any non-invasive method to entirely replace PaCO2 measurements.

Ideally, non-invasive techniques of measurement should be available for continuous monitoring of oxygenation and ventilation status. For instance, pulse oximetry has proven to be a rapid, reliable and non-invasive measurement of oxygen saturation by using a non-adhesive skin sensor, but there is no similar method for measuring CO2 levels transcutaneously [3]. Transcutaneous PCO2 (PTCCO2) and end-tidal PCO2 (PETCO2) measurements serve as alternatives to PaCO2 measurement and provide continuous and non-invasive monitoring of subject. The essence of non-invasive gas monitoring is to provide information about alveolar ventilation and circulatory gas levels without the need for repetitive blood sampling [1,2,3,4].

PETCO2 monitoring via capnometer provides information on the adequacy of ventilation and displays the waveform of PCO2 in exhaled air [4]. Detection of exhaled PCO2 has proven to be a valuable mechanism to confirm tracheal intubation and recognize accidental esophageal intubations, among other critical patient safety benefits [2]. The safety enhancements provided by CO2 monitoring also include the detection of invasive airway disconnection, dislodgement or obstruction, postoperative monitoring of respiratory depression, prediction of underlying airway or lung pathologies, and monitoring the effectiveness of cardiopulmonary resuscitation [5, 6].

PTCCO2 monitors perform measurements based on the capillary bed and provide continuous information about transcutaneous CO2 through the local application of heat and measurement by electrodes [7]. Transcutaneous monitors have been more widely used in neonates because of their thinner skin which minimizes resistance to gas diffusion [8]. There are numerous studies which show good correlations between non-invasive carbon dioxide measurement methods and PaCO2 values, both in the pediatric [9,10,11,12,13] and adult population [14, 15]. However, other authors have not been able to confirm these results, while some studies demonstrate conflicting findings [16,17,18].

The objective of this study was to evaluate the relationships between the PTCCO2 and PETCO2 methods and the gold standard ABG analysis in mechanically ventilated children in the pediatric intensive care unit. The secondary objective was to assess the variability of PTCCO2 measurements in relation to subject-related factors, such as skin and subcutaneous adipose tissue thickness and pulmonary diseases.

2 Materials and methods

2.1 Subjects

This is a single-center, prospective and comparative study approved by the Clinical Research Ethics Committee of Istanbul Medeniyet University Goztepe Training and Research Hospital (study registration number: 2017-9375).

The study evaluated all children aged between 1 month and 17 years that had been intubated with cuffed ETT due to a definite indication for mechanical ventilation. The intubations were performed with single-lumen cuffed ETT with appropriate size for age and weight. Among these patients, those who accepted invasive monitoring of arterial blood pressure and provided informed consent (from the parents or legal guardians) were included in the study. The presence of any one of the following characteristics was defined as grounds for exclusion from the study: sampling performed with venous blood, non-compliance to the study protocol (premature discontinuation of measurement, incorrect installation of sensor or signal abnormality of monitor or backup), use of uncuffed endotracheal tubes, determination of any type of air leakage in the lung (pneumothorax, pneumomediastinum etc.).

2.2 Measurements

We used two non-invasive CO2 measurement methods (end-tidal CO2; PetCO2 and transcutaneous CO2; PtcCO2) and an invasive CO2 measurement method (PaCO2) via ABG, in mechanically ventilated children admitted to Istanbul Medeniyet University Hospital, Pediatric Intensive Care Unit (PICU) between November 2017 and June 2019.

2.2.1 Transcutaneous CO2 measurements

Transcutaneous CO2 was measured by using a TCM4 PTCCO2/PTCO2 device (Radiometer™, Copenhangen, Denmark, TCM4® series CombiM). The electrode membrane device was cleaned and calibrated at the beginning of measurement and repetitive calibration was applied every four hours. A small drop of sensor gel was applied to the center of the sensor membrane’s surface. The purpose of using sensor gel is to enable gas diffusion by moisturizing the skin. The electrode temperature was set to 44.0 °C to enhance sufficient blood flow in the capillaries to allow for PCO2 measurement in accordance with the manufacturer’s recommendations. There are three preselected locations in the supine position: (a) second intercostal space in the midclavicular line, (b) lateral surface of abdomen at the level of the umbilicus in the midclavicular line and (c) inner surface of the thigh. New fixation rings were used at each transcutaneous CO2 measurement location. The transcutaneous sensor was applied to the child’s chest, thigh or abdomen and was allowed to stabilize for at least 15 min prior to data recording.

2.2.2 End-tidal CO2 measurements

The CO2 sensor (Mainstream Capnostat 5 EtCO2 Sensor, Philips Healthcare, Eindhoven, Netherlands) was placed next to the tracheal cannula or intubation tube and was connected to the monitor (MX 600 Philips Intellivue™, Amsterdam, The Netherlands) for display. Calibration of the PETCO2 sensor was performed by zeroing of the sensor in room air. Calibration was done prior to measurements for each subject.

2.2.3 Arteriel blood gas analysis

PCO2 measurements from ABG were analyzed at the bedside using an ABL 90 FLEX blood gas analyzer (Radiometer, Medical ApS, Copenhagen, Denmark) within 3 min of collection. As soon as blood samples were taken for ABG analysis, PTCCO2 and PETCO2 measurements were recorded simultaneously.

2.2.4 Measurement of skin and subcutaneous adipose tissue thickness

The same radiologist performed skin and subcutaneous adipose tissue thickness measurements via ultrasonography at the point where transcutaneous CO2 sensors were placed. A linear L12-3 probe was used (EPIQ 7C, Philips, Bothell, Seattle, WA, USA). Patients were in the supine position and measurements were performed without applying pressure to the probe at the CO2 ring localizations (chest, abdomen and thigh).

2.3 Study procedure

Transcutaneous CO2 measurement was initiated from chest location in each subject. Then, thigh and abdomen measurements were taken respectively. At the 15th minute and 3rd hour after sensor fixation and calibration, PTCCO2–PETCO2 and PaCO2 measurements were recorded simultaneously for each location starting from the chest location (Fig. 1). The measurement protocol was planned to be performed in two cycles per subject –each cycle containing 6 readings (chest, thigh and abdomen readings on the 15th minute and 3rd hour), unless subjects expired or were extubated before the two cycles were complete. Subjects who could not complete at least one cycle protocol (at least two measurements per location with a total of 6 readings) were excluded from all analyses (Fig. 2). The results were recorded after sensor fixation at three locations sequentially and were compared with PaCO2 and PETCO2 results that were measured simultaneously.

Fig. 1
figure 1

Flow chart of PCO2 monitoring in mechanically ventilated subjects. PCO2 Partial pressure of carbon dioxide, PaCO2 arterial PCO2, t recording time (in minutes)

Full size image
Fig. 2
figure 2

Flow chart showing description of the trial

Full size image

Finally, a total of 1118 pairs of measurements were recorded for each measurement method. The maximum acceptable difference between PaCO2 and non-invasive CO2 measurements (PTCCO2 and PETCO2) was defined as ± 4 mmHg [19].

The following demographic characteristics, clinical features and laboratory parameters of subjects were identified: sex, age (month) and core body temperature (sensor in the esophagus). Parameters of mechanical ventilation were also recorded, including FiO2, peak pressure (Ppeak) and mean airway pressure (MAP). Measurement of the non-invasive CO2 values (PTCCO2 and PETCO2), parameters of ABG analysis (pH, PaCO2, PaO2, HCO3ˉ, base excess, haemoglobin and lactate level), inotropic index (inotropic index = dose of dopamine + dobutamine + [100 × epinephrine] + [100 × norepinephrine] + [15 × milrinone] [in microgram/kg/min]) and oxygenation index (OI) (OI = [FiO2 × MAP × 100)/PaO2]) [20, 21] were also included among the parameters of the study.

For subgroup analysis, subjects were divided into two groups according to presence of pulmonary disease. In these two groups, subjects with pulmonary disease (PD) were defined as MAP ≥ 14 mmHg and/or OI ≥ 10, and subjects with non-pulmonary disease (Non-PD) were defined as MAP < 14 mmHg and/or OI < 10). PaCO2 values were compared with PTCCO2 and PETCO2 values in both groups.

Finally, we also determined the variability in transcutaneous CO2 measurement results and various parameters, including skin and subcutaneous adipose tissue thickness, presence of PD, measurement time, and measurement location.

2.4 Statistical analysis

Analyses were performed by using the SPSS version 21 (SPSS, Inc., Chicago, IL) or Med Calc v15.8 (Med Calc Software, Ostend, Belgium) software. Subject characteristics are described using qualitative variables (using frequencies and percentages) and quantitative variables (using means and standard deviation [SD] or median with interquartile range [IQR] depending on type of distribution). Simple linear regression analysis was performed and Pearson correlation coefficients were calculated for the assessment of the relationships between PaCO2, PTCCO2 and PETCO2. CO2 values of the different methods were compared by using Friedman’s test with Bonferroni correction method for all measurements and the Wilcoxon Signed Rank test for specific locations of PTCCO2 values. Bland–Altman plots were created to evaluate the agreement between measurements. We also performed multiple linear regression analysis with stepwise selection method to determine factors affecting PTCCO2 values. Variables with a p-value lower or equal to 0.10 in univariate analysis were included into the model. P < 0.05 values were accepted as statistically significant.

3 Results

The study was performed in 102 subjects with 1118 measurements for each method. The descriptive factors of the study are shown in Table 1. The tolerance of skin to the electrode was quite good; there were no signs of skin irritation or erythema at the end of monitoring. The trial flow chart is shown in Fig. 2.

Table 1 Demographic, clinical and laboratory characteristics of subjects
Full size table

3.1 Comparison of the two non-invasive PCO2 methods with ABG analysis results

The median PaCO2, PTCCO2 and PETCO2 values were 38.9 (IQR: 34.2–44.4), 38 (IQR: 34–43) and 37 (IQR: 32–44) mmHg, respectively. Results of the Bland–Altman analysis comparing PTCCO2/PaCO2 and PETCO2/PaCO2 pairs are summarized in Table 2 and illustrated in Fig. 3 with regard to all subject groups and also subgroups. In all subjects, the mean difference between PTCCO2 and PaCO2 was − 0.78 (± 7.29) (95% limits of agreement − 15.06 to 13.51 mmHg) with moderate correlation (r = 0.66, p < 0.001) (Fig. 3a). Similarly, the mean bias between PETCO2 and PaCO2 was -2.10 (± 8.39) (95% -18.54 to 14.33 mmHg) with moderate correlation (r = 0.51, p < 0.001) (Fig. 3b). Although both PTCCO2 and PETCO2 were moderately correlated, the correlation coefficient of PTCCO2 was higher.

Table 2 Results of the Bland–Altman analysis comparing PTCCO2/PaCO2 and PETCO2/PaCO2 pairs
Full size table
Fig. 3
figure 3

Bland–Altman plots for mean PTCCO2 versus PaCO2 and mean PETCO2 versus PaCO2. PaCO2 and PTCCO2 for all subjects (a), PaCO2 and PETCO2 for all subjects (b), PaCO2 and PTCCO2 for the subjects with non-pulmonary disease (c), PaCO2 and PETCO2 for the subjects with non-pulmonary disease (d), PaCO2 and PTCCO2 for the subjects with pulmonary disease (e), PaCO2 and PETCO2 for the subjects with pulmonary disease (f). The mean difference is represented as a continuous line, and 95% limits of agreement are represented as dotted lines

Full size image

According to our findings, reliable PCO2 measurements (within the predefined, clinically acceptable range of ± 4 mmHg) could be achieved by the PTCCO2 method, but not by the PETCO2 method. The difference between PaCO2 and PTCCO2 was ≤ ± 4 mmHg in 662 measurements out of 1118 (59.2%) while the difference between the PaCO2 and PETCO2 was ≤ ± 4 mmHg in 471 measurements (42.1%) (p = 0.001).

In Fig. 4, PTCCO2 and PETCO2 measurements are illustrated for all subjects. It was found that a 1 mm Hg increase in PTCCO2 values was associated with a 0.55 mm Hg increase in PETCO2 values.

Fig. 4
figure 4

The relationship between PTCCO2 and PETCO2 measurements. A 1 mm Hg increase in PTCCO2 values was associated with a 0.55 mm Hg increase in PETCO2 values

Full size image

3.2 Subgroup analyses and comparisons

Among the subjects, 46.1% (n = 47) had PD and 53.9% (n = 55) of subjects were without pulmonary disease (non-PD). In the non-PD group, the mean bias between PTCCO2 and PaCO2 was -0.29 (± 6.05) (95% limits of agreement − 12.15 to 11.57 mmHg) (Fig. 3c), while the mean bias between PETCO2 and PaCO2 was 0.44 (± 6.83) (95% limits of agreement − 12.95 to 13.83 mmHg) (Fig. 3d). Correlation coefficients were r = 0.67 (p < 0.001) and r = 0.52 (p < 0.001), respectively. In the PD group, the mean bias between PTCCO2 and PaCO2 was − 1.27 (± 8.32) (95% limits of agreement − 17.57 to 15.04 mmHg) (Fig. 3e). Whereas the mean bias between PETCO2 and PaCO2 was − 4.65 (± 9.01) (95% limits of agreement − 22.30 to 13.01 mmHg) (Fig. 3f). Although the mean bias for PTCCO2 and PETCO2 were increased in the presence of PD, PTCCO2 was better correlated with PaCO2, compared to PETCO2 (respectively: r = 0.61, p < 0.001 vs. r = 0.53, p < 0.001).

We found that the absolute values of PTCCO2–PaCO2 were significantly lower than the absolute values of PETCO2–PaCO2 for all subjects (p < 0.001), the non-PD group (p < 0.001) and also the PD group (p < 0.001) (Table 2).

3.3 The variability in PTCCO2 measurements in relation to subject-related factors

We performed multiple linear regression analysis with PTCCO2–PaCO2 as a dependent variable to determine factors affecting differences between the measurements. We found that increased subcutaneous adipose tissue thickness (p = 0.007), body temperature (p < 0.001) and inotropic index (p = 0.002) were related with higher PTCCO2 values relative to actual PaCO2 values (Table 3). The other factors included in the model, such as age (p = 0.061), gender (p = 0.151), skin tissue thickness (p = 0.571), PaO2 (p = 0.725), presence of PD (p = 0.134), measurement time (p = 0.299), and measurement location (p = 0.121) were found to be non-significant.

Table 3 Significant factors of the differences between measurement methods (PTCCO2 and PaCO2), multiple linear regression analysis
Full size table

4 Discussion

To our knowledge, this is the most comprehensive comparison between two non-invasive techniques for continuous measurement of CO2 in pediatric subjects undergoing invasive mechanical ventilation in the PICU. It is also the largest cohort study of PTCCO2 and PETCO2 measurement in mechanically ventilated subjects with 1118 measurements for each method. We also compared PTCCO2 values with subjects’ characteristics to determine their effects on methods of PaCO2 measurement. Our results demonstrated the superiority of PTCCO2 monitoring over PETCO2 in mechanically-ventilated critically ill subjects, as demonstrated by the differences between PaCO2 values and the two methods’ results (PTCCO2 and PETCO2).

In all subject groups, the mean bias between PTCCO2 and PaCO2 was − 0.78 mmHg (± 7.29) (95% limits of agreement − 15.06 to 13.51 mmHg). In regard to PETCO2 and PaCO2 difference, the value was − 2.10 mmHg (± 8.39) (95% − 18.54 to 14.33 mmHg) in all subjects. There was a higher correlation between PaCO2 and PTCCO2 values when compared to PaCO2 and PETCO2 (respectively, r = 0.66, p < 0.001; r = 0.51, p < 0.001). Various other studies have also found better correlations between PaCO2 and PTCCO2 values (correlation coefficients between 0.83 and 0.99) [22,23,24,25]. The rather lower level of correlation in our study may be explained by the inclusion of only critically ill children who required endotracheal intubation, whereas, healthy patients may have demonstrated relatively stable levels throughout comparisons performed with different methods.

In practice, the differences in the range of non-invasive CO2 measurement methods should be within the acceptable range [2, 19]. Accordingly, our results show that PCO2 measurements within the predefined, clinically acceptable range of ± 4 mmHg could be achieved by PTCCO2, but not by PETCO2. The difference between PaCO2 and PTCCO2 was ≤ ± 4 mmHg in 662 measurements out of the complete set of 1118 values (59.2%) while the difference between PaCO2 and PETCO2 was ≤ ± 4 mmHg in 471 out of overall 1118 values (%42.1). In other studies with acceptable bias (3 to 4.5 mmHg), it was found that 29–55% of PETCO2 measurements and 61–83% PTCCO2 measurements were within the acceptable level of bias [13, 26, 27].

There are few studies comparing the accuracy of non-invasive CO2 measurement methods. Tobias-Meyer et al. [11] studied intubated subjects in the PICU and found that the mean bias between PETCO2 and PaCO2 was 6.84 mmHg (± 5.1), whereas the mean bias between PTCCO2 and PaCO2 was 2.3 mmHg (± 1.3). Transcutaneous CO2 monitoring is also used in subjects with spontaneous breathing or non-invasive mechanical ventilator support, in addition to its use in those with invasive mechanical ventilation. In a study of non-intubated subjects in spontaneous respiration, simultaneous PETCO2, PTCCO2 and PaCO2 measurements were performed and showed very high correlation values between PTCCO2 and PaCO2 (r = 0.97), while moderate correlation (r = 0.62) was observed between PETCO2 and PaCO2 values [28].

Another strength of the current investigation lies in the subgroup analysis, where CO2 monitoring techniques were performed similarly in subjects with regard to the presence or absence of PD. When compared with PETCO2, PTCCO2 has been shown to be equally as accurate in children with normal respiratory function (non-PD group). The mean differences observed in the comparison of both methods with PaCO2 values were found to be similar. This is in line with a recent investigation in mechanically ventilated subjects without parenchymal lung disease [29]. Therefore, it could be postulated that, even though PTCCO2 determination seems to be better overall, PETCO2 monitoring is sufficient and accurate in subjects receiving MV, particularly if pulmonary disease is not present.

In contrast, the differences between each method and PaCO2 values increased in the presence of PD; however, PTCCO2 values were much more accurate compared to PETCO2 values. The present and previous trials have clearly demonstrated that monitoring with PETCO2 poorly estimates PaCO2 in subjects with PD [9, 19, 30,31,32,33]. This is most often explained by ventilation-perfusion mismatching and dead-space ventilation, as these two factors are associated with inadequate gas exchange that cannot be identified via PETCO2 [34, 35]. Therefore, it is apparent that the results of PaCO2 measurements in such patients will result in a lower value relative to actual CO2 levels [36, 37].

Previous reports have shown that PaCO2 measurements tend to be higher than the corresponding PETCO2 measurements [19, 38, 39] and the presence of PD further increases the PaCO2 and PETCO2 measurement gradient [40]. The results of our study are similar to the literature. However, 95% ULA values of the PETCO2–PaCO2 gradient were determined in the range of 13.01–15.04 mmHg, and these results are quite high compared to the literature [19]. In diseases that cause hemodynamic instability, such as sepsis and shock, PETCO2 measurements tend to be higher than corresponding PaCO2 measurements [41]. High 95% ULA values in our study may be associated with the presence of patients with hemodynamic instability (such as shock and multi-trauma diagnoses) in our study, and the analysis of the highest number of measurement values in the literature so far (1118 pairs).

Although agreement was good between PTCCO2 and PaCO2, it was still limited; most possibly due to the characteristics of our patient group. We performed multiple linear regression analysis with PTCCO2-PaCO2 as the dependent variable to determine factors affecting differences between measurements. We found that increased body temperature (p < 0.001) is related with falsely high PTCCO2 values. Compared to previous studies, we had a higher number of measurements that demonstrated similar results, somewhat contrasting to previously published findings [17, 25]. Despite frequent measurement of body temperature in these critically ill patients and setting the sensor to appropriate temperature before measurements, it is still possible that the actual local pressure at the measurement site was different from patient to patient (especially since these were all critically ill patients),thereby causing differences in results. This hypothesis is directly related to the operating principle of the sensor [42].

In this study, inotropic index was found to affect the accuracy of PTCCO2 measurements. There are concerns about the accuracy of PTCCO2 in situations that may compromise CO2 washout from the tissue, such as poor skin perfusion and low cardiac output [36]. In the current study, increased inotropic index (p < 0.001) was related to higher PTCCO2 values. Although some investigators have suggested that shock does not affect PTCCO2 accuracy [12, 16, 17], others have confirmed that the gradient between PTCCO2 and PaCO2 increases as tissue perfusion decreases [43,44,45]. In our study, an objective marker (inotropic index) was used as a marker of shock, therefore, enabling more accurate analysis compared to other studies. We think that inotropic-induced vasoconstriction could be expected to reduce the accuracy of transcutaneous monitoring.

This is the first study to assess the associations between PTCCO2 – PaCO2 measurements with regard to their correlation to skin and subcutaneous adipose tissue thickness. While measurements were not affected by skin thickness (p = 0.57), they were significantly influenced by an increase in subcutaneous fat tissue thickness (p = 0.007). Several studies reported conflicting results regarding the influence of skin thickness by indirect estimation of body mass index (BMI) on the diffusion of CO2 to the skin and therefore the values of PTCCO2 [4, 16, 46, 47]. In our study, skin thickness and subcutaneous adipose tissue thickness (at sites of transcutaneous CO2 sensor placement) were measured directly by using ultrasonography—leading to comparisons based on actual measurements rather than estimates. Since we were not able to find such evaluations in previous studies, we believe our study adds important data to the existing literature pertaining to transcutaneous CO2 measurement. Based on the results of our study, we may speculate that local conditions at the site of sensor placement, including the skin-subcutaneous adipose tissue thickness and conductivity of the skin, are more important for PTCCO2 measurement than whole body composition. Similarly, local edema increases the distance over which CO2 molecules travel to the probe; therefore, it could affect PTCCO2 measurements.

The results from our analyses have important implications for how transcutaneous CO2 monitoring should be applied. No specific recommendations for a preferred site or sites are provided by manufacturers. Similarly, guidelines on transcutaneous CO2 monitoring from the American Association for Respiratory Care do not provide a recommendation for the optimal site to place a transcutaneous CO2 sensor [42]. In addition, transcutaneous CO2 measurement was obtained from three different locations (chest, thigh and abdomen) in current study. In accordance with the literature, it was found that the measurement locations do not affect the accuracy of PTCCO2 measurements [4].

Although we have reached a large series of mechanically ventilated pediatric subjects and maximum number of transcutaneous CO2 measurement in the literature, there are some limitations in the study. Firstly, transcutaneous CO2 measurements were obtained from three different body locations of the subjects at separate times. It would be possible to compare much more collected data by increasing the number of time-points for measurement, and possibly, the number of body locations. Secondly, no evaluation was made regarding the effects of the thickness of muscle tissue at the measurement site. Thirdly, we limited our study to the TCM4 Radiometer PTCCO2 monitor. It is possible that other monitors perform with higher or lower accuracy. Finally, in this study, we did not record ventilation tidal volumes during PETCO2 measurements. Particularly low tidal volumes that are not sufficient to flush the anatomic dead volume may result in gas samples that do not represent the alveolar gas status. This is quite often a cause of low PETCO2 measurements.

5 Conclusion

The PTCCO2 method has higher reliability than the PETCO2 method for non-invasive monitoring of PCO2 in children undergoing invasive MV. Especially in children with PD, it is more reliable than PETCO2. However, PTCCO2 measurement is affected by subcutaneous fat (adipose) tissue thickness, core body temperature and inotropic index. PTCCO2 cannot replace ABG analysis in mechanically ventilated pediatric subjects, but it may be very useful to define early changes in ventilation, ease clinical management, and reduce the number of invasive procedures performed for arterial blood sampling.