The early detection of hypoventilation in patients helps to prevent hypoxemia. Monitoring the partial pressure of carbon dioxide (PCO₂) is an essential means of assessing alveolar ventilation in patients during anesthesia, procedural sedation and emergency care [1-6]. Although the arterial blood gas(ABG) measurement of PCO₂ (PaCO₂) is regarded as the gold standard technique for PCO₂ assessments [7-9], it is an invasive and painful method that requires arterial blood gas analysis. Importantly, PaCO₂ also disrupts patient sleep patterns if performed at night.

In general, two different techniques for continuous non-invasive PCO₂ monitoring have been introduced into clinical practice, namely, end-tidal partial pressure of carbon dioxide (PetCO₂) and transcutaneous partial pressure of carbon dioxide (PtcCO₂) measurements [8,9]. These monitoring systems are noninvasive and display real-time data. PetCO₂ and respiratory waveforms from a capnograph can provide vital information about CO₂ retention and respiratory depression and can serve as an apnea monitor in spontaneously breathing patients [3,4,10]. PtcCO₂ sensors are not only widely used in neonatology and in critically ill infants but also used in adult patients during noninvasive mechanical ventilation, the transportation of critically ill adults, bronchoscopy etc [7,11,12].

However, according to the early studies, neither PetCO₂ nor PtcCO₂ monitoring has provided accurate estimates of PaCO₂ [13-15]. PtcCO₂ overestimates PaCO₂ and PetCO₂ significantly underestimates PaCO₂. Both the PtcCO₂ and PetCO₂ provide just an approximate estimation of PaCO₂. Clinical use of these monitors cannot be proposed under actual conditions but will be advantageous after correction of the limiting errors.

Therefore, if PetCO₂ and PtcCO₂ monitoring are determined to be just as accurate as invasive, intermittent PaCO₂ measurements, this continuous and non-invasive technique could become the preferred means for monitoring spontaneously breathing patients. The aim of this study was to investigate combining use of the PetCO₂ and PtcCO₂ methods to offset their deviation with PaCO₂. The new model namely Pet_tcCO₂ would take advantage of both PetCO₂ and PtcCO₂ methods to reduce deviations and improve measurement accuracy. The accuracy (i.e., mean difference of the measurements) and precision (i.e., standard deviation (SD) of measurements) of Pet_tcCO₂, PetCO₂, and PtcCO₂ techniques in monitoring PCO₂ were evaluated.

This study was approved by the local ethics committee for human research and was performed in accordance with ethical standards outlined in the October 2008 Declaration of Helsinki. Patients were informed of the planned study procedures and were asked for their informed consent. 26 ASA physical status I-III postoperative patients who recovering from general anesthesia after surgery were selected for this prospective clinical-experimental study. Patients with severe cardiovascular or respiratory diseases, such as coronary heart disease, congestive heart failure, or chronic obstructive pulmonary disease were excluded from this study.

A 16-gauge (16-G) IV catheter was placed into the median cubital vein for fluid transfusion, and a 20-G arterial catheter was inserted into the left radial artery under local anesthesia for ABG sampling. The catheters were flushed using a pressure bag with 500 mL of heparinized saline. Before anesthesia, patients’ heart rate (HR) and arterial blood pressure were recorded as the preinduction values. Anesthesia was induced with propofol (4 μg/mL using the target concentration infusion), fentanyl ( 2 μg/kg), and atracurium (0.6 mg/kg).

After surgery, patients were kept in the postanesthesia care unit where routine monitoring consisted of electrocardiogram and noninvasive arterial blood pressure. Patients’ lungs were ventilated with 100% oxygen (2 L/min), and their tidal volume and respiratory rate were adjusted to maintain the PetCO₂ within 35 to 45 mm Hg. Anesthesia was maintained with propofol (2 g/mL with target concentration infusion model), desflurane (6%), and fentanyl to keep the blood pressure and HR within 20% of the preinduction values. Additional analgesia was provided by IV paracetamol and morphine as required. The study period started with patients’ breathing spontaneously and receiving oxygen (3L/m-1) through a facemask.

PetCO₂ was calculated by a portable capnometer (Tidal Wave Sp, Model 715, Novametrix Medical System Inc.) from a facemask. The device had a 5-Hz internal sampling rate and identified the highest PCO₂ level during the last third of expiration by using a built-in proprietary algorithm.

To monitor PtcCO₂, a TINA TCM4 device (Radiometer Copenhagen, Brønshøj, Denmark) was used. The TCM4 device provides PtcCO₂ information continuously and noninvasively. Before the sensor being placed on patients’ ear lobes, the electrode membrane and skin at the sensor site were cleaned with alcohol, and the electrode was calibrated. The working temperature of the electrode was kept between 41 and 42°C.

A stationary ABL625 blood gas analyzer (Radiometer Copenhagen Brønshøj, Denmark) was used to determine arterial CO₂. ABG samplings were intermittently obtained for every patient at six time points (i.e., 15, 30, 45, 60, 90, and 120 min). The measurement started after the transcutaneous sensor had been on the patients’ ear lobes for 15 minutes and the patients’ blood pressure, HR, tidal volume, and respiratory rate were constant to obtain the stable PtcCO₂ and PetCO₂. PetCO₂ and PtcCO₂ were simultaneously and continuously monitored, and values were recorded at 1-min intervals before each arterial blood sample was taken.

Previous studies [5-7] showed that PetCO₂ and PtcCO₂ correlate well with PaCO₂. Most time PtcCO₂ values are higher than PaCO₂ and PetCO₂ values are lower than PaCO₂. We supposed the new Pet_tcCO₂ method could combine PetCO₂ and PtcCO₂ techniques to offset the deviation. Thus, this study integrated PetCO₂ and PtcCO₂ methods into a single approach, namely Pet_tcCO₂, and its accuracy was compared with that of PaCO₂ techniques. Because PetCO₂ and PtcCO₂ are linear correlation with PaCO₂, the combined value Pet_PtcCO₂ could be also linear correlation with PaCO₂. PaCO₂ was the dependent variable, while PetCO₂ and PtcCO₂ values were considered arguments. Experimental data were used to calculate the coefficient of the linear model.

Statistical analysis was performed with Statistical Product and Service Solutions (SPSS) software version 18. Linear relationships alone (i.e., correlations and correlation coefficients) do not adequately demonstrate the consistency between two clinical measurement techniques [16-18]. Bland-Altman analysis [19] was applied to assess the mean bias and limits of agreement (LOA) (± 2 standard deviation of bias) of PaCO₂ and PtcCO₂, PaCO₂ and PetCO₂, PaCO₂ and Pet_ tcCO₂. According to the quality criteria for blood gas analyzer devices, variation in the mean bias of PtcCO₂ and PetCO₂ and PaCO₂ of 3.5% and a variation of LOA of 12.5% could be accepted. A mean bias ≤ ± 2.5 mm Hg and LOA ≤ ± 5 mm Hg was defined as good agreement. Differences were considered statistically significant when p<0.05.

Mahalanobis distance [20] was used to compare the similarity of PCO₂ measurements by PaCO₂, PtcCO₂, PetCO₂ and Pet_tcCO₂. The smaller Mahalanobis distance of the two methods means the more similarity of the two methods.

The following equation is defined to calculate the deviation of the different methods.

(1)

Note that X=PaCO₂ and Y∈{PetCO₂, PtcCO₂, Pet_tcCO₂}. S is a covariance matrix. The variable i represents the time point of the measurements. And D_X,Y represents the Mahalanobis distance between the measure method of X and Y.

A total of 156 data sets (i.e., PaCO₂ (immediate response mobile analyzer (IRMA)), PetCO₂, and PtcCO₂) from 26 patients (18 males, 8 females) were obtained after receiving informed consent. Patients were 63 ± 17 years old (range, 31-81 years) and had body mass indices of 28.0 ± 7.9 kg.m^-2.

All data sets (n=156) were complete, and the different measurement techniques were performed without complications. Table 1 shows the results of PaCO₂, PtcCO₂, and PetCO₂ measurements at six time points: 15, 30, 45, 60, 90, 120 min.

Table 1: Comparison of PCO₂ values determined using the various methods.

In these samples, both PtcCO₂ and PetCO₂ were correlated with PaCO₂. As shown in Figure 1A and Figure 2A, the linear regression equations between PtcCO₂ and PetCO₂ and PaCO₂ were obtained (PaCO₂=0.63×PtcCO₂+12.89, r2=0.53, p<0.01; PaCO₂=0.59×PetCO₂+19.35, r2=0.45, p<0.05). In addition, PtcCO₂ was also correlated with PaCO₂ at each time point (r2=0.59, 0.65, 0.51, 0.61, 0.66 and 0.54, respectively, p<0.05). And PetCO₂ was also correlated with PaCO₂ at each time point too (r2=0.47, 0.44, 0.53, 0.65, 0.57 and 0.56, respectively, p<0.05).

Figure 1: (A) Correlation between PaCO₂ and PtcCO₂ shows a coefficient r=0.73. Regression analysis: PaCO₂=0.63×PtcCO₂+ 12.89, r=0.73, p<0.01. (B) Bias and LOA between PaCO₂ and simultaneously measured PtcCO₂ measurements. The solid line represents the mean of the differences (= bias) between PaCO₂ and PtcCO₂ (3.8±1.85 mmHg). The dashed lines are equivalent to 2SDs of the means (= LOA).

Figure 2: (A) Correlation between PaCO₂ and PetCO₂ shows a coefficient R2=0.84. Regression analysis: PaCO₂ =0.51 × PetCO₂+19.35. (B) Bias and LOA between PaCO₂ and simultaneously measured PetCO₂ measurements. The solid line represents the mean of the differences (= bias) between PaCO₂ and PetCO₂ (–3.9 mmHg). The dashed lines are equivalent to 2SDs of the means (= LOA).

Using a Bland-Altman analysis to determine LOAs (i.e., mean ± 2SD), the three techniques showed different accuracies (Figures 1B and 2B). The bias and precision of transcutaneous measurements (3.8 ± 1.85 mm Hg, not significant (n.s.)) were comparable and revealed no significant differences. The bias and precision of endtidal measurements (-3.9 ± 1.95 mm Hg, n.s.) were comparable and revealed no significant differences too. Results showed a linear relationship between PetCO₂ and PaCO₂, and the PetCO₂ method seemed to underestimate PaCO₂ values. The PetCO₂ technique resulted in 6 outliers with a negative bias. Therefore, this technique was below the given interval (mean ± 2 SD), which causes an underestimation of actual PCO₂ values determined by reference measurements. An analysis of the transcutaneous measurement revealed 10 outliers above the given interval with a positive bias. This finding suggests that the PtcCO₂ technique systematically overestimated PCO₂ levels.

The previous results showed that PetCO₂ and PtcCO₂ values were significantly correlated with the PaCO₂. SPSS was used to calculate multiple linear regression equation Parameters. The linear regression module was selected. PetCO₂ and PtcCO₂ acted as the independent variables, and PaCO₂.acted as the dependent variable. A best-fit formula (Pet_tcCO₂=7.24+0.36×PetCO₂+0.46×PtcCO₂, R2=0.66, p<0.01; Figure 3) and curve were obtained. Pet_tcCO₂ value was obtained from the formula. The difference between Pet_tcCO₂ and PaCO₂ was shown in Figure 4. The bias and precision of Pet_tcCO₂ measurement (2.1 ± 1.25 mm Hg, n.s.) were comparable and revealed no significant differences to PaCO₂.

Figure 3: Pet_tcCO₂ matched curve of PtcCO₂ and PetCO₂ was added to Average course of PaCO₂ and PtcCO₂ and PetCO₂ measurements during sedation. Arterial samples were drawn at six time points: 15, 30, 45, 60, 90, 120 min after the start of data recording; PtcCO₂ and PetCO₂ were measured continuously.

Figure 4: Correlation between PaCO₂ and Pet_tcCO₂ shows a coefficient R2=0.66. Regression analysis: PaCO₂ = 0.83 × Pet_tcCO₂+5.47. (B) Bias and LOA between PaCO₂ and simultaneously measured Pet_tcCO₂ measurements. The solid line represents the mean of the differences (= bias) between PaCO₂ and PetCO₂ (2.0 mmHg). The dashed lines are equivalent to 2SDs of the means (= LOA).

Figure 5: The Mahalanobis distance between PetCO₂ and PaCO₂ was 83.4 , PtcCO₂ and PaCO₂ was 51.3, Pet_tcCO₂ and PaCO₂ was 21.6, respectively. Pet_tcCO₂ was found to be more similar than PetCO₂ and PtcCO₂ methods.

In order to compare PtcCO₂ and PetCO₂ and Pet_tcCO₂ which had the smallest deviation to the PaCO₂, Mahalanobis distances from PaCO₂ to PtcCO₂ and PetCO₂ and Pet_tcCO₂ were calculated through the experiment data. The result was showed in Figure 4. The Mahalanobis distance between PetCO₂ and PaCO₂ was 83.4, PtcCO₂ and PaCO₂ being 51.3, Pet_tcCO₂ and PaCO₂ being 21.6, respectively. Pet_tcCO₂ was found to be more similar to PetCO₂ and PtcCO₂ methods.

The purpose of this study was to combine the use of PetCO₂ and PtcCO₂ to improve the acquisition and reliability of PCO₂ measurements of spontaneously-breathing patients after general anesthesia. Some of the earlier studies have confused the role of capnography or transcutaneous in predicting PaCO₂. Russel et al. [21] found significant correlation between capnography and arterial blood gas measurements. Belpomme et al. [22] reported that wide variations in gradients exist between PaCO₂ and PetCO₂ values that depend on patient conditions. They concluded that the PetCO₂ technique is not useful in pre-hospital ventilation management (bias up to 8.6 mm Hg). On the contrary, Yanagidate et al. [10] concluded capnograph can provide vital information about CO₂ retention and respiratory depression and can serve as an apnea monitor in spontaneously breathing patients. Some research reported that transcutaneous PtcCO₂ monitoring can continuously and reliably measure PaCO₂ in pediatric and geriatric patients during surgery [8,23,24]. But some research found the results from transcutaneous PtcCO₂ have been shown to be delayed [7,8]. Furthermore, some studies were controversial in terms of the reliability of PtcCO₂ [7,25,26]. Therefore the relationships and differences between PetCO₂ and PtcCO₂ methods and PaCO₂ method should be made clear.

Our study compared PetCO₂ and PtcCO₂ methods with PaCO₂ method. Results showed that measurements by transcutaneous methods on patients undergoing general anesthesia were in good agreement with PaCO₂ measurements (3.8 ± 1.85 mm Hg, n.s.). The PetCO₂ method also had acceptable results (-3.9 ± 1.95 mm Hg, n.s.). We found the trend PtcCO₂> PaCO₂>PetCO₂ from the result. It was the overall trend, which was based upon means of PCO₂ at each time point. Although some individual trend differed from the trend listed above. We found that PtcCO₂ was over estimated compared with PCO₂ and PetCO₂ was underestimated compared with PCO₂ most of the time. PtcCO₂ measurements were almost all in the positive direction beside PaCO₂ while PetCO₂ were in the negative direction. So we supposed that combining these two methods measured value would offset the error between them. This is the original idea when we carried out the experiment. We synthesized PetCO₂ and PaCO₂ measurements into a linear regression model and obtained the equation Pet_tcCO₂=7.24+0 .36×PetCO₂+0.46×PtcCO₂, R²=0.66, p<0.01.

The experiment result showed the trend that PtcCO₂ was higher than PaCO₂. The main reason was that the sensor temperature improved local perfusion (capillary arterialization) and increased local production of carbon dioxide in the tissue, which leads to falsely high measurements [14]. On the contrary, PetCO₂ measurement was always underestimated compared with PCO₂ because gas from lung units that were ventilated but not perfused (alveolar dead space or high ventilation-perfusion ratio units) contains little or no carbon dioxide. When this gas was mixed with the gas from “normal” lung units, the resultant concentration of carbon dioxide was reduced in the expired gas. This phenomenon accounts for the fact that PetCO₂ was lower than PaCO₂ [27]. Many authors have reported the relationship between PetCO₂ and PaCO₂ during anesthesia. Nunn and Hill [28] reported that mean PaCO₂ exceeded mean PetCO₂ by 4.5 to 4.7 mm Hg during anesthesia, with either controlled or spontaneous respiration.

Most investigators focused their reports on the accuracy and precision of PtcCO₂ or PetCO₂, whereas the reliability of them were not or less discussed. We combined used of PtcCO₂ and PetCO₂ these two methods and this can not only help to improve the accuracy and precision but also help to improve the reliability. Bland-Altman analysis was used in the present study to measure the agreement between Pet_ tcCO₂ and PaCO₂. Previous studies showed that a difference of 5 mm Hg or less between PaCO₂ and other noninvasive measurements was a clinically acceptable difference [19,29,30]. The mean of Pet_tcCO₂ sub PaCO₂ difference was 2 mm Hg, which indicated that there was no difference between Pet_tcCO₂ and PaCO₂. The Mahalanobis distance between Pet_tcCO₂ and PaCO₂ was 21.6 which was smaller than PtcCO₂ and PetCO₂, which meant Pet_tcCO₂ was more similar to PtcCO₂ and PetCO₂. The scatter diagram of Pet_tcCO₂ and PaCO₂ showed that 7 points were outside of the acceptable range. While the scatter diagram of PetCO₂ and PaCO₂ showed that 6 points were outside of the acceptable range. In fact, no new equipment was used to measure the PCO₂ so that the precision could not be improved by this method.

The combined Pet_tcCO₂ technology can be easily applied to Clinical application. Pet_tcCO₂ model could be constructed as software embedded into the original anesthesia machine or ventilator equipment. Most of the existing monitoring devices have a digital output interface. PetCO₂ and PtcCO₂ data can be easily collected and put into the processing equipment such as the anesthesia machine or ventilator equipment. PetCO₂ and PtcCO₂ act as the input variables, clinicians can see the Pet_tcCO₂ value shown from the software. Because the measure methods of PtcCO₂ by transcutaneous and PetCO₂ by endtidal carbon dioxide are monitored normally, no additional cost will be needed to carry out the new approach.

Although this combine method could improve higher accuracy for the measurement of PCO₂ in general anaesthesia on adult patients, the two different devices need to be used on the patients. For those who do not need these devices for real time monitor, the costs will be increased. But for the serious patients who want to have a ideal way to monitor PCO₂ and release pain, this technology is acceptable.

When preparing to use this new method of measuring PCO₂, it is important to consider the patients’ basic health condition. Care should be taken to ensure that the patients’ face mask should not leak gas. Transcutaneous capnometry has already been established in clinical applications. In order to make the combine module simple, findings from this study were based on patients with stable cardiac statuses, stable body temperatures, no lung disease, and normal capnograms. Thus, further experiment should be carried out on patients with severe respiratory failure to be monitored using this approach. The sample size of this study was relatively small. Comparisons of complication rates for the two different monitoring techniques will require larger patient numbers because complications in patients undergoing procedural sedations were rare in the emergency department studied. Thus, this study would need to be conducted with a larger sample size to prove its results. With small data sets, the use of artificial intelligent technology, such as support vector machines, can be used to train and obtain model parameters that realize individuation monitors and improve accuracy.

The combined use of transcutaneous and end-tidal methods could help improve the accuracy and precision of PCO₂ measurements. Results of this study demonstrate that the Pet_tcCO₂ approach of combining PetCO₂ and PtcCO₂ techniques is a new technology for PCO₂ measurement in general anaesthesia on adults patients who have no serious lung disease.