Endoscopic submucosal dissection (ESD) has emerged as the standard treatment for early gastric cancer. However, the procedure is time-consuming and technically demanding; furthermore, unexpected patient movement increases the risk of complications related to procedural errors. Therefore, ESD requires deeper and more prolonged sedation than diagnostic endoscopy [1]. However, deep sedation carries a risk of hypoventilation and hypoxia. In the context of sedation-related respiratory complications during ESD, a randomized clinical trial comparing midazolam and propofol reported hypoxemia incidences of 21.1% and 14.8%, respectively [2]. Thus, standard patient monitoring protocols mandate pulse oximetry and blood pressure measurement.

However, monitoring peripheral capillary oxygen saturation (SpO₂) does not ensure adequate ventilation. In the absence of hypoxemia, ventilation is primarily driven by arterial carbon dioxide (CO₂) tension, a parameter undetectable by pulse oximetry. Capnography measures the concentration of end-tidal carbon dioxide (EtCO₂), offering a noninvasive, continuous, and quantitative display of respiratory activity via waveforms representing expiration (crests) and inspiration (troughs). However, the clinical utility of capnographic monitoring during endoscopic procedures remains a subject of debate [3]. The benefit and cost-effectiveness of capnography in this setting have been questioned [4,5]. This uncertainty arises because EtCO₂ monitoring poses technical challenges during endoscopic procedures. Unlike general anesthesia, moderate sedation may lead to voluntary or passive body movements; furthermore, frequent belching caused by endoscopic air insufflation results in extreme EtCO₂ values, leading to false alarms. Our preliminary internal data also indicated a suboptimal level of patient monitoring using capnography. Given these limitations, measuring absolute EtCO₂ values—such as transient increases or decreases—may fail to accurately capture a patient’s ventilatory status during endoscopy. This is particularly relevant for procedures involving frequent body movements or belching, which introduce substantial artifacts. Consequently, we hypothesized that the pattern and magnitude of EtCO₂ fluctuation over time—termed “EtCO₂ instability”—might more accurately reflect true respiratory compromise.

This pilot observational study examined the correlation between EtCO₂ instability (and other EtCO₂-related variables) and hypoxic events during ESD. We also assessed the efficacy of nasal high-flow oxygenation (NHFO) in patients with hypoxia, aiming to maintain sufficient oxygenation without discontinuing propofol infusion or awakening the patient.

Between September 2018 and January 2019, 103 patients underwent eligibility screening. Five patients were excluded (two for pre-existing hypotension and three for acute respiratory illness); consequently, 98 eligible patients underwent both capnographic and conventional monitoring. Approximately 17.7% of EtCO₂ data points recorded within the first 30 minutes were excluded from analysis to ensure signal fidelity, primarily due to artifacts attributed to belching or oral breathing. Of the included population, 28 (28.5%) patients developed intraprocedural hypoxia (Fig. 1).

Table 1 presents the comparison of baseline characteristics between the nonhypoxic and hypoxic groups. Although most variables were comparable, the hypoxic group exhibited a significantly higher prevalence of Mallampati class III–IV (35.7% vs. 10.0%; p=0.002), sleep apnea (21.4% vs. 1.4%; p=0.002), and obstructive lung disease (21.4% vs. 2.9%; p=0.002). Procedural comparisons revealed that EtCO₂ instability was significantly higher in the hypoxic group (5.449 vs. 2.811; p<0.001), accompanied by a higher incidence of altered ventilation and transient oral respiration (Table 2). Receiver operating characteristic curve analysis identified an EtCO₂ instability cutoff value of 4.33 (SD) for predicting hypoxia (area under the curve=0.806), yielding a specificity of 87.1% and sensitivity of 64.3%.

In patients progressing from altered ventilation to hypoxia, the mean interval between events was 84 seconds. Logistic regression analysis identified EtCO₂ instability (OR=2.417; p=0.002), higher Mallampati score, oral respiration, lower baseline SpO₂, and underlying obstructive lung disease as independent predictors of hypoxia (Table 3). Furthermore, Mallampati score, oral respiration, and lower baseline SpO₂ were significantly associated with increased EtCO₂ instability (Table 4).

In patients treated with NHFO (n=16), SpO₂ levels rapidly normalized (Fig. 2). Paired t-test analysis confirmed a significant increase in mean SpO₂ from 95.77±1.49 to 99.25±0.48 (p<0.001); moreover, sedation quality was maintained or improved overall (Fig. 3). Comparison between the NHFO (n=16) and non-NHFO (n=12) groups revealed significantly greater oxygenation improvement in the NHFO group (99.25±0.48 vs. 97.75±1.54; p=0.006). However, sedation quality did not differ significantly between groups (0.564±0.428 vs. 0.766±0.389; p=0.212). No adverse events were observed during or following NHFO therapy.

Procedural sedation improves patient tolerance and enhances procedural stability during endoscopy; however, depending on dosage and individual susceptibility, it may induce sedation-related respiratory depression. Early detection and prompt management are paramount for ensuring patient safety. Nevertheless, early episodes of respiratory depression often remain undetected by visual assessment and pulse oximetry [7].

Hypoventilation is accompanied by simultaneous decreases in EtCO₂, which precede SpO₂ desaturation [7]. Similarly, our study demonstrated that altered ventilation precedes hypoxia in most patients by a quantifiable interval. However, distinguishing whether a specific fluctuation in EtCO₂ indicates a clinically significant ventilation disturbance poses challenges [8,9]. This issue is particularly pronounced in gastric ESD, specifically when trans-endoscopic CO₂ insufflation is employed. Although EtCO₂ is sensitive to hypoventilation, it is highly susceptible to artifacts caused by belching or body movement, resulting in frequent false alarms. Consequently, monitoring EtCO₂ values may have limited utility during endoscopy. Reliance on EtCO₂ values alone may generate multiple false alarms well in advance of a significant hypoxic event, largely due to procedural artifacts such as belching. In contrast, monitoring EtCO₂ instability identifies patients at increased risk of subsequent hypoxia. This provides a therapeutic window to reposition the patient, increase oxygen supply, or initiate NHFO. Rather than focusing on individual capnographic values and reacting to each alarm as in conventional capnographic monitoring, our approach emphasizes respiratory instability over time.

In the present study involving patients undergoing moderate-to-deep sedation for ESD, the calculated parameter “EtCO₂ instability” (measured within the first 30 minutes) effectively predicted hypoxia risk. Given that higher Mallampati scores and oral respiration were associated with increased EtCO₂ instability, we suggest that pronounced EtCO₂ fluctuations reflect recurrent transient upper airway obstruction, particularly at the nasopharyngeal level. Notably, factors associated with EtCO₂ instability were also strongly correlated with the occurrence of hypoxia. This indicates that EtCO₂ instability serves as a measurable metric during the procedure to predict hypoxia.

Furthermore, employing NHFO to provide positive airway pressure effectively corrects hypoxia during endoscopy without compromising sedation quality. When hypoxia occurs during ESD, chin-lift maneuvers and withholding additional sedatives are among the first decisions made. However, these measures often destabilize sedation depth, potentially disrupting the procedure. Initially, we hypothesized that sedation quality would be superior in the NHFO group; we anticipated that the rapid restoration of oxygenation by NHFO would allow the endoscopist to proceed without interrupting propofol infusion. Nevertheless, our results indicated no statistically significant difference between the groups. This lack of difference may stem from factors such as patient discomfort associated with the NHFO and its high-flow oxygen. However, given the trend toward improved sedation in the NHFO group, the lack of statistical significance may reflect an insufficient sample size.

This raises the question of whether NHFO should be applied from the onset of sedation. This approach would likely aid in maintaining ventilatory stability. Furthermore, this system could be utilized for other high-risk procedures, such as endoscopic retrograde cholangiopancreatography [10]. This would allow endoscopists to focus on the procedure with fewer concerns regarding sedation-induced hypoventilation. However, given cost-effectiveness considerations, routine use may not be justifiable. Instead, NHFO should be targeted at high-risk patients, such as those with sleep apnea, a history of sedation-related hypoxia, or high Mallampati scores. Therefore, a comprehensive preoperative review of medical history and respiratory function is essential.

Several limitations of this study should be acknowledged. First, complete blinding of observers regarding EtCO₂ measurement and sedation parameters was not feasible, potentially introducing observer bias. Second, despite strict waveform criteria, the exclusion of artifactual EtCO₂ data (e.g., belching) relied on subjective judgment. Third, our definition of hypoxia (>5% drop) focused on relative desaturation to facilitate early detection, differing from the absolute thresholds used in other studies. Fourth, the sample size for assessing NHFO efficacy was limited. Fifth, we lacked objective sedation depth indicators, such as Bispectral Index (BIS) scores. Gotoda et al. [11] demonstrated that BIS monitoring in elderly patients facilitated safe gastric ESD with lower propofol doses. Although we utilized a clinical sedation index tailored to ESD, this metric remains subjective and lacks external validation. Furthermore, additional studies are warranted to determine the optimal timing for assessing EtCO₂ instability (e.g., post-induction vs. endoscope insertion).

In conclusion, our study demonstrates that EtCO₂ instability is independently associated with hypoxic events during ESD. Capnographic monitoring is valuable, as increased EtCO₂ instability precedes hypoxia, enabling clinicians to intervene proactively. Additionally, NHFO therapy improves oxygenation and sedation stability, allowing for uninterrupted propofol administration during ESD.