CO₂ narcosis, also known as CO₂ poisoning or intoxication, is defined as a state where patients with hypercapnia develop frank depressed mental status, including confusion, somnolence and lethargy, which may progress to coma and death [1]. While the mechanism and even the existence of this condition have been controversial for decades, recent studies have confirmed the syndrome of CO₂ narcosis, and have demonstrated that the main contributor is ventilation/perfusion (V/Q) mismatch due to a rapid reversal of hypoxic pulmonary vasoconstriction, as opposed to the long-held belief of suppression of hypoxia-driven respiratory drive [2-6].

CO₂ retention may develop rapidly with use of supplemental oxygen in susceptible patients [2,7]. CO₂ narcosis, however, often requires significant CO₂ retention, 80 mmHg for acute accumulation, and 90-100 mmHg in chronic settings [1]. Acutely confused or somnolent hospital patients frequently trigger rapid response teams, ICU admissions, invasive procedures, prolonged hospital stay and higher costs. The recognition of CO₂ narcosis is often not straightforward.

Identification and risk stratification of susceptible patients by using a scoring system might aid medical decision-making to reduce the incidence of iatrogenic CO₂ narcosis and its associated hospital costs. We therefore sought to develop a scoring system for prediction and risk stratification of CO₂ narcosis in the hospital settings.

Study population

We conducted a single-center, retrospective cohort study of patients admitted to St. Luke’s Hospital, a community hospital located in St. Louis, MO, between June 2013 and June 2016. The study was approved by the institutional review board of St. Luke’s Hospital (20160925.1).

Inclusion criteria were: (1) Partial arterial pressure of carbon dioxide (PaCO₂)>50 mmHg on arterial blood gas (ABG) testing; (2) adults ≥ 18 years old; and (3) inpatient status. Exclusion criteria were: (1) venous blood gas samples; (2) erroneous data entry; (3) index PaCO₂ obtained immediately after cardiac arrest; (4) actively dying patients; and (5) intubation with mechanical ventilation.

Data collection

Patients’ electronic health records (Cerner Millennium Power Chart, Cerner Corp, Kansas City, MO) were electronically scanned for the indicator variable PaCO₂>50 mmHg. Identified patients were assigned to individual investigators for data extraction. Since most patients with CO₂ narcosis did not have the diagnosis code (R06.89 in ICD-10 and 786.09 in ICD-9) on file, investigators reviewed all the inpatient notes around the time of the highest PaCO₂ values. Key words, such as confusion, somnolence, lethargy, hard to arouse, and altered mental status, were used to identify potential cases. CO₂ narcosis was confirmed when no other conditions might explain the altered mental status. The presence or absence of CO₂ narcosis was also reviewed by one investigator (CY). There were no classification disagreements between reviewers. For patients with multiple ABGs or multiple admissions, the one with the highest PaCO₂ was used. Fraction of inspired oxygen (FiO₂) was calculated according to the standard oxygen flow rate [8].

Statistical analysis and score development

SAS Statistical package v9.4 (SAS Institute, Cary, NC) was used. A two-sided p2 narcosis. These variables included age, sex, body mass index (BMI), COPD status, home oxygen, opioids use within 24 hours prior to the highest PaCO₂, first serum bicarbonate on admission, and FiO₂ at the time of the highest PaCO₂. Cut-off points for variable stratification were initially determined based on the univariate analysis on variables in the derivation group. Stepwise backward selection was performed until only variables with p2 narcosis of each score.

Patient characteristics

Eight hundred and eighty-two patients with PaCO₂>50 mmHg were identified, representing 1.8% of all inpatients (Figure 1). Two patients with very low PaO₂ were excluded due to probable venous blood samples. Similarly, 5 patients were excluded due to inexplicable values. Four pediatric patients were also excluded. In an effort to avoid confounders to the development of hypercapnia, ABGs during cardiac arrest (25 patients), patients actively dying (40 patients) and patients already on mechanical ventilation (199 patients) were not included in the analysis. Mental status changes in these excluded patients were likely not caused by their own respiratory derangements, but extrarespiratory catastrophes from those conditions. Among the 607 patients who underwent analysis, 188 (31%) were determined to have CO₂ narcosis, representing 0.38% of all inpatients during the observation period.

Figure 1: Flowchart of patient inclusion.

Baseline characteristics of the CO₂ narcosis and non-narcosis cohorts are summarized in Table 1. There are no statistically significant differences in age, sex or ethnicity. Compared with the patients without CO₂ narcosis, narcosis patients had higher BMI (32.9 vs 29.4, p<.001), higher rate of home oxygen use (50% vs 31%, p<0.001), a higher percentage of respiratory diseases as the principle diagnosis on admission (65% vs 49%, p<0.001), higher levels of initial serum bicarbonate on admission (36.5 vs 31.3 mmol/L, p<0.001), higher rates of supplemental oxygen (FiO₂ 52% vs 43%, p<0.001), higher levels of PaCO₂ when altered mental status was first noticed (86.9 mmHg vs 58.6 mmHg, p<0.001), and were more likely to be transferred to ICU (63% vs 34%, p<0.001) and be intubated (28% vs 4%, p<0.001). Narcosis patients had longer stays in the ICU (4.1 vs 1.8, p<0.001) and in the hospital (10.7 vs 8.9, p<0.001).

Table 1: Characteristics of the study cohorts.

Score derivation

In order to select appropriate variables for the development of a scoring system during the early hours of admission, the variables were first examined for statistical significance and availability on admission. Only about half the patients had pulmonary function tests, therefore forced expiratory volume in one second (FEV1) was not used even it appeared to be statistically significant between the 2 cohorts. Despite virtually all patients had chest X-ray, many were obtained later during the hospitalization. Thus, presence of pleural effusion was not included in the scoring system. For similar considerations, ABG results, ICU admission and intubation were also excluded. The following 8 variables were used for developing the scoring system: age, sex, BMI, COPD status, home oxygen, opioids use, first serum bicarbonate and FiO₂.

The study population was randomly divided into two groups, 300 patients in the derivation group and 307 patients in the validation group. Univariate analysis was performed on variables to identify variables that are associated with CO₂ narcosis (Table 2). No significant differences were found between the two groups. Independent variables were derived through stepwise backward selection. Points were assigned to each individual variable based on the corresponding odds ratio and p values from the logistic regression test. FiO₂ (OR 8.91, 95% CI 2.74- 28.99) and opioids use (OR 0.28, 95% CI 0.14-0.57) demonstrate the strongest association (Table 3).

Table 2: Comparison of derivation and validation groups.

Table 3: Multivariate logistic regression in the derivation group.

Score validation

Validation was performed on the validation group by evaluating the likelihood of CO₂ narcosis of each score (Table 4). As the score increased, the likelihood also increased. There are 0% and 100% likelihood of developing CO₂ narcosis with a score of 0 and 7, respectively. Among patients stratified as high risk, the likelihood of developing CO₂ narcosis during the hospitalization is more than 50%. The area under the curve of receiver-operating characteristics in the derivation group (0.7865, 95%CI 0.7293-0.8436) is similar to the one in the validation group (0.7762, 95% CI 0.7168-0.8356), respectively (Figure 2).

Table 4: The CO₂ narcosis score.

Figure 2: Comparison of the CO₂ narcosis score in the derivation and validation groups.

CO₂ narcosis is a condition of decompensated hypercapnia that can lead to depressed mental status, coma and death if not recognized and expeditiously treated. Unfortunately, confusion and lethargy are relatively common conditions among hospitalized patients. Diagnosis is often delayed until overt mental status changes occur, which can result in ICU transfer and increased hospital costs. This study analyzed 607 inpatients with hypercapnia to develop a scoring system using readily available clinical data in the early hours of admission to predict development of CO₂ narcosis. PaO₂ and pH have previously been proposed as predictors,9 but have not been widely adopted due to low availability of ABG results. Based on the likelihood in our scoring system, patients can be stratified into low, intermediate and high-risk categories on admission, therefore appropriate interventions can be initiated early for prevention.

Worsening hypercapnia and ensuing CO₂ narcosis can develop insidiously often as a result of injudicious use of supplemental oxygen, i.e., oxygen-induced hypercapnia [9]. The previously accepted hypothesis of diminished hypoxic ventilatory drive has been proved to play only a minor role. Instead, the major contributors during the acute phase are increased dead space (48%), Haldane effect (30%) and decreased minute ventilation (22%) [1,10]. V/Q mismatch as a result of lung parenchymal damage other than alveolar hypoventilation is a chronic contributor [11,12]. Development of hypercapnia due to oxygen use may occur rapidly. PaCO₂ increases by 6-23 mmHg within 15 minutes of exposure to 100% oxygen in COPD patients [2,3]. 42 mmHg after 30 minutes of an FiO₂>28% among COPD exacerbation patients [13] and 52 mmHg over 20 min among patients with neuromuscular disorders [14] likely due to rapid V/Q mismatch from a loss of hypoxic pulmonary vasoconstriction. Many patients in our cohort developed narcosis with nasal oxygen at 2-3 liters/minute. Low flow continuous oxygen has been associated with significant hypercapnia among patients with severe COPD, neuromuscular disorder and obesity [7,15-17].

Importantly, CO₂ narcosis did not develop with the acute CO₂ retention observed in the abovementioned studies [3,7,11,13-17], suggesting a sustained elevation of PaCO₂ is required for narcosis to occur. In our cohort, most cases of CO₂ narcosis develop 1-2 days after admission. The mechanism of increased blood CO₂ and depressed level of consciousness remains largely elusive [1].

With the scoring system developed here, patients can be stratified into low, intermediate and high-risk categories during the early hours of hospitalization (Table 4). For patients at low risk (0-2 points), no intervention is required; for patients with intermediate risk (3-4 points), transcutaneous PaCO₂ sensor [18], closed-loop oxygen delivery system [19], and more frequent mental status check may reduce the incidence; and for patients in the high-risk category (5-7 points), early use of noninvasive ventilation such as BiPAP maybe indicated. Since the majority of hospitals have adopted electronic health record systems, this algorithm could be readily embedded in the system to alert clinicians automatically.

A common misconception is that supplemental oxygen or high oxygen saturation is beneficial for patients with cardiac or respiratory diseases. Despite recommendations that maintaining oxygen saturation at 88-92% balances the risks of oxygen induced hypercapnia and tissue hypoxia [10,20,21] many hospitals still have policies to keep it greater than 92%. Another common error is accepting the face value of peripheral pulse oximetry while not appreciating that it has a wide range of variations depending on severity of vasculopathy, hypotension, vasospasm, sensor positioning or manicure.

Predicting and avoiding CO₂ narcosis would likely reduce hospital costs. Patients developing CO₂ narcosis are more likely to be transferred to ICU, be intubated and mechanically ventilated, and have longer ICU and hospital stays, compared with the patients without CO₂ narcosis [13,22].

Serum bicarbonate levels represent renal compensation for respiratory acidosis and reflect the severity of chronic CO₂ retention. We observed higher levels of bicarbonate in the CO₂ narcosis cohort (36.5 vs 31.3 mmol/L, p<0.0001). We further demonstrated a stepwise association of higher baseline serum bicarbonate with higher likelihood of developing CO₂ narcosis. The lack of a significant difference between baseline bicarbonate and ABG bicarbonate during CO₂ narcosis (36.5 vs 37.5 mmol/L, p=0.516), indicates that acute renal compensation was not sufficient to prevent narcosis. When arterial blood CO₂ values were compared, much higher levels of CO₂ were found in the CO₂ narcosis cohort (86.9 vs 58.6 mmHg, p<.001), consistent with previous findings that PaCO₂ often exceed 80 mmHg for narcosis to develop.1 The distribution of PaCO₂ values between the two cohorts substantially differed (Figure 3). The highest level of PaCO₂ in our study was 138 mmHg. The highest value reported in the literature is 233 mmHg [23].

Figure 3: Distribution of PaCO₂ values between CO₂ narcosis and non-CO₂ narcosis cohorts.

Surprisingly, age appears not to be a risk factor, as CO₂ narcosis rarely occurs in young patients. Oxygen therapy has previously been associated with development of hypercapnia in elderly patients but not young adults or children [24]. Development of CO₂ narcosis might require additional insults, which likely accumulate with aging.

The levels of FiO₂ in both cohorts may not represent baseline or pre-treatment conditions. The ABGs were often obtained after patients’ clinical conditions had undergone rapid changes, and supplemental oxygen had been increased by the initial responders, before arterial blood was drawn. Therefore, the true levels of FiO₂ causing CO₂ narcosis are likely to be lower than observed, as suggested by previous studies [7,15,16]. Prospective randomized controlled trials are required to determine the true narcosis-inducing levels of FiO₂.

There are additional limitations of this study. It is a single center study with a slightly older patient population than other areas. Retrospective studies and chart review are subject to limited information and insufficient controls. Smoking history and FEV1 might be independent risk factors, but they were not well documented or only available in portions of our patients. Similarly, if morphine milligram equivalents could have been accurately calculated, low dose opioids might have been shown to be safe. Prospective randomized controlled studies will be needed to refine and validate the CO₂ narcosis scoring system.

This is the first study to investigate risk factors of CO₂ narcosis among hospitalized patients. Four independent risk factors were identified, baseline serum bicarbonate, supplemental oxygen use, BMI and opioids use. The CO₂ narcosis scoring system developed here stratifies patients into low, intermediate, and high risk groups during early hours of hospitalization using readily available clinical data. When this scoring system is applied, appropriate interventions might be employed for the prevention of CO₂ narcosis, improving patient safety and reducing costs.

Author contributions: C. Y. is the guarantor, and contributed to study conception, design, and manuscript drafting. J. X. contributed to data analysis and manuscript drafting. S. S., A. H., S. N., S. D., M. M., and O. N. contributed to data collection. D. M. contributed to study design and manuscript revision. All authors have reviewed the manuscript, warrant accuracy, and consent to publication.

None.

None.

No conflicts exist for all authors.