Risks of hypercapnia induced by oxygen therapy have been recognized for almost 70 years.1,2,3 Indeed, hyperoxia-induced hypercapnia has been thoroughly documented in chronic obstructive pulmonary disease (COPD) patients, with a significant impact in the acute setting.3,4 Consequently, international guidelines recommend an oxygen saturation target of 88–92% for patients at risk of hypercapnia in this population.5,6,7

Obese patients may also be at risk for hyperoxia-induced hypercapnia,5,6 but the literature is sparse. Two small trials have been conducted in morbidly obese patients with obesity-hypoventilation syndrome (OHS) and an elevated baseline partial pressure of carbon dioxide (PCO2). They have, respectively, shown a mean increase in PCO2 of 5 mmHg (transcutaneous) and 3.8 mmHg after a 20-min exposure to hyperoxia.8,9 Weak recommendations for a peripheral capillary oxygen saturation (SpO2) target of 88–92% have been issued for selected patients such as the morbidly obese (grade D)5 and those with OHS (grade C).6

Whether hyperoxia-induced hypercapnia occurs in patients with milder obesity and without hypoventilation remains unknown. There is a gap in the literature concerning the impact in the acute setting as well. Meanwhile, obese patients represent no less than 41% of patients undergoing cardiac surgery10 and are known to have increased postoperative morbidity and mortality.10,11 Furthermore, hyperoxemia during and after cardiac surgery has been associated with coronary and systemic vasoconstriction, direct myocardial injury, and reduced cardiac index.12

This study aimed to test the hypothesis that a liberal oxygen saturation target ≥ 95% would be associated with higher PaCO2 values compared with a conservative oxygen saturation target of 90% in obese patients after coronary artery bypass grafting (CABG).

Methods

Study design and protocol

We conducted a single-centre randomized crossover trial at Quebec’s Heart and Lung Institute between October 2016 and January 2018. This study was approved by our Institutional Review Board (IRB #21366, September 15 2016) and written informed consent was obtained from all subjects participating in the trial. The trial was registered prior to patient enrollment at Clinicaltrials.gov (NCT02917668, Principal investigator: François Lellouche, Date of registration: September 25 2016). The manuscript adheres to the applicable CONSORT guidelines.13 Potentially eligible patients were prospectively screened through the cardiac surgery operating schedule by a research team member. Patients were invited to participate in the study the day before their procedure if they were obese (body mass index [BMI] ≥ 30 kg·m−2) and scheduled for CABG as a sole intervention. The relatively low BMI criteria was chosen to include patients with mild and moderate obesity, as this was identified as a gap in the literature. Another inclusion criterion was a SpO2 ≥ 95% before extubation. We excluded patients with either COPD, a restrictive syndrome other than one associated with obesity, or obstructive sleep apnea requiring continuous positive airway pressure (CPAP) to focus on obesity as a sole potential risk factor for hyperoxia-induced hypercapnia.

The trial was conducted in the post-cardiac surgery intensive care unit, starting within 30 min after extubation. Each patient took part in two consecutive 30-min periods in a randomized order: SpO2 target of 90 ± 2% (conservative) achieved with automated oxygen flow titration by FreeO2 (Oxynov, Quebec City, QC, Canada), and a SpO2 target ≥ 95% (liberal) achieved with routine manual oxygen flow titration by nurses as per local protocol. There was no wash-out period because of safety concerns, since most patients could not be completely weaned from oxygen therapy in this early postoperative state. Of note, in previous trials, a 20-min exposure to hyperoxia was sufficient to cause hypercapnia in patients with obesity-hypoventilation syndrome8,9 and in patients with exacerbated COPD.14 The FreeO2 device had been approved by Health Canada for research purposes (January 25 2017; approval #260546) by the time the study started, and it is currently under evaluation for commercialization in Canada. It is an automatic oxygen titration device that relies on a closed-loop system, allowing oxygen flow to be adjusted every second according to the patient’s real-time SpO2 and a target set by the physician on the controlling screen (eFig. 1 as Electronic Supplementary Material [ESM]).15 FreeO2 has proven superior to manual titration for keeping patients within a saturation target.15,16,17,18 We decided to use it in the conservative period only, to keep the liberal period as close to institutional protocol as possible. Non-rebreathing and Venturi masks were used for conservative and liberal periods respectively; neither non-invasive ventilation nor CPAP were applied during either study period. Bed head angulation was kept as constant as possible; only changes necessary for patient comfort were made by the nursing staff. All other aspects of patient care were based on institutional protocol. Protocol compliance was closely monitored by a research team member who remained at bedside for the whole protocol duration. Sequence randomization was computer generated by a research team member with randomization.org and concealment was preserved with opaque sealed envelopes. Neither the patients, the care providers, or the outcome assessors could be blinded because of the study design.

Data collection

A baseline capillary blood gas was obtained the day before surgery. On the study day, three arterial blood gases were drawn via arterial lines placed in the operating room: one before extubation and one after each study period. For the whole study, FreeO2 continuously recorded heart rate, SpO2, and oxygen flow with one value per second for each parameter. A Nonin oximeter embedded in the FreeO2 device provided non-averaged SpO2 values. FreeO2 was set in the closed-loop mode during the conservative period and in the recording mode during the liberal period. Respiratory rate, fraction of inspired oxygen (FiO2) (liberal period), arterial blood pressure, pulmonary artery pressure, as well as narcotics, vasopressor, and nitrate dosages were manually recorded by a research team member every ten minutes.

Outcomes

The primary outcome was end-of-period PaCO2. We defined as retainers the patients with any increase in PaCO2 during liberal oxygen administration. Secondary outcomes were end-of-period pH and partial oxygen pressure on arterial blood gas (PaO2); the percentage of time spent within the SpO2 target value (88–92% in conservative period and ≥ 95% in liberal period), with hypoxemia (SpO2 < 88%), with severe hypoxemia (SpO2 < 85%) and with hyperoxemia (SpO2 > 96%) in each period; and hemodynamic parameters (arterial blood pressure, pulmonary artery pressure, vasopressor dosage, lactates) measured at ten-minute intervals.

Statistical analyses

Sample size was first empirically set to 15 patients, since there was no previous study on the effect of hyperoxia on obese patients in acute care. An interim analysis of 15 patients showed that half were retainers, with a mean increase in PCO2 of 4 mmHg associated with liberal oxygen administration. In light of this tendency, the protocol was amended to increase the sample size to 30 patients to detect a 5-mmHg difference in PaCO2 between periods, with a power of 0.80 and an alpha error of 0.05.

Data were expressed as means (SD) to summarize baseline characteristics and outcomes. Sex and comorbidities were expressed as proportions. Continuous variables were analyzed using a mixed model. This statistical approach was involved to analyze a two-period crossover trial.

Estimated treatment comparison was obtained from a model with carry-over effect. Type I sum of square was used to perform tests on fixed effects in this order: carry-over effect, treatment effect, and period effect. Whereas the adjustment had no effect on standard error of the estimated treatment differences, degrees-of-freedom calculations proposed by Kenward and Roger were involved. The normality assumption was verified with the Shapiro–Wilk test on the error distribution from the statistical model after a Cholesky factorization. The Brown and Forsythe’s variation of Levene’s test statistic was used to verify the homogeneity of variances. The PaCO2 at the end of 30-min periods (primary outcome) was analyzed after adjusting for age, BMI, baseline preoperative PCO2, and cumulative narcotics dosage before study entry as independent variables in a multivariable analysis. Four measurements were recorded for respiratory rate, FiO2 (control period), arterial blood pressure, pulmonary artery pressure, as well as narcotics, vasopressor and nitrates dosages. These supplemental factors were analyzed as repeated factors with an autoregressive correlation structure among measurements. Interaction effects were investigated between these factors with the crossover model effects (three factors). The results were considered significant when P values were ≤ 0.05. All analyses were conducted using the statistical packages R v3.3.2 (R Foundation for Statistical Computing, Vienna, Austria) and SAS v9.4 (SAS Institute Inc, Cary, NC, USA).

Results

Subjects

The study was conducted between September 2016 and January 2018. Three hundred and sixty patients scheduled for CABG were screened, of which 30 were included and analyzed (Fig. 1). Subjects characteristics are shown in Table 1.

Fig. 1
figure 1

Patient selection flow chart. CABG = coronary artery bypass grafting

Full size image
Table 1 Subjects characteristics at baseline (n = 30)
Full size table

Intervention

Two groups with distinct oxygenation interventions were obtained. The mean (SD) SpO2 was 91 (1)% in the conservative period and 97 (2)% in the liberal period (P < 0.001). The randomization consisted of 17 patients beginning with conservative oxygen administration and 13 patients beginning with liberal oxygen administration. Mean (SD) bed head angulation was 38 (6) and 37 (6) degrees in the liberal and conservative periods, respectively. It was kept constant (± 2 degrees) between periods for 25/30 patients.

Primary outcome

Arterial blood gases data were complete for all 30 included patients (Table 2 and Fig. 2). Mean (SD) end-of-period PaCO2 was 42.0 (5.4) mmHg in the conservative period, compared with 42.6 (4.6) mmHg in liberal period (mean difference, − 0.6; 95% confidence interval, − 2.2 to 0.9 mmHg; P = 0.4). Multivariable analysis for age, BMI, baseline PCO2, and narcotics received before the start of the study resulted in an identical effect estimate. The only variable that had a significant interaction in the univariate analyses was baseline PCO2, but this became non-significant in the multivariable analysis. Since two patients did not reach a SpO2 of 95% in the liberal period, and electronic data were not logged for one patient, per-protocol analysis was conducted and showed no difference in the effect estimate (eTable 1 as ESM). A comparison of PaCO2 values for individual patients can be found in Fig. 2. Fourteen (47%) patients were retainers, showing a higher end-of-period PaCO2 in the liberal period compared with the conservative period with a mean (SD) increase of 3.3 (4.0) mmHg (eTable 2 as ESM; Fig. 2). Eleven (37%) patients had the opposite response, with a lower end-of-period PaCO2 in the liberal period that decreased by 1.8 (2.2) mmHg. Five (17%) patients did not show any change in end-of-period PaCO2. Looking at the PaCO2 change from baseline, the mean (SD) change in PaCO2 was 0.9 (4.3) mmHg in the liberal period, compared with − 0.6 (4.2) mmHg in the conservative period. Despite the crossover design, the sequence effect never had an observed P value below 0.15, and thus did not affect the results.

Table 2 Arterial blood gases and oxygenation during the two study conditions (intention-to-treat analysis; n = 30)
Full size table
Fig. 2
figure 2

Mean PaCO2 at the end of each period for individual patients. Data of retainers are presented with black dotted lines and data of non-retainers are presented with black solid lines. Mean values are presented in red. PaCO2 = partial pressure in CO2 (arterial blood gas)

Full size image

Secondary outcomes

There was no difference in pH or bicarbonate concentration between the two study periods (Table 2), except for slight decreases in the retainers’ population in the liberal period (eTable 2 as ESM). Pulse oximetry data recorded by FreeO2 were available for 29 patients (Table 2 and Fig. 3). Electronic data of one patient were not logged. FreeO2 maintained the patients within the conservative SpO2 target for 73 (4)% of the time. Considering the time when oxygen flow rate was > 0 L·min−1 only, this percentage (SD) rose to 82 (22)%, compared with 84 (27)% during the liberal period. During the conservative period, the percentage of time spent in hypoxemia (SpO2 < 88%) slightly increased (P = 0.04), while the percentage of time spent in hyperoxemia (SpO2> 96%) markedly decreased (P < 0.001). eFigure 2 (available as ESM) shows the oxygenation profiles (oxygen saturation, flow, and FIO2) over time during conservative and liberal oxygen administration for a typical patient.

Fig. 3
figure 3

Mean percentage of time spent within different saturation ranges in the different study periods. Conservative period is represented in blue (SpO2 target 90 ± 2%) and liberal period (SpO2 target ≥ 95%) in yellow bands. Black bars are standard deviations. Constant SpO2 monitoring and recording were done by FreeO2 with one value per second, using the FreeO2 with Nonin oximeter embedded. The device was set in FreeO2 mode (conservative period) or recording mode (liberal period). SpO2 = peripheral oxygen saturation

Full size image

Hemodynamic parameters were available for all 30 patients, except for a single missing serum lactate measurement and heart rate values for one patient, whose electronic data were not logged. A reduction in systemic blood pressure was noted after the conservative oxygenation period (Table 3), with a significant period effect that was not correlated to narcotics or nitrates dosage, nor change in PaCO2. No difference in vasopressor dosage resulted. Heart rate, pulmonary artery pressures, and lactates did not change significantly.

Table 3 Hemodynamic parameters during the two study conditions (n = 30)
Full size table

Discussion

This is the first trial to evaluate the risk of hyperoxia-induced hypercapnia in mildly to moderately obese cardiac surgery patients. We did not find a clinically important increase of PaCO2 in response to liberal oxygen administration in obese patients without other risk factors for hypoventilation, as opposed to results from two previous studies on OHS patients.8,9 This might be explained by several factors, namely a lower mean BMI of 33.6 kg·m−2 compared with 52.4 kg·m−2 and 53 kg·m−2 in the two previous studies on OHS patients,8,9 as well as moderate hyperoxia and a short duration of the exposure. Mean FiO2 was 58.6% in the present study compared with 100% in Wijesinghe’s study, which showed the greatest PaCO2 increase induced by hyperoxemia. Wijesinghe and al. studied 20 subjects with stable obesity-associated hypoventilation, comparing a FiO2 of 21% and 100% for 20 min each, and showed a mean increase in transcutaneous tension of CO2 of 5 mmHg in the hyperoxia period.8 The PaCO2 rise noted in retainers during liberal oxygen administration in the present study is similar to the overall mean difference observed in another trial, which exposed 28 stable OHS patients to a FiO2 of 28% and 50% for 20 min each.9

The percentage of time spent within oxygenation target with FreeO2 was lower than what was observed in previous studies on patients with acute respiratory failure in the emergency room (81.3%)17 and patients hospitalized for an acute exacerbation of COPD (81.2%)18 mainly because of an increase in the percentage of time spent above the SpO2 target while patients did not require any supplemental oxygen. This reflects the overall minimal oxygen requirements of our population to meet the SpO2 target of 88–92%. The percentage of time spent within target when FreeO2 was giving oxygen only was closer to the percentages seen in previous studies.17,18

A slight reduction in blood pressure was associated with lower SpO2 in the conservative oxygen administration period, which was an unexpected finding. This was not clinically significant since no increase in vasopressor requirements nor adverse outcomes occurred during the observation period. A period effect was found for systolic blood pressure only, which was probably due to chance given the small sample size. A similar effect of conservative oxygenation has been observed in a trial on oxygen therapy in myocardial infarction,19 in which a higher percentage of patients received intravenous inotropes in the ambient air group than in the oxygen group (2.1 vs 1.4%, P = 0.02). This might come from increased nitrous oxide (NO) production and systemic vasodilatation caused by hypoxemia, while hyperoxemia has been linked to decreased NO production and an increase in systemic resistances.20 A SpO2 target of 92% or 94% might be safer in obese patients after cardiac surgery, given the absence of induced hypercapnia with SpO2 above 95% and a possible detrimental effect of hypotension after CABG, but this is only speculative since we achieved moderate hyperoxia.

Strengths of this study include the crossover design, which allowed comparison of two groups, the population homogeneity, and the protocol compliance with constant bedside monitoring of the procedures. The protocol was completed for all 30 patients and there were very few missing data. We examined all available plausible factors to better characterize retainers and non-retainers but did not find any. This study compared two well separated oxygenation targets and it was possible to prove with continuous SpO2 monitoring that patients remained within oxygenation target range more than 80% of the time in both groups. The use of automated oxygen titration is another methodological strength of the study.

There are several limitations to the study. A type II error is possible because of the small sample size, the moderate hyperoxia to which the patients were exposed, and the short duration of the study. Nonetheless, a 20-min exposure to hyperoxia was sufficient to cause hypercapnia in previous trials.8,9,14 Of note, the crossover design precluded longer study periods because of the rapidly evolving postoperative state. Also, systematic preoperative spirometry and polysomnography could have provided further insight into the characteristics of retainers. Furthermore, the study was not blinded, so differential nursing care could have occurred between periods. Finally, this was a carefully selected population and extrapolation is limited.

This trial showed that moderate hyperoxia in postoperative obese patients without risk factors for hypoventilation does not affect short-term outcomes. Moreover, the results do not support a strict 88–92% saturation target in this population. More studies are needed to clarify the risk of acute hyperoxia-induced hypercapnia in obese patients with higher BMIs or risk factors for hypoventilation. The impact of higher levels and longer duration of hyperoxia should also be investigated in future trials.