Introduction

Electronic cigarettes (e-cigarettes) and vaping were developed to be a new substitute for smoking and were advertised as a safer option than conventional cigarettes despite the paucity of safety data. Electronic cigarettes entered the USA market around 2007 and have become the most common tobacco product used by American youths since 2014.1 According to a 2018 Centers for Disease Control report, 3.2% of USA adults (around 8.1 million) were current e-cigarette users.2

Fewer toxic chemicals have been detected in the aerosol from e-cigarettes compared with conventional cigarette smoking.3 Nevertheless, these new devices are not entirely harmless, and emerging pulmonary complications increase concerns about their safety. The most known harmful substance is vitamin E acetate, which is present in tetrahydrocannabinol (THC)-containing products, and has been associated with the development of e-cigarette or vaping product use associated lung injury (EVALI).4 Many other chemicals, flavours, and metals are also used in these devices and can lead to pulmonary toxicities.5,6,7

Hypoxemia and pulmonary complications are among the most common preventable postoperative complications independent of smoking and vaping.8,9 Postoperative hypoxemia is common and prolonged in patients recovering from major noncardiac surgery, with around 20% having at least 10 min·hr−1 of SpO2 < 90%.10 Postoperative hypoxemia can be caused by atelectasis, ventilator-induced lung injury, ventilation/perfusion mismatch, hypoventilation, and pulmonary edema.9,11,12 Hypoxemia is associated with prolonged hospitalization, intensive care unit admissions, mortality, and increased cost of care.

Available evidence suggests that vaping/e-cigarette use should be of concern to anesthesiologists, with a possible higher risk of postoperative hypoxemia and pulmonary complications. There are apparently no data on perioperative outcomes for patients who vape or use e-cigarettes. Therefore, we sought to undertake an evaluation of perioperative outcomes for vapers. Our primary hypothesis was that preoperative vaping/e-cigarette use, in adults undergoing noncardiothoracic surgery, is associated with increased hypoxemia (defined by the peripheral oxygen saturation divided by the fraction of inspired oxygen [SpO2/FIO2 ratio], a surrogate measure of oxygenation) during the first postoperative hour. Our secondary hypothesis was that vaping is associated with a higher risk of intraoperative and postoperative pulmonary complications compared with nonvaping.

Methods

The current analysis was a retrospective, single-centre, cohort study using data from the Cleveland Clinic Perioperative Health Documentation System (Cleveland, OH, USA) and was conducted after approval by the Institutional Review Board of the Cleveland Clinic Foundation on 7 April 2021, with waived individual consent.

We included data from all adults undergoing noncardiothoracic surgeries lasting more than 1 hr under general anesthesia with mechanical ventilation at the Cleveland Clinic Main Campus between January 2015 and April 2021. We excluded data from patients who had missing postanesthesia care unit (PACU) SpO2 and FIO2 data, as well as patients who had another surgery during the same hospitalization or received mechanical ventilation during the 48 hr before surgery. Patients who were intubated within 24 hr after surgery were excluded from the primary analysis.

The exposure of interest was vaping/e-cigarette use within one year before surgery. It was identified from providers’ notes on social history in the electronic health record and by searching for the International Classification of Diseases (ICD)-10 code (F17.29) for vaping nicotine. We also planned to record the specific type of vaping and e-cigarettes, what substance was used (e.g., nicotine, THC, etc.), the frequency of use, and dual smoking/vaping status when these data were available. Control group patients were identified as patients who did not use e-cigarettes in the year before surgery. No restrictions were placed on the absence of vaping screening as it was unavailable for most patients. Conventional cigarette smokers were part of both the treatment and control groups, but we adjusted for smoking status as a confounder in our analysis.

Our primary outcome was time-weighted average (TWA) SpO2/FIO2 ratio in the PACU during the first postoperative hour. Peripheral oxygen saturation data were collected from the electronic record. Fraction of inspired oxygen was estimated from the type of device and the oxygen flow based on the conversion table (Electronic Supplementary Material [ESM] eTable 1), assuming that FIO2 was unchanged between recordings. The SpO2/FIO2 ratio has been used as a reliable continuous and noninvasive surrogate for the partial pressure of arterial oxygen to FIO2 (PaO2/FIO2) ratio in adults with acute lung injury and acute respiratory distress syndrome (ARDS),13,14 and accepted as a replacement for the PaO2/FIO2 ratio in the respiratory part of the Sequential Organ Failure Assessment score.15 A 10% difference in the SpO2/FIO2 ratio between the two groups, corresponding to a 10% change in SpO2 at a fixed FIO2, was considered significant, a threshold based on previous studies that showed a 10% decrease in PaO2/FIO2 from baseline was clinically meaningful for lung injury.16,17

Our secondary outcome was a collapsed composite of intraoperative and postoperative pulmonary complications that occurred at any time between the beginning of the surgery and 72 hr postoperatively or discharge, whichever came first. This outcome was defined as the presence of at least one of the following complications as identified by their ICD-9/ICD-10 codes, including but not limited to: pulmonary infection and pneumonia, respiratory failure, bronchospasm, atelectasis, pulmonary oedema, pneumothorax, ARDS, pulmonary embolism, and all vaping-related disorders (ESM eTable 2).

Statistical analysis

For our primary hypothesis, we estimated the average treatment effect on the treated (ATT) for the TWA SpO2/FIO2 ratio. The ATT represents the effect of being exposed to vaping in our current vaping population and thus, correspondingly, how much harm could be prevented if patients were prevented from vaping. We used entropy balancing to adjust for confounding. This is similar to inverse probability of treatment weighting using logistic regression models, but offers certain advantages.18 Inverse probability of treatment weighting using logistic regression models is often an iterative process in which the propensity score model is tweaked and modified until satisfactory balance is achieved on confounders. Entropy balancing, on the other hand, uses optimization techniques to directly find weights that balance covariates between the two groups, thus obviating the need to perform an ad hoc search for the correct model specification.19

To calculate the ATT, we first estimated the weights using entropy balancing. All patients in the vaping group received a weight of 1, while patients in the nonvaping group received the estimated weight wi. Intuitively, the idea is to give more weight to nonvapers who are similar to vapers on confounder distribution, and less weight to those who are dissimilar. The distribution of weights was examined, and extreme weights were removed by trimming to the first and 99th percentile. Then, we evaluated the balance on the specified covariates using the absolute standardized difference (ASD), with an ASD > 0.10 indicating imbalance. Weighted outcome regression models were then fit to estimate ATT.

In the primary analysis, ATT mean difference was estimated using a weighted linear regression model with the TWA SpO2/FIO2 ratio as the outcome and vaping/e-cigarette use as the primary covariate of interest. Robust standard errors were calculated using the sandwich estimator. We adjusted for smoking status, comorbidities, and demographic factors (Table 1).

Table 1 Baseline characteristics of primary analysis patients
Full size table

For the secondary analysis, we used a similar procedure. We fitted a weighted logistic regression model to estimate the ATT odds ratio with a composite of intraoperative and postoperative pulmonary complications as the outcome and vaping/e-cigarette use status as the primary covariate of interest. Robust standard errors were calculated using the sandwich estimator.

We conducted two sensitivity analyses: the first one was conducted using the minimum SpO2/FIO2 as the outcome instead of TWA SpO2/FIO2. In the second one, we defined certain confounders to be treated as mediators. This is because a limitation of our analysis is that the patients were measured at only one time point. Thus, it is possible that some of the listed confounders are in fact mediators (e.g., a patient could have developed chronic obstructive pulmonary disease [COPD] after they started vaping). Lung cancer, COPD, and asthma were identified as potential mediators (Figure). These mediators were not used when calculating the new weights, allowing us to estimate the total effect of vaping.

Figure
figure 1

Balance on confounders between vapers and nonvapers, before and after weighting using entropy balancing

ASA = American Society of Anesthesiologists; CAD = coronary artery disease; CHF = cardiac heart failure; COPD = chronic obstructive pulmonary disease; PVD = pulmonary vascular disease

Full size image

All analyses were conducted at a significance level of 0.05. R version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria) was used for all statistical analyses.

Sample size justification

Planned

Based on a preliminary query, about 300 out of 10,000 surgery patients at the Cleveland Clinic used vaping/e-cigarettes per year. Assuming a TWA PACU SpO2/FIO2 ratio mean of 300 and a residual standard deviation of 80 after adjusting for other confounders and without considering any interaction, we planned to have more than 80% power to detect a difference of 10 or larger, assuming a minimal final sample size of 600 vapers and 19,400 nonvapers.

Actual

The final primary analysis population had 1,907 vapers out of a total of 109,217 patients. Weighting procedures, such as entropy balancing or inverse probability of treatment weighting, generally increase the variance of statistical estimates, which should be taken into account when estimating the power. The effective sample size (ESS) is a metric that quantifies this loss of precision and represents the number of unweighted observations that the weighted observations would be equivalent to. After weighting, the ESS was 1,907 vapers and 40,608 nonvapers. Keeping other parameters the same as before, we had more than 80% power to detect a mean difference of 10 or larger for TWA SpO2/FIO2 ratio as planned.

Results

We identified a total of 110,940 patients (1,941 vapers) who met the specified inclusion and exclusion criteria for the study. Of these, 109,217 (1,907 vapers) patients were included in the primary analysis after excluding patients with missing outcomes data, and those who were intubated in the 24 hr following surgery. There were no meaningful differences in the rates at which vapers were excluded for postoperative intubation (0.9%) compared with nonvapers (0.8%). All patients were included in the secondary analysis. For the primary analysis, we achieved satisfactory balance (ASD < 0.1) on all variables using entropy balancing (Table 1, Figure).

In the PACU, oxygen was delivered with a nasal cannula in 85% of the patients, with simple face masks in 14% of the patients, and with other devices in less than 1% of the patients (ESM eTable 3). This distribution was similar between both groups.

The unadjusted median [interquartile range] TWA SpO2/FIO2 ratio (N = 109,217) was 350 [302–396] in the vaping group and 348 [297–379] in the nonvaping group. The ATT mean difference [vapers − nonvapers] for the TWA SpO2/FIO2 ratio was estimated to be 4 (95% confidence interval [CI], 1 to 8; P = 0.007). This is equivalent to a 4% change in SpO2 at a 30% FIO2 (or at a fixed FIO2). The estimated treatment effect was statistically significant (Table 2). The interaction between vaping and smoking status was not statistically significant (P = 0.22). Similar results were observed when repeating the analysis with a more fine-grained categorization of smoking status, i.e., current smoker, former smoker, and never smoker (ESM eTable 4).

Table 2 Summary of primary and secondary analysis
Full size table

The intraoperative and postoperative pulmonary complications (N = 110,940) incidence was 1.4% (n = 27) in the vaping group and 1.3% (n = 1,447) in the nonvaping group. The ATT odds ratio (vapers/nonvapers) for experiencing pulmonary complications was estimated to be 1.04 (95% CI, 0.71 to 1.54; P = 0.84). The estimated treatment effect was not statistically significant (Table 2).

Our first sensitivity analysis showed that, when we used the minimum SpO2/FIO2 ratio as the outcome, the ATT mean difference was estimated to be 2 (95% CI, −2 to 5; P = 0.35). The average minimum SpO2/FIO2 ratio was estimated to be 309 (95% CI, 303 to 315) in the vaping group and 307 (95% CI, 301 to 303) in the nonvaping group.

In the second analysis, we excluded predefined mediators, i.e., asthma, COPD, and lung cancer. The ATT mean difference for the TWA SPO2/FIO2 ratio was estimated to be 4 (95% CI, 1 to 7; P = 0.01). The mean TWA SpO2/FIO2 ratio was estimated to be 348 (95% CI, 344 to 352) in the vaping group and 344 (95% CI, 342 to 346) in the nonvaping group.

Discussion

Our primary outcome was hypoxemia (SpO2/FIO2 ratio), a reliable surrogate for the PaO2/FIO2 ratio.13,14 We used this outcome to power the detection of vaping/e-cigarette use effect on the lungs even without having a definite documented diagnosis of a pulmonary complication. Although there is no other available perioperative data regarding vaping/e-cigarette use and hypoxemia that we can refer to or establish a comparison with, it is worth reporting that, in the nonoperative setting, hypoxia was the main presenting vital abnormality in vapers/e-cigarette users that were subsequently diagnosed with EVALI.20 Based on these data, we expected to observe more severe hypoxemia in vapers/e-cigarette users than in nonusers. The results of our analysis ultimately did not support this hypothesis.

Vaping/e-cigarette use was also not associated with a statistically significant increase in intraoperative and postoperative pulmonary complications. As with hypoxemia, there are no perioperative data on the effects of vaping on postoperative pulmonary complications. Nevertheless, a previous study in the nonoperative setting compared the lung function of 30 healthy individuals who had vaped in the last six months to 30 control individuals who had not. In the vaping group, they observed a significant reduction in lung function secondary to peripheral obstructive airway disease.21

Multiple factors could explain the lack of evidence for an association between vaping and pulmonary complications observed in our study. One important reason might be that our population was at lower risk of developing pulmonary complications. Only 28% had intra-abdominal surgeries, where complications are common. Another reason might be that we did not account for duration or intensity of vaping. It is reasonable to hypothesize that higher levels of exposure to vaping would result in greater lung damage. It has been shown that prolonged exposure to toxic substances and more severe lung damage is needed to detect postoperative pulmonary complications in middle age.22,23,24 Nevertheless, these details were not available for our analysis.

Our results should be interpreted cautiously considering the many limitations of our study. One limitation is that we used SpO2 as a continuous and noninvasive surrogate, instead of arterial blood samples, to assess postoperative lung function in the PACU. Nevertheless, the SpO2/FIO2 ratio is a validated measure of oxygenation that allows continuous and early detection of impaired oxygenation and lung injury, and it is a predictor of early development of ARDS and hospital mortality.13,25,26 Another limitation is that we assessed oxygenation only within the first hour of admission to the PACU. Therefore, delayed effects of vaping on oxygenation would be overlooked. A third important limitation is the risk of exposure underdocumentation since it is a new entity in preoperative screening. This would explain the low prevalence of vaping in our patient population compared with estimates from other studies. In general, providers are not used to completing and documenting vaping and e-cigarette use, or they may erroneously document it as smoking, which could be confused with cigarette smoking. Nevertheless, the Cleveland Clinic has already added a specific screening tool for vaping to its electronic medical record system, and we also used a free-text search to enhance the detection of exposure. A related limitation is that the specific type of substance used does not need to be documented in the screening tool. It is important to differentiate between nicotine, cannabidiol, THC, butane hash oils (dabs), and many other illicit products that can cause a different range of complications.27 Unfortunately, we did not have sufficient data to disaggregate the effects of different substances. Moreover, the extent to which patients abstained from vaping before surgery, which would impact the harmful effect on lungs, was also not documented. Finally, it is worth mentioning that most of the patients in the exposure group were either current or former smokers, and although the interaction between vaping and smoking status was not significant, we had only limited power to assess differential effects of vaping in nonsmokers compared with smokers.

In summary, our analysis of patients undergoing noncardiothoracic surgery did not show evidence of a clinically meaningful association between vaping/e-cigarette use, and either hypoxemia during the first hour in PACU or an increased risk of postoperative pulmonary complications. Nevertheless, our findings cannot definitively exclude the deleterious effects of vaping and e-cigarette use on the lungs, and anesthesiologists should consider potential perioperative complications.