Approximately 20% of adult Americans smoke cigarettes, and the fraction of smokers is probably even greater among surgical patients.1 Smoking is clearly an independent risk factor for intraoperative and postoperative complications, including cardiopulmonary and wound complications. Current smokers are 1.4 [95% confidence interval (CI), 1.1 to 1.7] times more likely to die after surgery than patients who never smoked.2 Active smokers also have significantly greater odds of pneumonia, unplanned tracheal intubation and need for mechanical ventilation, and are also significantly more likely to experience a cardiac arrest, myocardial infarction or stroke. Finally, superficial and deep infections, sepsis and septic shock are all more likely in smokers.2
Preoperative smoking cessation might reasonably be expected to reduce postoperative smoking-related complications, with longer periods of preoperative abstinence conferring stronger benefits. However, the optimal duration of preoperative smoking cessation remains controversial. Several randomised trials and reviews conclude that 3 to 8 weeks of preoperative cessation reduces wound and overall complications.3–5 In contrast, briefer periods of preoperative smoking cessation do not appear to reduce pulmonary complications.6 However, a recent meta-analysis demonstrated that each additional week of smoking cessation has a significant impact on the reduction of postoperative complications.7
Prospectively evaluating the effects of smoking cessation on postoperative outcomes is challenging because recruiting sufficient patients well in advance is difficult or impossible because most surgery is not predictable over the long term. Furthermore, a substantial number of patients refuse to stop smoking and even among those attempting to stop in the context of a defined protocol, the success rate is only about 50%.4,8 There are, therefore, only a few studies on the topic, and those that are available have limited numbers of patients, have evaluated few outcomes and were limited to certain surgical procedures and postoperative outcomes.4,5 For example, a recent meta-analysis was able to include only two prospective and seven retrospective studies with a total of just 400 patients.9 Perhaps unsurprisingly given the limited number of patients, the investigators were unable to demonstrate benefit in wound healing or other aspects of postsurgical recovery. There is, thus, clearly a need for well powered clinical trials that evaluate a broad range of patients and procedures.
There are, therefore, reasons to believe that preoperative smoking cessation generally improves postoperative outcomes, but the effect of various abstinence periods remains unclear. Consequently, we tested the primary hypothesis that preoperative smoking cessation reduces a collapsed 30-day composite of mortality and major complications in patients recovering from noncardiac surgery. Second, we tested the hypothesis that longer preoperative nonsmoking periods are associated with a progressively reduced risk of composite in-hospital morbidity and mortality compared with current smokers, and that there is no increase in respiratory complications in patients who stop smoking within the year before surgery.
Ethical approval for this study [Instituional review board (IRB)#12–159] was provided by IRB and Human Research Committee of Cleveland Clinic, Cleveland, Ohio, USA (Executive Director, Daniel Beyer, MS, MHA, CIP) on 2 October 2012. With IRB approval, written informed consent was waived for this retrospective cohort analysis of adults who had noncardiac surgery at the Cleveland Clinic Main Campus between May 2007 and December 2013. Demographic characteristics, medical history, perioperative data and postoperative outcomes were obtained from the Cleveland Clinic Perioperative Health Documentation System. All data were prospectively collected in a standardised fashion according to strict definitions of preoperative characteristics, intraoperative information and postoperative outcomes. Smoking history was obtained from a self-administered preoperative Health Quest questionnaire, which was completed by each patient prior to surgery.
We excluded patients with American Society of Anesthesiologists’ (ASA) physical status exceeding four, patients who did not have general anaesthesia, and patients with missing outcomes and/or covariables. When multiple procedures were performed within the study period, only the first operation for each patient was included in the analysis.
Our primary goal was to characterise the relationship between smoking cessation and a collapsed composite (i.e. any of the following outcomes vs. none) of in-hospital mortality and morbidities, including cardiac, haemorrhagic, infectious, neurological, peripheral venous vascular, pulmonary/respiratory, urinary/renal and wound disruption (Appendix 1) Our secondary hypotheses was that prolonged preoperative nonsmoking periods are associated with a progressively reduced incidence of pulmonary and respiratory complications compared with current smokers, and that there is no increase in respiratory complications in patients who quit smoking less than 1 year before surgery.
We assessed the relationship between smoking cessation and the collapsed composite of in-hospital mortality/morbidity after propensity score matching patients who stopped smoking before surgery (former smokers) to current smokers. We first estimated the propensity score (probability of stopping smoking prior to surgery) for each patient using a logistic regression based on all potential confounding variables listed in Table 1. We also grouped each current procedural terminology code using the Agency for Healthcare Research and Quality's Clinical Classification Software for services and procedures (AHRQ-CCS). Among a total of 244 mutually exclusive, clinically appropriate categories defined by AHRQ-CCS, our study cohort contained 146 categories. Next, a 1-to-1 greedy matching algorithm (SAS macro: gmatch) was used for the matching.10 Successful former and current smoker matches were restricted to patients with the same AHRQ-CCS category, as well as with estimated propensity score logits [i.e.
: estimated propensity score] within 0.2 SDs of the propensity score logit across all patients (i.e. within 0.2 × 0.176 = 0.035).11
The matched variables were summarised using mean ± SD for normally distributed continuous variables, median (Q1 to Q3) for nonnormally distributed continuous variables and number (percentage) of patients for categorical variables, as appropriate. The balance between former and current smokers on matched variables before and after matching were assessed using the absolute standardised difference (ASD), that is, the absolute difference in means for continuous variables or proportions for categorical variables divided by the pooled SD. The ASD was calculated based on their ranks for nonnormally distributed variables. To account for even minimal potential confounding, we used a conservative criterion of greater than 0.028 (i.e.
ASD as an indication of imbalance). Any imbalanced variables after the propensity matching were included into the multivariable models when comparing the matched former and current smokers on outcomes to reduce potential confounding.
Propensity score-matched former and current smokers were compared on the collapsed composite of in-hospital mortality/morbidity using a multivariable logistic regression, adjusting for any residual imbalanced covariables after the propensity score matching. As a sensitivity analysis, we used conditional logistic regression, accounting for the matching pairs. We also assessed the interaction between smoking cessation and amount of smoking (measured as pack-years).
Second, we assessed the heterogeneity of the association between smoking cessation and nine individual components of our composite endpoint by testing the cessation-by-outcome interaction in a ‘distinct-effects’ generalised estimating equation model.12 As no significant heterogeneity was observed, we also estimated the ‘common effect’ (i.e. global) odds ratio (OR) of smoking cessation across the nine individual components using a generalised estimating equation model with unstructured covariance matrix. For informational purposes, we also reported the effect of smoking cessation on each specific component; the significance criterion for each individual component was P value less than 0.006 (i.e. 0.05/9, Bonferroni correction). We adjusted for any covariables that remained imbalanced after propensity-score matching.
We assessed the impact of smoking cessation duration on our composite primary outcome. Specifically, we separated former smokers into four groups according to their smoking cessation duration defined by the Cleveland Clinic preoperative Health Quest questionnaire, including less than 1 year, 1 to 5 years, 5 to 10 years and more than 10 years. We collapsed categories 1 to 3 years and 3 to 5 years into 1 to 5 years, because of the small number of patients in the category of 3 to 5 years. Next, we 1-to-1 matched patients who stopped smoking within 1 year prior to surgery, 1 to 5 years and to 10 years to patients who had stopped for more than 10 years, in which patients were exactly matched on type of surgery and propensity score matched on all the other potential confounding variables as in the primary analysis. We then compared matched patients on the composite of in-hospital morbidity/mortality as well as the individual components of the composite, by a multivariable logistic regression adjusting for any residual imbalanced confounding variables after the matching. The overall significance level for comparing each matched subset was P value less than 0.017 (i.e. 0.05/3, Bonferroni correction); the significance criterion for comparing each individual component was further adjusted, P value less than 0.002 (i.e. 0.017/9, Bonferroni correction).
We also matched and compared patients who stopped smoking less than 1 year before surgery and current smokers on pulmonary/respiratory morbidity. The detailed cessation duration categories for patients, who stopped within 1 year prior to surgery, including 1 or 2 days, 2 days to 8 weeks and 8 weeks to 1 year, were not collected on the Cleveland Clinic preoperative Health Quest questionnaire until March 2013. For information purposes, we summarised outcome incidence by the detailed cessation categories for patients who underwent surgery from March 2013 to December 2013 (about 15% of our study cohort).
Post hoc power calculation
With 18 964 propensity score-matched patients, we had more than 90% power to detect an OR of 0.8 or less for the collapsed composite outcome (any/all in-hospital morbidities/mortality vs. none) at the 0.05 significance level, using the observed incidence of 7.8% in current smokers. However, our study was not powered for any individual component of the composite outcome.
SAS software version 9.3 (SAS Institute, Cary, North Carolina, USA) and R software version 2.15.3 (The R Foundation for Statistical Computing, Vienna, Austria) were used for all statistical analyses.
A total of 37 511 patients, including 26 269 (70%) former smokers and 11 242 (30%) current smokers, met our inclusion criteria. We successfully matched 9482 current smokers (84%) with 9482 former smokers (36%) for a total of 18 964 matched patients. Among a total of 244 mutually exclusive clinically appropriate categories defined by AHRQ-CCS, our study cohort contained 146 categories. Our propensity score-matched subset retained 122 AHRQ-CCS categories (84% of 146 prematch categories, Appendix 2). The 24 (16%) unmatched AHRQ-CCS categories represented only 0.3% of our study population (107 of 37 511 patients).
Potential confounding factors used for the propensity score matching were much better balanced in the matched subset (Table 1). Only sex (ASD 0.05), BMI (ASD 0.168), ASA physical status (ASD 0.078) and amount of smoking (ASD 0.197) remained imbalanced as defined by an ASD more than 0.028 between the matched former and current smokers. These variables were, therefore included in our multivariable model comparing former smokers and current smokers.
Within the propensity score-matched patients, fewer former smokers [6.92% (95% CI 6.42 to 7.45%)] experienced the composite outcome than current smokers [7.83% (7.29 to 8.38%, P = 0.02, unadjusted)]. The estimated OR of having the composite of in-hospital morbidity/mortality was 0.83 (95% CI 0.74 to 0.93, former vs. current smokers, P = 0.001), after adjusting for sex, BMI, ASA grade and amount of smoking (Table 2). The sensitivity analysis accounting for the matching pairs gave a similar estimation of 0.81 (0.72 to 0.92). Furthermore, the impact of smoking cessation did not depend on amount of smoking (measured as pack-years, with a cessation-by-amount interaction P = 0.75).
The impact of smoking cessation was consistent across the nine components of the primary outcome (cessation-by-outcome interaction P = 0.19). The estimated common effect OR of smoking cessation across the individual components in the composite was 0.84 (95% CI 0.75 to 0.95; former vs. current smokers, P < 0.001). However, smoking cessation was significantly associated only with decreased odds of cardiac morbidity [OR 0.61 (99.4% CI 0.38 to 0.98), P = 0.004]. Apart from haemorrhage and wound disruption, the estimated OR for all the other individual components in the composite were in the direction of smoking cessation being protective, which helps explain why the overall common effect OR was significantly less than 1.0 (Table 2 and Fig. 1). Lack of significance for individual components may be because of the low incidence of the individual components (Appendix 3).
The observed incidence of the primary composite of in-hospital morbidity/mortality was 6.9% (656/9482) for all former smokers; the incidence was 7.8% (152/1951) for patients who stopped smoking less than 1 year before surgery, 6.3% (118/1977) for 1 to 5 years, 7.2% (115/1596) for 5 to 10 years and 6.9% (271/3457) for more than 10 years (Appendix 3). To assess the impact of smoking cessation durations, we 1-to-1 matched patients who stopped smoking within 1 year, 1 to 5 years or 5 to 10 years to patients who had stopped smoking for more than 10 years. No significant difference on the in-hospital morbidity/mortality was observed within any of the matched subsets, after adjusting for all the variables used for the matching (Table 3). The estimated OR was 1.07 (98.3% CI 0.70, 1.63) for less than 1 year vs. more than 10 years (P = 0.66), 0.83 (0.54 to 1.28) for 1 to 5 years vs. more than 10 years (P = 0.26) and 1.04 (0.68 to 1.58) for 5 to 10 years vs. more than 10 years (P = 0.82), respectively. Furthermore, smoking cessation duration was not associated with any in-hospital morbidities or in-hospital mortality (Table 3).
The incidence of the composite of in-hospital morbidity/mortality did not differ in patients who stopped smoking less than a year before surgery (7.5%) and current smokers (8.8%); the estimated OR was 0.84 (97.5% CI 0.64 to 1.10) for patients who stopped within a year vs. current smokers (P = 0.16). Furthermore, there was no significant difference between patients who stopped within 1 year and current smokers on in-hospital pulmonary/respiratory complication [OR 0.78 (99.3% CI 0.42 to 1.43) for <1 year vs. current, P = 0.22]. Finally, there was no significant difference on any in-hospital morbidities or mortality (Table 4).
Appendix 4 displays outcome incidences by the detailed cessation categories for patients who quit within 1 year prior to surgery, including 1 or 2 days, 2 days to 8 weeks and 8 weeks to 1 year, which were collected on the Cleveland Clinic preoperative Health Quest questionnaire since March 2013 (about 15% of our study cohort). The observed incidences of the primary composite outcome were 8% for patients who stopped smoking 1 or 2 days before surgery, 13.1% for 2 days to 8 weeks and 5.5% for 8 weeks to 1 year. A similar trend was observed for pulmonary/respiratory complications, which were 2% for 1 or 2 days, 4.6% for 2 days to 8 weeks and 1.6% for 8 weeks to 1 year.
Our analysis shows that former smokers experienced less composite mortality, fewer serious complications and less cardiac morbidity than current smokers – independent of the amount of smoking. Importantly, there was no association between preoperative nonsmoking duration and adverse outcomes including pulmonary/respiratory morbidity.
Previous publications related to smoking cessation are contradictory and difficult to interpret. For example, Groth et al.13 and Saxena et al.14 reported similar complication rates in current and former smokers, whereas others report that smoking cessation decreases risk.15 Furthermore, Hawn et al.16 found that past smokers have similar morbidities when compared with current smokers. However, most previous studies evaluated only a limited number of patients, and often in specific surgical subgroups (e.g. lung resection).
Yoshida et al.,17 for example, addressed the impact of duration of smoking cessation on incidence of postoperative complications in just 246 oesophagectomy patients and found that the incidences of pneumonia and severe morbidities decreased as the duration of smoking cessation increased. In particular, smoking cessation of at least 30 days was an independent risk factor for pneumonia, whereas smoking cessation for at least 90 days was an independent risk factor for severe morbidities. This result contrasts with ours, possibly because of differences in study populations and the types of surgical procedure that were considered. The incidence of pulmonary morbidity in patients having oesophagectomy was much higher compared with our general surgical population. Furthermore, duration of smoking cessation before surgery did not affect outcomes in our general surgical population.
We previously reported that cigarette smoking is a strong risk factor for perioperative mortality and morbidity.2 As might be expected, our current analysis also shows that the incidences of composite in-hospital morbidity and mortality were greater in current (7.8%) than former (6.9%) smokers. Specifically, current smokers who underwent noncardiac surgery had an estimated 17% increase in the odds of developing in-hospital morbidity and mortality. However, the magnitude of the effect of smoking cessation in this analysis was considerably less than the 1.38 OR we previously reported.2 The observed lower OR for smoker vs. nonsmokers in our current study might be related to differences in the studied population, including surgical procedure and associated comorbidities. A relatively small effect in our current analysis might also result from two other factors: we only analysed in-hospital outcomes, but some complications clearly occur thereafter15; and improvements in perioperative management and more aggressive prevention and treatment of various in-hospital complications, such as universal implementation of postoperative incentive spirometry and aggressive pulmonary toilet, may be especially beneficial in smokers.
Consistent with our current findings, two other investigations suggest little or no association between smoking and short-term postoperative complications.18,19 Although the overall collapsed composite and the common effects models were both highly significant, the only component that differed significantly was cardiac morbidity, which was reduced by 39%. However, the ORs were roughly similar among components and there was no cessation-by-outcome interaction, suggesting that lack of significance was consequent to insufficient power rather than heterogeneity.
It is likely that the risk of some smoking-related complications revert toward baseline (nonsmoker) risk faster than others15; cardiovascular risk appears to take especially long, reverting to baseline risk only after 10 or more years. Furthermore, smoking is associated with adverse postoperative outcomes even after adjusting for the presence of smoking-related injury, including cardiovascular disease, chronic obstructive pulmonary disease and cancer.15
Hawn et al.16 showed that pulmonary complications, such as pneumonia, diminished over time after smoking cessation. However, the incidence of in-hospital pulmonary complications did not differ significantly in current and former smokers. Furthermore, it appears that duration of smoking cessation before surgery is not of great importance. Our results are, thus, consistent with those of Mason et al.,20 who used the Society of Thoracic Surgeons database to evaluate 8000 patients scheduled for lung cancer resection and demonstrated that mortality and pulmonary complications were similar in current smokers and those who had stopped smoking at least a year preoperatively. In contrast, other studies suggest that mortality remains elevated for decades and never decreases to the level of ‘never’ smokers. Prolonged risk may result from irreversible damage, such as altered gene expression21 and interstitial structure destruction, which overwhelms any benefits conferred by short-term abstinence prior to surgery.
Surprisingly, we also found that, among smokers, postsurgical mortality and morbidity did not depend on the amount of smoking (in packs per day) or the duration of preoperative smoking. Smoking at any time, thus, appears to increase postsurgical in-hospital morbidity and mortality, largely independent of amount or duration of use. In accordance, Hawn et al.16 previously reported that exposure of more than 20 pack-years of cigarettes significantly increased smoking-related surgical complications. Musallam et al.15 found that for current smokers the increased adjusted odds of mortality were evident in light (<10 pack-years, OR 1.34) and heavy (≥50 pack-years, OR 1.17) smokers. In contrast, past smokers did not have higher mortality regardless of number of pack-years which is consistent with our results for former smokers.
Several studies reported that smoking cessation 3 to 8 weeks before surgery might reduce postoperative pulmonary complications.3–5 Furthermore, a recent meta-analysis including six randomised trials and 15 observational studies, demonstrated that longer periods of smoking cessation reduce postoperative complications.7 Specifically, the authors concluded that each additional week of smoking cessation reduces postoperative complications. Interestingly, these findings could not be confirmed by our study, as we observed similar numbers of complications in patients who stopped smoking less than 1 year before surgery and in those who stopped 1 to 3 years, 3 to 5 years, 5 to 10 years or more than 10 years before surgery.
Retrospective analyses potentially suffer from selection bias and confounding factors that are largely prevented by randomisation. Randomisation prevents both sources of error, but smoking status cannot be randomly allocated. We, therefore, used multivariable analysis to adjust for differences in known and potential confounding factors. However, our list of potential confounding factors was surely incomplete; similarly, we have only crude estimates for the magnitude of many. The extent to which selection bias and confounding factors contribute to our conclusions remains unknown, but could well be substantial. Smoking status was determined by patient self-report, which may be biased toward longer interval of cessation or lower amount of smoking.
In summary, smoking cessation was associated with reduced in-hospital morbidity and mortality, which was – surprisingly – independent of the cessation interval. From a clinical perspective, patients should, thus, be encouraged to stop smoking regardless of the time period before surgery.
Acknowledgements relating to this article
Assistance with the study: none.
Financial support and sponsorship: none.
Conflicts of interest: none.