Postoperative pulmonary complications (PPCs) have been reported to occur in 5%-10% of the general patient population [1,2] and in 4%-22% of patients undergoing abdominal surgery [3-5]. PPCs may be associated with significant mortality . Many authors have studied the incidence of PPCs in both thoracic and nonthoracic surgery patients [1,2,4-8]. They have shown that PPCs have multiple risk factors, such as anesthesia duration, incision site, anesthesia type, and chronic obstructive pulmonary disease (COPD). However, different authors have defined PPCs differently. Some have defined PPCs as atelectasis and pneumonia [1-3,9]; others have included all PPCs together into one group for statistical analysis, sometimes defining pleural effusion , pulmonary embolism , cough , or aspiration pneumonia  as a PPC. However, we believe that all PPCs should not be analyzed together as one group, because including all the different kinds of PPCs together effectively treats them all as equivalent, and they are not. Atelectasis, for example, is less clinically significant than respiratory failure and pneumonia which, in turn, are less clinically significant than death. Each PPC should be analyzed individually for risk factors.
One problem with this kind of analysis is sample size. For example, the number of postoperative pneumonias will be smaller than total number of all the different kinds of PPC. Therefore, determining the risk factors for pneumonia requires a larger sample size than determining the risk factors for a group of PPCs that includes atelectasis and pneumonia. A larger sample size is also required when studying risk factors for the more clinically significant PPC, since the more severe PPCs are less common than atelectasis. For example, only eight instances of respiratory failure occurred in a study of anesthetic complications in 13,693 routine surgery patients .
COPD patients have been found to be at higher risk for postoperative atelectasis or pneumonia [2,7,14], and death [6,15-17]. Therefore, perhaps the risk factors for the more severe PPCs may be determined by studying a relatively small group of COPD patients.
The purpose of this study was to determine the incidence of different PPC in patients with severe COPD. We also wanted to determine the specific risk factors for each kind of PPC.
In compliance with institutional review board guidelines and without written patient informed consent, demographic and clinical data were collected on every patient with a clinical diagnosis of COPD, a forced expiratory volume in 1 s (FEV1) <or=to1.2 L and a FEV1/forced vital capacity (FVC) ratio <75% who underwent noncardiothoracic surgery at the Long Beach Veteran's Administration Medical Center from April 1986 to May 1990. A FEV1 threshold of 1.2 L was selected because it has been reported previously that these patients are at higher risk for PPC . The predicted FEV1 for all our patients was 2.50 L or larger; thus all our patients had FEV1 <50% of their calculated predicted normal value , which is one indicator of severe pulmonary impairment . Patients with preexisting tracheostomies and thoracic surgery patients were excluded. If a patient had more than one operation during the study, only data from the first operation was included.
Potential preoperative risk factors obtained were: age, smoking history (pack-years), sputum production and color, ASA physical status grade, a postoperative pulmonary complication risk score as described by Shapiro (Shapiro score, see Table 5) , FEV1, FEV1/FVC, and arterial blood gas (ABG). The anesthesia care team assigned the patient's ASA physical status grade. A research team member determined and recorded the patient's Shapiro score. FEV1 and FVC were measured with a Vitalograph Registered Trademark spirometer (Vitalograph, Buckingham, England). Potential intraoperative risk factors recorded were: type of anesthesia, duration of anesthesia, duration of surgery, and incision site. Each patient's attending anesthesiologist selected the anesthetic technique. We defined an operation as emergent if the patient had a suffix "E" assigned to their ASA physical status grade by the attending anesthesiologist. The occurrence of five different postoperative complications was specifically sought: death, pneumonia, prolonged intubation, refractory bronchospasm, and prolonged intensive care unit (ICU) stay. Death as a postoperative complication was defined as death occurring during the same hospital stay. Postoperative pneumonia was defined as a new infiltrate on a chest radiograph combined with fever, leukocytosis, and a positive sputum Gram stain or culture. Prolonged intubation was defined as the failure to extubate within 24 h after the end of the operation. Refractory bronchospasm was defined as wheezing upon auscultation that acutely required additional parenteral drug administration (intravenous or subcutaneous injection; not inhaled) in addition to the patient's preoperative drug regimen. Prolonged ICU stay has been proposed to be a more objective indicator of perioperative complications than other PPCs  and was defined as an ICU stay > or=to4 days. Atelectasis (i.e., isolated fever or abnormal chest radiograph without a positive sputum culture) was not considered a PPC.
Patients who received combined regional and general anesthesia were included in the general anesthesia group. Similarly, patients with lower abdominal and upper abdominal incisions were combined into one group, defined as patients with abdominal incisions. Only patients who had received general anesthesia were analyzed for the risk of prolonged postoperative intubation.
Statistical analysis was done in three phases. First, univariate analysis regarding the PPC incidence for each potential risk factor was evaluated by Fisher's exact test. Optimal cutoff points for univariate analysis were determined using reduced monotonic regression analysis (M. J. Schell, B. Singh, personal communication, 1994). We considered a P value <0.05 as statistically significant. All potential risk factors were also analyzed by Spearman rank (nonparametric) correlation coefficient analysis to determine whether the potential risk factors were truly independent variables. Most of the variables were not highly correlated. For example, Shapiro score and ASA physical status only had a correlation coefficient (r) of 0.22 with each other. Only three variables had a r >0.50 with another variable: ASA physical status and emergency operation (r = 0.56), prebronchodilator FEV1 and postbronchodilator FEV1 (r = 0.62), and duration of anesthesia and duration of surgery (r = 0.93). Since duration of anesthesia and duration of surgery were highly correlated, duration of anesthesia was included in the logistic regression model and duration of surgery was excluded.
Next, the univariate risk factors with P < 0.25 were then evaluated by forward stepwise multiple logistic regression using SAS computer programs (SAS Institute, Cary, NC). A P < 0.05 was required to stay in the logistic regression model. Age, smoking (pack-years), duration of anesthesia, FEV1, and FEV1/FVC were examined both as continuous variables and categorized by cutoff points as in Table 2. Optimal cutoff points were determined using reduced monotonic regression analysis (M. J. Schell, B. Singh, personal communication, 1994). No optimal cutoff points for age, smoking, or FEV1/FVC could be determined so the cutoff points for these variables listed in Table 2 were selected arbitrarily. Anesthesia duration >2 h was a unique cutoff point; almost all the PPCs occurred in patients with anesthesia duration >2 h. This caused the effect of anesthesia duration >2 h to overwhelm the effect of the other single-risk factors in the regression models. Therefore anesthesia duration >2 h was accepted as a risk factor for PPC and excluded as a cutoff point in the logistic regression models. It may be unfair to compare composite classification systems such as ASA physical status and Shapiro score with individual risk factors because several individual risk factors may be incorporated into one classification system. Therefore, the logistic regression models were computed two ways: with and without the composite classification systems as risk factors. Lastly, to evaluate the effect of second order interactions, another stepwise logistic regression analysis was computed which included second order interaction terms for all dichotomous (two-valued) risk factors.
Long-term followup for determining survival was accomplished by examining hospital admission and discharge records, clinic visit records, personal telephone contact, and Social Security Administration office registration. The long-term survival curve was created using Epistat 4.0 (Epistat Services, Richardson, TX).
One hundred five patients were studied; 53 had general anesthesia (38 had general anesthesia alone and 15 had combined [general and epidural] anesthesia), 51 had regional anesthesia (38 spinal, 13 epidural), and 1 had local anesthesia. All but four were male; other patient characteristics are summarized in Table 1. Information about all preoperative and intraoperative risk factors studied, except for sputum quality, ABG data, and postbronchodilator lung spirometry data, were available in all patients. Room air ABG analyses were available in 77 patients. Of 60 patients who had spirometry before and after bronchodilator therapy, 35 increased their FEV (1) > or=to15% after bronchodilator therapy.
Thirty-nine patients (37%) had one or more PPCs. Refractory bronchospasm (17 patients) and prolonged ICU stay (24 patients) were the most common PPCs. Seven patients died during their hospitalization, seven patients developed pneumonia, and eight patients remained intubated postoperatively.
Emergency operation was the univariate risk factor most frequently associated with a PPC. The incidence of all the potentially significant univariate risk factors for predicting individual PPCs is summarized in Table 2 All other potential risk factors failed to show any relationship with any PPC by univariate analysis.
ASA physical status > or=toIV, Shapiro score > or=to 5, and FEV1 were the only significant preoperative risk factors for different individual PPC by multivariate analysis. Emergency operation, abdominal incision, anesthesia duration, and general anesthesia were the intraoperative risk factors which were significant independent risk factors for different individual PPCs by multivariate analysis. The significant independent risk factors for each PPC and their odds ratios are listed in Table 3.
When the composite classification systems (ASA physical status and Shapiro score) were excluded from the multiple logistic regression models, FEV1 became the only significant preoperative risk factor for different individual PPCs. Emergency operation, anesthesia duration, and general anesthesia remained significant intraoperative risk factors. The significant independent risk factors for PPCs and their odds ratios, when the composite classification systems are excluded from the models, are listed in Table 4. No statistically significant second order interactions were found in any of the multiple logistic regression models.
The long-term survival of patients with severe COPD undergoing noncardiothoracic surgery was poor. Figure 1 shows the study group's long-term survival curve. At least 2 yr of followup was available on all 105 patients in the study. The 2-yr mortality rate was 47%.
This study shows that patients with severe COPD undergoing surgery and anesthesia have a high incidence of PPCs and poor long-term survival. Pulmonary risk factors alone do not predict the risk of PPCs. A high numerical value on a composite classification system was the most common preoperative risk factor for any individual PPC. Including composite classification systems in the multiple logistic regression models improves the ability of the models to predict PPCs. Composite classification systems are probably more useful than single-risk factors in predicting PPCs because nonpulmonary variables are also important.
It is not surprising that patients with severe COPD have a high incidence of PPCs; they have underlying pulmonary disease. In addition, both surgery and anesthesia can adversely affect lung function [9,23]. In general, the risk factors associated with PPCs in severe COPD patients are some of the same factors that have been also been associated with postoperative atelectasis, i.e., ASA physical status [4,5], duration of surgery , duration of anesthesia , and abdominal incision . Since higher ASA physical status has also been associated with postoperative pneumonia , prolonged postoperative intubation , and higher mortality [24,25], it is not surprising that high ASA physical status would be associated with PPCs.
A high numerical value on a composite classification system, either a ASA physical status > or=toIV or a Shapiro score > or=to5, was generally the most common and strongest risk factor for PPCs. Besides anesthesia duration >2 h, the single-risk factors that were associated with individual PPC were mostly nonpulmonary, intraoperative risk factors (emergency surgery, abdominal incision, and general anesthesia). Pulmonary risk factors (i.e., smoking history, FEV1, or FEV1/FVC) were less common and less significant risk factors. This implies that nonpulmonary risk factors were important in predicting the risk of PPC. Nonpulmonary risk factors may have been important in our study because our study group had a moderately high incidence of nonpulmonary disease Table 1. The presence of both pulmonary and nonpulmonary disease in our patients may have favored selecting composite classification systems as risk factors because classification systems can include both pulmonary and nonpulmonary risk factors.
Composite classification systems enhanced the logistic regression models, as evidenced by the chi squared values associated with the models. chi squared values were generally higher when the composite classification systems were included in the model than when they were not Table 3 and Table 4. Including Shapiro score > or=to5 as a risk factor particularly improved the logistic regression model for bronchospasm. Why is a Shapiro score > or=to 5, which includes spirometry measurements, more useful than FEV1 values in predicting death and bronchospasm? A Shapiro score >or=to5 signifies that at least two organ systems are diseased: the pulmonary system and one other. Thus, a Shapiro score >or=to5 becomes more like an ASA physical status >or=toIV, i.e., an indicator of severe systemic disease. Of the preoperative risk factors that were predictive of PPCs, the Shapiro score is the hardest to determine, spirometry is easier, and the ASA physical status is the easiest to determine. The ASA physical status classification system also provides for emergency surgery, which was the best single risk factor for PPC in this study. Composite classification systems such as the ASA physical status should be included in the logistic regression models for PPCs.
The most serious PPC is death. Besides same-hospitalization mortality, we looked at long-term mortality. The long-term mortality of COPD patients undergoing noncardiothoracic surgery is very high. Patients with severe COPD undergoing nonthoracic surgery have long-term mortality rates comparable to similar patients with severe coronary artery disease who undergo noncardiac surgery . Our mortality rates are higher than those reported by other authors who performed longitudinal mortality studies of severe COPD patients who did not undergo surgery at the time of study entry [27-29]. The added mortality in our patients may be due to the risk of surgery and anesthesia, or due to the underlying disease that made the operation necessary.
Knowing the incidence of PPCs, the risk factors for PPCs, and the long-term survival of patients with severe COPD is useful for three reasons. First, although there may be no spirometry value or specific risk factor which absolutely contraindicates surgery and anesthesia, knowing the incidence of PPCs and long-term mortality allows the clinician to better weigh the risk versus benefit of a surgical procedure in any given patient preoperatively. Second, knowing the risk factors for PPC helps identify which patients are at risk for severe PPCs so that their management can be individualized. Patients who are identified to be at higher risk for severe PPCs might deserve extra preoperative or postoperative care (e.g., preoperative pulmonary toilet or postoperative ICU observation), although this study did not specifically evaluate the benefit of these interventions. Regarding the intraoperative period, our results suggest that avoiding general anesthesia should decrease the risk of bronchospasm and that shortening the duration of surgery (and anesthesia) and avoiding abdominal incision may decrease the risk of prolonged ICU stay in these patients. General anesthesia has been reported previously to be a risk factor for bronchospasm , probably because endotracheal tubes can directly irritate airways. Anesthesia duration >2 h is a risk factor for PPCs because only patients having simpler surgeries (n = 32) had anesthesia duration <2 h. The duration of surgery (and anesthesia) probably depends on the underlying surgical disease, and the true risk factor for PPCs may be severe surgical disease. Unfortunately, avoiding general anesthesia with tracheal intubation, shortening the duration of surgery (and anesthesia), and avoiding abdominal incision may not always be possible. Last, knowing the incidence of PPCs in severe COPD patients provides some basis for observed versus expected outcome comparisons.
Expected outcome comparisons look at more than the overall outcome (e.g., overall PPC rates). Expected outcome comparisons also account for expected risk. For example, the relatively high rate of PPCs in our study could be interpreted two ways: either the patients were at a relatively high risk for PPCs or they received relatively low quality perioperative medical care. We believe that our patients received good perioperative medical care. The way to determine whether our patients received good or poor quality health care would be to study several institutions with comparable risk patients and compare our incidence of PPCs with the other institutions' incidence of PPCs. Lower quality health care would be associated with worse outcomes than expected (an incidence of PPCs greater than the other institutions' incidence of PPCs).
Before the risk factors reported here can be used to predict the risk of PPCs in other patients at other institutions, several potential problems with this study should be considered: sample bias, assumptions inherent with our logistic regression model, small sample size, and a nonrandomized study design.
Our logistic regression model has several inherent problems. First, our identification of PPC risk factors is limited to the risk factors that we studied. In reality, coexisting diseases that we did not examine may have contributed to PPCs, particularly death and prolonged ICU stay. Another problem with the logistic regression model is that it sequentially selects factors one at a time. The logistic regression model includes the most significant factor first and then discounts (excludes) the lesser significant factors. This could exclude a variable that was really a significant risk factor. For example, both ASA physical status >or=toIV and a Shapiro score >or=to5 indicate severe systemic medical disease which could increase the risk of death. Selection of the most significant scoring system first (i.e., Shapiro score >or=to5) in the logistic regression model excluded selection of ASA physical status >or=toIV as a risk factor even though ASA physical status >or=toIV, which other authors have reported to be a risk factor for death [24,25], might still be a risk factor for death in our patient group. Another problem with the logistic regression model is that the risk factors found using logistic regression are only associated with PPCs; they do not necessarily cause PPCs. Therefore, shortening surgery and anesthesia duration and avoiding general anesthesia may not actually decrease the incidence of PPCs.
Two other potential weaknesses of this study are the lack of a randomized control group and a relatively small sample size. Although this was not a prospective, randomized controlled study, it was a longitudinal study that included consecutive patients, and the data was collected concurrently during the hospital stay. The small sample size results in large confidence intervals for our odds ratios. A larger sample size would have resulted in smaller confidence intervals for our odds ratios and also might have allowed us to evaluate more factors simultaneously. However, a larger sample would have required a study period longer than 4 yr or a larger, multicenter trial. Despite our relatively small sample size, we noted almost as many serious PPCs as larger studies with a wider spectrum of pulmonary disease: 8 instances of respiratory failure and 17 instances of bronchospasm were noted in 13,696 patients in a study by Forrest et al.  and 16 instances of pneumonia, using our definition of pneumonia, were noted in 1000 patients in a study by Hall et al. .
In conclusion, patients with FEV1 <or=to1.2 L undergoing noncardiothoracic surgery have a 37% incidence of PPCs, excluding atelectasis, and have a 47% 2-yr mortality rate. However, preoperative pulmonary factors alone do not predict the likelihood of PPCs in severe COPD patients. Composite classification systems, such as the ASA physical status, probably predict the likelihood of PPCs better than individual risk factors because they include both pulmonary and nonpulmonary factors. During the intraoperative period, avoiding general anesthesia with tracheal intubation may decrease the risk of postoperative bronchospasm. Shortening the duration of surgery and anesthesia may decrease the risk of prolonged ICU stay.
The authors would like to thank Ms. Diana Lugo for excellent secretarial support.
1. King DS. Post-operative pulmonary complications. Surg Gynecol Obstet 1933;56:43-50.
2. Wightman JAK. A prospective survey of the incidence of postoperative pulmonary complications. Br J Surg 1968;55:85-91.
3. Dripps RD, Deming MV. Post-operative atelectasis and pneumonia. Ann Surg 1946;124:94-110.
4. Garibaldi RA, Britt MR, Coleman ML, et al. Risk factors for post-operative pneumonia. Am J Med 1981;70:677-80.
5. Hall JC, Talala RA, Hall JL, Mander J. A multivariate analysis of the risk of pulmonary complications after laparotomy. Chest 1991;99:923-7.
6. Tarhan S, Moffitt EA, Sessler AD, et al. Risk of anesthesia and surgery in patients with chronic bronchitis and chronic obstructive pulmonary disease. Surgery 1973;74:720-6.
7. Latimer RG, Dickman M, Day WC, et al. Ventilatory patterns and pulmonary complications after upper abdominal surgery determined by pre-operative and post-operative computerized spirometry and blood gas analysis. Am J Surg 1971;122:622-32.
8. Lockwood P. The principles of predicting the risk of post-thoracotomy function-related complications in bronchial carcinoma. Respiration 1973;30:329-44.
9. Tisi GM. Pre-operative evaluation of pulmonary function. Am Rev Respir Dis 1979;119:293-310.
10. Gold MI, Schwam SJ, Goldberg M. Chronic obstructive pulmonary disease and respiratory complications. Anesth Analg 1983;62:975-81.
11. Kroenke K, Lawrence VA, Theroux JF, Tuley MR. Operative risk in patients with severe obstructive pulmonary disease. Arch Intern Med 1992;152:967-71.
12. Pedersen T, Eliasen K, Henriksen E. A prospective study of mortality associated with anaesthesia and surgery: risk indicators of mortality in hospital. Acta Anaesthesiol Scand 1990;34:176-82.
13. Forrest JB, Rehder K, Cahalan MK, Goldsmith CH. Multicenter Study of General Anesthesia, Anesthesiology 1992;76:3-15.
14. Stein M, Koota GM, Simon M, Frank HA. Pulmonary evaluation of surgical patients. JAMA 1962;181:765-70.
15. Gracey DR, Divertie MB, Didier ED. Pre-operative pulmonary preparation of patients with chronic obstructive pulmonary disease, a prospective study. Chest 1979;76:123-9.
16. Stein M, Cassara EL. Preoperative pulmonary evaluation and therapy for surgery patients. JAMA 1970;211:787-90.
17. Grover FL, Hammermeister KE, Burchfiel C. Initial report of the Veterans Administration preoperative risk assessment study for cardiac surgery. Ann Thorac Surg 1990;50:12-28.
18. Manfreda J, Nelson N, Cherniak RM. Prevalence of respiratory abnormalities in a rural and an urban community. Am Rev Respir Dis 1978;117:215-6.
19. Gold WM, Boushey HA. Pulmonary function testing. In: Murray JF, Nadel JA, eds. Textbook of respiratory medicine. Philadelphia: WB Saunders, 1988:667-70.
20. Shapiro BA, Harrison RA, Kacmarek RM, Cane RD. Clinical application of respiratory care. Chicago: Year Book Medical Publishers, 1985:518.
21. Zibrak JD, O'Donnell CR, Marton K. Indications for pulmonary function testing. Ann Intern Med 1990;112:763-71.
23. Craig DB. Postoperative recovery of pulmonary function. Anesth Analg 1981;60:46-52.
24. Vacanti CJ, Van Houten RJ, Hill RC. A statistical analysis of the relationship of physical status to postoperative mortality in 68,388 cases. Anesth Analg 1970;49:564-6.
25. Marx GH, Matteo CV, Orkin LR. Computer analysis of post anesthetic deaths. Anesthesiology 1973;39:54-8.
26. Browner WS, Li J, Mangano DT. In hospital and long-term mortality in male veterans following noncardiac surgery. JAMA 1992;268:228-32.
27. Burrows B, Earle RH. Course and prognosis of chronic obstructive lung disease. A prospective study of 200 patients. N Engl J Med 1969;280:397-404.
28. Postma DS, Burema J, Gimeno F, et al. Prognosis in severe chronic obstructive pulmonary disease. Am Rev Respir Dis 1979;119:357-67.
29. Renzetti AD, McClement JH, Litt BD. The Veterans Administration cooperative study of pulmonary function. Mortality in relation to respiratory function in chronic obstructive pulmonary disease. Am J Med 1966;41:115-29.
30. Olsson GL. Bronchospasm during anaesthesia. A computeraided incidence study of 136,929 patients. Acta Anaesthesiol Scand 1987;31:244-52.