RSS Feeds
Preoperative Pulmonary Risk Stratification for Noncardiothoracic Surgery: Systematic Review for the American College of Physicians
- Gerald W. Smetana, MD;
- Valerie A. Lawrence, MD; and
- John E. Cornell, PhD
- From the Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, and South Texas Veterans Health Care System and The University of Texas Health Science Center at San Antonio, San Antonio, Texas.
Abstract
Background: The importance of clinical risk factors for postoperative pulmonary complications and the value of preoperative testing to stratify risk are the subject of debate.
Purpose: To systematically review the literature on preoperative pulmonary risk stratification before noncardiothoracic surgery.
Data Sources: MEDLINE search from 1 January 1980 through 30 June 2005 and hand search of the bibliographies of retrieved articles.
Study Selection: English-language studies that reported the effect of patient- and procedure-related risk factors and laboratory predictors on postoperative pulmonary complication rates after noncardiothoracic surgery and that met predefined inclusion criteria.
Data Extraction: The authors used standardized abstraction instruments to extract data on study characteristics, hierarchy of research design, study quality, risk factors, and laboratory predictors.
Data Synthesis: The authors determined random-effects pooled estimate odds ratios and, when appropriate, trim-and-fill estimates for patient- and procedure-related risk factors from studies that used multivariable analyses. They assigned summary strength of evidence scores for each factor. Good evidence supports patient-related risk factors for postoperative pulmonary complications, including advanced age, American Society of Anesthesiologists class 2 or higher, functional dependence, chronic obstructive pulmonary disease, and congestive heart failure. Good evidence supports procedure-related risk factors for postoperative pulmonary complications, including aortic aneurysm repair, nonresective thoracic surgery, abdominal surgery, neurosurgery, emergency surgery, general anesthesia, head and neck surgery, vascular surgery, and prolonged surgery. Among laboratory predictors, good evidence exists only for serum albumin level less than 30 g/L. Insufficient evidence supports preoperative spirometry as a tool to stratify risk.
Limitations: For certain risk factors and laboratory predictors, the literature provides only unadjusted estimates of risk. Prescreening, variable selection algorithms, and publication bias limited reporting of risk factors among studies using multivariable analysis.
Conclusions: Selected clinical and laboratory factors allow risk stratification for postoperative pulmonary complications after noncardiothoracic surgery.
Postoperative pulmonary complications contribute importantly to the risk for surgery and anesthesia. The most important and morbid postoperative pulmonary complications are atelectasis, pneumonia, respiratory failure, and exacerbation of underlying chronic lung disease. Since the publication of the first cardiac risk index in 1977 (1), clinicians have been aware of the importance of, and the risk factors for, cardiac complications. Clinicians who care for patients in the perioperative period may be surprised to learn that postoperative pulmonary complications are equally prevalent and contribute similarly to morbidity, mortality, and length of stay. For example, in a large retrospective cohort study of 8930 patients undergoing hip fracture repair, 1737 (19%) patients had postoperative medical complications (2). Serious pulmonary complications occurred in 229 (2.6%) patients and serious cardiac complications occurred in 178 (2.0%) patients.
Similarly, in a study of 2964 patients undergoing elective noncardiac surgery, postoperative pulmonary and cardiac complications occurred in 53 patients and 64 patients, respectively (3). Rates of postoperative cardiac and pulmonary complications are similar in other large cohort studies of patients undergoing noncardiac surgery (4-6). Pulmonary complications may also be more likely than cardiac complications to predict long-term mortality after surgery. For example, among postoperative complications in a recent study of patients older than 70 years of age who were undergoing noncardiac surgery, only pulmonary and renal complications predicted long-term mortality (7). In another report of patients undergoing esophagectomy for cancer, postoperative pneumonia was second only to tumor stage in predicting long-term survival after surgery and predicted long-term mortality to a greater degree than postoperative cardiac, renal, or hepatic complications (8).
Office and hospital consultation for patients preparing for surgery is an important activity for internists. While guidelines and consensus statements for perioperative cardiac evaluation have been published (9, 10), no similar guideline is available to assist in perioperative pulmonary evaluation. The quality and number of studies that estimate perioperative pulmonary risk have increased in the past 2 decades, and this is no longer a neglected area of inquiry. We prepared this 2-part systematic review 1) to guide clinicians on clinical and laboratory predictors of perioperative pulmonary risk before noncardiothoracic surgery and 2) to evaluate the efficacy of strategies to reduce the risk for postoperative pulmonary complications (11). Risk factors for postoperative venous thromboembolism differ substantially from those for postoperative pulmonary complications, and they are not the subject of our review.
Methods
Literature Search and Selection Criteria
The Appendix contains a detailed description of our methodology. We performed a MEDLINE search to identify relevant publications from 1 January 1980 through 30 June 2005. We used the following Medical Subject Heading (MeSH) terms and specified that they be the article's primary focus: intraoperative complications, postoperative complications, preoperative care, intraoperative care, and postoperative care, plus the text term perioperative complications in the title or abstract. We identified additional MeSH and text terms by a review of the MEDLINE indexing for the retrieved articles. These included terms for pulmonary, respiratory, or cardiopulmonary diseases, conditions, or complications and terms for oxygenation and chest roentgenography. We performed additional searches specific to preoperative chest radiography and preoperative spirometry. We identified additional references by reviewing bibliographies of retrieved studies.
We included only English-language publications and excluded publication types without primary data (that is, letters, editorials, case reports, conference proceedings, and narrative reviews). We excluded 1) studies with fewer than 25 participants per study group; 2) studies that used only administrative data (for example, International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes) or lacked explicit criteria or definitions for pulmonary complications; 3) studies from developing countries (because of potential differences in respiratory and intensive care technology); 4) studies of ambulatory surgery; 5) studies of physiologic (for example, lung volumes and flow, oximetry) rather than clinical outcomes; 6) studies of gastric pH manipulation; 7) studies of complications unique to a particular type of surgery (for example, upper airway obstruction after uvulectomy); 8) studies of cardiopulmonary or pediatric surgery; and 9) studies of organ transplantation surgery (because of profoundly immunosuppressive drugs).
Of 16 959 citations identified by the search, 1223 citations were duplicates and 14 793 citations were not relevant by title and abstract review (Figure). Of the remaining 943 potentially relevant citations, we excluded 626 citations after review of the full publication and abstracted 145 citations in detail.
Assessing Study Quality
We used the U.S. Preventive Services Task Force (USPSTF) criteria for assigning hierarchy of research design, grading a study's internal validity as our basis for assessing study quality, and assigning summary strength of recommendations for each risk factor and laboratory test (12).
Statistical Analysis
Our literature search yielded primarily unadjusted estimates for most laboratory factors of interest. Limited multivariable, adjusted studies were available for serum albumin level less than 30 g/L and elevated blood urea nitrogen level. However, rather than attempt to compute potentially biased summary estimates, we provided narrative descriptions of the pattern of results for these potential risk factors.
The eligible multivariable risk factor studies varied considerably in the number and type of competing risks and confounders included in the analyses. Extensive use of prescreening methods and variable selection algorithms often limited reporting to the subset of risk factors that were determined to be statistically significant in a given sample. The result is a subtle form of publication bias, which we verified by examination of the funnel plots and trim-and-fill estimates for each risk factor.
We extracted odds ratios from each study, along with their respective SEs, 95% confidence limits, or both. We used the I2 statistic (13) and the Cochran Q statistic (14) to assess study heterogeneity. We also recomputed pooled estimates with and without studies that produced extreme results. An I2 statistic of 50% or more indicates substantial heterogeneity among study estimates. We used the DerSimonian–Laird method to compute random-effects estimates when the set of studies was heterogeneous (15). In cases where 3 or more studies contributed estimates for a risk factor, we used the trim-and-fill method to adjust pooled estimates of a risk factor's effect on postoperative pulmonary complications for publication bias (16). Trim-and-fill estimates check the sensitivity of pooled estimates to potential publication bias (17). We used meta-analysis procedures available in Stata software, version 8 (Stata Corp., College Station, Texas), to conduct these analyses (18).
Role of the Funding Source
The Veterans Evidence-based Research, Dissemination, and Implementation Center (VERDICT) (Veterans Affairs Health Services Research and Development, HFP 98-002) provided the research librarian and administrative support for the study. The funding source had no role in the design, conduct, or reporting of the study or in the decision to submit the manuscript for publication.
Results
Eighty-three publications provided univariate data on clinical predictors of postoperative pulmonary complications. Appendix Table 1 summarizes the characteristics of these studies (2, 3, 19-99). Seventy-three (88.0%) publications were cohort studies; 3 (3.6%) were randomized, controlled trials; 2 (2.4%) were case–control studies; and the remaining 5 (6.0%) were case-series studies. Slightly less than half (45.8%) of the cohort studies used a prospective design. Ten studies were of good quality, 18 studies were of fair quality, and 55 studies were of poor quality. Eligible studies included 11 851 postoperative pulmonary complication events among 173 500 patients.
Twenty-seven studies reporting multivariable analyses (10 960 postoperative pulmonary complication events among 324 648 patients) met our inclusion criteria (Appendix Table 2) (100-126). These studies form the principal basis of our review. Most studies (96%) were prospective cohort studies, and only 1 report was a case–control study. The 3 largest studies (118, 120, 123) used subsets of patients from the Veterans Affairs National Surgical Quality Improvement Project (NSQIP) (127). These 3 studies accounted for 89.8% of all patients included in the multivariable studies and 82.3% of the observed postoperative pulmonary complications. The crude postoperative pulmonary complication rate among the cohort studies was 3.4%.
The studies were heterogeneous with respect to study objectives, study samples, and criteria for defining a postoperative pulmonary complication. Seventeen of the 27 (63.0%) studies aimed to identify potential risk factors for postoperative pulmonary complications. The objective in 3 studies was to develop a risk index for postoperative pulmonary complications (113, 118, 120). The remaining studies focused on high-risk subgroups, such as patients undergoing aortic surgery (104, 125), patients with smoking histories (114), elderly patients (102, 117, 121), or patients with chronic obstructive pulmonary disease who required prolonged stays in the intensive care unit (108).
Postoperative pulmonary complication definitions varied considerably across studies. While 16 (59.3%) of the studies included some combination of pneumonia or respiratory infection along with respiratory insufficiency or failure, the studies varied in the inclusion of other complications. Two studies included pulmonary edema (102, 121) and 2 studies included pulmonary embolus in addition to traditional postoperative pulmonary complication definitions (102, 125). Three studies used atelectasis as an exclusion criterion (110, 118, 120). One study explicitly excluded patients who required postoperative mechanical ventilation (128). The 2 largest studies derived from the NSQIP limited their analyses to either postoperative respiratory failure (118) or pneumonia (120). Appendix Table 3 details the principal results of the studies that reported multivariable analyses.
Patient-Related Risk Factors
We considered patient- and procedure-related risk factors separately and divided the patient-related risk factors into the following general categories: age, chronic lung disease, cigarette use, congestive heart failure, comorbid condition measures, functional dependence, obesity, obstructive sleep apnea, impaired sensorium, and other factors. Table 1 displays the DerSimonian–Laird pooled and trim-and-fill estimate odds ratios for the patient-related risk factors.
Age
The influence of age on postoperative pulmonary complication rates is not well established. Most previous reviews have considered age to be a minor risk factor for the development of postoperative pulmonary complications. Studies that reported postoperative pulmonary complications by age categories varied with respect to the cutoff ages used to define age strata (Appendix Table 4). Ten studies reported unadjusted postoperative pulmonary complication rates by age strata for patients older than 65 years of age (5 studies) and for patients older than 70 years of age (5 studies). Unadjusted postoperative pulmonary complication estimates for patients older than 65 years of age ranged from 1% to 34%, with a median postoperative pulmonary complication rate of 14%. For patients 70 years of age and older, the unadjusted postoperative pulmonary complication estimates ranged from 4% to 45%, with a median postoperative pulmonary complication rate of 15%.
Eleven multivariable risk factor studies reported statistically significant effects for age. This was the second most commonly identified risk factor in our review. Seven of these studies—4 good-quality studies, 2 fair-quality studies, and 1 poor-quality study—provided odds ratios along with SEs or CIs. The remaining studies did not report values, used age as a continuous or ordered categorical variable, or reported results from a multivariate discriminant function analysis. Three studies reported age-related odds ratios in several strata. We organized study estimates into 4 age strata based on deciles (50 to 59 years, 60 to 69 years, 70 to 79 years, and ≥80 years). While several studies used this stratification scheme for age, some studies used 1 age category stratification, such as 65 years and older or 60 years and older. We grouped these study values within the 60 to 69 years of age stratification in our analyses. Two studies reported age strata as 50 to 69 years and 70 years and older. We assigned the 50 to 69 years of age category to the 60 to 69 years of age group and the 70 years and older age category to the 70 to 79 years of age group. Sensitivity analyses showed that reassigning the 50 to 69 years of age category to the 50 to 59 years of age group or reassigning the 70 years and older age category to the 80 years and older age group had little effect on the results. The odds that patients experienced a postoperative pulmonary complication increase systematically with age (Table 1), with older age categories conferring higher postoperative pulmonary complication risk, even after trim-and-fill correction for publication bias. Odds ratios for the 60 to 69 years of age group and 70 to 79 years of age group are 2.09 (95% CI, 1.66 to 2.64) and 3.04 (CI, 2.11 to 4.39), respectively.
While unadjusted risk due to increasing age was previously believed to be due to accumulating comorbid conditions, our review indicates that advanced age is an important independent predictor of postoperative pulmonary complications even after adjustment for comorbid conditions.
Chronic Lung Disease
Among studies reporting multivariable analyses, chronic obstructive pulmonary disease was the most frequently identified risk factor for postoperative pulmonary complications. Thirteen of 15 studies that entered this factor into a multivariate model reported it to be a statistically significant predictor of postoperative pulmonary complications. Eight studies—2 good-quality studies, 4 fair-quality studies, and 2 poor-quality studies—provided odds ratios with SEs, 95% CIs, or both. The trim-and-fill bias-corrected odds ratio for chronic obstructive pulmonary disease is 1.79 (CI, 1.44 to 2.22).
Two small, poor-quality studies reported unadjusted postoperative pulmonary complication rates for patients with and without abnormal findings on chest examination (40, 73). Postoperative pulmonary complications occurred in 35 of 57 patients with abnormal findings. Only 1 of these 2 studies reported postoperative pulmonary complication rates for patients with chronic obstructive pulmonary disease (40). One multivariable study reported that abnormal findings on chest examination (defined as decreased breath sounds, prolonged expiration, rales, wheezes, or rhonchi) were the strongest predictor of postoperative pulmonary complication rates (odds ratio, 5.8 [CI, 1.04 to 32.1]) (110). While the data indicate a higher postoperative pulmonary complication risk for patients with abnormal findings, the magnitude of this effect is uncertain because of the small number of studies. One study evaluated the cough test as a potential tool to stratify risk (126). To perform a cough test, the patient takes a deep breath and coughs once. A positive test result is recurrent coughing. The adjusted odds ratio for postoperative pulmonary complication was 3.8 (P = 0.01).
No eligible study determined the incremental postoperative pulmonary complication risk for patients with chronic restrictive lung disease or restrictive physiology due to neuromuscular disease or chest wall deformity, such as kyphoscoliosis.
Cigarette Use
Five multivariable studies (3 good-quality and 2 fair-quality studies) provided odds ratios, with SEs, 95% CIs, or both, of the effect of cigarette use on postoperative pulmonary complication rates. The trim-and-fill bias-adjusted odds ratio for cigarette use is 1.26 (CI, 1.01 to 1.56), suggesting a modest increase in postoperative pulmonary complication risk among patients with a smoking history.
Studies evaluating the effect of smoking cessation on postoperative pulmonary complication rates have generally evaluated patients undergoing pulmonary or cardiac surgery, which we excluded from our review. One multivariable study of 410 patients undergoing elective general, orthopedic, urologic, or cardiovascular surgery reported an adjusted odds ratio of 5.5 (CI, 1.9 to 16.2) for the risk for postoperative pulmonary complications in current smokers versus nonsmokers (114). Of interest, current smokers who attempted to reduce cigarette use shortly before surgery were more likely to develop a postoperative pulmonary complication than those who continued usual smoking habits. The adjusted odds ratio was 6.7 (CI, 2.6 to 17.1). Possible explanations include selection bias (patients who correctly perceived themselves as being at high risk for complications may have been more likely to attempt to reduce cigarette use before surgery) or a transient increase in cough and sputum production in the first 1 to 2 months after cigarette cessation. In a study of self-reported duration of smoking cessation before minor surgeries, 2 months of preoperative smoking cessation was necessary for intraoperative sputum volume to decrease to the baseline levels of nonsmokers (129).
Congestive Heart Failure
Three good-quality multivariable risk factor studies identified congestive heart failure as a statistically significant risk factor for postoperative pulmonary complications. While the estimates are variable (I2 = 91%), the DerSimonian–Laird random-effects estimate for the risk associated with congestive heart failure is 2.93 (CI, 1.02 to 8.43). Both the standard and trim-and-fill bias-adjusted methods produce similar estimates (Table 1).
Comorbid Condition Measures
Investigators have evaluated several integrated measures of comorbid conditions as potential predictors of postoperative pulmonary complications. The American Society of Anesthesiologists' (ASA) classification aims to predict perioperative mortality but has since been proven to predict both postoperative pulmonary and cardiac complications (102). The 5 ASA classes are 1) a normally healthy patient (class I), 2) a patient with mild systemic disease (class II), 3) a patient with systemic disease that is not incapacitating (class III), 4) a patient with an incapacitating systemic disease that is a constant threat to life (class IV), and 5) a moribund patient who is not expected to survive for 24 hours with or without operation (class V) (130). In our review, 12 studies (1 good-quality study, 2 fair-quality studies, and 9 poor-quality studies) stratified postoperative pulmonary complication rates by ASA class. Since ASA class is a subjective composite clinical judgment based on several risk factors, we organized the data according to 2 criteria, ASA class II or higher versus ASA class lower than II and ASA class III or higher versus ASA class lower than III, and we generated pooled odds ratios for each criterion (Table 1 and Appendix Table 5). Using either approach, higher ASA class is associated with a substantial increase in postoperative pulmonary complication risk (odds ratios, 4.87 [CI, 3.34 to 7.10] and 2.55 [CI, 1.73 to 3.76], respectively).
In 1 eligible nested case–control study of patients undergoing elective abdominal surgery, the authors studied the predictive value of the Charlson comorbidity index for postoperative pulmonary complications (110). This is a multidisease-specific (incorporating 19 medical conditions), weighted summary measure that considers both number and severity of diseases. Possible summary scores range from 0 to 37 (131). Among 82 patients with postoperative pulmonary complications and controls without postoperative pulmonary complications matched by operation type and age, the Charlson comorbidity index score was 1 of 4 independent, statistically significant risk factors in a multivariable analysis (odds ratio, 1.6 [CI, 1.0 to 2.6] per point).
Functional Dependence
The 2 largest eligible trials from the NSQIP evaluated functional dependence as a potential risk factor for postoperative pulmonary complications (118, 120). Total dependence was the inability to perform any activities of daily living (for example, a dependent patient in a nursing home). Partial dependence was the need for equipment or devices and assistance from another person for some activities of daily living. Our pooled estimates of odds ratios for total and partial dependence are 2.51 (CI, 1.99 to 3.15) and 1.65 (CI, 1.36 to 2.01), respectively.
Obesity
Decreased lung volumes after surgery is a principal cause of postoperative pulmonary complications. Obesity may lead to restrictive pulmonary physiology and may further reduce lung volumes and the ability to take a deep breath after surgery. However, studies evaluating clinically meaningful pulmonary complications after surgery have generally found no increased risk, even for patients with morbid obesity (132, 133). In our review, 9 studies (4 fair-quality and 5 poor-quality studies) reported only unadjusted data (7134 total patients [range, 114 patients to 2964 patients]) and 2 of 8 multivariable studies (1 good-quality study and 1 fair-quality study) determined postoperative pulmonary complication rates for obese patients (Appendix Table 6). Definitions of obesity varied from a body mass index (BMI) of 25 kg/m2 or greater to “morbid obesity.” Of the 8 studies that reported multivariable models, obesity was an independent risk factor in only 1 study. In studies that reported only univariate results, postoperative pulmonary complication rates are similar in obese and nonobese patients (6.3% and 7.0%, respectively).
Even among obese patients, those with greater obesity did not seem to have an increased postoperative pulmonary complication risk. In a study of 197 morbidly obese patients undergoing gastric bypass surgery, authors stratified patients by BMI (99). Postoperative pulmonary complication rates were 10% for patients with a BMI of 43 kg/m2 or less and 12% for those with a BMI greater than 43 kg/m2. This difference was not statistically significant.
Obstructive Sleep Apnea
Obstructive sleep apnea increases the risk for airway management difficulties in the immediate postoperative period, but its influence on postoperative pulmonary complication rates has not been well studied. We identified 1 univariate study that evaluated the risk due to obstructive sleep apnea among patients undergoing hip or knee replacement (92). The case–control study (101 patients with obstructive sleep apnea and 101 matched controls) found non–statistically significant trends toward higher rates of reintubation, hypercapnia, and hypoxemia for patients with obstructive sleep apnea. The authors did not measure rates of postoperative pneumonia or respiratory failure. However, differences for unplanned intensive care unit transfers (20% vs. 6%), all serious complications (24% vs. 9%), and length of stay (6.8 days vs. 5.1 days) were statistically significant. While we await further research, these findings suggest that postoperative pulmonary complication rates may be higher among patients with obstructive sleep apnea.
Impaired Sensorium
Two large trials from the NSQIP evaluated the influence of impaired sensorium on respiratory failure (118) and pneumonia (120) after major noncardiac surgery. The authors defined impaired sensorium as 1) an acutely confused or delirious patient who can respond to verbal or mild tactile stimulation or both or 2) a patient with mental status changes, delirium, or both in the context of current illness. This definition excluded patients with stable chronic mental illness or dementia. Our pooled odds ratio estimate for impaired sensorium is 1.39 (CI, 1.08 to 1.79).
Other Patient-Related Factors
Among eligible studies in our review, diabetes and asthma did not influence postoperative pulmonary complication rates (see Appendix for details). Five studies (2 fair-quality and 3 poor-quality studies) provided unadjusted estimates for postoperative pulmonary complication rates among patients with diabetes. Postoperative pulmonary complication rates for diabetes varied from 6% to 40% among these studies, with a median rate of 21%. Among 4 studies that provided unadjusted data on postoperative pulmonary complication rates for patients with asthma (n = 895), the unadjusted postoperative pulmonary complication rate was 3.0%, which is similar to the crude adjusted postoperative pulmonary complication rate for all studies in our review (3.4%). For 2 additional patient-related factors, exercise capacity and HIV infection, the evidence (on the basis of 1 study each) was insufficient to determine the influence on postoperative pulmonary complication rates (see Appendix).
Procedure-Related Risk Factors
Table 2 displays the unadjusted and adjusted summary estimates for procedure-related risk factors, including surgical site, duration of surgery, anesthetic technique, and emergency surgery.
Surgical Site
We obtained unadjusted postoperative pulmonary complication rates for upper abdominal, lower abdominal, and any abdominal surgery from 43 studies. These were 19.7%, 7.7% and 14.2%, respectively (Appendix Table 7). The unadjusted postoperative pulmonary complication rate for 11 studies of patients undergoing esophagectomy was 18.9%. Among 16 studies of patients undergoing abdominal aortic aneurysm repair, the unadjusted postoperative pulmonary complication rate was 25.5%. Head and neck surgery (6 studies) carried an intermediate risk (unadjusted postoperative pulmonary complication rate, 10.3%). Low-risk procedures were hip surgery (5 studies) and gynecologic or urologic procedures (2 studies), and the unadjusted postoperative pulmonary complication rates were 5.1% and 1.8%, respectively.
The multivariable risk factor studies were heterogeneous in how each handled type of surgery, surgical site, or both. The I2 index for surgical sites ranged from 66.4% to 98.7% for all surgical sites except head and neck surgeries. Seven studies included all noncardiac surgeries but provided only crude postoperative pulmonary complication rates for each type of surgery. Among the 14 studies providing information on abdominal surgeries, 5 studies restricted their sample to patients undergoing upper or lower abdominal surgery, 2 studies compared major abdominal surgery with minor abdominal surgery, 3 studies compared general abdominal surgery with other noncardiac surgeries, and 4 studies compared upper abdominal surgery with lower abdominal or other noncardiac surgeries. Two studies focused on specific high-risk surgical procedures, such as esophagectomy (124) and thoracoabdominal aortic (104) surgeries. Only the 2 largest good-quality NSQIP studies provided a comprehensive assessment of the effect of type of surgery on postoperative pulmonary complication rates (118, 120). These 2 studies are the only source of adjusted estimates for aortic, head and neck, neurologic, and peripheral vascular surgeries (Appendix Table 3).
Patients undergoing open aortic surgeries are at the highest risk for postoperative pulmonary complications (odds ratio, 6.90 [CI, 2.74 to 17.36]). One cohort study compared postoperative pulmonary complication rates for open surgical repair of abdominal aortic aneurysms and endovascular repair (125). After multivariable adjustment for patient-related confounders, the hazard ratio for endovascular repair was 0.14 (CI, 0.04 to 0.47) compared with open surgery. Other high-risk surgeries include thoracic (odds ratio, 4.24 [CI, 2.89 to 6.23]) and upper abdominal operation (odds ratio, 2.91 [CI, 2.35 to 3.60]). Three good-quality and 3 fair-quality multivariable risk factor studies provided estimates of the effect of any abdominal surgery on postoperative pulmonary complication rates. The trim-and-fill bias-corrected odds ratio for any abdominal surgery is 3.01 (CI, 2.43 to 3.72). For surgical procedures with 3 or more studies, trim-and-fill estimates for the surgical procedures differ little from the original random-effects estimates. Much of the observed heterogeneity (I2 = 59.5%) is attributable to differences in composition of the reference group or criteria for defining a postoperative pulmonary complication.
Duration of Surgery
Five fair-quality multivariable risk factor studies provided odds ratios, with SEs, CIs, or both, for prolonged surgery. The definition of prolonged surgery ranged from 2.5 hours to 4 hours. Publication bias for estimates of the effect of prolonged surgery on postoperative pulmonary complication rates was not very evident. The pooled odds ratio for prolonged surgery is 2.26 (CI, 1.47 to 3.47). This finding contrasts with data on postoperative cardiac complications, where duration of surgery is not an independent predictor and does not appear in any commonly used cardiac risk index (1, 134).
Anesthetic Technique
Two good-quality and 4 fair-quality studies provided estimates for postoperative pulmonary complication risk attributable to the use of general anesthesia. The studies were heterogeneous (I2 = 81.7%). The trim-and-fill bias-corrected odds ratio is 1.83 (CI, 1.35 to 2.46).
Emergency Surgery
Six multivariable risk factor studies—2 good-quality, 2 fair-quality, and 2 poor-quality studies—provided odds ratios, with SEs and CIs, for emergency versus elective surgery. The trim-and-fill bias-corrected odds ratio for emergency surgery is 2.21 (CI, 1.57 to 3.11). Patients undergoing emergency surgery incur a modest risk for the development of postoperative pulmonary complications.
Laboratory Testing To Estimate Risk
Spirometry
The first systematic review of the predictive value of preoperative spirometry, published in 1989, concluded that its value was unproven (135). A subsequent economic evaluation found that estimated annual real costs for preoperative spirometry are $25 million to $45 million in 1991 U.S. dollars (136). If use of spirometry were reduced to meet current guidelines, potential savings to third-party payers would range from $29 million to $111 million.
We identified 14 additional eligible studies for our review that evaluated the ability of preoperative spirometry to stratify postoperative pulmonary complication risk (Appendix Table 8) (28, 50, 52, 57, 60, 73, 102, 108, 112, 115, 137-140). Ten studies provided unadjusted univariate data for postoperative pulmonary complications on the basis of particular laboratory findings. In 1 study, 6 of 22 (27%) patients with abnormal results on spirometry had a postoperative pulmonary complication, while fewer patients with normal spirometry results (16 of 100 [16%] patients) had a postoperative pulmonary complication (28). In 3 of 4 studies that determined mean FEV1 values and 3 studies that determined mean FVC values, the value was lower for patients who developed a postoperative pulmonary complication than for those who did not. These differences were, however, small and were unlikely to help clinicians undertake risk stratification.
Three studies (n = 505) provided categorical groupings of FEV1 values. The postoperative pulmonary complication rates for patients in the highest and lowest FEV1 categories were 14.6% and 31.4%, respectively. One study each (n = 324 total) performed a similar analysis by using either FVC or FEV1–FVC ratio and reported similar results (52, 102). None of these studies compared the predictive value of abnormal spirometry results with that of abnormal findings on history or physical examination.
Only 4 eligible studies used multivariable analysis to adjust for potentially relevant clinical variables to determine the independent predictive value of spirometry. Wong and colleagues (108) studied 105 patients undergoing noncardiothoracic surgery who had severe chronic obstructive pulmonary disease (as defined by an FEV1 < 1.2 L and FEV1–FVC ratio < 75%). In their small, select cohort, FEV1:FVC less than 50% was 1 of 5 independent risk factors. Three other factors (abdominal surgery, ASA class IV or V, and general anesthesia) conferred higher odds ratios in the multivariable model. In a study of 460 patients undergoing abdominal surgery, FEV1 less than 61% predicted, Pao2 less than 9.33 kPa (70 mm Hg), FEV1 of 61% to 79% predicted, ischemic heart disease, cancer operation, and age were each independent predictors (140). The single strongest factor was FEV1 less than 61% predicted.
In a third study of 361 patients undergoing upper abdominal surgery, residual volume, diffusing capacity of carbon monoxide (% predicted), and FEV1 (% predicted) were statistically significant independent predictors of postoperative pulmonary complications (112). Chronic mucus hypersecretion (sputum production for at least 3 months of each year) was the strongest factor and predicted risk to a greater degree than any spirometric value. Finally, the fourth study used a case–control design to study 116 patients undergoing elective abdominal surgery and matched controls (110). In this report, FEV1 results were similar between case-patients and controls (2.4 L vs. 2.6 L, respectively) and FVC results were identical (3.6 L). Abnormal findings on chest examination, abnormal results on chest radiography, Goldman cardiac risk index, and Charlson comorbidity index were independent predictors of postoperative pulmonary complication. In contrast, spirometry results were not statistically significant in the final model.
No eligible studies provided data on the use of spirometry to stratify risk for patients with restrictive pulmonary disease or restrictive physiology due to chest wall or neuromuscular disease.
The available literature suggests that spirometry may identify patients at higher risk for development of postoperative pulmonary complications; however, the data are mixed. An additional problem is that while spirometry diagnoses obstructive lung disease, this diagnostic clarity does not translate into effective risk prediction for individual patients. Furthermore, the few studies that have compared spirometric data with clinical data have not consistently shown spirometry to be superior to history and physical examination. While consensus exists on the value of spirometry before lung resection and in determining candidacy for coronary artery bypass, its value before extrathoracic surgery remains unproven. Finally, the data do not suggest a prohibitive spirometric threshold below which the risk for surgery is unacceptable. For example, in a study of 107 operations in patients with severe chronic obstructive pulmonary disease (FEV1 < 50% predicted and FEV1–FVC ratio < 70%), 6 deaths and 7 severe postoperative pulmonary complications occurred (50). While this risk is substantial, it may be acceptable when contemplating life-saving surgery.
Chest Radiography
Clinicians frequently order chest radiography as part of a routine preoperative evaluation. This practice is often due to local institutional guidelines requiring chest radiography for all patients older than a particular age. Only 2 univariate studies that met our inclusion criteria stratified postoperative pulmonary complication rates on the basis of the finding of a normal or abnormal preoperative chest radiograph (28, 139). In their small, pooled patient sample (n = 150), 46% of patients with an abnormal preoperative chest radiograph had a postoperative pulmonary complication, and the rate for patients with a normal preoperative study was 25%. Two eligible studies used multivariable analysis to determine the effect of an abnormal chest radiograph, and both reported that it was a statistically significant predictor of postoperative pulmonary complication rates (110, 114).
Most studies of the value of preoperative chest radiography, however, have not studied postoperative pulmonary complication as the primary outcome measure but have evaluated the frequency with which an abnormal study changes perioperative management. While these studies do not meet the inclusion criteria for our review, we discuss them in our report to provide additional insight into the value of this commonly ordered test. In a recent review of the value of routine preoperative testing, the authors identified 8 studies (n = 14 650) published from 1980 to 2000 of the frequency with which preoperative chest radiography results influenced perioperative management (141). While 23.1% of preoperative chest radiographs in the sample were abnormal, only 3.0% of studies influenced management. Only 4.9% of chest radiographs among patients younger than 50 years of age were abnormal. In an earlier review of 21 studies (n = 14 390) published between 1966 and 1993 (2 of the studies were included in both reviews), 10% of all routine preoperative chest radiographs were abnormal (142). However, only 1.3% of all studies showed unexpected abnormalities and only 0.1% of all studies influenced management.
From these observations, we conclude that clinicians may predict most abnormal preoperative chest radiographs on the basis of the history and physical examination and that chest radiography only rarely provides unexpected information that influences preoperative management. While existing data on clinical outcomes do not allow firm conclusions, the incremental value of the test in estimating postoperative pulmonary complication risk is small. Limited evidence from multivariable risk factor studies supports the use of preoperative chest radiography for patients with known cardiopulmonary disease and those older than 50 years of age who are undergoing upper abdominal, thoracic, or abdominal aortic aneurysm surgery.
Serum Measures of Renal Function
Two studies using NSQIP data identified a serum blood urea nitrogen level of 7.5 mmol/L or greater (≥21 mg/dL) as a statistically significant predictor after multivariable adjustment (118, 120). The risk increased with increasing blood urea nitrogen levels. One study identified serum creatinine level greater than 133 µmol/L (>1.5 mg/dL) as a risk factor after multivariable analysis (123).
Serum Albumin Measurement
Four studies that reported univariate analyses (n = 56 050) stratified postoperative pulmonary complication rates by serum albumin level and used a threshold of 36 g/L to define low serum albumin level (62, 88, 143, 144). Unadjusted postoperative pulmonary complication rates for patients with low and normal serum albumin levels were 27.6% and 7.0%, respectively. Our review of studies reporting multivariable analyses confirms the value of a low serum albumin level as an important predictor of postoperative pulmonary complications. In 4 of 5 eligible studies that considered albumin level, it was an independent risk factor for postoperative pulmonary complications (low level values defined variably from 30 g/L to 39 g/L) (100, 101, 109, 118, 145). In the 1 study that provided an adjusted estimate of risk, the odds ratio was 2.53 (CI, 2.04 to 2.56) (118).
This is consistent with the NSQIP report that a low serum albumin level was also the most important predictor of 30-day perioperative morbidity and mortality (88). In the report, the relationship between serum albumin levels and mortality was continuous for values less than approximately 35 g/L without a clear threshold value.
Oropharyngeal Culture
One eligible univariate study evaluated the value of preoperative oropharyngeal culture to predict postoperative pulmonary complication risk before upper abdominal surgery (51). The evidence is insufficient to determine the influence of this test on postoperative pulmonary complication rates (see Appendix for details).
Pulmonary Risk Indices
While clinicians have used preoperative cardiac indices for more than 3 decades (1), early efforts to develop perioperative pulmonary risk indices for pulmonary and nonpulmonary surgery were limited by conflicting results in validation cohorts (113, 146-148). More recently, Arozullah and colleagues (118, 120) developed 2 indices on the basis of NSQIP data. In the multifactorial postoperative respiratory failure index, Arozullah and colleagues (118) identified statistically significant risk factors in a multivariable analysis of 81 719 male veterans undergoing major noncardiac surgery and validated the index in an additional cohort. The definition of respiratory failure was mechanical ventilation for more than 48 hours or unplanned intubation. The final weighted index included 7 factors. Procedure-related factors dominated the index, which included type of surgery (abdominal aortic aneurysm [27 points]; thoracic [21 points]; neurosurgery, upper abdominal, or peripheral vascular [14 points]; neck [11 points]), emergency surgery (11 points), albumin level less than 30 g/L (9 points), blood urea nitrogen level greater than 10.71 mmol/L (>30 mg/dL) (8 points), partially or fully dependent functional status (7 points), chronic obstructive pulmonary disease (6 points), and age (≥70 years [6 points] or 60 to 69 years [4 points]).
A similarly derived postoperative pneumonia index differed by greater relative weight to age and the inclusion of weight loss, general anesthesia, impaired sensorium, history of cerebrovascular accident, transfusion of more than 4 units, emergency surgery, steroid use for chronic condition, current smoker within 1 year, and alcohol intake of more than 2 drinks per day in the past 2 weeks (120).
These rigorously derived indices from a large cohort advance the field of pulmonary risk stratification. The prominence of unmodifiable risk factors was notable in both indices. These indices, however, allow clinicians to reconsider the indications for surgery in a high-risk patient and suggest patients who will most benefit from strategies to reduce the risk for postoperative pulmonary complications.
Discussion
Postoperative pulmonary complications are common and are an important cause of perioperative morbidity. We present data from a systematic review of the literature on preoperative risk stratification for postoperative pulmonary complications after noncardiothoracic surgery. Table 3 provides the summary strength of the evidence and odds ratios for the association of patient, procedure, and laboratory factors with postoperative pulmonary complications. Among patient-related risk factors, good evidence supports advanced age, ASA class II or greater, functional dependence, chronic obstructive pulmonary disease, and congestive heart failure. Fair evidence, based on fewer studies or a lower odds ratio, supports impaired sensorium, abnormal findings on chest examination, cigarette use, alcohol use, and weight loss. Good evidence suggests that obesity and well-controlled asthma are not risk factors. Fair evidence, based on a single large study, suggests that poorly controlled asthma confers an increased risk. Evidence is insufficient to estimate risk due to obstructive sleep apnea, corticosteroid use, or poor exercise capacity.
The surgical site is an important factor for predicting postoperative pulmonary complication risk. Surgeons and medical consultants have recognized for decades that certain procedures incur inherently higher risk for postoperative pulmonary complications. Our review confirms this observation. The major procedure-related risk factors confer higher risk for pulmonary complications than that of patient-related risk factors. Good evidence supports aortic aneurysm repair, thoracic surgery, abdominal surgery, neurosurgery, vascular surgery, and head and neck surgery. Good evidence also supports the procedure-related risk factors of emergency surgery and prolonged surgery. While good evidence supports general anesthesia as a risk factor on the basis of adjusted observational data, randomized, controlled trials have not consistently reported an effect of anesthetic type on postoperative pulmonary complication rates (see accompanying review [11]). Fair evidence supports esophageal surgery and perioperative transfusion as risk factors.
The value of preoperative testing to estimate pulmonary risk is perhaps the most controversial area in the field of preoperative pulmonary evaluation. While some reports have suggested that certain tests, such as spirometry, identify a subset of high-risk patients, few studies have systematically compared the incremental risk attributable to abnormal preoperative testing with that obtained by history and physical examination. Among potential laboratory tests to stratify risk, a serum albumin level less than 35 g/L is the most powerful predictor and predicts risk to a similar degree as the most important patient-related risk factors. Fair evidence supports a serum blood urea nitrogen level of 7.5 mmol/L or greater (≥21 mg/dL) as a risk factor, but the magnitude of the risk seems to be less than that for low serum albumin level.
While spirometry may provide some risk stratification, most patients identified as high risk by spirometry can be identified equally well by clinical evaluation. Evidence is insufficient to determine whether spirometry provides incremental value as a tool to estimate postoperative pulmonary complication risk. The evidence does not support the use of routine spirometry to stratify risk before noncardiothoracic surgery. Reasonable, although untested, indications include the evaluation of dyspnea when the cause is not apparent by history and physical examination (that is, cardiac causes, pulmonary causes, and deconditioning are all considerations) and in the patient with chronic obstructive pulmonary disease or asthma only if it is uncertain whether airflow obstruction has been maximally reduced before elective surgery. No eligible study in our review provided data on the predictive value of arterial blood gas analyses. Clinicians may anticipate most abnormal preoperative chest radiographs on clinical grounds, and the test infrequently changes preoperative management. Fair evidence supports that an abnormal chest radiograph identifies a cohort of patients with an increased risk for postoperative pulmonary complications.
Research methods in the literature have improved over the past 2 decades, but substantial problems remain. Notable limitations include study sample sizes that are too small to measure clinically relevant pulmonary outcomes, unblinded outcome assessment, inconsistencies in definitions of postoperative pulmonary complication, dependence on observational studies, and statistical issues. The preponderance of the literature consists of observational studies that focus on the discovery of potential risk factors rather than direct hypothesis testing.
Most studies use univariate prescreening and multivariable selection methods to identify a subset of statistically significant risk factors. The net effect is the introduction of a subtle form of publication bias into effect estimates. We used the random-effects trim-and-fill estimates to adjust for publication bias. The trim-and-fill estimates provide an assessment of the sensitivity of our results to the effects of publication bias. Recent methodologic studies, however, suggest that the trim-and-fill method may inappropriately adjust for publication bias where none exists, particularly when study estimates are heterogeneous (149). The net result may be an overcorrection for publication bias; therefore, the reader must exercise care in interpreting the meaning of the actual values.
The literature has developed to a point that we have, in our review, identified a set of potentially important risk factors. Future studies should be large enough to adjust for most, if not all, of these factors to move from exploratory studies to hypothesis testing and confirmatory studies. Investigators should strive to use outcome assessment that is blinded to preoperative risk status.
Appendix
Details of Literature Search and Selection Criteria
We used a broad literature search strategy to identify relevant publications from 1 January 1980 through 8 October 2003. We tested 3 search strategies, as MeSH terms in this area are imprecise. First, we tested an inclusive strategy by using the following MEDLINE indexing terms: intraoperative complications, postoperative complications, preoperative care, intraoperative care, anesthesia, or analgesia. The result was an unmanageable 63 000 citations with an estimated yield less than 2%. The second approach was more tailored, using any of the following terms and specifying them as the article's primary focus: intraoperative complications, postoperative complications, preoperative care, intraoperative care, and postoperative care, plus perioperative complications as text in the title or abstract. This strategy produced 15 466 citations. Third, we tested a narrow strategy by combining the second strategy with terms specific to pulmonary diseases and complications and identified 2860 potentially relevant citations.
We determined the degree of agreement between the second and third searches. Title review of the second, broader strategy results indicated 395 potentially relevant citations of 15 466 (2%) citations. Of these, the third, narrow strategy identified only 157 citations, leaving 238 potentially relevant citations that were missed by the narrow strategy. We reviewed the abstracts of the 238 missed citations in more detail, and 120 of them were potentially relevant. We then reviewed MeSH terms for these 120 citations to identify additional terms to improve the specificity of the second search without sacrificing sensitivity. These included terms for pulmonary, respiratory, or cardiopulmonary diseases, conditions, or complications and terms for oxygenation, chest roentgenography, and lung expansion modalities, such as incentive spirometry.
On the basis of recent systematic reviews and seminal randomized trials, we performed 7 additional focused searches that were tailored to specific topics: preoperative chest radiography, preoperative spirometry, laparoscopic versus open major abdominal procedures, general versus spinal or epidural anesthesia, intraoperative neuromuscular blockade, management of postoperative pain, and postoperative lung expansion techniques. We identified additional references by reviewing bibliographies of retrieved studies.
We limited the review of intervention strategies to randomized, controlled trials and previously published systematic reviews. We subsequently updated searches for studies of risk factors that included multivariable results and randomized, controlled trials of interventions to prevent postoperative pulmonary complications through 30 June 2005. We included only English-language publications and excluded publications without primary data from detailed abstraction (that is, letters, editorials, case reports, conference proceedings, and narrative reviews). We excluded 1) studies with fewer than 25 participants per study group; 2) studies that used only administrative data (for example, ICD-9-CM codes) because of recent data showing poor validity of administrative data for postoperative complications (150-152); 3) studies from developing countries because of potential differences in respiratory and intensive care technology (according to lists from the Organisation for Economic Co-operation and Development and the International Monetary Fund) (153, 154); 4) studies that lacked explicit criteria or definitions for pulmonary complications; 5) studies of ambulatory surgery; 6) studies in which outcomes were physiologic rather than clinical (for example, lung volumes and flow, oximetry); 7) studies of gastric pH manipulation; 8) studies of complications unique to a particular type of surgery (for example, upper airway obstruction after uvulectomy); 9) studies of cardiopulmonary or pediatric surgery; and 10) studies of organ transplantation because of profoundly immunosuppressive drugs. For studies using multivariable logistic regression analysis, we required at least 5 outcome occurrences for each covariate entered into the model. We based this criterion on evidence for minimum thresholds for model stability and reliability when estimating odds ratios and CIs (155). We excluded the few studies that used discriminant analysis because we could not compare the results with odds ratios generated by logistic regression. We did not require eligible studies to provide explicit boundary criteria for risk factors (for example, severity of chronic obstructive pulmonary disease) or for severity of postoperative pulmonary complications (for example, severity of atelectasis). When studies provided such information, we included this in our summary of study characteristics (Appendix Tables 1, 2, and 8).
An investigator evaluated each citation according to the following strategy: title and abstract review, then review of the full reference if necessary. If a reviewer was uncertain, we made the final decision by consensus.
Of 16 959 citations identified by the search, 1223 were duplicates and 14 793 were not relevant on title and abstract review (Figure). Of the remaining 943 potentially relevant citations, we excluded 626 citations after review of the full publication, abstracted 145 citations in detail, and used 172 citations as background references. We systematically abstracted data from eligible studies into standardized electronic data forms. Eligible studies varied in their definitions of the postoperative period. Most commonly, authors defined the postoperative period as the hospital stay, ranging from 4 hours to 3 months after surgery.
Assessing Study Quality
We used the USPSTF criteria for assigning hierarchy of research design and grading a study's internal validity as our basis for assessing study quality (12). A good-quality cohort or case-series study, at a minimum, adjusted for key confounders of age, chronic obstructive pulmonary disease, and surgical type; showed little or no differential loss to follow-up; explicitly masked outcome assessment, and provided explicit definitions for what constituted a postoperative pulmonary complication. A fair-quality cohort or case-series study adjusted for key confounders, showed little or no differential loss to follow-up, and provided a clear definition for a postoperative pulmonary complication but was unclear about masking of outcome assessment. A poor-quality cohort or case-series study did not include 1 or more of the key confounders, showed statistically significant differential loss to follow-up, provided vague or no definitions for a postoperative pulmonary complication, explicitly did not mask outcome assessments, or a combination of these criteria. We assigned summary strength of recommendations for each risk factor and laboratory test according to modified criteria proposed by the USPSTF (12). We modified the criteria for the review on preoperative risk stratification to reflect the absence of a risk–benefit equation when considering a risk factor rather than an intervention. When both univariate and multivariate data were available about a potential risk factor, we considered most strongly the effect of the multivariate data when assigning strength of recommendations. When the effect of a risk factor was based on only 1 multivariate study or was limited to univariate data, we considered the evidence to be insufficient to determine that the factor contributed to postoperative pulmonary complication risk.
A good-quality randomized, controlled trial met all of the following criteria: comparable groups assembled initially and maintained throughout the study, follow-up of at least 80% of participants, reliable and valid measurement instruments applied equally to all groups, clearly described interventions, consideration of important and relevant outcomes, appropriate attention to confounders in analysis, and intention-to-treat analyses. We graded studies as fair if any of the following problems occurred: generally comparable groups assembled initially but some (although not major) possible differences occurring in follow-up, generally acceptable (although not the best) measurement instruments generally applied equally, some but not all important outcomes considered, some but not all potential confounders accounted for in analysis, and intention-to-treat analyses. Studies were poor if any of the following “fatal” flaws occurred: sufficiently comparable groups not assembled initially nor maintained throughout the study, unreliable or invalid measurement instruments, measurement instruments not applied equally among groups during follow-up (including unblended outcome assessment), follow-up of less than 80% of participants, little or no attention to key confounders, and intention-to-treat analyses not done.
We also graded systematic reviews as good, fair, or poor on the basis of extent of literature searched, inclusion or exclusion of non–English-language publications, statements of inclusion and exclusion criteria, protocols for appraisal of study quality and data abstraction, data synthesis methods, presentation of results, and discussion of clinical inferences and future research needs. Components of good quality included searching of MEDLINE plus other important databases (for example, EMBASE, Cochrane Library, or Clinical Trials Registry), inclusion of non–English-language publications, and a clear statement of acceptable inclusion criteria (for example, population, intervention, primary outcomes, study design, and assessment of agreement among reviewers). Good-quality reviews had good protocols for appraisal of study quality (for example, randomization, allocation concealment, blinding, independent assessment by ≥ 2 reviewers, assessment of interreviewer agreement, and process for resolution of agreement stated) and data abstraction (for example, independent assessment by ≥ 2 reviewers, interreviewer agreement, resolution process for disagreement, and standardized data abstraction forms). Components of good-quality quantitative synthesis included random-effects models, assessment of statistical heterogeneity, handling of missing data, rationale for a priori sensitivity and subgroup analyses, and assessment of publication bias. Good-quality presentation of results included a flow diagram for results of the literature search with numbers and reasons for exclusions, adequate reporting of characteristics of included studies (for example, study design, participant characteristics, quality score, details of intervention, outcome definitions, and assessment of clinical heterogeneity), and summary results with effect sizes and CIs. A good-quality discussion included summarization of key findings, clinical inferences based on internal and external validity of studies, interpretation of results on the totality of the evidence, potential biases, and a future research agenda. We graded systematic reviews and meta-analyses as fair if they were of fair to good quality on most components and as poor if they achieved only poor quality on most components.
Statistical Methods
Our literature search yielded primarily unadjusted estimates for most laboratory factors of interest. Limited multivariable, adjusted estimates were available for albumin level less than 30 g/L and elevated blood urea nitrogen level. However, rather than attempt to compute a potentially biased summary estimate, we provided narrative descriptions of the pattern of results for these potential risk factors.
The eligible multivariable risk factor studies varied considerably in the number and type of competing risks and confounders included in the analyses. Extensive use of prescreening methods and variable selection algorithms often limited reporting to the subset of risk factors that were determined as statistically significant in a given sample. The result is the introduction of a subtle form of publication bias, which we verified by examination of the funnel plots for each risk factor.
We extracted odds ratios from each study, along with their respective SEs, 95% confidence limits, or both. When necessary, we estimated SEs from the 95% confidence limits (156). We used the I2 statistic (13) and the Cochran Q statistic (14) to assess study heterogeneity. We also recomputed pooled estimates with and without studies that produced extreme results. The I2 statistic is the proportion of the total variance in the pooled estimate that is attributable to between-study variance. It is the maximum of (0, (Q − df)/Q), where the degrees of freedom (df) are the number of studies minus 1. This situation occurs when we have only 2 studies. An I2 statistic of 50% or greater indicates substantial heterogeneity among study estimates. We used the DerSimonian–Laird method to compute random-effects estimates when the set of studies was heterogeneous (15). In cases where 3 or more studies contributed estimates for a risk factor, we used the trim-and-fill method to adjust pooled estimates of a risk factor's effect on postoperative pulmonary complications for publication bias (16).
For the review on intervention strategies, we performed simple means and chi-square testing when eligible studies did not provide CIs or P values for statistical significance. We did not perform quantitative pooling or meta-analyses because we identified previously published meta-analyses and found insufficient additional evidence to warrant repooling. In other cases, studies were too few or were too clinically heterogeneous for meta-analysis.
Other Factors That Do Not Influence Risk
Exercise Capacity
We identified only 1 eligible study that evaluated the value of self-reported exercise capacity as a predictor of postoperative pulmonary complication rates (84). In the report, investigators asked 600 consecutive patients who were referred to a preoperative medical consultation clinic to estimate the number of blocks on level ground that they could walk and the number of flights of stairs that they could climb without rest. By using a definition of 4 blocks or 2 flights of stairs as good exercise capacity, the investigators found that postoperative pulmonary complication rates for patients with good and poor exercise capacity were 6.3% and 9.0%, respectively (P = 0.21). The unadjusted odds ratio showed a non–statistically significant trend toward poor self-reported exercise capacity as a predictor of postoperative pulmonary complication rates (odds ratio, 1.43 [CI, 0.81 to 2.53]).
Diabetes
Five studies (n = 1017) reported the unadjusted (univariate) postoperative pulmonary complication risk attributable to diabetes (Appendix Table 6). The postoperative pulmonary complication rate among patients with diabetes was 15.9%. Event rates in 1 study (91) were too low to reliably estimate the odds ratio. The unadjusted summary estimate after pooling the remaining 4 studies is not statistically significant (odds ratio, 1.7 [CI, 0.9 to 3.1]) for diabetes. No eligible study using multivariable analyses reported diabetes to be an independent, statistically significant risk factor for postoperative pulmonary complication.
Asthma
Only 1 of 4 studies examining postoperative pulmonary complication rates among patients with asthma included a control group of patients without asthma; therefore, we could not calculate odds ratios. However, the postoperative pulmonary complication rate among patients with asthma (n = 895) was 3.0%, which is similar to the crude adjusted postoperative pulmonary complication rate for all studies in our review (3.4%). The low postoperative pulmonary complication rate probably resulted in part from a younger, healthier population than among unselected surgical patients. For example, in the largest study of patients with asthma, by Warner and colleagues (74), the mean age was 18 years (interquartile range, 9.9 years to 36.4 years) and 87% of patients were ASA class I or II. The study also included minor postoperative pulmonary complications of little clinical significance and minimal morbidity, such as bronchospasm and laryngospasm.
In Warner and colleagues' study, asthma severity, defined by need for asthma medications in the 30 days before surgery, and asthma control, defined by emergency department or office visits in the 30 days before surgery, both correlated with postoperative pulmonary complication rates. Postoperative pulmonary complication rates stratified by recent asthma medication use were 5.1% vs. 0.4%, and those stratified by recent emergency department or office visit were 28% vs. 1.3%. These differences were statistically significant.
HIV Infection
We identified only 1 eligible study that reported postoperative pulmonary complication risk among patients with HIV infection (76). Among 15 059 patients, 5 of 89 (5.6%) HIV-infected patients required an unplanned postoperative critical care admission for mechanical ventilation. The overall rate of unplanned critical care admissions was 0.3%, and HIV infection was the only comorbid condition that statistically significantly predicted risk. This observation, however, was based on a small number of patients.
Oropharyngeal Culture
One eligible univariate study evaluated the value of preoperative oropharyngeal culture to predict postoperative pulmonary complication risk before upper abdominal surgery (51). In the report, investigators obtained throat swabs before elective surgery. Nine of 17 patients whose preoperative cultures grew Haemophilus influenzae developed postoperative chest infection, while only 15 of 91 patients with negative cultures developed postoperative chest infection. While provocative, the evidence suggests that clinicians should await confirmation by other studies before considering this as a strategy for postoperative pulmonary complication risk stratification.
Article and Author Information
-
Acknowledgments: The authors gratefully acknowledge the tremendous contribution of medical librarian Martha R. Harris, MA, for her time and expertise in searching the medical literature and managing the resulting project database.
-
Grant Support: By the Veterans Evidence-based Research, Dissemination, and Implementation Center (VERDICT) (Veterans Affairs Health Services Research and Development, HFP 98-002).
-
Potential Financial Conflicts of Interest: Stock ownership or options (other than mutual funds): G.W. Smetana (SafeMed Harvard Imaging); Other: G.W. Smetana (Novartis Pharma Schweiz).
-
Requests for Single Reprints: Gerald W. Smetana, MD, Division of General Medicine and Primary Care, Beth Israel Deaconess Medical Center, 330 Brookline Avenue, Boston, MA 02215; e-mail, gsmetana{at}bidmc.harvard.edu.
-
Current Author Addresses: Dr. Smetana: Division of General Medicine and Primary Care, Beth Israel Deaconess Medical Center, 330 Brookline Avenue, Boston, MA 02215.
-
Drs. Lawrence and Cornell: Medicine/General Medicine, The University of Texas Health Science Center at San Antonio, 7703 Floyd Curl Drive, Mail Code 7879, San Antonio, TX 78229-3900.
References
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.↵
- 94.↵
- 95.↵
- 96.↵
- 97.↵
- 98.↵
- 99.↵
- 100.↵
- 101.↵
- 102.↵
- 103.↵
- 104.↵
- 105.↵
- 106.↵
- 107.↵
- 108.↵
- 109.↵
- 110.↵
- 111.↵
- 112.↵
- 113.↵
- 114.↵
- 115.↵
- 116.↵
- 117.↵
- 118.↵
- 119.↵
- 120.↵
- 121.↵
- 122.↵
- 123.↵
- 124.↵
- 125.↵
- 126.↵
- 127.↵
- 128.↵
- 129.↵
- 130.↵
- 131.↵
- 132.↵
- 133.↵
- 134.↵
- 135.↵
- 136.↵
- 137.↵
- 138.↵
- 139.↵
- 140.↵
- 141.↵
- 142.↵
- 143.↵
- 144.↵
- 145.↵
- 146.↵
- 147.↵
- 148.↵
- 149.↵
- 150.↵
- 151.↵
- 152.↵
- 153.↵
- 154.↵
- 155.↵
- 156.↵
Related Clinical Content
