Flexman, Alana M. MD, FRCPC*; Merriman, Bradley MD*; Griesdale, Donald E. MD, FRCPC*; Mayson, Kelly MD, FRCPC*; Choi, Peter T. MD, FRCPC*; Ryerson, Christopher J. MD, FRCPC†
Postoperative respiratory failure significantly increases length of hospital stay and mortality.1–4 Patients requiring reintubation or prolonged ventilation following surgery are vulnerable to multiple complications including hypoxemia and ventilator-associated pneumonia, and they consume scarce critical care resources.5–8 Although previous studies have identified risk factors for respiratory failure in a broad surgical population, predictors of respiratory failure and death may vary in specific surgical populations such as neurosurgery.
Neurosurgical patients are at an increased risk of respiratory failure, pneumonia, and death compared with other surgical specialties.2,9,10 Potential reasons for this increased risk include a higher incidence of altered mental status, prolonged surgical procedures, and central respiratory dysfunction secondary to postoperative swelling (ie, retraction near the brainstem).11 Patients undergoing resection of infratentorial mass lesions may have an even higher risk of respiratory failure and death due to several unique factors. The infratentorial compartment is relatively limited in size and contains vital structures, such as the brainstem and cranial nerves, which are susceptible to compression. Furthermore, intraoperative prone positioning may precipitate mechanical airway obstruction from macroglossia, whereas resection near the brainstem may precipitate central respiratory dysfunction and cranial nerve injury.1,10,12 Although neurosurgical procedures are associated with an increased risk of respiratory failure in multiple studies, the incidence and predictors in neurosurgery have not been adequately defined.
We retrospectively analyzed a large international surgical outcomes database, the American College of Surgeons National Surgical Quality Improvement Program (NSQIP) database, to identify the incidence and independent predictors of respiratory failure and death in patients undergoing intracranial neurosurgery. Our specific objective was to determine whether infratentorial procedures have a higher risk of respiratory failure and death compared with supratentorial procedures. Identifying predictors of these clinically important outcomes in neurosurgical patients may improve perioperative decision making and potentially reduce the morbidity, mortality, and resource consumption.
Data were retrospectively extracted from the NSQIP database following approval from the University of British Columbia Clinical Research Ethics Board (REB H12-01444). The NSQIP database is a prospective multicenter database of adult patients that has been described previously in detail.13 The most recent year of the database in this study (2010) includes data from 258 sites. Participating sites range from small, rural hospitals to large academic teaching medical centers. Data were obtained for all 240 variables in the 2005 to 2010 NSQIP databases, inclusive. Variables include preoperative risk factors, intraoperative variables, and morbidity and mortality outcomes up to 30 days postoperatively.14 A trained clinical reviewer enters data using a web-based interface at each participating site. High-quality data collection and entry are ensured through rigorous training of the clinical reviewer and the interrater reliability audits to ensure accurate and consistent data entry. The overall disagreement rate is 1.8%.14 Cases are sampled using an 8-day cycle to ensure even case mix, and operating room logs are audited to ensure representative sampling of cases.
Patients undergoing supratentorial or infratentorial craniotomy for excision of a brain tumor were included in the analysis. Patients undergoing combined supratentorial and infratentorial procedures were not included in the study population as we were unable to accurately analyze the effect of surgical site on our primary outcome. The type of surgery and its location (supratentorial or infratentorial) were determined using Common Procedural Terminology codes (Supplemental Digital Content, Table 1, http://links.lww.com/JNA/A11). Patients were excluded from the analysis if they were ventilated preoperatively, comatose, quadriplegic, pregnant, American Society of Anesthesiologists physical status class 5, or underwent an anesthetic procedure other than general anesthesia (Supplemental Digital Content, Table 2, http://links.lww.com/JNA/A11). Patients older than 90 years were excluded as these patients did not have an accurate age listed in the database to maintain deidentification of these few patients.
The primary outcome was a composite of death within 30 days after surgery, unplanned reintubation within 30 days after surgery, and failure to wean from mechanical ventilation within 48 hours after surgery (Supplemental Digital Content, Table 3, http://links.lww.com/JNA/A11). Only the first event was included in the analysis if a patient had >1 event in the composite outcome. For example, if a patient died following unplanned reintubation, the patient would be assigned the outcome for unplanned intubation only.
Candidate predictor variables were identified on the basis of the literature review and clinical experience, and we considered variables that were available to NSQIP subjects throughout all available years of the database. We included predictors of respiratory failure in the general surgical population that were identified in a previous study, which evaluated a broad range of surgical specialties (emergency surgery, albumin level <30 g/L, blood urea nitrogen [BUN] level >30 mg/dL, dependent functional status, chronic obstructive pulmonary disease [COPD], and age). In addition, we evaluated potential predictor variables for respiratory failure that may be specific to the neurosurgical population. Variables included age, sex, body mass index, smoking history, alcoholism, dyspnea, functional status, history of COPD, preoperative pneumonia, cardiac disease, impaired sensorium, neurological disease, preoperative BUN, preoperative albumin, location of tumor (axial or extra-axial), emergency surgery, and anesthetic duration (Supplemental Digital Content, Table 4, http://links.lww.com/JNA/A11). Several related variables were combined to produce composite neurological and cardiac disease predictor variables (Supplemental Digital Content, Table 4, http://links.lww.com/JNA/A11).
Baseline features in patients with respiratory failure and death were compared with those without using a χ2 test, the Fisher exact test, t test, or the Wilcoxon rank sum test. Independent predictors of the primary outcome were identified using multivariate logistic regression. The multivariate model was built in several steps, starting with variables that independently predicted respiratory failure in a broader surgical population.1 All candidate predictors associated with the primary outcome (P<0.20) on bivariate analysis in the current study were subsequently added to this model. The P value threshold of 0.20 was chosen to rule out confounding with greater certainty. Further model simplification was based on relative accuracy and complexity of potential multivariate models, considering change in the area under the receiver operating characteristic curve, goodness-of-fit, and likelihood-ratio testing. Continuous predictor variables were transformed to approximate a normal distribution if necessary to improve model performance and were assessed for a linear relationship with the outcome variable. Included variables were assessed for collinearity and interactions. The consistency of the multivariate model was tested using alternative model-building strategies, including bootstrap and jack-knife procedures, to reduce the impact of potential outliers, as well as forward and backward stepwise regression. A P value of 0.05 was used as the threshold for retention of variables in the stepwise models. All data analyses were performed using STATA 11.2 (StataCorp, TX).
A total of 1808 patients met the inclusion criteria, selected from 1,334,886 patients in the NSQIP database. Following exclusion of 109 patients, 1699 remained in our cohort for analysis (Supplemental Digital Content, Fig. 1, http://links.lww.com/JNA/A11). Table 1 summarizes the patient characteristics stratified by the presence or absence of respiratory failure and death. Of the 1699 cases, 1336 (79%) were supratentorial and 363 (21%) were infratentorial.
Postoperative respiratory failure and death occurred in 75 of the 1699 patients (4.4%, 95% confidence interval [CI], 3.4%-5.4%). The frequency ranged from 2.8% in 2008 (7 of 253 cases) to 8.3% in 2006 (6 of 72 cases), with no association between our composite outcome and year of surgery (P=0.23). The primary outcome was defined as death in 26 cases (35%), reintubation in 30 (40%), and failure to wean from mechanical ventilation in 19 (25%). Sixteen additional patients died following reintubation or failure to wean from mechanical ventilation. Total mortality was 2.4% (95% CI, 1.7%-3.1%). The median time to reintubation was 2 days (interquartile range, 1 to 6 d), and the median time to death was 19 days (interquartile range, 10 to 26 d) from the time of surgery (Fig. 1).
Bivariate Predictors of Postoperative Respiratory Failure and Death
Variables associated with respiratory failure and death on bivariate analysis included older age, worse dyspnea severity, impaired sensorium, worse functional status, neurological disease, COPD, cardiac disease, emergency surgery, infratentorial surgical site, and longer anesthetic duration (Table 1). Respiratory failure and death occurred in 3.8% of supratentorial surgeries and 6.6% of infratentorial surgeries (unadjusted OR, 1.77; 95% CI, 1.08-2.91; P=0.02). Mortality was higher following infratentorial surgery (4.1%; 95% CI, 2.1%-6.2%) than that following supratentorial neurosurgery (1.9%; 95% CI, 1.2%-2.7%; P=0.02).
Urea and albumin were not associated with the primary outcome on unadjusted analysis in our study population (P=0.10 for BUN, P=0.07 for albumin). These variables were associated with a high rate of missing data in our study population (BUN, 15% and albumin, 51%) in a nonrandom manner. Patients with missing data for these variables were younger, more often smokers, less likely to have a comorbid disease (eg, cardiac disease, neurological disease, and diabetes), and tended to have lower rates of respiratory failure and death (P=0.07 for BUN, P=0.03 for albumin). Less than 0.5% of data were missing for all other variables included in the analysis.
Multivariate Predictors of Respiratory Failure and Death
We describe our multivariate model-building process in the Supplemental Digital Content (Table 5, http://links.lww.com/JNA/A11). We started with the original model of respiratory failure published by Arozullah et al,1 substituting continuous variables for age, BUN, and albumin, which were categorized in the Arozullah model. We labeled this as model 1. We next added all candidate predictor variables that were associated with our primary outcome on bivariate analysis with a P<0.20 in the present study (model 2). Urea and albumin were not independently associated with the primary outcome in model 1 or 2 and had a high rate of missing data. Exclusion of BUN and albumin from model 2 improved the c-index and goodness-of-fit test (model 3). We next evaluated removal of potentially collinear or overlapping variables, including dyspnea versus COPD and neurological disease versus impaired sensorium versus functional status. Removal of COPD (model 4) and functional status (model 5) did not adversely affect the model performance or goodness-of-fit (likelihood-ratio test, P=0.21 and 0.32, respectively) and resulted in the final multivariate model (Table 2). Removal of surgical site (infratentorial vs. supratentorial) significantly worsened the model performance, and this variable was therefore retained in the final model (likelihood-ratio test, P<0.05).
Although anesthetic duration deviated slightly from a normal distribution, we retained this variable in its natural form, as substitution of the normally distributed logarithm of anesthetic duration did not improve the model performance. An unbiased test for interaction among included variables revealed a statistically significant interaction only between anesthetic duration and dyspnea severity (P=0.02; unadjusted for multiple comparisons testing). Inclusion of this interaction term neither improved the model performance nor altered the associations observed in the original model. Similar results were produced using stepwise variable selection as an unbiased model-building strategy, as well as the bootstrap and jack-knife procedures to mitigate the influence of outliers (data not shown). Backward stepwise regression included the same variables as the original final model. Forward stepwise selection added functional status to the final model (P=0.14), with no meaningful change in point estimate or confidence interval for the other variables.
The final model included surgical site, age, dyspnea, cardiac disease, neurological disease, impaired sensorium, emergency surgery, and anesthetic duration (Table 2). The independent risk of postoperative respiratory failure and death attributed to age and anesthetic duration (the 2 continuous variables in the final model) is shown in Figure 2 and in the Supplemental Digital Content (Fig. 2, http://links.lww.com/JNA/A11). Infratentorial surgery was associated with a similar increased risk for each component of the composite outcome with the strongest association seen between infratentorial site and death (adjusted OR, 2.44; 95% CI, 1.23-4.87) (Fig. 3).
We retrospectively analyzed a large international prospective database to determine the incidence of and risk factors for respiratory failure and death following intracranial tumor resection. We identified several novel risk factors for postoperative respiratory failure and death in patients undergoing neurosurgical procedures. Specifically, we identified an infratentorial surgical site as a novel predictor of a poor outcome following intracranial tumor resection when compared with supratentorial craniotomy. This increased risk persists even with adjustment for potential confounders, confirming the robustness of this finding. These results have important implications for perioperative risk stratification in patients undergoing intracranial tumor resection.
The overall incidence of respiratory failure and death in our study population (4.4%) is higher than previously reported in large unselected surgical populations in similar databases (3.0% to 3.4%).1,2,15 These findings may be in part due to a higher risk of postoperative respiratory failure and death in neurosurgical patients compared with a broader surgical population. In addition, we a priori included mortality in our composite outcome, although not all previous studies investigating respiratory failure have done so. We believe that failure to include death as a component of respiratory failure introduces bias by excluding events that rapidly resulted in death before the patient could be reintubated. Although death may result from etiologies other than respiratory failure, reintubation and prolonged ventilation can similarly be the consequence of multiple diverse etiologies that are not directly the result of respiratory failure. The limitations of the NSQIP data set did not permit us to analyze the events leading up to any of these outcomes (eg, cause-specific mortality). As death, reintubation, and prolonged ventilation are all clinically meaningful outcomes on the continuum of respiratory failure, we therefore believe that death should be included as a component of respiratory failure.
An interesting, although unexpected, finding in our study was that, although all 3 components of our composite outcome had a similar direction of effect, the increased risk associated with infratentorial tumor resection was most pronounced for death. Intraoperative variables are less likely to have contributed to this outcome as death typically occurred later in the postoperative period (median 19 d). A previous single-center study reported that infratentorial neurosurgery for intra-axial tumor resection was associated with a significantly higher rate of postoperative neurological complications,16 and similar complications may be responsible for the increased mortality in our study. We were unable to address this possibility due to the absence of this level of detail in the NSQIP database. The significantly increased mortality rate in patients undergoing infratentorial neurosurgery is an important, although preliminary, observation. An additional study is required to determine the reasons for this finding and to identify potential strategies that could minimize this risk.
Previous studies have consistently demonstrated that age and emergency surgery are the risk factors for respiratory complications in unselected surgical populations, and our results are consistent with these findings.1–3 Similarly, 2 previous studies have identified impaired sensorium as a risk factor for respiratory failure in unselected surgical populations with a pooled OR of 1.39 (95% CI, 1.08-1.79).15 Our results not only provide further validation for the association between impaired sensorium and respiratory failure and death (OR, 2.28; 95% CI, 1.19-4.37) but also suggest that this association is stronger in the neurosurgical population. Although the limitations of the NSQIP data set do not permit us to draw conclusions about the etiology of respiratory failure and death in these neurosurgical patients, we can speculate that altered neurological status may have contributed to these outcomes.
Preoperative albumin and BUN did not predict postoperative respiratory failure and death following neurosurgery despite previous studies showing that these variables predict respiratory failure and postoperative mortality in a variety of surgical settings.1 Differences between patients with and without values for one or both of these variables may have influenced the prognostic value of the 2 variables. Patients with unmeasured BUN and albumin were younger and had fewer comorbid diseases, suggesting that inclusion of these variables in a prediction model may introduce bias and would prohibit generalization of these results to patients in whom these variables would not have been prospectively measured for clinical purposes.
Our study has several strengths compared with the previously published literature. We used a large, international, multicenter database that prospectively and rigorously collects preoperative variables and outcome data in surgical patients. We restricted our analysis to neurosurgical patients undergoing intracranial tumor resection to produce a specific set of risk factors that are unique to this patient population. Our results are therefore based upon and generalizable to neurosurgical patients undergoing intracranial tumor resection. Unlike a previous study looking at respiratory complications after neurosurgery,17 we used objective and clinically important outcomes to define respiratory failure (reintubation, failure to wean from mechanical ventilation) and included mortality in our analysis. Overall, we were able to confirm the increased postoperative risks of respiratory failure and death following infratentorial neurosurgery using a rigorous, transparent, and reproducible methodology. Furthermore, we showed that the association between infratentorial surgery and our composite outcome is most influenced by increased mortality in this population.
Our study has several limitations. First, the number of cases and events within the NSQIP neurosurgical population was insufficient for both derivation and validation cohorts. However, our study population was derived from a large, multicenter, international database, suggesting that the risk of bias is low and our results are generalizable to a large number of clinical settings. Second, our analysis was limited to the variables included in the NSQIP database, but there may be other unmeasured predictors of respiratory failure and death. For example, we were unable to analyze the relationships between preoperative oximetry and intraoperative factors related to anesthetic management, and we did not have access to details about the tumor characteristics or intracranial pressure. Similarly, we were unable to evaluate the potential impact of surgeon experience and hospital volume as NSQIP does not record surgeon-specific or site-specific variables. In addition, some authors have suggested that scheduled procedural time is a more appropriate variable than actual anesthetic duration for preoperative risk stratification18; however, scheduled procedural time is not reported in the NSQIP database. We instead included actual anesthetic duration, as this variable is similarly important in risk prediction and may be used to guide postoperative disposition at the conclusion of surgery. Finally, our results are only applicable to patients undergoing intracranial tumor resection under general anesthesia and we are unable to extrapolate these results to other types of neurosurgical procedures or those performed with local anesthesia.
Overall, our study demonstrates that infratentorial craniotomy is associated with an increased risk of postoperative respiratory failure and death when compared with supratentorial craniotomy and death is the most important contributor to the elevated risk experienced by this population. Our results highlight that the potential life-prolonging benefits of intracranial tumor resection need to be carefully weighed against the risk of death in patients undergoing infratentorial procedures. Future studies are required to identify the etiology of this elevated mortality and to evaluate potential strategies to minimize the risk of death following neurosurgery.
1. Arozullah AM, Daley J, Henderson WG, et al .Multifactorial risk index for predicting postoperative respiratory failure in men after major noncardiac surgery. The National Veterans Administration Surgical Quality Improvement Program.Ann Surg. 2000; 232:242–253.
2. Johnson RG, Arozullah AM, Neumayer L, et al .Multivariable predictors of postoperative respiratory failure after general and vascular surgery: results from the patient safety in surgery study.J Am Coll Surg. 2007; 204:1188–1198.
3. Canet J, Gallart L, Gomar C, et al .Prediction of postoperative pulmonary complications in a population-based surgical cohort.Anesthesiology. 2010; 113:1338–1350.
4. Gupta H, Gupta PK, Fang X, et al .Development and validation of a risk calculator predicting postoperative respiratory failure.Chest. 2011; 140:1207–1215.
5. Beauregard CL, Friedman WA .Routine use of postoperative ICU care for elective craniotomy: a cost-benefit analysis.Surg Neurol. 2003; 60:483–489
6. Dimick JB, Chen SL, Taheri PA, et al .Hospital costs associated with surgical complications: a report from the private-sector National Surgical Quality Improvement Program.J Am Coll Surg. 2004; 199:531–537.
7. Shander A, Fleisher LA, Barie PS, et al .Clinical and economic burden of postoperative pulmonary complications: patient safety summit on definition, risk-reducing interventions, and preventive strategies.Crit Care Med. 2011; 39:2163–2172.
8. Vidotto MC, Sogame LC, Gazzotti MR, et al .Implications of extubation failure and prolonged mechanical ventilation in the postoperative period following elective intracranial surgery.Braz J Med Biol Res. 2011; 44:1291–1298.
9. Arozullah AM, Khuri SF, Henderson WG, et al .Development and validation of a multifactorial risk index for predicting postoperative pneumonia after major noncardiac surgery.Ann Intern Med. 2001; 135:847–857.
10. Cata JP, Saager L, Kurz A, et al .Successful extubation in the operating room after infratentorial craniotomy: the Cleveland Clinic experience.J Neurosurg Anesthesiol. 2011; 23:25–29.
11. Bruder N, Ravussin P .Recovery from anesthesia and postoperative extubation of neurosurgical patients: a review.J Neurosurg Anesthesiol. 1999; 11:282–293.
12. Pivalizza EG, Katz J, Singh S, et al .Massive macroglossia after posterior fossa surgery in the prone position.J Neurosurg Anesthesiol. 1998; 10:34–36.
13. Mashour GA, Shanks AM, Kheterpal S .Perioperative stroke and associated mortality after noncardiac, nonneurologic surgery.Anesthesiology. 2011; 114:1289–1296.
15. Smetana GW, Lawrence VA, Cornell JE .Preoperative pulmonary risk stratification for noncardiothoracic surgery: systematic review for the American College of Physicians.Ann Intern Med. 2006; 144:581–595.
16. Brell M, Ibanez J, Caral L, et al .Factors influencing surgical complications of intra-axial brain tumours.Acta Neurochir (Wien). 2000; 142:739–750.
17. Sogame LC, Vidotto MC, Jardim JR, et al .Incidence and risk factors for postoperative pulmonary complications in elective intracranial surgery.J Neurosurg. 2008; 109:222–227.
18. Dexter F, Dexter EU, Ledolter J .Statistical grand rounds: importance of appropriately modeling procedure and duration in logistic regression studies of perioperative morbidity and mortality.Anesth Analg. 2011; 113:1197–1201.