Early Extubation following Cardiac Surgery in a Veterans Population 

All patients undergoing cardiac surgery requiring CPB at the Denver Veterans Affairs Medical Center, a university-affiliated teaching institution, were managed on an FT clinical pathway instituted October 1, 1993. Organizational and operational characteristics of our FT pathway have been described before. [15] 

A “light” anesthetic technique incorporating volatile anesthesia (primarily isoflurane or desflurane) and low doses of fentanyl (intravenous bolus) or sufentanil (intravenous bolus or continuous infusion), supplemented with midazolam (intravenous bolus) or propofol (continuous infusion), was used in all patients. Isoflurane was administered via a vaporizer on the CPB machine. Steroids (methylprednisolone sodium succinate) were administered during operation and for 24 h afterward [dagger, dagger] Hypothermic CPB (either mild [32 [degree sign]C] or moderate [28 [degree sign]C] systemic hypothermia) with a membrane oxygenator and intermittent anterograde or retrograde (or both) blood cardioplegia during a single period of aortic cross clamping was used routinely. Patients were considered for early tracheal extubation if they met standard clinical criteria (adequate mental status, strength, oxygenation, normal body temperature, and lack of mediastinal bleeding). [15] 

After we received institutional review board approval, we began a retrospective data collection. We analyzed data from 304 consecutive patients managed on our FT protocol from October 1, 1993 to September 30, 1995 previously presented as a single cohort. [15]One patient was included twice (initial valve replacement with semi-elective replacement for endocarditis several months later). Two patients who died, one during operation and one <10 h after operation, were excluded because they did not reach the time to extubation cut point used for logistic regression modeling (explained here subsequently). Three more patients were eliminated because of missing data for one or more of the study variables. Consequently, 299 patients were included in the final multivariate analyses.

Preoperative risk factors known or postulated to be associated with postoperative outcome were selected for univariate screening for an association with timing of extubation based on published evidence and risk models developed by the DVA's Continuous Improvement in Cardiac Surgery Program (CICSP) in the DVA patient population (Table 1). The CICSP logistic regression modeling procedure generates a “risk-adjusted” estimate of operative mortality based on mandatory reporting of a battery of preoperative risk factors and clinical outcomes for each patient undergoing cardiac surgery in the DVA. [18,19]Other risk factors not specifically included in the CICSP database, but believed by the investigators to influence timing of postoperative extubation, were obtained from the preanesthesia evaluation record. Perioperative processes of care specific to the FT pathway, and others known to influence outcome based on literature review and clinical consensus of the investigators (i.e., surgical procedural details, perfusion details, blood product administration, and intraoperative surgical complications), were considered (Table 2and Table 3). [5,15,17]Time to extubation was defined as the time from arrival in the ICU until (first) removal of the endotracheal tube. Total postoperative LOS was substituted for time to extubation in patients not extubated (i.e., death or tracheostomy) as long as they met the inclusion criteria of survival past the 10-h cut point used for the logistic regression (see Modeling of Time to Extubation). Although not a primary endpoint of this analysis, postoperative complications common to cardiac surgery and LOS (preoperative, intensive care, intermediate care and postoperative ward), commonly used as markers of resource utilization, were also considered. [17,18] 

Various archival sources were used for data collection, including the medical record (discharge summaries, anesthesia records, nursing and physician progress notes, perfusion records), the DVA hospital computer system database (demographic data, discharge summaries, operative dictations, laboratory data), the CICSP database, specialized software and reports developed by the hospital's Information Resources Management Service for LOS in each hospital ward, and the National Beneficiary Indicator Records Locator System for 6-month mortality data. [15] 

To determine the appropriate statistical method to correlate time to tracheal extubation with the selected risk and intraoperative clinical process variables, the frequency distribution of time to extubation was examined. Given that we rarely extubate patients after midnight or before 5:00 A.M. because nurse staffing ratios are low then and fewer residents are available, the distribution is bimodal with a nadir in the frequency of extubation occurring 10 - 12 h after operation. This distribution is not amenable to Gaussian transformation required for analysis of time to extubation as a continuous variable. Similarly, the relation between time to extubation and the resource utilization outcome (i.e., time of transfer out of the ICU) is not continuous in our center given that transfer out of the ICU occurs at relatively set times of the day centered around nursing shift changes. Although continuous regression techniques may yield more accurate coefficient estimates (particularly for physiologic variables), [20]we chose dichotomous logistic regression because it is a widely used, conservative approach for this type of distribution.

Our choice of a 10-h rather than a 12-h cut point for modeling in the logistic regression model was based on the trend in our current clinical practice toward even earlier tracheal extubation times than used in this cohort, anecdotal clinical observations of similar trends at other centers around the country, and a median time to extubation in the total cohort closer to 10 than to 12 h (10.6 h).

Study variables were initially evaluated by univariate screening. Categorical variables were tested using a chi-squared test or, when appropriate, Fisher's exact test. Normally distributed continuous variables were compared using a Student's t test; continuous variables not normally distributed were compared nonparametrically using the Wilcoxon rank sum test. Variables with a univariate significance level <or= to 0.20 were considered for inclusion in the multivariate model.

An initial model (model R) was constructed considering preoperative risk factors alone, using stepwise logistic regression with entry and exit criteria for variables of P < 0.05. Clock time of arrival into the ICU was included in the initial model to control for the lack of extubation from midnight to 5 A.M. To adjust for temporal trends, the number of days from the institution of the FT protocol for that particular patient was also included. The type of surgery (CABG only vs. valve or valve - CABG) was included to adjust for the different underlying risks of these procedures. Variables identified in the first model were included in a second model (model R-P) to determine which intraoperative clinical process variables were significant in the presence of the risk factors identified in model R. The type of opioid used for anesthesia (fentanyl vs. sufentanil) was included to interpret the relative risk associated with varying opioid dosage. Quadratic functions of the continuous variables meeting the univariate screening criteria were included in the multivariate model to relax the assumption of constant risk. However, none was significant.

The Wald chi-squared test statistic was used to determine the significance of the study variables in the multivariate model. For significant variables, odds ratios and corresponding 95% confidence intervals were obtained. An odds ratio <1.0 indicates a significant association between the variable and extubation <or= to 10 h (early). An odds ratio >1.0 indicates a significant association between the variable and extubation >10 h (late). Model discrimination was assessed using the c-index, which corresponds to the area under the receiver operating characteristic curve relating the sensitivity and specificity of the model in predicting the outcome studied. [21]The maximal value of 1.0 represents perfect discrimination, whereas a value of 0.5 reflects chance discrimination only. Model calibration or “goodness of fit” was assessed using the Hosmer-Lemeshow statistic, which has a chi-squared distribution and tests the null hypothesis that there is no difference in the average and predicted outcome rate across subgroups of patients (assigned by deciles of expected risk). A lower Hosmer-Lemeshow value (with a larger P value) is associated with better model calibration. [22]The likelihood ratio test (a chi-squared test calculated as twice the difference of the log likelihoods between the model with and without the effect being tested) was used to determine the statistical significance of adding the intraoperative clinical process variables to the preoperative risk model. [23]Data were analyzed using SAS Version 6.10 (Cary, NC).

The median duration of intubation for the entire study population was 10.6 h. One hundred forty-six patients (48.3%) were extubated within 10 h after surgery.

(Table 1) shows preoperative risk factors. Variables significant on univariate screening entered into the first logistic regression analysis (model R) included age, hematocrit, surgical priority, preoperative intra-aortic balloon pump (IABP), angina class, left main coronary stenosis >or= to 70%, left ventricular contraction score, cardiomegaly, diuretic use, beta blocker use, intravenous nitroglycerin, chronic obstructive pulmonary disease, bronchodilator use, creatinine >or= to 2 mg/dl, cerebrovascular disease, peripheral vascular disease, and the CICSP risk estimate. Clock time of arrival into the ICU and days since institution of the FT protocol were also entered. Although statistically significant on univariate analysis, forced expiratory volume in one second was not considered in the multivariate analysis because of a large proportion of missing values (13.2%).

(Table 2) shows “general” intraoperative clinical process variables. Variables significant on univariate screening that were entered into the second logistic regression analysis (model R-P) included use of an arterial graft, post-CPB use of nitroglycerin, pre-CPB inotrope use, major intraoperative inotrope use, IABP for weaning, intraoperative transfusion of fresh frozen plasma, homologous erythrocytes or platelets, total urine output, duration of CPB and ischemic time, total operating room time, and serum glucose level on arrival in the surgical ICU. Clock time of arrival into the ICU was also entered. Although univariately significant, body temperature at the conclusion of surgery was not considered because a large proportion (9.9%) of values were missing.

(Table 3) shows FT “specific” clinical processes of care. Variables entered into the second logistic regression analysis (model R-P) included total intraoperative crystalloid volume, volatile agent use, type of opioid, doses of fentanyl, sufentanil, and midazolam, percentage of total CPB time isoflurane was used, and use of the mixed venous oxygen saturation pulmonary artery catheter.

Multivariate associations are presented in Table 4and Table 5along with the corresponding odds ratios and confidence intervals. Significant variables in model R (Table 4) included age and preoperative IABP along with type of surgery (CABG alone vs. valve or valve - CABG) and the clock time of arrival into the ICU. In model R-P (Table 5), of the initial risk variables, age and the clock time into the ICU remained significant. Significant intraoperative clinical process variables included use of an arterial graft, major inotrope use, platelet transfusion, and the dose of fentanyl or sufentanil. Of these variables, those related to early extubation (odds ratio <1.00) were use of an arterial graft, and those associated with late extubation (odds ratio >1.00) were major inotrope use, platelet transfusion, dose of opioid, and time of arrival in the ICU. Of the continuous variables, increasing doses of either sufentanil or fentanyl and later time of arrival into the ICU were associated with later tracheal extubation.

Model performance, assessed by the c-index (model R, 0.764 vs. model R-P, 0.853) and the Hosmer-Lemeshow statistic P value (model R, 0.463 vs. model R-P, 0.782), was superior in model R-P. The likelihood ratio test between model R and model R-P (P < 0.001) indicated a significant benefit of adding intraoperative clinical process variables to the baseline risk model.

(Table 6) shows patient outcomes. Only one patient extubated early required emergent reintubation (0.68%). The frequency of late reintubation for medical reasons (i.e., cardiac, respiratory failure, sepsis, and other conditions not temporally related to the initial extubation) was lower in the early extubation group (2.1% vs. 7.7%, P < 0.024). For each of the major postoperative complications, there were trends toward lower frequencies in the early extubation group, although statistical significance was reached only for adult respiratory distress syndrome (0.0% vs. 3.9%, P < 0.03). Similar patterns were present for minor postoperative complications, with significance reached only in the frequency of nosocomial pneumonia (3.4% vs. 11%, P < 0.012). The frequency of additional postoperative surgical procedures (3.4% vs. 10.3%, P < 0.05), 30-day (0.0 vs. 7.7%, P < 0.001) and 6-month mortality (1.4% vs. 10.9%, P < .001) were also significantly lower. There were no documented episodes of recall of intraoperative events in either group.

Preoperative LOS was not significantly different between groups. The lengthy preoperative LOS is due to a predominance of patients who underwent cardiac catheterization on the same admission (150 of 302, or 49.7%), most of whom were admitted with unstable angina. Statistically significant differences were noted in postoperative LOS in both monitored surgical ICU - intermediate care units and unmonitored ward settings, with the greatest difference in the monitored settings (33% for the surgical ICU alone, 25% for surgical ICU - stepdown unit, 19% for total postoperative stay, P < 0.001 for each). The median postoperative day of discharge was significantly shorter in the early extubation group (day 6 vs. 8, P < 0.001).

Nearly half of consecutive, unselected patients undergoing primary or redo CABG, valve surgery, or both in a DVA medical center managed on an FT clinical pathway were extubated within 10 h of surgery. Previously we found a reduction in time to extubation and LOS without significant change in morbidity or mortality rates in sequential consecutive cohorts, before and after institution of our FT pathway. [15]Considering the FT cohort from that report, the present analysis suggests that with the exception of patient age and preoperative use of IABP, intraoperative clinical “process” variables (i.e., “acts of care”) are more highly associated with the timing of postoperative extubation than are preoperative risk factors alone. [16,17] 

Our observations, if validated in a prospective study, may facilitate development of clinical prediction models that would allow clinicians to triage “high risk” patients (based on preoperative factors) to early extubation if the clinical process variables indicate a “low risk” or “successful” operation. This modeling approach may better facilitate internal benchmarking, comparison of results from different institutions, or both. Because our approach uses intraoperative clinical process variables assessed until the time when patients leave the operating room, preoperative prediction of the timing of extubation is not the ultimate goal. However, it emphasizes the importance of selected variables related to patient selection, surgical and anesthetic management, and overall system efficiency in FT management. Future studies could target specific factors that can be modified to maximize the percentage of patients that have early tracheal extubation.

Of the preoperative risk factors univariately screened, only age and preoperative IABP were independently associated with later extubation. Time of arrival in the ICU was also significant. This factor, clearly not a preoperative risk factor, was required in the statistical analysis to “control” for the bimodal distribution of time to extubation caused by a lack of extubation in the late night or early morning hours.

The adverse effects of increasing age on outcome and resource use in patients undergoing cardiac surgery are documented. [18,24]Increasing age not only has adverse effects on pharmacokinetics and pharmacodynamics of anesthetic agents [25]but also increases the risk of adverse central nervous system effects of CPB. [26]Recently, Ott et al. [9]showed that elderly patients (167 patients >or= to 70 yr old) may be safely managed on an “aggressive” FT protocol. However, only 19% of the elderly were discharged before the fifth postoperative day, in contrast to 48% of the younger cohort.

The adverse association of preoperative IABP was expected given that this was one of the few exclusion criteria followed by most clinicians at this center. However, two patients with preoperative IABP were extubated early with good results. The primary concerns in these patients, aside from potential hemodynamic instability, are the conflicting nursing considerations of having the patient in a sitting position (advantageous for weaning and extubation) rather than supine (advantageous for IABP management and removal). Although most centers share these same concerns, Ott et al. [9]recently presented data from an FT cohort in which approximately 50% of patients had preoperative IABPs placed. It is implied that most of these patients were extubated 4 - 8 h after operation.

Time of arrival in the ICU was significant, with later time of arrival successively decreasing the probability of early tracheal extubation. This is likely related to concerns regarding safety of extubation in the late evening or early morning hours. Our current experience is that this effect is less prominent. However, we believe it particularly important in the developmental and earlier history of an FT program. It is also one that can be emphasized in quality improvement efforts.

Use of an arterial graft (in this study a single left internal mammary artery was used predominantly) was associated with early tracheal extubation. The beneficial effects of this artery as a bypass conduit on operative mortality risk and long-term patient outcome are documented. [27]However, internal mammary artery dissection has been associated with postoperative decline in expiratory flow rates and lung volumes and with increased postoperative pain. [28,29] 

We cannot explain the beneficial association of arterial grafting that we observed. Neither Arom et al. [6]nor Habib et al. [8]included this variable in their analyses. In contrast, all patients in the randomized study of Cheng et al. [11]had internal mammary artery grafts. It is possible that our surgeons felt that internal mammary artery grafting was contraindicated in the patients at highest risk. [27]It is also possible that patients receiving an internal mammary artery graft had a greater increase in early post-CPB cardiac function, facilitating earlier extubation.

Several variables were independently associated with later extubation. Of the preoperative risk variables, age and time of arrival in the ICU remained statistically significant in model R-P.

The adverse association of major inotrope use (defined as the use of two or more catecholamines or a phosphodiesterase inhibitor inotrope for at least 30 min) is likely related to the greater degree of hemodynamic instability and poor cardiac function in these patients. Potential statistical interaction of this variable with preoperative ejection fraction and post-CPB IABP may account for the lack of significance of these variables.

An adverse association with intraoperative platelet transfusion, rather than homologous erythrocyte or fresh frozen plasma transfusion, was observed. Longer CPB times are generally considered to induce a greater degree of coagulopathy due to platelet activation and degranulation, although the precise cause is controversial. [30]Thus platelet transfusion could be more closely associated with total CPB duration than either erythrocyte or fresh frozen plasma requirements. It is also possible that platelet transfusion has deleterious physiologic effects on postoperative pulmonary function. A limitation of our analysis was lack of consideration of additional blood products transfused after operation.

The type of surgery was not significant, suggesting that patients undergoing valve repair or replacement or combined procedures were not at increased risk for later tracheal extubation. Although the percentage of valve patients in this study was small (11% of < 10 h and 15.4% of >or= to 10 h), this finding is of interest, particularly because similar data have been presented in only one other study. [10]In the DVA, valve surgery has been associated with a twofold increase in complications related to the timing of extubation (mechanical ventilation >48 h, low cardiac output syndrome, and reoperation for bleeding) compared with CABG alone. [18]A more recent private sector study suggests improved outcomes in the past decade. [24] 

Our data related to anesthetic technique must be interpreted cautiously given a nonrandomized, retrospective study design. The odds ratio for the use of fentanyl (3.41) suggests an adverse effect of this opioid, although a borderline P value and wide confidence interval (the lower boundary crossing 1.0) suggests inadequate statistical power. In patients who were hemodynamically unstable immediately after CPB, it is likely that a decision not to extubate early was made in the operating room and additional doses of opioid were administered specifically for postoperative sedation.

Early extubation was associated with shorter postoperative LOS, particularly in the intensive care and intermediate care units. We could not determine actual cost or charge differences between groups because these data are not widely available in the DVA system. Furthermore, inefficiencies in dynamic reallocation of resources (i.e., nursing and ancillary staff ratios in a particular clinical unit) in response to changes in patient volume complicate the relation between LOS and costs of health in surgical settings. [13,31]Private sector studies have documented monetary savings associated with early extubation, primarily by reductions in direct variable components of cost. [5–7,13] 

There were trends toward lower postoperative morbidity (minor or major) in patients who were extubated early, including a reduction in the frequency of nosocomial pneumonia. In our previous report, considering all FT patients together, we noted a reduction in nosocomial pneumonia compared with our pre-FT cohort. [15]However, as noted in that study, limitations of retrospective analysis in coding this variable are substantial. The frequency of several “minor” complications (i.e., need for supplemental nasal oxygen, arrhythmias, and so forth) was not different between early and late extubated patients.

The observed difference in both 30-day and 6-month mortality between early and late extubated patients was striking. However, our study has inadequate statistical power to determine the association of early extubation with mortality risk. Furthermore, it is unlikely that early extubation has any direct causal role in influencing the mortality risk in most patients.

Limitations of our study include retrospective data collection, use of a single center, a predominantly male population, limited sample size, possible confounding effects of steroid administration, and temporally associated changes in other processes of care unrelated to our FT pathway. Several of these have been addressed previously. [15]Data from a single center are subject to unmeasured, unreported, or chance processes and structures of care (who provides the care) unique to that center. [16]Our status as a university-affiliated institution with residents providing care may differentiate us from nonaffiliated DVA medical centers and private sector institutions.

Our models were developed using a relatively small sample size. Although our ratio of outcome events to significant variables is greater than 10, above the threshold considered acceptable to avoid model overfitting [32]and our goodness-of-fit indices suggest that model R-P has high discriminative ability, [33]we do not recommend use of these data for clinical prediction because of the study design limitations already noted.

Use of perioperative steroids may have had an independent effect on early extubation. Steroid administration (usually methylprednisolone sodium succinate) was advocated in two influential early FT studies. [4,5]Purported benefits include reduction or blockade of inflammatory mediators generated by CPB implicated in the postoperative “systemic inflammatory response.”[34]Because our FT pathway included routine steroid administration, we cannot assess its impact.

Our models failed to consider physician-specific factors (structures of care), particularly qualifications, training, and volume of procedures performed. These factors influence the choice and quality of clinical processes performed. In addition, the degree of resident supervision by attending physicians may be important. These factors are being considered in a larger, prospective study. [35] 

Finally, time to extubation is a surrogate outcome among the primary outcomes of cardiac surgery: clinical safety, resource consumption (including LOS), patient satisfaction, long-term functioning, and so forth. Given our study design, we could not address these. However, because anesthesiologists are most involved in this portion of the FT, we believe our data are of interest.

Our findings are consistent with previously reported non-DVA studies of early extubation. [4–8,10,11,13]The percentage of patients extubated early (within a 8 - 12 h time frame) and the mean time to extubation are comparable, although comparison is complicated by varying definitions for preoperative risk variables and exclusion of valve procedures. Our study had a distinctly higher incidence of chronic obstructive pulmonary disease and a near absence of female patients. Female sex may impart additional operative risk in CABG surgery [36]and was an adverse risk factor for early extubation univariately in one study and multivariately in another. [6,8]Increasing age and transfusion of blood products, factors we found significant, were also identified in other studies. [6,8]Although no study has conclusively modeled clinical safety, outcomes are relatively similar. Although LOS in monitored settings was shorter in other studies, our total postoperative LOS is similar.

Our results suggest that the conduct of the surgical procedure as measured by intraoperative clinical process variables (including markers of inotrope use, transfusion, opioid dose, and use of an arterial graft for revascularization) adds significantly to information obtained from preoperative risk factors alone (age and preoperative IABP) in the timing of postoperative extubation. Some of these factors are likely markers of intraoperative problems (i.e., multiple inotropes, need for transfusion), whereas opioid dosages may indicate inefficient management or, more likely, a response to decision making against early extubation in the operating room. The “protective” effect of use of an arterial graft is unclear and is possibly related to preoperative risk factors, although a “better” cardiac response to revascularization cannot be discounted. Valve surgery (with or without CABG) was not an adverse factor, although the number of patients is small. Later times of arrival in the ICU were associated with longer times to tracheal extubation. Given our study's limitations, we cannot determine causal relations between these variables and timing of extubation, although we have identified factors worthy of further study (such as modifiable factors that could result in quality improvement). Early extubation in this cohort appeared to be well tolerated, although a larger, prospective study is required to establish its clinical safety and relation to clinical outcomes and lengths of stay.

The authors thank Sharon Del Hotal for secretarial assistance and Charles P. Gibbs, M.D., Chairman, Department of Anesthesiology, University of Colorado Health Sciences Center, for support.

[dagger, dagger] The intraoperative methylprednisolone sodium succinate dose was increased from 100 to 500 mg given intravenously in June 1994; postoperative administration (125 mg given intravenously every 8 h for three doses) was sporadic until April 1994.