Temporal Onset, Risk Factors, and Outcomes Associated With Stroke After Coronary Artery Bypass Grafting FREE

Stroke is a devastating complication after coronary artery bypass graft (CABG) surgery, occurring in 1.3% of US Medicare beneficiaries each year.¹ Because CABG increasingly is being reserved for elderly patients with extensive coronary disease and comorbid conditions, prevalence of stroke after CABG is likely to remain substantial. Many studies have identified patient factors associated with post-CABG stroke^2- 5; however, information about timing of perioperative stroke and influence of different surgical techniques remains limited.^4,6 Some patients develop stroke intraoperatively, with neurologic deficit apparent immediately after awakening from anesthesia; others develop it postoperatively, after initially awakening without neurologic deficit. Some recent studies have observed decreased occurrence of stroke when CABG is performed without cardiopulmonary bypass.^7- 8

The purposes of this study were to examine prevalence and timing of perioperative stroke, identify patient and surgical factors associated with stroke and its timing, and assess association of stroke with hospital mortality and morbidity and long-term survival.

From 1982 to 2010, 45 432 consecutive patients underwent isolated primary or reoperative CABG surgery at Cleveland Clinic (Table 1 and eFigure 1). Patients who underwent concomitant cardiovascular procedures (valve replacement, repair, aneurysmectomy, carotid endarterectomy) were excluded. Preoperative, operative, and postoperative variables were retrieved from the Cardiovascular Information Registry, a database populated concurrently with patient care. It has been approved for use in research by the Cleveland Clinic institutional review board, with patient consent waived. During the study time frame, details of in-hospital postoperative neurologic events were recorded in standardized fashion. Medical records and data were reviewed by one author (K.G.T.) for all patients recorded as having experienced an adverse neurologic event.

Perioperative strokes were diagnosed by physicians involved in the daily care of patients and confirmed by staff neurologists and imaging (computed tomography or magnetic resonance imaging). Perioperative stroke was defined as any new focal or global neurologic deficit lasting more than 24 hours that could not be explained by another medical process. This included persistent neurologic deficit for more than 72 hours and reversible ischemic neurologic deficit with recovery within 72 hours; it excluded transient ischemic attack with recovery within 24 hours. Temporal onset of stroke was classified as intraoperative if the neurologic deficit was present when the patient awoke from anesthesia and as postoperative if the deficit developed after the patient emerged from anesthesia. Some patients underwent 2 operations on the same day; in that case, timing of stroke was determined with respect to the second operation unless a stroke was already manifest after the first.

Four different operative strategies were used to perform CABG: cardiopulmonary bypass (on-pump) with arrested heart; on-pump with beating heart; on-pump with hypothermic circulatory arrest; and off-pump (eFigure 2). For most of this series, on-pump arrested-heart surgery was the standard strategy. From 1999 to 2004, off-pump CABG was performed routinely by some surgeons. However, throughout the series a surgical revascularization strategy was used to lower risk of atheroembolism in patients with ascending aortic arteriosclerosis; this included on-pump beating-heart CABG, on-pump CABG with hypothermic circulatory arrest, and off-pump CABG. Additional details about these techniques are included in eAppendix 1.

End points included hospital mortality and morbidity, resource utilization reflected in intensive care unit and postoperative length of stay after revascularization, and long-term mortality.

Hospital morbidities (perioperative myocardial infarction, respiratory failure, sepsis, and renal failure) were recorded prospectively, according to definitions of the Society of Thoracic Surgeons National Adult Cardiac Database.⁹ It was often not possible to differentiate complications as definitely preceding the perioperative stroke or following it; therefore, these postoperative morbidities were considered irrespective of their timing.

In addition, new-onset postoperative atrial fibrillation was considered a time-varying covariable for postoperative stroke. Among patients experiencing a postoperative stroke, documentation of its timing (which was prospectively recorded from continuous postoperative rhythm monitoring throughout the postoperative hospital course) was reviewed by one author (K.G.T.) to identify its occurrence prior to postoperative stroke.

For long-term mortality, vital status was obtained by query of the Social Security Death Master File.^10- 11 Common closing date for survival analysis was February 26, 2010 (6 months prior to its query to account for delays in posting). The interval from stroke—or date of operation in patients without a stroke—to death or follow-up was calculated for each patient. Nonparametric survival estimates were obtained by the Kaplan-Meier method and hazard functions (instantaneous risk of death) from multiphase parametric modeling.¹² Briefly, the latter method was devised to account for nonproportional hazards often present after an intervention. It consists of additive hazard phases, each with a highly flexible shaping function characteristic of early peaking, constant, and late-peaking hazard phases, as characterizes human populations. Each data-driven phase is modulated by risk factors. Because of the method's mathematical properties (orthogonality), the same set of risk factors can be considered simultaneously in each hazard phase. Shaping parameters for each of these phases are provided in eAppendix 2. (For additional details, see http://my.clevelandclinic.org/professionals/software/hazard/default.aspx.)

Analysis of perioperative stroke was performed separately for intraoperative and postoperative stroke. In theory, the multiphase temporal decomposition model used for postoperative stroke could have been used to assess all strokes simultaneously. However, intraoperative stroke represents interval-censored data occurring at an unknown time between anesthesia induction and awakening. Because awakening time was not reliably recorded, separate analyses were performed for intraoperative stroke (non–time-related logistic regression analysis) and postoperative stroke (time-related multiphase parametric modeling). This separate-analyses strategy is a type of time-segmented analysis.¹³

All analyses were performed using SAS version 9.1 (SAS Inc, Cary, North Carolina). P ≤ . 05 (2-sided) is used throughout as indicating statistical significance (with at least 50% reliability by bagging in multivariable analyses).

Risk Factors for Intraoperative Stroke. Multivariable logistic regression was used to identify factors (eAppendix 3) associated with intraoperative stroke. Bagging was used for variable selection.¹⁴ Briefly, for this machine-learning technique, 250 data sets of size 45 432 were formed by random sampling of the original data set with replacement (bootstrap samples); on average, about two-thirds of patients are present in each sample, and the remaining are duplicates. Automated forward stepwise regression with backward elimination was performed for each bootstrap data set, with an entry criterion of P ≤ .15 (2-sided) and retention criterion of P ≤ .05. These results were stored. Frequency of occurrence of variables across all models constituted the aggregation step. In addition, some variables were clustered (such as linearizing transformation of scale of continuous variables formulated in preparatory data analysis) and counted as a single concept. Variables or clusters of variables that entered more than 50% of the 250 models (median rule) were chosen for the final model.

Targeted interaction analyses, including age and type of surgical support technique, were performed to identify possible modulation of main effects associated with intraoperative stroke.

Some data were not routinely assessed, recorded, or abstracted for blocks of years. Thus, we analyzed the data initially in blocks of years. Missing-value indicator variables were then created to manage these variables as interaction terms in the final comprehensive analyses. Otherwise, missing values were so sparse that simple-means imputation was used.

Risk Factors for Postoperative Stroke. Patients experiencing an intraoperative stroke were excluded from this analysis. For the remaining patients, the interval from end of surgery to documented occurrence of postoperative stroke was calculated from medical record review. If a patient did not experience a postoperative stroke, this interval was censored at time of hospital discharge or hospital death. Nonparametric estimates of freedom from postoperative stroke were obtained by the Kaplan-Meier method and instantaneous risk by multiphase parametric hazard modeling.¹² This method yielded information about the time-varying instantaneous risk of stroke; identified 2 phases of postoperative stroke risk, an early peaking phase and a constant-hazard phase; and permitted identification of different risk factors for these 2 phases (so-called nonproportional hazards). Bagging (as described earlier) was used to identify factors (eAppendix 3) associated with postoperative stroke simultaneously for each hazard phase.

Both hospital death and hospital discharge are competing risks for postoperative stroke. Among the 45 136 patients who did not have an intraoperative stroke, 713 died in hospital without a stroke (1.6%); 285 deaths (40%) occurred within 5 days, the point at which 95% of postoperative strokes had occurred. Although large in number, we considered this to be a sufficiently small proportion of the total sample to ignore.

Propensity Matching. To assess association of stroke with hospital outcomes and time-related survival, propensity matching was used to assemble groups of patients having similar characteristics with and without perioperative stroke. Initially, a parsimonious model was developed using multivariable logistic regression and bagging, as previously described (eAppendix 3). Then, clinically relevant variables not found to be significantly associated with perioperative stroke were added to the parsimonious model to form a variable-rich propensity model (54 variables, C = 0.73) (eAppendix 3). Using this semisaturated model, a propensity score (probability of stroke) was calculated for each patient. Greedy matching was used to identify patients from the no-stroke group to form a 1:4 propensity-matched control group, well matched according to standardized differences (eFigure 3).¹⁵ All patients with stroke were matched. The matched pairs spanned the spectrum of CABG (eFigure 4).

Hospital outcomes were compared using the χ² test for categorical variables, Cochran-Mantel-Haenszel test for trend for ordinal variables, and Wilcoxon rank-sum test for length-of-stay (skewed) variables. Comparisons were between all patients experiencing and not experiencing a perioperative stroke (unadjusted comparison) and between propensity-matched patients experiencing and not experiencing stroke (adjusted comparison). Unadjusted and propensity-matched survival was similarly compared using the log-rank test.

Presentation. Continuous variables are summarized by mean (SD) and categorical variables by frequency, percentage, and, for outcomes, 95% asymmetric binomial confidence intervals (CIs). For consistency, CIs used throughout are 95%, equivalent to ±2 SEs. All P values are 2-sided.

Among 45 432 patients, 705 (1.6% [95% CI, 1.4%-1.7%]) experienced a stroke. Occurrence of stroke peaked in 1988 at 2.6% (95% CI, 1.9%-3.4%), then slowly declined at 4.69% (95% CI, 4.68%-4.70%) per year (P = .04) (Figure 1), despite increasing patient risk profile, such as higher prevalence of preoperative stroke (eFigure 5), hypertension, and diabetes (eFigure 6), as reflected in Society of Thoracic Surgeons risk scores ranging from 1.4 in 2000 to 1.8 in 2008. Of patients experiencing stroke, intraoperative stroke occurred in 279 (40%) and postoperative stroke in 409 (58%); timing was indeterminate in 17 (2.4%).

Instantaneous risk of postoperative stroke peaked 40 hours after surgery, and by day 6 it had decreased to a constant hazard of 0.055%/d (95% CI, 0.047%–0.065%) (Figure 2). The hazard function comprised 2 phases: an early peaking hazard phase and a constant-hazard phase.

Risk factors common to both intraoperative and postoperative stroke were older age, smaller body surface area, previous stroke, preoperative atrial fibrillation, and on-pump CABG with hypothermic circulatory arrest (Table 2 and Table 3, eTable 1). Overall risk of intraoperative stroke was 0.35% (95% CI, 0.27%-0.44% [59/16 852]) for patients younger than 60 years, 0.70% (95% CI, 0.59%-0.84% [121/17 167]) for those between 60 and 70 years, and 0.87% (95% CI, 0.71%-1.1% [99/11 396]) for those 70 years and older (Figure 3). Overall risk of postoperative stroke was 0.49% (95% CI, 0.39%-0.61% [83/16 793]) for those younger than 60 years, 0.93% (95% CI, 0.79%-1.1% [158/17 046]) for those between 60 and 70 years, and 1.5% (95% CI, 1.3%-1.7% [168/11 297]) for those 70 years and older (eFigure 7). As number of arteriosclerotic comorbid conditions increased, stroke risk increased (eFigure 8).

Intraoperative Stroke. Specific additional risk factors for intraoperative stroke were peripheral and carotid arterial disease, previous cardiac operation, worse preoperative clinical condition, left ventricular dysfunction, circumflex stenosis greater than 70%, and on-pump arrested-heart CABG (as well as on-pump CABG with hypothermic circulatory arrest [Table 2, eTable 1]). Different surgical techniques were associated with different risks of intraoperative stroke. Unadjusted rates of stroke were highest among patients who had on-pump CABG with hypothermic circulatory arrest (5.3% [95% CI, 2.0%-11%]) and lowest among those who had off-pump CABG (0.14% [95% CI, 0.029%-0.40%]) and on-pump beating-heart CABG (0% [95% CI, 0%-1.6%]). Risk of intraoperative stroke was intermediate for those undergoing on-pump arrested-heart CABG (0.50% [95% CI, 0.41%-0.61%]).

Although demographics of patients differed among these surgical technique groups, particularly with respect to older age in the groups undergoing on-pump CABG with hypothermic circulatory arrest and on-pump beating-heart CABG (eTable 2), these differences in intraoperative stroke risk according to surgical technique remained statistically significant in multivariable analysis (Table 2, eTable 1). In targeted analysis, intraoperative stroke risk increased with advancing age when either on-pump arrested-heart CABG or on-pump CABG with hypothermic circulatory arrest was used (on-pump arrested-heart: 0.32%, 0.60%, and 0.58% for patients aged ≤60, 60-70, and >70 years, respectively; on-pump with circulatory arrest: 0%, 3.0%, and 8.2% for patients aged ≤60, 60-70, and >70 years, respectively) but not when either off-pump CABG or on-pump beating-heart CABG was used (off-pump: 0.15%, 0.32%, and 0% for patients aged ≤60, 60-70, and >70 years, respectively; on-pump beating-heart: 0% at all ages) (eTable 3, eTable 4, eFigure 9).

Postoperative Stroke. An additional factor specific to the early hazard phase for postoperative stroke that appeared to lower the risk of such strokes was new-onset postoperative atrial fibrillation. That is, occurrence of new-onset postoperative atrial fibrillation was not associated with increased risk of postoperative spoke. Additional risk factors specific to the constant-hazard phase of postoperative stroke included left main stenosis, treated diabetes, less use of internal thoracic artery grafts, and earlier date of operation (Table 3, eTable 1).

Patients who experienced a stroke had substantially worse hospital outcomes, even after propensity adjustment for preoperative factors: 19% mortality (95% CI, 16%-22% [136/705]) vs 3.7% (95% CI, 3.1%-4.5% [105/2820]), P < .001; 44% prolonged ventilation (95% CI, 32%-57% [31/70]) vs 15% (95% CI, 12%-20% [53/343]), P < .001; and 13% renal failure (95% CI, 11%-16% [92/705]) vs 4.3% (95% CI, 3.6%-5.1% [122/2820]), P < .001 (eTable 5). They also experienced substantially longer intensive care unit (median, 120 vs 48 hours; P < .001) and postoperative (median, 14 vs 7 days; P < .001) lengths of stay (eTable 5).

Long-term survival was assessed from 469 644 patient-years of follow-up, with mean follow-up among survivors of 11 (SD, 8.6) years and 10% of patients undergoing follow-up for more than 24 years. Patients experiencing a perioperative stroke had an unadjusted survival of 70% (95% CI, 69%-72%), 37% (95% CI, 35%-39%), and 12% (95% CI, 11%-14%) at 1, 10, and 20 years, respectively, compared with 95.3% (95% CI, 95.2%-95.4%), 68% (95% CI, 67%-68%), and 35% (95% CI, 34%-35%) among those not experiencing a perioperative stroke (log-rank P < .001) (Figure 4). Propensity-matched survival among patients not experiencing a perioperative stroke was 91.9% (95% CI, 91.4%-92.4%), 54% (95% CI, 53%-55%), and 22% (95% CI, 21%-22%) at the same time intervals (log-rank P < .001) (eFigure 10A). This was mainly attributable to a higher, more prolonged early risk of death to about 3 months (eFigure 10B); patients with stroke surviving beyond this point had a risk of death only somewhat higher than those without stroke (hazard, 7.2%/y [95% CI, 6.6%-8.0%], 9.1%/y [95% CI, 8.5%-9.7%], and 12%/y [95% CI, 11%-14%] at 1, 10 and 20 years, respectively, vs 3.3%/y [95% CI, 3.1%-3.5%], 7.8%/y [95% CI, 7.6%-8.0%], and 10%/y [95% CI, 10%-11%], respectively).

Stroke remains one of the most devastating and disabling complications of CABG surgery, with great clinical and economic implications for patients and the health care system.^4,16 These implications are expected to remain substantial, despite improving surgical and anesthetic techniques, because of the increasing risk profile of patients undergoing CABG. Previous attempts to identify the subset of patients at increased risk of stroke have focused on identifying clinical risk factors, whereas few studies have examined timing of stroke after CABG.^3,16- 17 Understanding the risk factors specific to the timing of postoperative stroke should be beneficial in identifying the cause of the stroke and developing preoperative, operative, and postoperative strategies to predict and prevent stroke.

This study demonstrated that stroke more commonly occurs postoperatively than intraoperatively; identified common as well as specific risk factors for strokes occurring intraoperatively, early postoperatively, and late postoperatively; compared not only the risk of stroke associated with off-pump CABG vs on-pump arrested-heart CABG, but with 4 different surgical strategies; noted that new-onset postoperative atrial fibrillation is not associated with increased risk of postoperative stroke; and observed that stroke increases early risk of morbidity, mortality, and resource utilization overall and among propensity-matched patients but that survivors of stroke have a late survival similar to that of patients not experiencing a stroke.

In this study, prevalence of stroke was 1.6% among patients undergoing CABG, lower than the 3% reported by Roach et al.¹⁶ However, over the last 3 decades, occurrence of stroke has decreased despite increasing patient risk profile. This is most likely the result of improving preoperative assessment, intraoperative anesthetic and surgical techniques, and postoperative care.

Contrary to the report by Salazar et al,¹⁸ we found, as have some other investigators,^4,16- 17 that most strokes occurred postoperatively rather than intraoperatively. Instantaneous risk of postoperative stroke peaked around postoperative day 2 and by day 5 had decreased to a constant hazard of about 0.055%/d.

Intraoperative strokes may result from emboli, thrombosis, or hypoperfusion. Previous studies have identified arteriosclerotic emboli from intraoperative manipulation of the aorta as the main culprit.^19- 20 We similarly identified factors associated with arteriosclerotic burden as major risk factors for intraoperative stroke. Preoperative and intraoperative imaging (eg, computed tomography, intraoperative transesophageal echocardiography, and epiaortic echocardiography) may be used to assess ascending aortic arteriosclerosis.^21- 23 If it is found, surgical techniques could be used that avoid or minimize aortic manipulation to eliminate risk of arteriosclerotic embolization. These include alternative cannulation sites, such as the axillary artery,^24- 26 off-pump CABG, or on-pump beating-heart CABG with minimal or no aortic manipulation.

In addition to atherosclerotic embolus, intraoperative hypoperfusion is a potential cause of stroke during CABG and highlights the importance of maintaining adequate blood pressure during both cardiopulmonary bypass and manipulation of the heart during off-pump CABG.^27- 28 Hypotension with resulting hypoperfusion may occur during retraction of the heart in off-pump CABG; hypotension also may occur at any time during on-pump surgery from vasodilatation.

The timing of postoperative stroke reflects 2 types of risk: a constant “background” risk that persists after surgery, reflecting arteriosclerosis in this population; and an unexplained peak risk around postoperative day 2. The inflammatory process and hypercoagulability after surgery might provide some explanation for this peak. Identifying the etiology of this postoperative risk factor for stroke may lead to better strategies to prevent it, whether through more aggressive use of antithrombotic and antiplatelet agents, prophylactic prevention of atrial fibrillation, or both. The constant background risk of stroke is demonstrated in this study to have decreased over time, and that risk contributed most to the downward trend of stroke over the last 20 years.

We found off-pump CABG and on-pump beating-heart CABG to be associated with the lowest risk of intraoperative stroke, on-pump arrested-heart CABG with slightly higher risk, and on-pump CABG with hypothermic systemic circulatory arrest with the greatest risk. Both off-pump CABG and on-pump beating-heart CABG can be performed with minimal aortic manipulation, and therefore they likely lower the risk of stroke by decreasing the risk of aortic arteriosclerotic embolization. Although these findings are nonrandomized, they suggest that off-pump and on-pump beating-heart CABG may reduce stroke risk. This potential benefit must be weighed against the greater risk of incomplete revascularization, lower graft patency, and worse 1-year outcomes reported for patients undergoing off-pump CABG.^29- 31 A surgical revascularization strategy should enable optimal surgical coronary revascularization while keeping the risk of stroke low. In patients at high risk of intraoperative stroke, such as the elderly or those with aortic arteriosclerosis, off-pump CABG or on-pump beating-heart CABG with no or minimal aortic manipulation may be best. However, in patients at low risk of stroke, such as those without aortic arteriosclerosis and minimal arteriosclerotic burden, on-pump CABG is likely the best option to provide optimal surgical revascularization and minimal risk of stroke. Future trials will be needed to investigate these proposed strategies.

Although atrial fibrillation continues to be a common postoperative problem with peak incidence on the second day,³² suggesting a possible mechanism for the early peaking hazard phase of postoperative stroke, an unanticipated finding was that there was no increased risk of stroke with new-onset postoperative atrial fibrillation. In fact, new-onset atrial fibrillation was associated with a lower risk of postoperative stroke. To treat new-onset atrial fibrillation, we initially try early medical conversion or electroconversion and, if atrial fibrillation recurs or is persistent, rate control and anticoagulation. This strategy appears to be associated with not only preventing an anticipated increased risk of postoperative stroke but perhaps with actually lowering the risk.

Patients experiencing a stroke after CABG also experienced substantially more hospital morbidity, mortality, and resource utilization than those not experiencing a stroke.¹⁶ However, late survival of patients who survived a stroke was similar.

This study has several strengths and limitations. Among the strengths are that we included operations performed over nearly 3 decades; data collection was concurrent, prospective, and detailed; and trends could be observed according to time of operation and surgical technique.

The study also has several limitations. First, the classification of the timing of stroke occurrence is arbitrary and sometimes challenging in the setting of anesthesia and postoperative delirium and other complications. Second, we were unable to determine whether each stroke was embolic, thrombotic, or hypoperfusive in etiology. Third, patients may experience competing risk of death prior to stroke. Fourth, our comparisons of the association of surgical technique and stroke risk were observational and may be confounded by unmeasured covariates. Fifth, these results represent findings at a single academic medical center and may not be generalizable to US practice.

Among patients undergoing CABG surgery at a single center during the past 30 years, the occurrence of stroke declined despite increasing patient risk profile, and more than half of strokes occurred postoperatively rather than intraoperatively. Clinical presentation and surgical technique were specific to intraoperative stroke, but age and arteriosclerotic burden were associated with both intraoperative and postoperative stroke. Further studies are needed to develop better strategies to minimize the occurrence of stroke among patients undergoing CABG.