Early Versus Delayed Stroke After Cardiac Surgery: A Systematic Review and Meta‐Analysis

Abstract

Background

Although it is traditionally regarded as a single entity, perioperative stroke comprises 2 separate phenomena (early/intraoperative and delayed/postoperative stroke). We aimed to systematically evaluate incidence, risk factors, and clinical outcome of early and delayed stroke after cardiac surgery.

Methods and Results

A systematic review (MEDLINE, EMBASE, Cochrane Library) was performed to identify all articles reporting early (on awakening from anesthesia) and delayed (after normal awakening from anesthesia) stroke after cardiac surgery. End points were pooled event rates of stroke and operative mortality and incident rate of late mortality. Thirty‐six articles were included (174 969 patients). The pooled event rate for early stroke was 0.98% (95% CI 0.79% to 1.23%) and was 0.93% for delayed stoke (95% CI 0.77% to 1.11%; P=0.68). The pooled event rate of operative mortality was 28.8% (95% CI 17.6% to 43.4%) for early and 17.9% (95% CI 14.0% to 22.7%) for delayed stroke, compared with 2.4% (95% CI 1.9% to 3.1%) for patients without stroke (P<0.001 for early versus delayed, and for perioperative stroke, early stroke, and delayed stroke versus no stroke). At a mean follow‐up of 8.25 years, the incident rate of late mortality was 11.7% (95% CI 7.5% to 18.3%) for early and 9.4% (95% CI 5.9% to 14.9%) for delayed stroke, compared with 3.4% (95% CI 2.4% to 4.8%) in patients with no stroke. Meta‐regression demonstrated that off‐pump was inversely associated with early stroke (β=−0.009, P=0.01), whereas previous stroke (β=0.02, P<0.001) was associated with delayed stroke.

Conclusions

Early and delayed stroke after cardiac surgery have different risk factors and impacts on operative mortality as well as on long‐term survival.

Clinical Perspective

What Is New?

• This is the first systematic review and meta‐analysis to examine the incidence of early and delayed stroke after cardiac operations.

What Are the Clinical Implications?

• Early and delayed stroke after cardiac surgery have different risk factors and impacts on operative mortality as well as on long‐term survival.

Perioperative stroke is a devastating complication after cardiac surgery, and the incidence has remained largely unchanged despite advances in surgical techniques.1 Data from administrative databases and observational registries suggest that the incidence of perioperative stroke after cardiac surgery ranges from 0.8% to 5.2%.2 In landmark trials comparing outcomes of coronary artery bypass graft (CABG) and percutaneous coronary intervention, including SYNTAX (SYNergy between percutaneous coronary intervention with TAXus and cardiac surgery) and FREEDOM (Future REvascularization Evaluation in patients with Diabetes Mellitus), the burden of stroke has been a major limitation for surgery.3, 4, 5

Stroke may occur intraoperatively (usually detected when patients initially awaken from anesthesia) or thereafter. The 2 types of strokes have different pathophysiologic mechanisms: early/intraoperative stroke occurs primarily from aortic manipulation and atheroembolism, whereas delayed/postoperative stroke is usually related to postoperative atrial fibrillation or cerebral vascular disease.6 The conceptual framework of early and delayed stroke is important because it facilitates implementation and evaluation of tailored preventative strategies for both. Greater understanding of the incidence, risk factors, and sequelae of early and delayed stroke will facilitate the continued improvement in the safety of surgical intervention.

Here we performed a systematic review and meta‐analysis to give an objective and weighted estimate of the incidence and risk of early and delayed stroke following cardiac surgery and their impact on operative and long‐term patient survival.

Methods

The data, analytic methods, and study materials will not be made available to other researchers for purposes of reproducing the results or replicating the procedure. This systematic review and meta‐analysis were performed according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) statement7 and the MOOSE (Meta‐Analysis of Observational Studies in Epidemiology) guidelines (Table S1).8

Search Strategy

A medical librarian (M.D.) performed comprehensive systematic searches to identify studies that evaluated perioperative stroke after cardiac surgery. Searches were run in April 2018 on the following databases: Ovid MEDLINE (In‐Process & Other Non‐Indexed Citations and Ovid MEDLINE 1946 to Present); Ovid EMBASE (1974 to present); and The Cochrane Library (Wiley, Hoboken, NJ). The search strategy included all appropriate controlled vocabulary and keywords for identified “cardiac surgical procedures” and “intra‐ and postoperative stroke.” Full details regarding the search strategy for Ovid MEDLINE are provided in Table S2.

Study Selection and Inclusion Criteria

Database searches were conducted and deduplicated by a qualified librarian (M.D.). Three preliminary reviewers screened the searched database for inclusion. A fourth independent reviewer confirmed the adequacy of studies based on predefined inclusion criteria for titles and abstracts. Inclusion criteria were full‐text English articles on adult patients who had undergone cardiac surgery and reported perioperative strokes and classified them as early or delayed. A full text of preliminary screened studies was then retrieved for a second round of eligibility screening. Reference lists of the included articles were also searched, and additional studies included (ie, backward snowballing). The full PRISMA flow chart outlining the study selection process is available in Figure S1. The Newcastle‐Ottawa Scale for quality assessment was used for the critical appraisal of included studies (Table S3).9 Studies with scores of 6 or more were included.

Clinical Outcomes/Definitions

The primary outcome was the rate of early and delayed stroke. Secondary outcomes were (1) rate of perioperative (early+delayed) stroke; (2) operative mortality among patients with perioperative stroke, early stroke, delayed stroke, and among patients without stroke; and (3) late mortality for the above groups of patients.

We used the original articles’ definitions for early and delayed stroke. The most common definitions used were stroke observed “on awakening” or “after extubation” for early stroke and stroke occurring after a symptom‐free interval for delayed stroke (Table S4).

Data Extraction and Statistical Analysis

Extracted variables included study name, publication year, study design, type of surgery, total sample size, number of patients with perioperative, early, and delayed stroke, mean age (years), percentages of women, diabetes mellitus, preoperative atrial fibrillation (AF), preoperative carotid disease, previous history of stroke or urgent or emergency surgery, peripheral vascular disease, chronic renal failure, redo surgery, in‐hospital mortality in the whole sample and in different subgroups, long‐term mortality, and mean follow‐up.

Measurement data were reported as mean±SD. Pooled event rates with 95% CI were calculated for binary outcomes. For late outcomes the incidence rate with an underlying Poisson process with a constant event rate was used to account for different follow‐up periods in different studies with the total number of events observed within a treatment group out of the total person‐time of follow‐up for that treatment group calculated from the study follow‐up. Additionally, for long‐term survival, individual patient survival data were reconstructed using an iterative algorithm that was applied to solve the Kaplan‐Meier equations originally used to produce the published graphs. This algorithm uses digitalized Kaplan‐Meier curve data obtained by the Graph Grabber software package (Quintessa, Oxfordshire, UK) to find numerical solutions to the inverted Kaplan‐Meier equation. Based on the published data in each included study, 4 different levels of information might be available (“all information,” “no numbers at risk,” “no total events,” and “neither”). The censoring pattern varied based on the numbers at risk published intervals as in Williamson.10 For the “no number at risk” case, the censoring pattern is assumed constant over the interval, and for the “neither” case, no censoring is assumed.11

The reconstructed patient survival data were then aggregated to obtain combined survival curves.

Subgroup analysis was used to compare early and delayed stroke for primary and secondary outcomes. Meta‐regression was used to assess the effect of age, sex, diabetes mellitus, preoperative AF, preoperative carotid disease, previous stroke, urgency or emergency surgery, off‐pump CABG, single or multiple aortic clamping, ascending aorta atheroma or calcification, cardiopulmonary bypass time, and aortic clamp time on the rate of early and delayed stroke.

Study heterogeneity was assessed using the Cochran Q statistic and the I² test. For the primary outcomes, if heterogeneity was significant (I² >75%), a leave‐one‐out sensitivity analysis was performed. Potential publication bias was assessed using a funnel plot and the Egger regression test.

A random‐effect model (inverse variance method) was used. In addition, prediction interval was calculated as described by Riley et al.12 Supplementary analyses using a fixed‐effect model were also performed, and τ² was provided as an inference on between‐study variability; we then used meta‐regression, which used covariates to explain some of this variability. A restricted maximum‐likelihood model was used for meta‐regression because it estimates parameters that maximize the likelihood of the error distribution while imposing restrictions to avoid overfitting, which makes it possible to obtain a better balance between the fractions of the variability captured by the fixed part versus the random part of the statistical model.13 Hypothesis testing for equivalence was set at the 2‐tailed P<0.05. Analyses were performed using R (version 3.3.3, R Project for Statistical Computing, Vienna, Austria) with the statistical packages “meta” and “metafor” within RStudio (0.99.489, http://www.rstudio.com).

Results

Characteristics of Eligible Studies

We retrieved 5212 articles and 3784 articles after deduplication. Thirty‐six articles met our inclusion criteria (list of the included studies provided in the supplemental references). The PRISMA flowchart is shown in Figure S1. The mean sample size for each study was 4860 patients (range: 245‐45 432), and the mean follow‐up time was 8.25 years (range 1.0‐11.0 years). The mean age was 65.5 years (range: 54.0‐74.0 years) (Table S5). Women represented 15% to 83% of the included patients, and diabetes mellitus, AF, carotid disease, and urgent/emergent procedures were reported in 5% to 47%, 1% to 19%, 6% to 32%, and 1% to 70% of patients, respectively. Sixty‐three percent of procedures were isolated CABG (Table S5). A total of 174 969 patients were included in the analysis, of whom 2.0% (3421 of 174 969 patients) had perioperative stroke, 1.0% (1767/174 969) had early stroke, and 1.0% (1654/174 969) had delayed stroke (P=0.68) (Table 1).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

Table 1.

Details of Outcomes in the Included Studies

Study/Year	Study Type	Cohort Size	Perioperative Stroke	Early Stroke	Delayed Stroke
Blossom 199214	R	3428	46	16	30
Boivie 200515	R	2641	98	76	22
Borger 200116	R	6682	98	90	8
Bull 199317	P	245	5	4	1
Calafiore 200218	R	4875	49	24	25
Cao 201119	R	430	32	4	28
Carrascal 201420	R	844	32	23	9
Chen 201521	R	1010	11	5	6
Doi 201022	R	611	8	0	8
Fessatidis 199123	R	1487	15	12	3
Filsoufi.A 200824	R	2808	63	35	28
Filsoufi.B 200825	R	2985	48	25	23
Gaudino 199926	R	2987	31	25	6
Goto 200327	P	463	18	13	5
Hedberg 200528	R	2641	77	58	19
Hedberg 201129	R	9122	245	146	99
Hedberg 201330	R	10 809	339	223	116
Hogue 199931	P	2972	48	17	31
Imasaka 201832	R	1134	20	8	12
Karhausen 201733	R	6130	110	35	75
Karkouti 200534	R	10 949	160	110	50
Kinnunen 201535	R	1314	23	7	16
Lahtinen 200436	R	2630	52	20	32
Lee 201137	P	1367	33	15	18
Lisle 200838	R	7201	202	46	156
Martin 198239	R	253	8	4	4
Marui 201240	R	2446	45	20	25
Murdock 200341	R	2104	68	18	50
Nishiyama 200942	P	2516	46	17	29
Peel 200443	R	10 573	211	57	154
Ridderstolpe 200244	R	3282	64	47	17
Salazar 200145	R	5971	214	158	56
Tarakji 201146	P	45 432	688	279	409
Toumpoulis 200847	R	4140	138	102	36
Weinstein 200148	P	2217	51	24	27
Wijdicks 199649	R	8270	25	4	21

P indicates prospective; R, retrospective.

The majority of stroke patients suffered from an ischemic event (88%). Transient ischemic attacks were reported in 1.38% of cases.

Mean bypass time was 81.6 minutes, and mean cross‐clamp time was 82.2 minutes.

Meta‐Analysis

Rate of Stroke

The pooled event rate of perioperative stroke was 2.03% (CI 1.75% to 2.35%) (Figure S2), early stroke was 0.98% (CI 0.79% to 1.23%; Figure 1), and delayed stroke was 0.93% (CI 0.77% to 1.11%; P=0.68 Figure 2; Table 2; a summary of the outcomes as analyzed by means of a fixed‐effect model is reported in Table S6).

Figure 1.

Pooled event rate for early stroke.

Figure 2.

Pooled event rate for delayed stroke.

Table 2.

Summary of the Outcomes (Random‐Effect Model)

Outcomes	No. of Studies	Proportion [CI]	Heterogeneity (I2, P Value)	τ2	Perioperative Stroke vs No Strokea	Early vs Delayed Stroke Differencea
Random‐effect model
Pooled rate of perioperative stroke	36	2.03% [1.75; 2.35]; PI=0.85‐4.74	94.1%, P<0.0001	0.1804	···	···
Pooled rate of early stroke	36	0.98% [0.79; 1.23]; PI=0.27‐3.49	94.7%, P<0.0001	0.3912	···	0.6774
Pooled rate of delayed stroke	36	0.93% [0.77; 1.11]; PI=0.33‐2.59	91.5%, P<0.0001	0.2568	···	0.6774
Pooled rate of operative mortality in the whole group	20	2.2% [1.8; 2.8]	96.9%, P<0.0001	0.2543	···	···
Pooled rate of operative mortality for patients with perioperative stroke	22	21.3% [18.3; 24.5]	58.8%, P=0.0003	0.0935	<0.0001	···
Pooled rate of operative mortality for patients without stroke	16	2.4% [1.9; 3.1]	96.9%, P<0.0001	0.2419	<0.0001	···
Pooled rate of operative mortality for patients with early stroke	12	28.8% [17.6; 43.4]	84.2%, P<0.0001	0.9440	···	<0.0001
Pooled rate of operative mortality for patients with late stroke	13	17.9% [14.0; 22.7]	20.1%, P=0.2407	0.0550	···	<0.0001
Incidence rate of late mortality in the “all” group	5	3.4% [2.3; 5.2]	99.3%, P<0.0001	0.2150	···	···
Incidence rate of late mortality in patients with perioperative stroke	5	10.9% [7.3; 16.2]	84.8%, P<0.0001	0.1600	<0.0001	···
Incidence rate of late mortality in patients without stroke	8	3.4% [2.4; 4.8]	99.7%, P=0	0.2426	<0.0001	···
Incidence rate of late mortality in patients with early stroke	5	11.7% [7.5; 18.3]	87.6%, P<0.0001	0.2194	···	0.5063
Incidence rate of late mortality in patients with delayed stroke	5	9.4% [5.9; 14.9]	71.2%, P=0.008	0.1771	···	0.5063

PI indicates prediction interval.

^a P value for subgroup difference.

Operative Mortality

Overall pooled event rates for operative mortality was 2.2% (CI: 1.8% to 2.8%). For patients with perioperative stroke, the operative mortality was 21.3% (CI: 18.3% to 24.5%), 28.8% (CI: 17.6% to 43.3%) for patients with early stroke and 17.9% (CI: 14.0% to 22.7%) for patients with delayed stroke (P<0.001 for early versus delayed stroke; Figures 3 and 4). The pooled event rate for operative mortality without stroke was 2.4% (CI 1.9% to 3.1%) (P<0.001 compared with patients with perioperative stroke, early stroke, and delayed stroke; Figure 4).

Figure 3.

Pooled event rate for operative mortality in patients with (top) and without perioperative stroke (bottom).

Figure 4.

Pooled event rate for operative mortality for patients with early stroke (top) and late stroke (bottom).

Late Mortality

The weighted mean follow‐up was 8.25 years. Overall incidence rate for late mortality for the entire cohort was 3.4% (CI 2.3% to 5.2%). The incidence rate for late mortality was 10.9% (CI 7.3% to 16.2%) for patients with perioperative stroke, 11.7% (CI 7.5% to 18.3%) for early stroke, and 9.4% (CI 5.9% to 14.9%) for delayed stroke (P=0.50). The incidence rate for late mortality for patients without stroke was 3.4% (CI 2.4% to 4.8%; P<0.0001 compared with patients with perioperative, early, and delayed stroke; Table 2; a summary of the outcomes as analyzed by means of a fixed‐effect model is reported in Table S6).

Reconstructed individual patient survival data from Kaplan‐Meier survival curves showed 1‐, 3‐, 5‐, and 10‐year survival of 80.2%, 73.0%, 63.3%, and 40.7%, respectively, in the early‐stroke group and 88.1%, 85.2%, 71.3%, and 30.2%, respectively, in the delayed‐stroke group (Figure S3). For patients who did not experience stroke, 1‐, 3‐, 5‐, and 10‐year survivals were 99.5%, 99.2%, 99.1%, and 97.1%, respectively.

The funnel plot of observed and imputed studies (trim‐and‐fill method) and leave‐one‐out analysis for the primary outcomes revealed absence of publication bias (the Egger intercept is −2.25±1.33 [P=0.10] for early and −1.47±1.04 [P=0.17] for delayed stroke; Figure S4). The cumulative analyses for the primary outcomes are shown in Figure 5.

Figure 5.

Cumulative analysis of incidence of (A) early stroke and (B) delayed stroke.

Meta‐Regression

Off‐pump surgery was inversely associated with early stroke (β=−0.009, P=0.01). Previous stroke (β=0.02, P<0.001) was associated with delayed stroke. Single versus multiple aortic clamping, ascending aortic atheroma or calcification, use of circulatory arrest, cardiopulmonary bypass, and aortic cross‐clamp times were not associated with either early or delayed stroke (Table S7).

Discussion

To our knowledge the present work is the first systematic review and meta‐analysis to examine the incidence, risk factors, and impact on clinical outcomes of early and delayed stroke after cardiac surgery.

Previous evidence was based on single‐center cohorts with variable sample size, incidence of events, and follow‐up duration so that a general and objective estimate of the incidence of the 2 types of stroke was difficult to ascertain. For this reason, the results of this meta‐analysis are of substantial relevance for patient counseling, clinical decision making, and planning of research for preventive interventions.

The main findings were as follows: (1) the rates of early and late stroke were similar at ≈1% each, (2) both early and delayed stroke were associated with a significant increase in operative as well as late mortality, (3) the impact on operative mortality was significantly higher for early versus delayed stroke, (4) a prior history of stroke was associated with delayed stroke, whereas (5) off‐pump CABG was inversely associated with early stroke.

Early stroke (defined as detected “on awakening” or “after extubation”) is directly linked to intraoperative events. Early strokes were inversely associated with off‐pump CABG but not with any patient characteristics, suggesting the technical/surgical nature of their etiology. Cerebral embolization is known to occur mainly due to aortic manipulation (cannulation, cross‐clamping, and performance of proximal aortic anastomoses during CABG).50, 51, 52 Early stroke has been reported to be usually located in the right hemisphere, consistent with the jet of the flow from the aortic cannula.15, 28

Although large randomized controlled trials have reported similar neurological outcomes after on‐ and off‐pump CABG (30‐day stroke incidence for on‐ versus off‐pump, respectively; 0.7% versus 1.3% [P=0.28] in the ROOBY [Randomized On/Off Bypass] trial53; 2.7% versus 2.2% [P=0.47] in the GOPCABE (German Off‐Pump Coronary Artery Bypass Grafting in Elderly Patients) trial54; 1.1% versus 1.0% [P‐value not reported] in the CORONARY (CABG Off or On Pump Revascularization Study) trial55), in our analysis off‐pump surgery was significantly and adversely associated with early stroke. Differences in sample size, treatment allocation, and surgeon expertise are the possible reasons for these differences.

Hedberg et al in a series of 10 809 patients reported that early strokes were predominantly located in the right hemisphere (P=0.009), whereas delayed stroke had a uniform spatial distribution. Authors suggested that the preponderance for right‐hemispheric lesions might suggest an embolic etiology via the brachiocephalic trunk.30 Higher stroke‐related mortality (odds ratio 9.16; P<0.0001) and greater rehabilitation needs for early versus delayed stroke were reported in a review of 7201 patients.38

Significant efforts have been aimed at intraoperative stroke reduction including minimizing or eliminating aortic manipulation, eliminating cardiopulmonary bypass, and using preoperative CT scan of the ascending aorta and duplex scanning of the carotid arteries as well as epiaortic ultrasound.56, 57, 58, 59, 60

Motallebzadeh et al randomized a total of 212 patients to receive on‐pump versus off‐pump coronary artery bypass and demonstrated reduced cerebral embolism with a better neurocognitive score at discharge in those undergoing off‐pump surgery (P<0.001 and P=0.01, respectively); there were 3 nonfatal strokes in the on‐pump group and 1 in the off‐pump group within 30 days of surgery.61

In a large series including more than 12 000 patients, the use an aortic facilitating device to perform the proximal anastomosis significantly reduced the postoperative stroke rate but was inferior to no–aortic touch technique (stroke rates 0.6%, 1.2%, and 1.5% in the no‐touch, clampless facilitating device, and the clamp group, respectively).62 Consistent with these results, Vallely reported that anaortic off‐pump coronary artery bypass resulted in 0.25% of neurological adverse events as compared with 1.1% in the groups with side‐clamping for proximal anastomoses.63

A multicenter randomized trial enrolling 383 patients undergoing surgical aortic valve replacement recently evaluated the potential neuroprotective role of 2 cannulation systems designed to capture aortic microemboli (Embol‐X Embolic Protection Device, Edwards Life Science, Irvine, CA; and CardioGard Cannula, CardioGard Medical Ltd, Or‐Yehuda, Israel). The rate of freedom from cerebral infarction at 7 days was 32.0% with suction‐based extraction versus 33.3% with control (ie, standard aortic cannula) (between‐group difference, −1.3%; 95% CI −13.8% to 11.2%) and 25.6% with intra‐aortic filtration versus 32.4% with control (between‐group difference −6.9%; 95% CI −17.9% to 4.2%); no significant differences in mortality (3.4% for suction‐based extraction versus 1.7% for control; and 2.3% for intra‐aortic filtration versus 1.5% for control) or clinical stroke (5.1% for suction‐based extraction versus 5.8% for control; and 8.3% for intra‐aortic filtration versus 6.1% for control) were detected.64

The effectiveness of early stroke reduction strategies was recently demonstrated by the EXCEL (Evaluation of XIENCE Versus Coronary Artery Bypass Surgery for Effectiveness of Left Main Revascularization) trial, in which surgeons were encouraged to use intraoperative adjunctive techniques for stroke reduction including epiaortic ultrasound and transesophageal echocardiography for assessment of ascending aortic calcification.65 The result was an overall incidence of stroke that did not differ between CABG and percutaneous coronary intervention (2.9% versus 2.3%, P=0.37).

In our study early stroke was associated with a 12‐fold increase in operative mortality (29% versus 2% without stroke) as well as much higher increases in the risk of late death (12% versus 3% without stroke), suggesting that addressing this potential complication can significantly improve the outcomes of cardiac surgery. Of note, the impact on operative mortality was significantly higher for early versus delayed stroke.

Delayed stroke (defined as stroke occurring after a normal awakening from anesthesia) is probably mostly related to postoperative AF or to cerebrovascular disease.66, 67, 68, 69, 70, 71, 72, 73, 74 In our analysis delayed stroke was associated with a 7‐ and 3‐fold increase in operative and late mortality, respectively. Late stroke was also associated with history of stroke, suggesting a greater influence of patient‐related factors such as vascular disease compared with early stroke. Indeed, contemporary cardiac surgical patients are older and have greater numbers of cardiovascular comorbidities including hypertension, diabetes mellitus, advanced age, kidney disease, peripheral artery disease, and cerebrovascular disease.75 Validated stroke risk prediction tools such as the CHA₂DS₂–VASC (congestive heart failure, hypertension, age, diabetes [mellitus], and stroke/TIA–vascular disease and female gender) scoring schema indicate that a substantial portion of cardiac surgical patients are at high risk for AF‐related stroke.76

The main strategies for delayed stroke prevention are (1) pharmacological or nonpharmacological AF prophylaxis,77 (2) anticoagulation for prevention and treatment of clot formation,78 and (3) elimination of the left atrial appendage.1 AF prophylaxis includes amiodarone, β‐blockers, magnesium, atrial pacing, and posterior pericardiotomy.77 Regarding left atrial appendage isolation, LAAOS III (the Left Atrial Appendage Occlusion Study) is an ongoing prospective, double‐blind, randomized trial comparing concomitant surgical left atrial appendage occlusion and no‐occlusion in patients with AF or flutter who are undergoing cardiac surgery (ClinicalTrials.gov Identifier: NCT01561651). Again, continued efforts to evaluate interventions to lower the risk of delayed stroke in prospective surgical trials are needed.

This study shares the common limitations of analyses of aggregate data. First, this analysis included a range of cardiac surgical procedures, although isolated CABG was the most common type of procedure (Table S5). There was heterogeneity in the definitions used by the different studies, in the surgical and postoperative protocols (Table S4), as well as in the follow‐up approaches, in the involvement of a neurologist in the diagnosis of stroke events, and in the documentation of these events by cerebral‐imaging studies. Moreover, postdischarge stroke might have been missed in some studies. Finally, most of the studies did not use continuous monitoring of postoperative cardiac rhythm, and thus, we have no solid information on the occurrence of postoperative AF, and we were unable to include this variable in our meta‐regression analysis. As in all meta‐analyses, ecological fallacy is a concern. Finally, it was not possible to determine whether early or late deaths were directly related to strokes.

Summary

This is the first systematic review and meta‐analysis to examine the incidence of early and delayed stroke after cardiac operations. There is a 1% risk for both early and delayed stroke after cardiac surgery. Early stroke is not associated with any patient‐level risk factors, suggesting a technical cause, and is associated with a significant increase in operative mortality as well as reduction in long‐term survival. The impact of early stroke on operative mortality is significantly higher than that of delayed stroke. Delayed stroke is associated with previous stroke and also negatively impacts survival. Continued targeted interventions to reduce the burden of both early and delayed strokes are imperative to improve overall surgical outcomes.

Disclosures

None.

Supporting information

Table S1. MOOSE Checklist for Meta‐Analyses of Observational Studies

Table S2. The Search Strategy for Ovid MEDLINE

Table S3. Critical Appraisal of Included Studies Using the Newcastle‐Ottawa Quality Assessment Scale for Cohort Studies

Table S4. Stroke Definitions in the Included Studies

Table S5. Demographics of the Included Studies

Table S6. Summary of the Outcomes (Fixed‐Effect Model)

Table S7. Meta‐Regression for Early and Delayed Stroke (Restricted Maximum‐Likelihood Model)

Figure S1. PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) flowchart.

Figure S2. Pooled event rate for perioperative stroke.

Figure S3. Reconstructed Kaplan‐Meier survival curves from derived individual patient data (IPD) for (A) no stroke vs perioperative stroke and (B) no stroke vs early and delayed stroke. Solid/dotted line represents aggregation of all available Kaplan‐Meier curves with 95% CI.

Figure S4. Leave‐one‐out analysis (top) and funnel plot (bottom) for incidence of (A) early stroke and (B) delayed stroke.

(J Am Heart Assoc. 2019;8:e012447 DOI: 10.1161/JAHA.119.012447.)