Patient Population

All patients who had undergone LVAD implantation at the reporting institution between January 2010 and June 2020 were retrospectively evaluated (N = 820). Exclusion criteria were as follows: patients who underwent LVAD implantation for a BTT indication; patients who underwent dual LVAD and right ventricular assist device procedures; patients with multiple LVAD implantation events, with only the most recent surgery maintained; and recipients of isolated right ventricular assist device, HeartMate XVE (Thoratec/St Jude), total artificial heart, or biventricular assist devices. Patients were further classified by the type of LVAD device placed: HeartMate II (Abbott), HeartMate 3 (Abbott), and HeartWare Ventricular Assist Device (HVAD; Medtronic). The approach to LVAD insertion was left to the discretion of the attending physician. The analyzed cohort included 311 patients (Fig. 1).

View Figure

• Download as Powerpoint
• Download Figure

Management of Perioperative AF: Institutional Protocols

It is not the reporting institution's protocol to treat POAF prophylactically in the preoperative setting. For patients with preoperative AF, β-blocker therapy is reinitiated after the patient has been weaned off inotropic support and remains hemodynamically stable. Home-based amiodarone therapy is initiated at the onset of POAF. With respect to anticoagulation, the institution's written and multidisciplinary approved protocols begin with the administration of aspirin 325 mg on postoperative day 0. Aspirin is continued for life. On postoperative day 1, intravenous heparin is initiated at a flat rate of 750 units per hour. This therapy transitions to administration using an intravenous heparin nomogram begun on postoperative day 2 with concomitant initiation of warfarin 1 mg. The target partial thromboplastin time range while on heparin is 60 to 90 seconds. The patient's warfarin dose is sequentially increased until a target international normalized ratio of 2.0 to 3.0 is reached. Heparin is discontinued once the target international normalized ratio has been reached. Patients with preoperative AF who required oral anticoagulation are converted to intravenous heparin preoperatively.

The 2014 American Association for Thoracic Surgery guidelines for management of new-onset POAF suggest the use of intravenous β-blockers, nondihydropyridine calcium channel blockers, digoxin, or amiodarone; nonpharmacologic management includes cardioversion with or without pretreatment with antiarrhythmic drugs.^23,26 Patient charts were reviewed for techniques to manage POAF, including the use of antiarrhythmic drugs and cardioversion. Patients requiring postoperative amiodarone on discharge were evaluated at 1-month and 3-month follow-up. If they were found to be in normal sinus rhythm at these visits, amiodarone was discontinued. If patients remained in AF at their 3-month follow-up, amiodarone was continued for life.

Diagnosis and Treatment of Stroke

Stroke was defined as any new-onset focal or global neurologic deficit determined by standard neurologic evaluation, with deficits lasting beyond 24 hours to distinguish stroke from transient ischemic attack. These findings were confirmed by neurology consult and verified using computed tomography diagnostic imaging. To determine the subtype of the stroke, anatomic and radiologic findings from computed tomographic images and neurology consults were evaluated. Chart review of electronic health records for radiographic imaging, neurology consult service, and follow-up notes were conducted to identify patients who may not have been captured by the national databases used in this study. Treatment of stroke was guided by stroke type: embolic or hemorrhagic. Embolic stroke management modalities included intra-arterial embolectomy and antithrombotic medication. Reversal of coagulopathy has been indicated as treatment for hemorrhagic strokes, with antithrombotic medications resuming within 2 weeks of stroke treatment.¹⁷ Antithrombotic regimens included antiplatelet medication, anticoagulant medication, combination therapies, or more aggressive regimens as indicated.

Statistical Analysis

Continuous variables were expressed as median (IQR) values. The Mann-Whitney U test was used to evaluate differences between the 2 groups. Kruskal-Wallis testing was used to evaluate differences among multiple groups, and post hoc Dwass-Steel-Critchlow-Fligner testing was used to adjust statistically significant results for multiple comparisons. Categorical variables were expressed as frequencies and percentages and compared using the χ² test. P < .05 was considered statistically significant.

The risk of stroke was evaluated using a competing-risks methodology, such that the proportion of patients who developed a stroke during longitudinal follow-up was represented visually by the cumulative incidence function (CIF). Patients were categorized into 1 of 3 mutually exclusive states at each time point. If a patient developed a stroke at any point during the follow-up period, then this was the terminal state, and the patient was classified as part of the stroke cohort. In patients who never developed a stroke during follow-up, the other terminal state was death during follow-up; otherwise, patients were considered to be alive and censored at their last follow-up date. Cumulative incidence functions were generated for the competing risks of death and stroke during the follow-up period. Because the 3 states are mutually exclusive, the probability of remaining alive and free from stroke was equivalent to the probability of not experiencing any of the competing risks and was therefore depicted as a composite Kaplan-Meier estimate. This composite end point was presented alongside the CIFs for each competing risk.

Predictors of stroke were identified using Fine-Gray subdistribution hazards regression. Potential predictors that were deemed clinically relevant based on the judgment of senior authors as well as on prior work indicating their association with the development of postoperative stroke,^14-16,27-30 which had sufficient events per variable, were selected for inclusion in the univariable model to increase the accuracy of the regression coefficients. Forward selection of univariable predictors with P < .20 was used to create the final multivariable model. Analysis was performed using SPSS Statistics, version 25 (IBM Corp), and R, version 3.6.3, statistical software using the cmprsk package (R Foundation for Statistical Computing) for the competing-risk analysis and graphical output of CIF. SAS OnDemand for Academics (SAS Institute Inc) was used to generate baseline comparisons.

Baseline Characteristics

Over the study period, the majority of recipients were male (n = 246 [79%]) and White (n = 246 [79%]), with a mean (SD) age of 60 (12) years. Patients implanted with a continuous-flow LVAD (n = 311) were classified by the type of device implanted: HeartMate II (n = 97 [63%]), HeartMate 3 (n = 72 [23%]), and HVAD (n = 42 [14%]). The most common Interagency Registry for Mechanically Assisted Circulatory Support profile for recipients was type 2: progressive decline (n = 196 [63%]). Dilated cardiomyopathy as a result of ischemia (n = 149 [48%]) or idiopathic causes (n = 146 [47%]) were the most common heart failure etiologies. The median (IQR) follow-up time was 1.9 (0.6-3.4) years. In the cohort, 42 patients had clinically significant improvement in functional status, crossed over to a BTT indication, and were subsequently transplanted. Detailed baseline characteristics can be found in Table I and Supplemental Table I.

TABLE I.Baseline Demographics and Perioperative Characteristics^a

View Fullscreen

Perioperative Results

Median (IQR) cardiopulmonary bypass time was 68 (56-89) minutes. The most common postoperative complications were the development of AF (n = 110 [35%]) and gastrointestinal (GI) events, including upper or lower GI bleeding, gastroenteritis, and unspecified GI disorders requiring hospitalization (n = 85 [27%]). The median (IQR) postoperative hospital length of stay was 22 (16-35) days. Thirty-three patients (11%) died within 30 days of their procedure. Comprehensive perioperative outcomes can be found in Table I and Supplemental Table I.

Rhythm Outcomes

A total of 35% (n = 110) of patients developed POAF. Among these patients, 67% (n = 74) developed POAF within the first 2 postoperative days (Fig. 2). New-onset AF, defined as POAF identified in patients who had not been known to have preoperative atrial arrhythmia, was discovered in 21% (n = 64) of patients. The median (IQR) day of POAF onset was postoperative day 2 (1-3). Incidence of POAF did not differ based on LVAD type (P = .61). Of the patients who developed POAF, 85% (n = 94) received amiodarone as treatment, and 20% (n = 22) underwent attempted cardioversion to restore normal sinus rhythm (Table II). One patient had an atrioventricular node ablation with an upgrade to a biventricular implantable cardioverter-defibrillator at an external institution that did not deem the patient a suitable candidate for catheter ablation because of failed amiodarone loading and cardioversion. Other treatment outliers included 6 patients who received lidocaine and 2 patients who received digoxin for medical management of their POAF in addition to amiodarone. Patients who developed POAF had a higher incidence than patients who did not of developing sepsis. There were no statistically significant differences in any other perioperative complications, including reoperation for bleeding, in-hospital stroke, GI events, and multisystem organ failure, between these patient groups (Table III).

View Figure

• Download as Powerpoint
• Download Figure

TABLE II.Postoperative Outcomes^a

View Fullscreen

TABLE III.Perioperative Outcomes in Patients Who Developed Postoperative AF

View Fullscreen

The resolution of POAF to sinus rhythm occurred in 91% (100/110) of patients within 90 days. Of the patients who recovered within 90 days, 85% (85/100) were treated with amiodarone, and 15% (15/100) were treated with successful cardioversion while hospitalized. Three patients who did not recover sinus rhythm in 90 days experienced stroke. Slightly more than half of patients (55% [55/100]) who recovered within 90 days were discharged on amiodarone. Of the 10 patients whose POAF did not resolve in 90 days, 50% (n = 5) had preoperative AF. Beyond 90 days, 7 out of 9 patients’ POAF never resolved because the patients experienced a complicated hospital course after LVAD implantation and died. The other 2 patients’ POAF resolved after heart transplantation or after explantation and subsequent implantation of a new LVAD. Rhythm management and outcomes are further delineated in Table II, Supplemental Table II, and Figure 3.

View Figure

• Download as Powerpoint
• Download Figure

Stroke Development

Incidence of stroke events was 23% (n = 71) among all LVAD recipients (Table II). Stroke was identified more frequently in patients who had received HeartMate II devices (n = 59 [29%]) than in recipients of newer-generation centrifugal flow devices (P = .001). Among patients who developed stroke, 45% (n = 32) of the cases were embolic and 55% (n = 39) were hemorrhagic. The type of stroke did not vary based on LVAD type (P = .97). The composite probability of remaining alive and free of stroke was estimated to be 68% at 1 year, 43% at 3 years, and 27% at 5 years in recipients of continuous-flow LVADs (Fig. 4). The incidence of stroke at the same time points was 15%, 24%, and 26%. Composite end point estimates and CIFs from competing risks analysis can be found in Table IV.

View Figure

• Download as Powerpoint
• Download Figure

TABLE IV.Composite End Point Kaplan-Meier Estimates and Cumulative Incidence Functions for the Development of Stroke and Death Following Continuous-Flow LVAD Implantation

View Fullscreen

Competing-Risks Analysis

Fine-Gray regression models of the subdistribution hazard function were used to evaluate potential predictors of stroke (Table V). Amiodarone continued at discharge (subdistribution hazard ratio, 0.55 [95% CI, 0.33-0.94]; P = .03) and HeartMate 3 implantation (hazard ratio, 0.41 [95% CI, 0.19-0.90]; P = .03) were associated with decreased stroke risk on multivariable analysis. Postoperative AF was associated with increased stroke development (0.55 [95% CI, 0.33-0.94]; P = .04). Preoperative AF (P = .08) and a history of stroke (P = .36) were not found to be statistically significant predictors of stroke.

TABLE V.Predictions for the Development of Stroke in Continuous-Flow LVAD Recipients on a Univariable and Multivariable Subdistribution Hazards Regression Model^a

View Fullscreen

Study Limitations

Although this is one of the first single-center studies to report a consistent and comprehensive analysis with long-term follow-up in patients who undergo continuous-flow LVAD with a destination therapy indication, there are limitations. This study was retrospective and nonrandomized; therefore, it was subject to inherent selection bias. For null results in particular, the sample size may have contributed to the increased risk of type II error. The sample size was too small to allow for evaluation of predictors for each device type. Variables were chosen for inclusion in the Fine-Gray regression based on perceived clinical relevance and prior work, and the number of variables included was limited by the events per variable; results are therefore subject to this inherent selection bias. Further evaluation of stroke in this patient population is recommended, with similarly robust statistical methodology and granularity in patient follow-up among larger patient cohorts. Referral patterns, destination therapy patient selection, and local demographics may have altered the patient population. Adverse events that occurred at other centers may not have been reported to national databases or to the reporting institution, so estimates for late complication rates may be underestimated. All patients were closely monitored according to the reporting center's protocol, however, and follow-up data were obtained from both national databases and institutional records.