Author + information
- Received March 15, 2017
- Revision received July 5, 2017
- Accepted July 11, 2017
- Published online November 5, 2018.
- Faraz Pathan, MBBSa,b,
- Eswar Sivaraj, MScb,
- Kazuaki Negishi, MD, PhDa,b,
- Rifly Rafiudeen, MBBSb,
- Shahab Pathan, MBBSc,
- Nicholas D’Elia, MBBSd,e,
- John Galligan, MBBSb,
- Samuel Neilson, MBBSb,
- Ricardo Fonseca, MBBSa and
- Thomas H. Marwick, MBBS, PhD, MPHa,d,∗ ()
- aMenzies Institute for Medical Research, University of Tasmania, Hobart, Australia
- bDepartment of Cardiology, Royal Hobart Hospital, Hobart, Australia
- cDepartment of Cardiology, Nepean Hospital, Sydney, Australia
- dBaker Heart and Diabetes Institute, Melbourne, Australia
- ePrincess Alexandra Hospital, Brisbane, Australia
- ↵∗Address for correspondence:
Dr. Thomas H. Marwick, Baker Heart and Diabetes Institute, 75 Commercial Road, Melbourne, Victoria 3004, Australia.
Objectives This study sought to identify whether atrial strain could be used as an imaging biomarker to predict atrial fibrillation (AF).
Background AF is found in up to 30% of cryptogenic cerebrovascular accidents (CVAs), which themselves account for 30% to 40% of ischemic CVA.
Methods This observational study evaluated all patients who had an echocardiogram (transthoracic echocardiogram [TTE]) following presentation with cryptogenic CVA from 2010 to 2014. The TTEs were evaluated for reservoir strain (ƐR), contractile strain (ƐCt), and conduit atrial strain (ƐCd) using speckle tracking. Baseline clinical and TTE characteristics of patients who developed AF over 5 years of follow-up and those who did not were compared. The independent and incremental predictive value of atrial strain over established clinical models was assessed. Discriminatory cutpoints were defined using a Classification and Regression Tree (CART) analysis to identify patients at risk of developing AF.
Results Of 538 patients, 61 (11%) developed AF, and this occurred within 2 years in 85% of patients. Patients who developed AF were older, had higher clinical risk scores, had higher LA volume, and had lower atrial strain than did those who did not develop AF. The area under the receiver-operating characteristic curve was 0.85 for ƐR, 0.83 for ƐCt, and 0.76 for ƐCd (all p < 0.001). The nested Cox regression model showed that ƐR (p = 0.03) and ƐCt (p < 0.001) demonstrated independent and incremental predictive value over the clinical risk. CART analysis identified ƐR ≤21.4%, ƐCd >10.4%, and CHARGE-AF (Cohorts for Heart and Aging Research in Genomic Epidemiology Atrial Fibrillation) score >7.8% as discriminatory for AF, with a 13-fold greater hazard of AF (p < 0.001) in patients with increased clinical risk and reduced ƐR. However, validation is needed for these strain cutoffs for detection of AF.
Conclusions Left atrial strain adds independent and incremental predictive value to current risk-prediction models for AF following cryptogenic CVA. Further studies should examine the implications of these findings for AF monitoring or empiric anticoagulation.
Ischemic cerebrovascular accidents (CVA) and transient ischemic attack (TIA) are associated with death and disability, with 20% to 30% attributed to cardioembolic and 30% to 40% to cryptogenic sources (1,2). Atrial fibrillation (AF) is found in 9%, 12%, and 30% of cryptogenic strokes at 6, 12 and 36 months respectively (3). Thus, in addition to brain, carotid, and cardiac imaging, the diagnostic work-up for cerebral ischemia includes 12-lead electrocardiography (ECG) and ECG monitoring (1,4,5). Current guidelines recommend that 24-h Holter monitoring be used when occult AF is suspected (5,6). However, as it is often asymptomatic and paroxysmal, AF can be difficult to detect; the detection rate of new AF after cerebral ischemia is approximately 2% to 5% from a standard 12-lead ECG (7,8) and 2.0% to 9.2% from 24-h Holter (9,10).
Long-term monitoring for AF is now possible with implantable loop recorders, but the best strategy for selecting patients for this intervention is unclear. A variety of clinical scores have been proposed to predict risk of AF, with C-statistic ranging from 0.72 to 0.78 (11). Although neuroimaging has been proposed to be of value in attributing strokes to AF, a blinded review of imaging data from the CRYSTAL-AF (Cryptogenic Stroke and Underlying Atrial Fibrillation) trial showed no evidence for an association between acute brain infarction pattern and AF detection using an invasive cardiac monitor in patients with cryptogenic strokes (12).
Although in the absence of cardiac disease, 3% of transthoracic echocardiograms (TTEs) performed after cerebral ischemia reveal a cardioembolic source (5), there is an association between left atrial (LA) size and cryptogenic or cardioembolic strokes (13). Atrial strain has been used to facilitate stroke risk calculation and has prognostic implications in AF (14). Patients with cryptogenic strokes have lower strain values compared with control subjects, possibly due to atrial myopathy and AF (15). However, the data are limited and heterogeneous; a reservoir strain (ƐR) <25.8% has a sensitivity of 70% and specificity of 75% for the prediction of AF after cerebral ischemia (16), but in another study, ƐR <14.5% had a sensitivity of 60% and a specificity of 95%. Nonetheless, atrial strain has incremental value over age, sex, CHA2DS2-VASc (congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, prior stroke or transient ischemic attack or thromboembolism, vascular disease, age 65 to 74 years, sex category), left atrial volume index (LAVi), and ratio of early mitral inflow velocity and mitral annular early diastolic velocity (E/e′) for prediction of AF (17). Although atrial contractile function is also impaired with AF due to atrial stunning (18), the predictive value of contractile strain has not been evaluated following cryptogenic stroke. Accordingly, we sought to assess the independent value of ƐR, ƐCt, and conduit strain (ƐCd); their incremental value to standard echocardiographic parameters and clinical risk scores; and their appropriate cutoffs for the prediction of AF after cerebral ischemia.
The aims of this study were to determine if different components of atrial strain add independent and incremental value for the prediction of AF (primary endpoint) in addition to risk scores and standard echocardiographic parameters, and to define optimal strain cutoff values that can be integrated within existing clinical risk scores to improve prediction of AF.
In this cohort study, we included consecutive patients admitted with a stroke or TIA, who underwent an echocardiogram at Royal Hobart Hospital from 2010 to 2014. We defined cryptogenic strokes according to the TOAST (Trial of Acute Stroke Treatment) guidelines (19). Patients were excluded if: 1) there was a history of AF or they were diagnosed with AF during admission or on outpatient Holter prior to the TTE being performed; 2) an alternative cause of cerebral ischemia was identified: ipsilateral carotid stenosis >70%, left-sided cardiac mass or thrombus, left sided endocarditis, or atrial septal aneurysm with patent foramen ovale; 3) only transesophageal echocardiography was performed; or 4) images were inadequate for strain analysis.
Patients were followed up until June 6, 2016, for the primary outcome of AF. Three investigators (S.P., J.G., and S.N.) collected clinical and outcome data by reviewing electronic medical records, admission codes, clinic correspondence and cardiac investigations (ECG, telemetry, Holter, pacemaker reports). Date of death was obtained for all patients who died by the completion of follow-up.
The study was approved by the Tasmanian Health and Medical Human Ethics Research Committee (HERC reference no. H0015502).
We documented all relevant clinical variables including demographics, height, weight, body surface area (BSA), cardiac risk factors, presence or absence of heart failure, ischemic heart disease, and vital signs. Using these measures, we computed the CHARGE-AF (Cohorts for Heart and Aging Research in Genomic Epidemiology Atrial Fibrillation) and CHA2DS2-VASc scores.
All TTEs were performed by accredited sonographers in accordance with American Society of Echocardiography guidelines (20). Three investigators (E.S., R.R., and N.D.) measured echocardiographic parameters in accordance with American Society of Echocardiography guidelines (21). Echocardiographic parameters included in analysis were: LA volume, left ventricular mass, left ventricular end-systolic and end-diastolic volume and systolic and diastolic function, and presence and severity of valvular heart disease. All volumetric measures were indexed to BSA.
Atrial strain was analyzed on commercial software (TomTec Image Arena, Munich, Germany). We used 2- and 4-chamber views with the onset of QRS complex used as the zero reference point (R-R gating) as has been previously described (22). A region of interest was drawn along the LA endocardial border in both views (Figure 1A), tracking was reviewed to ensure that it was appropriate and a true representation of atrial motion, and strain results were taken as an average of the 2 views. The resulting atrial strain curve provided 2 peaks consistent with reservoir and contractile strain. The difference between these was conduit strain. (Figures 1B and 1C).
In cases where only one view (4-chamber or 2-chamber) was available, this view was used. We allowed exclusion of 2 segments in 2 views or 1 segment from one view if tracking was inadequate; if more segments were inadequate, the patient was excluded. One investigator experienced with strain imaging (F.P.) was blinded to the outcome and echocardiographic characteristics and evaluated atrial strain in all patients. Reproducibility of measurements was assessed in 20 randomly selected cases (10 AF, 10 non-AF) by repeated measurements performed by FP and a second operator ES experienced in strain analysis. Both observers were blinded to one another’s strain measurements and the clinical endpoint.
Baseline categorical variables are shown as number and percentage, and computed using the chi-square test. Continuous variables are presented as mean ± SD, and compared using the independent samples Student’s t test. If not normally distributed, they are presented as median (IQR), and compared with the Kruskal-Wallis test.
Multiple receiver-operating characteristic (ROC) curve analysis was performed on echocardiographic parameters with respect to the prediction of AF. We performed a multivariable nested Cox regression model to evaluate incremental predictive value. The 2 risk scores were included as covariates in separate models, which also included 4 echocardiographic covariates. Collinearity diagnostics were performed on variables to ensure stability of the multivariable model with variance inflation factor >3 used as the threshold.
We performed a sensitivity analysis by comparing the above model with a competing risk model that included death as a competing outcome variable and AF as the primary endpoint.
Finally, we examined incremental value by comparing CHARGE-AF and CHA2DS2-VASc scores with CHARGE-AF score with ƐR and ƐCt as well as CHA2DS2-VASc score with ƐR and ƐCt by examining the incremental area under the ROC curve (AUC).
The optimal strategy to predict AF was evaluated using a Classification and Regression Tree (CART) analysis (23). All baseline demographic, clinical, echocardiographic, and strain measurements were used as input variables. The result of the CART analysis was used to define cut-off values to predict future development of AF. The net reclassification improvement was calculated for the CART model compared with the CHARGE-AF model.
Interobserver and intraobserver variability was computed using the intraclass correlation coefficients (ICCs) using a 2-way mixed model with absolute agreement between measures. Reliability was also assessed using the Bland-Altman plots for interobserver and intraobserver differences. A p value <0.05 was considered statistically significant. All statistical analysis was performed on IBM SPSS Statistics version 24 (IBM Corporation, Armonk, New York) and R version 3.2.2 (R Foundation for Statistical Computing, Vienna, Austria).
Of 1,545 patients who had echocardiographic examinations following a CVA between January 1, 2010, and December 30, 2014, 929 were excluded, as they were known to be or found to be in AF before the echocardiogram or had alternative explanations for the stroke or TIA. Of the remaining 616 patients whose echocardiograms were reviewed, 4 were in AF at the time of the TTE, 5 only had a transesophageal echocardiogram, and 69 patients were excluded due to images being unsuitable for strain computation.
Of 538 patients included in our analysis, 61 (11%) went on to develop AF during follow-up (Figure 2). Eighty-five percent of patients who developed AF did so within 2 years of the stroke or TIA and 93% within 3 years.
Table 1 shows the baseline characteristics of both the patients who developed AF and those who did not (n = 477). The patients who developed AF were older (p < 0.001), with a greater prevalence of hypertension (p = 0.04) and heart failure (p < 0.001) than those patients in the non-AF group. Both clinical risk scores, CHARGE-AF (p < 0.001) and CHA2DS2-VASc (p < 0.001), were significantly higher in patients who developed AF. LA volume was significantly higher in the patients who developed AF (p < 0.001). All atrial strain indices were significantly lower in the patients who went on the develop AF (p < 0.001). Nonsignificant TTE variables are shown in Online Table 1.
A higher proportion of the patients who developed AF died compared with those who did not develop AF 39% versus 12% (p < 0.001).
Table 2 summarizes the ROC curve analyses. ƐR, ƐCt, ƐCd, E/e′ ratio, average e′, and LA volume (ml/m2) were identified as the echocardiographic predictors with the highest AUC (Online Figure 1). ƐCd and average e′ were excluded from the multivariable models because of collinearity.
Incremental value of LA strain
Two models were computed: 1) CHARGE-AF versus CHARGE-AF, ƐR, and ƐCt; and 2) CHA2DS2-VASc versus CHA2DS2-VASc, ƐR, and ƐCt. There was improvement for the AUC for both models by addition of ƐR and ƐCt for the prediction of AF (for CHARGE-AF from 0.78 to 0.86, p < 0.001; and for CHA2DS2-VASc from 0.74 to 0.86, p < 0.001).
Figure 3 illustrates the results of the nested Cox regression algorithm, which demonstrates independent and incremental predictive value for the CHARGE-AF score, ƐR, and ƐCt. LA volume and E/e′ ratio provided incremental predictive value though this was not independent of the remaining covariates. Similar independent and incremental value was shown when using the CHA2DS2-VASc score (Online Figure 2). Variance inflation factor values for all variables in models were <3.
Competing risk models
There were more deaths in the patients who developed AF (p < 0.001). The sensitivity analysis with AF as the primary endpoint and death as a competing endpoint showed similar results to the initial Cox regression model (Table 3). CHARGE-AF score, ƐR, and ƐCt were significantly associated with future AF. LAVi, ejection fraction, and E/e′ ratio were not significant markers of future AF. Similar results were seen using the CHA2DS2-VASc score (Online Table 2).
The CART decision tree analysis identified 3 discriminatory nodes as predictors of AF (Figure 4): ƐR ≤21.4%, ƐCd >10.4%, and CHARGE-AF >7.8%. The strain nodes (ƐR ≤21.4% and ƐCd >10.4%) demonstrated a specificity of 99% and a sensitivity of 36% for development of AF. Based on these results we defined patients with ƐR >21.4% and CHARGE AF ≤7.8% as low risk and the remainder as high risk. The sensitivity and specificity for high-risk patients were 92% and 67%, respectively, and these patients had a 21-fold increment of hazard (p < 0.001). Even after adjustment for age, ejection fraction, LAVi, E/e′ ratio, and CHA2DS2-VASc score, high-risk status showed a 13-fold increment of hazard (p < 0.001) (Figure 5). Compared with an existing clinical approach to risk (CHARGE-AF >5%), high-risk status showed a net reclassification improvement of 12% (95% confidence interval [CI]: 4% to 20%) for predicting AF.
The ICC for intraobserver variability was 0.96 (95% CI: 0.91 to 0.99) for ƐR and 0.90 (95% CI: 0.77 to 0.96) for ƐCt. The mean difference and limits of agreement from the Bland-Altman plots for ƐR and ƐCt were −0.70 (95% CI: −7.20 to 5.70) and −0.41 (95% CI: −5.20 to 4.40), respectively (Online Figure 3). The ICC for interobserver variability was 0.95 (95% CI: 0.87 to 0.98) for ƐR and 0.86 (95% CI: 0.68 to 0.94) for ƐCt. The mean difference and limits of agreement for ƐR and ƐCt were −0.6 (95% CI: −7.9 to 6.6) and −0.6 (95% CI: −6.0 to 4.8), respectively (Online Figure 3).
In this study, 11% of cryptogenic stroke patients developed AF during follow-up—almost all within 3 years. We have demonstrated independent and incremental predictive value for atrial strain over and above all other clinical and echocardiographic variables with respect to AF prediction following cryptogenic CVAs. Our study is the first to utilize all phases of the atrial strain curve to predict risk of AF. Furthermore, to our knowledge this is the first study combining atrial strain with clinical risk prediction models (CHARGE-AF and CHA2DS2-VASc) to develop a sensitive algorithm with which to predict AF. As expected both CHARGE-AF and CHA2DS2-VASc were higher in those who developed AF. Although LA volume and E/e′ ratio were higher and average e′ lower in patients who developed AF, they did not provide independent and incremental predictive value over the clinical risk scores. This result was stable despite adjustment for death as a competing outcome variable. Furthermore, an analysis of all clinical characteristics, risk scores, and echocardiographic parameters identified 3 discriminatory nodes (CHARGE-AF, ƐR, and ƐCd), which in composite provide both sensitive and specific measures for evaluating patients at risk. Earlier studies employing atrial speckle tracking have not included ƐCd or ƐCt (16,17).
Clinical implications of AF prediction
Multiple clinical risk scores have been developed which facilitate the prediction of AF. Recently the addition of B-type natriuretic peptide to the CHARGE-AF score demonstrated incremental predictive value (24).
Atrial strain using speckle tracking has been evaluated in multiple conditions including hypertension, diabetes, heart failure, ischemic or valvular heart disease, and AF in which the test has facilitated stroke risk calculation (25) and assessment of prognostic implications (14). Atrial strain is a novel imaging biomarker that has been shown to predict AF following CVA and maintenance of sinus rhythm post–catheter ablation for AF (16,17,26).
Despite differences relating to age, BSA, diastolic blood pressure, and the prevalence of hypertension and heart failure between the AF and non-AF patients, the discriminative value of these differences was limited. We did not assess individual risk factors, as the CHARGE-AF score calculated a combined prediction model adjusted for age, race, height, weight, systolic blood pressure or diastolic blood pressure, antihypertensive treatment, smoking, diabetes, heart failure, and myocardial infarction (27). Similarly, despite differences in LAVi, left ventricular end-systolic volume, and left ventricular mass, only atrial strain indices provided independent and predictive value.
Use of strain indices alone (ƐR ≤21.4% and ƐCd >10.4%) identified a very high-risk subset of patients who may be a target for empirical anticoagulation, particularly given the independent and incremental value of LA strain in predicting thromboembolic risk over and above the CHA2DS2-VASc score (25). A combined strain and clinical score decision algorithm (ƐR ≤21.4% or CHARGE-AF >7.8%) identified 56 of 61 patients who went on to develop AF, providing for a sensitive tool which could guide long-term AF monitoring. Given the underuse of anticoagulation particularly in those at high risk of stroke (28), an imaging-guided approach to anticoagulation may facilitate targeted therapy.
Clinical trials (NCT02313909 and NCT02239120) are currently looking at the role of novel oral anticoagulants for embolic strokes of undetermined source. The potential problem of this approach is that the majority of these patients do not actually have AF, so they will be exposed to bleeding risk without benefit. Whatever the outcome of these studies, the results of our study may serve to improve patient selection by identifying those at highest risk of AF.
First, the rate of AF is likely underestimated given the absence of long term monitoring. However, this result is consistent with the control arm of the CRYSTAL-AF study, with a 2% incidence of AF at 12 months (3). Second, the modest number of events (n = 61) poses a challenge to modeling. Third, the 3 CART-derived cutpoints need to be externally validated in future studies. Fourth, the incidence of AF in patients with TIA is lower than in those patients with strokes, resulting in a lower AF prevalence compared with studies that enrolled only patients with strokes. The recent paradigm shift to embolic strokes of undetermined source represents an enriched population where the prevalence of AF is higher and may represent a more appropriate target for future studies (29).
Finally, images of the atrium were not optimized for strain analysis. Nevertheless, strain computation was feasible in 87% of echocardiograms. There is variability in strain measurement, lack of standardization between vendors and few dedicated atrial strain packages. These limitations can and should be overcome.
LA strain is a novel imaging biomarker that adds independent and incremental predictive value to established clinical risk prediction tools for AF. A composite of ƐR and CHARGE-AF score may provide a sensitive predictor of AF following cryptogenic CVAs, but validation is needed to verify the exact strain cutoffs used for this purpose. Further studies should seek whether an algorithm which incorporates imaging biomarkers with existing clinical risk scores can improve patient selection for long term rhythm monitoring strategies, and the use of anticoagulation in those at highest risk.
COMPETENCY IN MEDICAL KNOWLEDGE: The paroxysmal nature of AF makes it difficult to recognize as a cause of cryptogenic CVAs. This study shows that LA strain adds independent and incremental predictive value to current risk prediction models for AF.
COMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS: Echocardiography is widely performed in patients presenting with stroke. The assessment of LA strain should be considered as an additional marker of possible cardiogenic stroke.
TRANSLATIONAL OUTLOOK: Further studies should examine the implications of these findings for AF monitoring or empiric anticoagulation.
The authors have reported that they have no relationships relevant to the contents of this paper to disclose. Nathaniel Reicheck, MD, served as the Guest Editor for this paper.
- Abbreviations and Acronyms
- atrial fibrillation
- Classification and Regression Tree
- confidence interval
- cerebrovascular accident
- early mitral inflow velocity and mitral annular early diastolic velocity
- conduit strain
- contractile strain
- reservoir strain
- intraclass correlation coefficient
- left atrium
- transient ischemic attack
- transthoracic echocardiogram
- Received March 15, 2017.
- Revision received July 5, 2017.
- Accepted July 11, 2017.
- 2018 American College of Cardiology Foundation
- Yang H.,
- Nassif M.,
- Khairy P.,
- et al.
- Kolominsky-Rabas P.L.,
- Weber M.,
- Gefeller O.,
- Neundoerfer B.,
- Heuschmann P.U.
- Saric M.,
- Armour A.C.,
- Arnaout M.S.,
- et al.
- Easton J.D.,
- Saver J.L.,
- Albers G.W.,
- et al.
- Kishore A.,
- Vail A.,
- Majid A.,
- et al.
- Alhadramy O.,
- Jeerakathil T.J.,
- Majumdar S.R.,
- Najjar E.,
- Choy J.,
- Saqqur M.
- Alonso A.,
- Norby F.L.
- Sanfilippo A.J.,
- Abascal V.M.,
- Sheehan M.,
- et al.
- Hoit B.D.
- Leong D.P.,
- Joyce E.,
- Debonnaire P.,
- et al.
- Pagola J.,
- Gonzalez-Alujas T.,
- Flores A.,
- et al.
- Kim D.,
- Shim C.Y.,
- Cho I.J.,
- et al.
- Adams H.P.,
- Bendixen B.H.,
- Kappelle L.J.,
- et al.
- Pathan F.,
- D'Elia N.,
- Nolan M.T.,
- Marwick T.H.,
- Negishi K.
- Hammerstingl C.,
- Schwekendiek M.,
- Momcilovic D.,
- et al.
- Alonso A.,
- Krijthe B.P.,
- Aspelund T.,
- et al.
- Alamneh E.A.,
- Chalmers L.,
- Bereznicki L.R.