Incremental value of the combined brain-cardiac CT protocol on prediction of atrial fibrillation after stroke

Abstract

Introduction:

Atrial fibrillation (AF) is one of the most common causes of ischemic stroke. It is essential to target patients at highest risk of AF detected after stroke (AFDAS), who should benefit from a prolonged rhythm screening strategy. Cardiac-CT angiography (CCTA) was added to the stroke protocol used in our institution in 2018. We sought to assess, for AFDAS, the predictive value of atrial cardiopathy markers by a CCTA performed on admission for acute ischemic stroke.

Patients and Methods:

From November 2018 to October 2019, consecutive stroke patients with no history of AF were included. Let atrial volume (LAV), epicardial adipose tissue (EAT) attenuation and volume, and LAA characteristics were measured on CCTA. The primary endpoint was the presence of AFDAS at follow-up, diagnosed by continuous electrocardiographic monitoring, long-term external Holter monitoring during hospital stay, or implantable cardiac monitor (ICM).

Results:

Sixty of the 247 included patients developed AFDAS. Multivariable analysis shows independent predictors of AFDAS: age >80 years (HR 2.46; 95%CI (1.23–4.92), p = 0.011), indexed LAV >45 mL/m² (HR 2.58; 95%CI (1.19–5.62), p = 0.017), EAT attenuation > −85HU (HR 2.16; 95%CI (1.13–4.15), p = 0.021) and LAA thrombus (HR 2.50; 95%CI (1.06–5.93), p = 0.037). Added consecutively to AFDAS prediction AS5F score (combining age and NIHSS >5), these markers had an incrementally better predictive value compared with the global Chi² of the initial model (p = 0.001, 0.035, and 0.015 respectively).

Discussion and conclusion:

Adding CCTA to the acute stroke protocol to assess markers of atrial cardiopathy associated with AFDAS may help to better stratify the AF screening strategy, including the use of an ICM.

Graphical abstract.

Introduction

Atrial fibrillation (AF) is the most prevalent cardiac arrhythmia, and it is associated with four to five fold increased risk of stroke. ¹ It is therefore critical to screen for AF detected after stroke (AFDAS). ² Current stroke and cardiac guidelines recommend an ECG-monitoring for at least 24 h ³ in the acute phase and prolonged cardiac monitoring, either external (EMBRACE study ⁴ ) or implantable (CRYSTAL-AF) in cryptogenic stroke. ¹ However, these devices are expensive and not universally available, and AF will be ultimately detected in one third of cases. Consequently, a better selection of patients most at risk of AFDAS is required.

Recently, the concept of atrial cardiopathy has emerged defined by a set of structural abnormalities that lead to thrombogenesis and AF. ⁵ Recent studies showed that the implementation of cardiac CT angiography (CCTA) in the acute stroke imaging protocol led to the diagnosis of LAA thrombus in 11% of confirmed ischemic strokes ⁶ and had a superior diagnostic yield compared to transthoracic echocardiography for the detection of high-risk cardio-aortic sources of embolism. ⁷ We thus hypothesized that CCTA parameters of atrial cardiopathy at admission for ischemic stroke could be used for the early identification of patients at high risk of developing AFDAS during follow-up.

We aimed to evaluate the association between AFDAS and imaging markers of atrial cardiopathy as assessed on cardiac CT added to the acute stroke imaging protocol.

Materials and methods

Patients

This post-hoc analysis is based on the population of the previous study by Bernard et al. ⁶ that included 324 patients with an imaging-confirmed diagnosis of stroke based on a cohort of 875 consecutive patients with suspected acute ischemic stroke or transient ischemic attack (TIA) who were referred to our University Hospital between November 2018 and October 2019 (Flow chart in Figure 1). Ethical approval was not sought for the present study according to institutional policy regarding retrospective studies, and this study was completed in accordance with the Helsinki Declaration as revised in 2013.

Figure 1.

Flow chart of the study.

AF: atrial fibrillation; CEM: continuous electrocardiographic monitoring; ICM: implantable cardiac monitor.

Exclusion criteria: standard exclusion criteria for contrast-enhanced CT were applied (i.e. allergic reaction to iodinated contrast, kidney disease, age less than 18 years, and pregnancy) and intracranial hemorrhage. Patients from peripheral hospitals not hospitalized at our University Hospital were excluded for missing follow-up data. Finally, patients were excluded if there was too much artifact or if the CCTA was not interpretable. In addition to the initial exclusion criteria, we also excluded patients with known AF before stroke for the present analysis, in order to focus only on the prediction of AFDAS (n = 68).

Cardiac computed tomography angiography

A CCTA was added to the acute stroke emergency imaging protocol for all patients admitted with suspected stroke by using a 320 detectors scanner allowing volume acquisition over 16 cm (Aquilion ONE GENESIS, Canon Medical Systems, Otawara, Japan). ⁶ This process provides acquisition of the cardiac mass in single rotation synchronized to one heartbeat.

Image analysis

CT post-treatment was performed by a radiologist using a workstation (Syngo.via^®, Siemens Healthineers, United States, version 5.2) that allows the volumetric tracing (Supplemental Figure 1). A prospective volumetric acquisition integrated with CANON AIDR technology was performed. All images were then reviewed by a cardiac imaging expert blinded to the previous data set.

Epicardial adipose tissue (EAT) volume and attenuation was determined as follows: on the axial slices, we contoured the pericardium semi-automatically with artificial intelligence and created a volume. Based on previous studies, ⁸ we defined a window from − 200 to − 50HU range, so only fat tissue was kept. EAT volume and attenuation were automatically calculated by the software. Interobserver agreement, measured on a sample of 50 CT, was excellent with an intraclass correlation coefficient of 0.91 (0.86–0.95) for the mean EAT attenuation and 0.98 (0.96–0.99) for the EAT volume.

LA volume (LAV) and LA surface were assessed using the same semi-automatically contouring method. ⁸ This measurement was adjusted to body surface area to obtain the indexed LAV (LAVI).

Data collection

Demographic data, cardiovascular risk factors, clinical and biological data collected within 48 h of admission were retrospectively collected after review of the electronical medical records. The NIHSS score was used to grade the severity of stroke. ⁹

AF was defined according to European guidelines. ¹⁰ Based on our institution’s AF screening protocol, AFDAS detection was performed as follows: continuous electrocardiographic monitoring during the stroke unit stay, and long-term Holter-ECG during hospitalization in the neurology department. If no arrhythmia was detected and no etiology found for the ischemic stroke after the diagnostic workup, an ICM was placed, as recommended in case of cryptogenic stroke. AFDAS was defined as a newly detected AF after ischemic stroke in patients without known AF, that is, any AF detected on admission ECG, inpatient cardiac monitoring, or outpatient AF screening technologies. ² We calculated the AS5F score, by taking into account age and NIHSS >5, and used the cut-off of 67.5 to classify patients at high risk of AF based on the work of Uphaus et al. ¹¹

Follow up

Patients were contacted by phone at 3 months and seen for an outpatient visit at 6 months. Data was collected for vital status, current treatments, and cardiovascular events.

Patients implanted with an ICM had a follow-up cardiology consultation at 6 weeks and then every 3 months or by remote monitoring.

Statistical analysis

Continuous data were expressed as medians (25th–75th percentile) and dichotomous data as numbers (percentages). A Mann-Whitney test or Student’s t-test was used to compare continuous data, and the Chi-square test or Fisher’s test was used for dichotomous data. The optimal threshold to discriminate AF from the continuous data of interest was obtained with the receiver-operating characteristic (ROC) curve with the best sensitivity and specificity according to the Youden index. The Kaplan-Meier survival statistics was used to assess AFDAS-free. Data was censored at the date of the AF episode, the last date of follow-up (12/31/2020), or death. We used a multiple-imputation procedure with five imputations to replace missing values for continuous covariates defined for adjustment. Variables entered into the multivariate model were chosen according to their univariate relationship, with an inclusion cut-off at 10% and exclusion cut-off at 5%. The proportional hazard’s assumption was met for all variables included in the Cox analysis. Univariate and multivariable Cox proportional hazard regression analysis were used to identify factors associated with AFDAS. The multivariate model was tested for multicollinearity and was found to be stable. Moreover, to test for the incremental predictive value of CCTA parameters with regard to classical predictors of AFDAS, models based the on the AS5F score and on clinical variables were built. ¹¹ We then consecutively added these the three CCTA parameters to these models and the changes in the global chi-square of the model were established. A p value <0.05 was considered statistically significant. Analyses were performed using SPSS software (26.0, IBM Inc., USA).

Results

Baseline characteristics

After a median follow-up of 639 (540–708) days, AF was diagnosed in 60 (24%) of the 247 included patients. AFDAS Kaplan-Meier rate estimates at 1 and 2 years of follow-up were 23% and 27% (Figure 2). Table 1 describes the clinical, biological and computed tomography characteristics of the 247 included patients according to AFDAS diagnosis.

Figure 2.

AFDAS Kaplan-Meier rate estimates at 1 and 2 years of follow-up were 23% and 27% rates among the 247 included patients.

Table 1.

Clinical, biological and CT characteristics of the controls and patients with AFDAS.

	SR (n = 187)	AFDAS (n = 60)	p-value
Demographic characteristics
Age (years)	71 ± 15	81 ± 10	<0.001
Age >80	55 (30%)	41 (68%)	<0.001
Female sex	78 (42%)	35 (58%)	0.025
Cardiovascular risk factors
Hypertension	121 (65%)	41 (68%)	0.607
Diabetes	42 (23%)	6 (10%)	0.039
BMI (kg/m2), (n = 218)	25.9 (23.1–27.8)	26.6 (23.1–29.2)	0.306
Current smoking	29 (16%)	4 (7%)	0.123
Dyslipidemia	63 (34%)	16 (27%)	0.334
Other comorbidities
Obstructive sleep apnea syndrome	13 (7%)	2 (3%)	0.533
Peripheral artery disease	14 (8%)	3 (5%)	0.770
Coronary artery disease	25 (13%)	11 (19%)	0.325
Previous anticoagulation therapy	10 (5%)	1 (2%)	0.304
Biological data at admission
NT-proBNP (ng/L), (n = 154)	314 (92–1 426)	1 263 (559–3 037)	<0.001
Troponin > 0.02 ng/L	56 (33%)	30 (2%)	0.011
CRP (mg/L)	4.7 (0–18.5)	5.6 (0–31.8)	0.244
Mean platelet volume (fL)	10.2 (9.8–11.0)	10.6 (10.0–11.2)	0.042
Clinical data at admission
Heart rate (bpm)	77 (65–89)	76 (66–90)	0.948
NIHSS score	4 (2–10)	7 (4–15)	0.006
AS5F score	67.9 (59.9–75.1)	75.9 (71.3–84.1)	<0.001
AS5F score >67.5	83 (51)	46 (87)	<0.001
TOAST classification			<0.001
1 – large-artery atherosclerosis	67 (36%)	0
2 – cardio-embolism	23 (12%)	44 (73%)
3 – small-vessel disease	10 (5%)	0
4 – stroke of other determined cause	27 (14%)	0
5 – stroke of undetermined cause	60 (32%)	16 (27%)
Computed tomography data
LAA thrombus	5 (3%)	10 (17%)	<0.001
LA surface (cm2)	20 (16.7–23.7)	26 (20.9–29.4)	<0.001
LA volume (cm3)	74.5 (58.3–89.3)	102 (80.6–131.5)	<0.001
LAVI (mL/m2), (n = 217)	41.5 (32.8–48.8)	55.5 (45.5–71.8)	<0.001
LAVI >45 mL/m2	63 (38%)	42 (81%)	<0.001
Mean EAT attenuation (HU)	−86 ± 5	−85 ± 6	0.086
EAT attenuation > −85HU	74 (40%)	33 (55%)	0.035
EAT volume (cm3)	61 (39.4–92.1)	60 (35.7–94.6)	0.749

AFDAS: atrial fibrillation diagnosed after stroke; BMI: body mass index (kg/m²); CT: computed tomography; EAT: epicardial adipose tissue; IQR: interquartile range; LAA: left atrial appendage; LAVI: left atrial volume indexed; NIHSS: national institutes of health stroke scale; SD: standard deviation; TOAST: trial of ORG 10172 in acute stroke treatment.

n (%), mean (± SD), median (IQR).

Compared to sinus rhythm patients (SR), AFDAS patients were significantly older (81 ± 10 vs 71 ± 15 years old, p < 0.001) and more often female (58% vs 42%, p = 0.025). Moreover, AFDAS patients had diabetes less often that SR patients (10% vs 22.5%, p = 0.039) and higher NIHSS score at admission (7 (4–15) vs 4 (2–10), p = 0.006).

The two groups differed according to several biological parameters: compared to SR patients, AFDAS patients had significantly higher NT-proBNP levels (1 263 (559–3 037) vs 314 (92–1 426) ng/L, p < 0.001) and higher mean platelet volume (10.6 (10.0–11.2) vs 10.2 (9.8–11.0) fL, p = 0.042), and they were more likely to have troponin above normal value (>0.02 ng/L) (51.7% vs 32.9%, p = 0.011).

Regarding CT data, compared to SR patients, AFDAS patients did not significantly differ on EAT volume (61 (39.4–92.1) vs 60 (35.7–94.6) cm³, p = 0.749) nor on mean EAT attenuation (− 86 ± 5 vs − 85 ± 5HU, p = 0.086). However, the AFDAS group had higher LAV (74.5 (58.3–89.3) vs 102 (80.6–131.5) mL, p < 0.001), LAVI (41.5 (32.8–48.8) vs 55.5 (45.5–71.8) mL/m² with p < 0.001) and LA surface (20 (16.7–23.7) vs 26 (20.9–29.4) cm², p < 0.001) than the SR group.

Thresholds determination for markers of atrial cardiopathy

We performed ROC curves analyses to assess the relationship and the best cut-off values between AFDAS and the clinical, biological and imaging markers of atrial cardiopathy. After ROC curve analysis, the best predictive values for AFDAS were 80 years for age, 85 mL for LAV, 45 mL/m² for LAVI, − 85HU for mean EAT attenuation, and 10.25 fL for mean platelet volume.

Multivariable analysis

Among the variables significantly associated with AFDAS in bivariate analysis, age >80 years old (HR 2.46; 95%CI: 1.23–4.92, p = 0.011), LAVI >45 mL/m² (HR 2.58; 95%CI: 1.19–5.62, p = 0.017), mean attenuation of EAT > −85HU (HR 2.16; 95%CI: 1.13–4.15, p = 0.021) and LAA thrombus (HR 2.50; 95%CI: 1.06–5.93, p = 0.037) were independently associated with AFDAS (Table 2). Moreover, when added consecutively to the AS5F score for AFDAS prediction (combining age and NIHSS score >5), the three CCTA markers had an incrementally better predictive value compared with the global Chi ² of the initial model (p = 0.001, 0.035, and 0.015, respectively) (Figure 3).

Table 2.

Univariate and multivariable Cox proportional hazards models of AFDAS-associated factors.

	Univariate			Multivariate
	HR	95%CI	p-value	HR	95%CI	p-value
Age >80 years	4.26	2.47–7.36	<0.001	2.46	1.23–4.92	0.011
Female sex	1.87	1.11–3.15	0.018	1.40	0.68–2.88	0.361
Diabetes	0.42	0.18–0.98	0.045	0.77	0.30–2.03	0.777
NIHSS score	1.05	1.02–1.09	0.005	1.02	0.96–1.07	0.531
Intracerebral thrombus	2.26	1.33–3.82	0.002	1.92	0.90–4.08	0.091
LAA thrombus	4.40	2.22–8.74	<0.001	2.50	1.06–5.93	0.037
Mean density of EAT > −85HU	1.92	1.10–1.33	0.021	2.16	1.13–4.15	0.021
Troponin >0.02 ng/mL	2.14	1.28–3.59	0.004	1.26	0.66–2.38	0.476
Mean platelet volume >10.25 fL	1.65	0.98–2.80	0.062	1.22	0.64–2.31	0.543
LAVI >45 mL/m2	5.47	2.74–10.9	<0.001	2.58	1.19–5.62	0.017

AFDAS: atrial fibrillation detected after stroke; CI: confidence interval; EAT: epicardial adipose tissue; HR: hazard ratio; LAA: left atrial appendage; LAVI: left atrial volume indexed.

Bold indicates the significance of p < 0.05

Figure 3.

Incremental value of atrial cardiopathy markers on cardiac CT atrial cardiopathy to predict AFDAS. Bar graph illustrating the change in global x² value by the addition of LAVI >45 mL/m², EAT density > −85HU and LAA thrombus to AS5F score. All models were significantly associated with AFDAS (p < 0.05). LAVI: left atrial volume index; EATa: epicardial adipose tissue attenuation; LAAt: left atrial appendage thrombus.

Sensitivity analysis

To confirm our results, we performed two sensitivity analyses using continuous data rather than dichotomized variables, and including variables with missing data not previously included in the main analysis:

• - Multivariate analysis on the 125 patients for whom the variables analyzed were complete:

• ○ An initial clinical model combining AF5S, diabetes, NT proBNP, troponins, female sex. After conditional stepwise regression, only the AS5F score remained significantly associated with AFDAS at this first stage (HR 1.06; 95%CI: 1.032–1.089, p < 0.001)

• ○ Addition of the CT variables (we did not included echocardiographic data at this stage because of very limited available echocardiographic data): consecutive addition of LAVI (p = 0.033) and mean EAT attenuation (p = 0.035) as continuous variables and of LAAt (p = 0.030) had incremental predictive value over AS5F.

• - A second approach based on the pooled dataset generated by multiple imputations methods for missing data:

• ○ Once again, only the AS5F score was retained at the first stage ((HR 1.064; 95%CI: 1.038–1.091, p < 0.001)

• ○ Addition of LAVI (p = 0.17) or LAV (p = 0.39) assessed by echocardiography did not improve the predictive performance of AS5F. However, the consecutive addition of LAVI (CT) (p < 0.001) and mean EAT attenuation (p < 0.001) as continuous variables and of LAAt (p < 0.001) had incremental predictive value over the AS5F score, confirming our previous results.

Discussion

We demonstrate in the present study that the addition of a CCTA to the acute stroke imaging protocol can be used to identify strong determinants of AFDAS at follow-up. Moreover, the association of EAT attenuation, LAVI and LAA thrombus with AFDAS at follow-up highlight their potential role as early markers of atrial cardiopathy.

AFDAS incidence

The incidence of AFDAS in our population (24%) was consistent with the rates described in a recent meta-analysis in which the overall AFDAS detection rate was 23.7% after all external monitoring phases. ¹² The diagnostic rate could be increased when ICM data is taken into account. For instance, in the SAFAS study, we found an AFDAS rate of 32% when all methods of AF monitoring were included. ¹³

Epicardial adipose tissue attenuation and atrial fibrillation

Epicardial fat plays a major role in AF and coronary artery disease progression. ¹⁴ The pathophysiological mechanism involves various mediators including free fatty acids ¹⁵ and atrial natriuretic peptide. ¹⁶

We found higher rates of EAT attenuation in patients developing AFDAS at follow-up compared to SR patients, and we described a cut-off of EAT attenuation > −85HU as independently associated with AFDAS in stroke patients, reflecting the inflammatory character of the EAT. This is corroborated by the study of Gaibazzi et al., ¹⁷ though with slightly different HU interval for EAT definition, ranging from −190 to −30HU.

Many studies have focused on the association between EAT volume and AF, but there have been considerable differences in their results. For example, a cut-off of 40.68 cm³ was found in Tsao et al. ⁸ compared with 68.1 cm³/m² in Shmilovich et al. ¹⁸ EAT attenuation appears to be more reproducible, which supported by our study in which EAT volume does not appear as a significant variable. ¹⁷ Others studies have also shown the impact of EAT localization, especially on the posterior wall of the LA, as a predictor of AF and cardiovascular events including stroke and tried to quantify it by different modalities (MRI, echocardiography, CCTA).^8,14,15,19

Left atrial dilatation and AFDAS

Several studies have demonstrated the independent association between LAV and AF. This variable has been included in some predictive scores, as the iPAB score. ²⁰ One of these studies also highlighted the association between LAVI measured on CT and AF in stroke patients. ²¹ Thus, our study is the first to propose a CCTA threshold (LAVI >45 mL/m²) for LAVI as an associated factor of AFDAS. In our work, this cut-off was associated with a more than doubled risk of AFDAS at follow-up, independently of the other predictors.

LAA thrombus and AFDAS

We found that the presence of a thrombus within LAA was independently associated with AFDAS. These results are consistent with previous studies suggesting that LAA thrombus results from atrial cardiopathy: dilatation of the atrium, endothelial dysfunction, and blood stasis are all factors that contribute to clot formation, as shown by Tsao et al. ⁸ Our previous work also suggested the benefit of early diagnosis of LAA thrombus in the initiation of anticoagulation therapy without additional risk of bleeding. ⁶ Randomized studies comparing early versus late anticoagulation in patients with AF related stroke are ongoing and will provide safety as well as efficacy robust data on these strategies (ELAN ²² and LASER ²³ trials)

Age, NIHSS score and AFDAS: the AS5F score

Several studies that looked for association between EAT and AF included an elderly population, with a mean age of about 70 years, which is in line with the result of our study.^8,14 Hsieh et al. demonstrated that the predictive score for AF on 72 h Holter monitoring post-stroke (called the AS5F) developed by Uphaus et al., ¹¹ integrating only two criteria: age and stroke severity based on the NIHSS score has an adequate discrimination for AFDAS compared to seven other scores. ²⁴ The successive addition of the three imaging markers of atrial cardiopathy highlighted by our study to the AS5F score resulted an incrementally better predictive value of AFDAS compared with the AS5F score. This result suggests that the current AFDAS predictive scores could be easily improved by taking into account the CCTA parameters assessed at patient admission.

Limitations

First, the single center and retrospective nature of this study could imply a selection bias. Second, only a small number of centers perform emergency CCTA in patients with suspected stroke, which limits the generalization of this practice for now. Moreover, our cut-off of the mean EAT attenuation is measured on enhanced cardiac CT, which cannot be extrapolated to a non-enhanced cardiac CT for calculating calcium score by example and creates variations according to the injection conditions. Third, we measured the total volume of EAT and not only the atrial or even posterior wall atrial EAT, which seems to be play a more specific pathological role in AF.^8,17 Fourth, electrocardiographic markers of atrial cardiopathy such as P-wave duration or P wave terminal force in V1 were not collected and could not be implemented in our models. Thus, we were not able to assess atrial cardiopathy as defined in the ongoing ARCADIA trial, that will test a systematic anticoagulation of cryptogenic stroke patients with atrial cardiopathy markers. ²⁵

Conclusion

The addition of cardiac CT to the acute stroke protocol provides an early assessment of markers of atrial cardiopathy that include left atrial indexed volume, epicardial fat attenuation and left atrial appendage thrombus. These parameters, in addition to an age >80 years old, were independently associated with AFDAS at follow-up. Our results could help clinicians to better stratifying the AF screening strategy to be implemented rapidly after stroke, including the use of an implantable cardiac monitor.