Identifying Patients at High Risk of Left Atrial Appendage Thrombus Before Cardioversion: The CLOTS‐AF Score

Abstract

Background

Transesophageal echocardiography–guided direct cardioversion is recommended in patients who are inadequately anticoagulated due to perceived risk of left atrial appendage thrombus (LAAT); however, LAAT risk factors remain poorly defined.

Methods and Results

We evaluated clinical and transthoracic echocardiographic parameters to predict LAAT risk in consecutive patients with atrial fibrillation (AF)/atrial flutter undergoing transesophageal echocardiography before cardioversion between 2002 and 2022. Regression analysis identified predictors of LAAT, combined to create the novel CLOTS‐AF risk score (comprising clinical and echocardiographic LAAT predictors), which was developed in the derivation cohort (70%) and validated in the remaining 30%. A total of 1001 patients (mean age, 62±13 years; 25% women; left ventricular ejection fraction, 49.8±14%) underwent transesophageal echocardiography, with LAAT identified in 140 of 1001 patients (14%) and dense spontaneous echo contrast precluding cardioversion in a further 75 patients (7.5%). AF duration, AF rhythm, creatinine, stroke, diabetes, and echocardiographic parameters were univariate LAAT predictors; age, female sex, body mass index, anticoagulant type, and duration were not (all P>0.05). CHADS₂VASc, though significant on univariate analysis (P<0.001), was not significant after adjustment (P=0.12). The novel CLOTS‐AF risk model comprised significant multivariable predictors categorized and weighted according to clinically relevant thresholds (Creatinine >1.5 mg/dL, Left ventricular ejection fraction <50%, Overload (left atrial volume index >34 mL/m²), Tricuspid Annular Plane Systolic Excursion (TAPSE) <17 mm, Stroke, and AF rhythm). The unweighted risk model had excellent predictive performance with an area under the curve of 0.820 (95% CI, 0.752–0.887). The weighted CLOTS‐AF risk score maintained good predictive performance (AUC, 0.780) with an accuracy of 72%.

Conclusions

The incidence of LAAT or dense spontaneous echo contrast precluding cardioversion in patients with AF who are inadequately anticoagulated is 21%. Clinical and noninvasive echocardiographic parameters may identify patients at increased risk of LAAT better managed with a suitable period of anticoagulation before undertaking cardioversion.

Nonstandard Abbreviations and Acronyms

AFL

atrial flutter

DCR

direct cardioversion

DOAC

direct oral anticoagulant

LAA

left atrial appendage

LAAT

left atrial appendage thrombus

LAVI

left atrial volume index

SEC

spontaneous echo contrast

TAPSE

tricuspid annular plane systolic excursion

TEE

transesophageal echocardiography/echocardiogram

Clinical Perspective.

What Is New?

• Noninvasive clinical and echocardiographic parameters can predict left atrial appendage thrombus risk in patients with atrial fibrillation/atrial flutter undergoing transesophageal echocardiography–guided direct cardioversion.

• Despite this, there is a distinct lack of a universal risk stratification schema to guide the need for preprocedural transesophageal echocardiographic imaging.

• The CLOTS‐AF risk model is readily applied in the clinical context and outperformed the CHADS₂VASc score with regard to left atrial appendage thrombus risk prediction.

What Are the Clinical Implications?

• A novel risk score incorporating noninvasive clinical and echocardiographic parameters may identify high‐risk individuals in whom a period of anticoagulation should be initiated rather than undertaking a strategy of early transesophageal echocardiography–guided cardioversion, with a high likelihood of left atrial appendage thrombus.

• External validation could clarify the broader clinical utility of the proposed risk model in diverse populations and clinical settings.

Atrial fibrillation (AF) is associated with a 5‐fold increased risk of stroke and thromboembolism. ¹ Left atrial appendage thrombus (LAAT) is a common source of emboli and has been implicated in up to 90% of AF‐related strokes. ² To minimize the risk of thromboembolism, guideline‐directed anticoagulation is required for 3 weeks before direct cardioversion (DCR) or a preprocedural transesophageal echocardiogram (TEE) is recommended to facilitate an expedited cardioversion. ³ LAAT is present in up to 2.7% of patients with AF/atrial flutter (AFL) despite guideline‐directed anticoagulation and in up to 23% in patients with inadequate anticoagulation. ⁴ , ⁵ , ⁶ , ⁷ Therefore, identifying patients with a high probability of LAAT and selecting a suitable period of therapeutic anticoagulation rather than early TEE‐guided DCR, may minimize procedural risk to the patient and promote better health care usage by reducing cost and resources associated with TEE, to reduce low‐value care.

The CHADS₂VASc score has been widely adopted to predict stroke risk among the AF population, although the predictive value of 0.67 is relatively modest. ⁸ Although a higher CHADS₂VASc score is a reliable indicator of thromboembolic risk in AF, a lower CHADS₂VASc score has a specificity as low as 30% to 55%. ⁹ Prior studies have demonstrated a modest predictive value for the CHADS₂VASc score in determining the presence of LAAT. ¹⁰ , ¹¹ , ¹² , ¹³

The aim of the present study was to identify noninvasive clinical and echocardiographic predictors of LAAT and dense spontaneous echo contrast (SEC) in a large population of consecutive patients with AF/AFL undergoing TEE in whom guideline‐directed anticoagulation recommendations were not satisfied.

Methods

Availability of Data

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Study Population

From January 8, 2002, to January 12, 2022, 1001 consecutive patients with AF/AFL undergoing preprocedural TEE before DCR (referred for TEE on the basis of inadequate anticoagulation or subtherapeutic international normalized ratio) were retrospectively identified from a large tertiary referral center (The Alfred Hospital, Melbourne, Australia). The study period was defined on the basis of the first available TEE study retrieved from the local institutional echocardiography database. The rationale for a TEE‐guided approach to cardioversion was attributed to the following reasons: an inadequate period of preprocedure systemic anticoagulation defined as <3 weeks of uninterrupted direct oral anticoagulant (DOAC) or subtherapeutic anticoagulation (international normalized ratio <2 within the 4 weeks before DCR in those on warfarin ¹⁴ ) in those with AF/AFL lasting >48 hours or of unclear duration (n=611); documented nonadherence with systemic anticoagulation (n=208), no systemic anticoagulation (n=104); subtherapeutic anticoagulation in the setting of a reported relative contraindication (n=30); and uncertain or undocumented reasons (n=48).

We excluded individuals undergoing TEE for alternate indications unrelated to AF/AFL and removed duplicate studies to form the final study cohort. Available baseline demographic, clinical, and transthoracic echocardiography parameters were obtained from electronic medical records. The CHADS₂VASc score was calculated for each individual according to standard definitions. ⁸ Anticoagulation prescribing was in accordance with guideline recommendations at the relevant time periods. Baseline transthoracic echocardiographic parameters were included if performed within 12 months of the TEE date. The study was approved by the institutional ethics review board at the Alfred Hospital, Melbourne, Australia, and a waiver of consent was granted on the basis of the retrospective nature of this study.

TEE Image Acquisition

TEE was performed using either an X7‐2t or X8‐2t Phillips transducer (Phillips iE33; Phillips Medical Systems, Andover, MA) at the time of DCR by an experienced echocardiologist to assess for the presence of LAAT or SEC. SEC was graded from 0 to 4 according to a published classification schema (SEC grade 0 corresponds to absence of echogenicity; 1 refers to minimal echogenicity; 2 refers to mild–moderate more dense swirling; 3 refers to moderate dense swirling pattern in the left atrial appendage (LAA) detected throughout the cardiac cycle; 4 refers to severe, “intense echogenicity and very slow swirling pattern in the LAA, usually with a similar density in the main cavity”). ¹⁵ Dense SEC was defined as a swirling echodensity present throughout the cardiac cycle with or without a viscid appearance (termed sludge) in the absence of a discrete mass with appropriate gain settings to avoid artifact. ¹⁵

The LAA was imaged in multiple views including X‐plane to identify thrombus. LAA emptying velocities were recorded from the formal TEE report and manually verified on the procedural imaging database. LAA emptying velocity was recorded by placing the pulse‐wave Doppler gate within 1 cm of the LAA orifice, and emptying velocity was considered reduced if the value was <0.4 m/s.

Patients with definite thrombus or dense SEC (defined as moderate or severe SEC according to previously published SEC classification schema ¹⁶ ) that precluded cardioversion were classified as the LAAT cohort. This was based on previous studies demonstrating higher rates of stroke and thromboembolism among patients with AF with moderate/severe SEC according to conventional classification systems. ¹⁷ The decision to alter anticoagulation after a diagnosis of LAAT was at the discretion of the treating physician.

A subgroup of individuals with LAAT underwent repeat TEE imaging a minimum of 4 weeks later at the treating physician's discretion. We sought to determine clinical and imaging characteristics associated with thrombus resolution in this subpopulation.

Statistical Analysis

The study population was divided according to the presence or absence of LAAT (comprising thrombus or dense SEC). Baseline characteristics are summarized as mean±SD for continuous variables (median and interquartile ranges are reported for nonparametric continuous data where applicable), and frequencies and percentages were presented for categorical variables. The chi‐square, Wilcoxon rank‐sum, or unpaired t‐test was used as appropriate to examine baseline differences between those with and without LAAT. Echocardiographic indices including left ventricular ejection fraction (LVEF), left atrial volume index (LAVI), and LAA emptying velocity were collected as continuous variables and were also dichotomized on the basis of clinically relevant thresholds for normal ranges. ¹⁸

Assessment of Covariates

Baseline clinical and echocardiographic characteristics were collected and individualized and were classified according to LAAT status. The variables of interest for evaluation were selected a priori on the basis of published data and clinical relevance and included clinical variables (AF duration; baseline rhythm [AF or AFL]; anticoagulation duration before TEE; body mass index; renal function; and components of the CHADS₂VASc score, namely, heart failure, hypertension, age, diabetes, vascular disease, and sex). and echocardiographic parameters, including left and right ventricular (RV) systolic function, atrial and ventricular chamber dimensions, and parameters of diastolic filling. Univariate logistic regression was used to identify independent predictors of LAAT, and significant predictors were then evaluated in a multivariable logistic model. In this model, LAAT was the dependent/outcome variable, whereas potential predictors of LAAT were the independent variables. LAAT risk was expressed as odds ratio (OR) with 95% CI.

Cross validation was performed using the set seed function to divide the study population into derivation (70%) and validation (30%) cohorts. AF rhythm, prior stroke, creatinine, LVEF, LAVI, and tricuspid annular plane systolic excursion (TAPSE) remained significant predictors of LAAT on multivariable logistic regression. TAPSE was selected as a recognized and readily measured marker of RV systolic function. Each variable was then dichotomized on the basis of universally recognized clinically relevant thresholds and confirmed using Youden's index as follows: LVEF <50%, TAPSE <17 mm, LAVI >34 mL/m², and creatinine >1.5 mg/dL to indicate abnormal ranges. ¹⁸ Creatinine >1.5 mg/dL was selected on the basis of the heightened thromboembolic risk in renal impairment and a readily recognized clinical threshold for renal DOAC dose reduction established from clinical trial data. ¹⁹

Regression analysis yielded beta coefficient values for each variable. The weighted CLOTS‐AF risk score (Creatinine >1.5 mg/dL, LVEF <50%, Overload (LAVI >34 mL/m²), TAPSE <17 mm, prior Stroke, and AF rhythm) was developed using the beta coefficient rounded to the nearest integer, with each dichotomous variable in the model assigned a weighted score with a total point range of 0 to 12. A new variable was created combining each weighted covariate for the purpose of multivariable regression. Generalized linear regression was used to evaluate the predictive performance of the weighted risk model and compute the receiver operating characteristic curve and calculate associated CIs and threshold sensitivity and specificity.

The CHADS₂VASc score (range, 0–9 points) was calculated by combining relevant individual variables, which were weighed according to standard criteria as previously described. ⁸ The CHADS₂VASc score was evaluated as a continuous variable using univariate regression analysis and then categorized as CHADS2VASc ≥2 according to the recognized clinical threshold for a heightened (2.9‐fold) thromboembolic risk where anticoagulation is recommended. ²⁰ Generalized linear regression for receiver operating characteristic curve analysis was used to evaluate the discriminatory capacity of the CHADS₂VASc score in predicting LAAT, which was expressed as area under the curve (AUC) and CI.

The discriminatory performance of both the unadjusted and weighted models were evaluated using the area under the receiver operating characteristic curve. The CLOTS‐AF risk model was compared with the CHADS₂VASc score to determine the discriminatory performance to predict LAAT risk using the area under the receiver operating characteristic and DeLong methods, ²¹ with a DeLong P value <0.05 considered a statistically significant difference between model performance. ²² A 2‐tailed P value was set at <0.05 for statistical significance. All analyses were performed using R version 4.2.0 (R Core Team).

Results

Baseline Characteristics

The study population consisted of 1001 individuals (mean age, 62±13 years; 25% women; LVEF, 49.8±14%) undergoing TEE before DCR. The rhythm at the time of TEE was AF in 688 (68.7%) and AFL in 314 (31.3%). LAAT was detected in 140 (14%) and dense SEC in an additional 75 (7.5%) precluding DCR. The LAAT cohort (LAAT or dense SEC resulting in an abandoned procedure) had a significantly higher prevalence of heart failure, renal impairment, diabetes, prior stroke or transient ischemic attack, peripheral vascular disease, and higher CHAD_S2VASc score (P<0.001). Baseline characteristics and pharmacotherapy according to LAAT status is outlined in Table 1.

Table 1.

Baseline Characteristics According to LAAT Status

Baseline characteristics	No LAAT (N=786)	LAAT (N=215)	P value
Age, y	61±13	63±13	0.15
Female, n (%)	195 (25)	57 (27)	0.60
BMI, kg/m2	29±6	30±7	0.70
Rhythm at TEE			<0.001
AF, n (%)	519 (66)	183 (85)	<0.001
AFL, n (%)	267 (34)	32 (15)
PsAF duration, d	118±237	216±387	0.005
Anticoagulation <30 d, n (%)	241 (57)	116 (62)	0.20
Heart failure, n (%)	205 (26)	119 (55)	<0.001
Prior MI, n (%)	122 (16)	44 (20)	0.084
Creatinine, μmol/L	91±40	108±65	<0.001
OSA, n (%)	65 (8.3)	17 (7.9)	0.90
Diabetes, n (%)	118 (15)	46 (21)	0.025
Obesity (BMI ≥30 kg/m2)	277 (35)	79 (37)	0.70
Hypertension, n (%)	385 (49)	112 (52)	0.40
Hyperlipidemia, n (%)	255 (32)	81 (38)	0.20
Cardiac device, n (%)	52 (6.6)	36 (17)	<0.001
Stroke/TIA, n (%)	34 (4.3)	29 (13.5)	<0.001
PVD, n (%)	24 (3.1)	17 (7.9)	0.001
CHADS2VASc score, mean±SD	1.9±1.4	2.3±1.3	<0.001
CHADS2VASc score ≥2, n (%)	435 (55)	157 (73)	<0.001
Baseline pharmacotherapy
Anticoagulant type			<0.001
Apixaban, n (%)	155 (19.7)	29 (13.5)
Dabigatran, n (%)	37 (4.7)	5 (2.3)
Rivaroxaban, n (%)	117 (15)	23 (10.7)
Warfarin, n (%)	170 (21.6)	133 (61.9)
None, n (%)	307 (39)	25 (11.6)
Beta blocker, n (%)	614 (78)	149 (69)	0.007
ACEi/ARB/ARNi, n (%)	398 (51)	103 (48)	0.50
Antiarrhythmics, n (%)	317 (40)	109 (51)	0.006
MRA, n (%)	96 (12)	26 (12)	0.98
Furosemide, n (%)	175 (22)	98 (46)	<0.001

ACEi indicates angiotensin‐converting enzyme inhibitor; AF, atrial fibrillation; AFL, atrial flutter; ARB, angiotensin receptor blocker; ARNi, angiotensin receptor blocker/neprilysin inhibitor; BMI, body mass index; LAAT, left atrial appendage thrombus; MI, myocardial infarction; MRA, mineralocorticoid receptor antagonist; OSA, obstructive sleep apnea; PsAF, persistent atrial fibrillation; PVD, peripheral vascular disease; and TIA, transient ischemic attack.

In those with LAAT, anticoagulation had not been commenced before TEE in 11.6%, was of inadequate duration or compliance in 64.1%, or required TEE in the setting of subtherapeutic international normalized ratio in 20%. Discrete thrombus was observed in 4.3% despite guideline‐directed anticoagulation. A histogram of pre‐TEE anticoagulation across the study population is depicted in Figure S1.

Echocardiographic Parameters

Echocardiography was available in 911 (91%) patients within 12 months of undergoing TEE. The presence of LAAT was associated with significantly lower LV and RV systolic function (LVEF, 41±14% versus 52±14; P<0.001), larger left ventricular (LV) dimension (left ventricular end‐diastolic diameter [55±16 mm versus 51±8; P<0.001]) and LV mass (LV mass index, 105±35 versus 95±27 g/m²; P<0.001) and left atrial (LA [LA diameter and LAVI, both P<0.001]) dimensions and higher filling pressures (E/e′ and RV systolic pressure both P<0.001). Most individuals with LAAT were in AF as opposed to AFL at the time of TEE (83% versus 66% in the non‐LAAT cohort; P<0.001). The LAAT cohort had lower LAA emptying velocity (25±12 versus 48±19 cm/s; P<0.001) and more transthoracic echocardiography–detected mitral regurgitation (moderate/severe, 19.8% versus 9.7%; P<0.001; Table 2).

Table 2.

Baseline Echocardiographic Characteristics

Echocardiographic characteristics	No LAAT (N=786)	LAAT (N=215)	P value
TTE parameters
SBP at TTE, mm Hg	127±18	121±18	<0.001
LVEF, %	52±14	41±14	<0.001
LV mass, g	194±60	215±74	0.001
LVEDD, mm	51±8	55±16	<0.001
E/e′	11±6	16±9	<0.001
TAPSE, cm	2.0±0.5	1.7±0.5	<0.001
RVSP, mm Hg	31±11	35±10	<0.001
LA diameter, mm	44±7	49±9	<0.001
LA area, cm2	26±6	30±7	<0.001
LAVI, mL/m2	43±14	55±21	<0.001
RA area, cm/2	20.9±5.6	24.3±7.1	<0.001
RA volume, mL	69±38	91±33	<0.001
MR grade, n (%)			<0.001
None	576 (73)	76 (35)
Mild	134 (17)	96 (45)
Moderate	68 (8.7)	37 (17)
Severe	8 (1)	6 (2.8)
TEE parameters
Rhythm at TEE			<0.001
AF, n (%)	515 (66)	179 (83)
AFL, n (%)	271 (34)	36 (17)
SBP, mm Hg	117±17	114±20	0.003
LAA velocity, cm/s	48±19	25±12	<0.001

AF indicates atrial fibrillation; AFL, atrial flutter; LA, left atrial; LAVI, left atrial volume index; LV, left ventricular; LVEDD, left ventricular end‐diastolic dimension; LVEF, left ventricular ejection fraction; MR, mitral regurgitation; RA, right atrium/atrial;

RVSP, right ventricular systolic pressure; SBP, systolic blood pressure; TAPSE, tricuspid annular plane systolic excursion; TEE, transesophageal echocardiography; and TTE, transthoracic echocardiogram.

Regression Analysis

Heart failure, diabetes, prior stroke or transient ischemic attack, peripheral vascular disease, creatinine, AF duration, CHADS₂VASc score (as a continuous variable), and warfarin use were univariate predictors of LAAT (Table 3). Age, sex, anticoagulation duration, and hypertension were not (all P>0.05). Echocardiographic indicators of cardiac remodeling were significant on univariable analysis (LV end‐diastolic diameter, LV mass, LVEF, TAPSE, LAVI, E/e′, and RV systolic pressure all P<0.001; Table 3).

Table 3.

Univariable Regression

	OR	95% CI	P value
Age	1.002	0.995–1.240	0.112
Female sex	0.982	0.885–1.090	0.611
BMI	1.001	0.993–1.008	0.869
Rhythm			<0.001
AF	1.107	1.061–1.150
AFL	0.927	0.823–1.043
PsAF duration	1.22	1.18–1.26	<0.001
Heart failure	1.333	1.218–1.459	<0.001
Hypertension	1.037	0.947–1.135	0.419
Prior MI	1.161	1.028–1.310	0.084
Diabetes	1.143	1.006–1.300	0.025
Stroke	1.359	1.112–1.661	<0.001
TIA	1.153	0.874–1.521	0.027
PVD	1.403	1.104–1.781	0.001
Creatinine	1.001	1.001–1.002	<0.001
CHADS2VASc score	1.098	1.060–1.137	<0.001
CHADS2VASc score ≥2	1.195	1.090–1.309	<0.001
Anticoagulant duration	1.02	0.922–1.130	0.182
Warfarin vs NOAC	1.51	1.340–1.699	<0.001
Rivaroxaban vs other NOAC	1.039	0.918–1.176	0.117
LVEF	0.987	0.983–0.990	<0.001
LAVI	1.007	1.004–1.009	<0.001
LVEDD	1.005	1.001–1.010	<0.001
E/e′	1.019	1.011–1.027	<0.001
LV mass	1.002	1.001–1.003	<0.001
TAPSE	0.965	0.933–0.999	<0.001
RVSP	1.014	1.008–1.020	<0.001

AF indicates atrial fibrillation; AFL, atrial flutter; BMI, body mass index; LAVI, left atrial volume index; LV, left ventricular; LVEDD, left ventricular end‐diastolic diameter; LVEF, left ventricular ejection fraction; MI, myocardial infarction; NOAC, novel oral anticoagulant; PsAF, persistent atrial fibrillation; PVD, peripheral vascular disease; RVSP, right ventricular systolic pressure; TAPSE, tricuspid annular plane systolic excursion; and TIA, transient ischemic attack.

On multivariable analysis, prior stroke/transient ischemic attack (OR, 1.432; P=0.002), creatinine (OR, 1.002; P=0.019), AF rhythm (OR, 1.210; P<0.001), LVEF (OR, 0.992; P<0.001), TAPSE (OR, 0.888; P<0.001), and LAVI (OR, 1.015; P<0.001) remained independent predictors of LAAT, whereas peripheral vascular disease (P=0.187), CHADS₂VASc score (P=0.124), warfarin (P=0.059), and other echocardiographic parameters were not (all P>0.05; Table 4). CHADS₂VASc score ≥2 did not independently predict LAAT after multivariable adjustment (P=0.269) when excluding its component covariates.

Table 4.

Multivariable Regression

	OR	95% CI	P value
Clinical parameters
AF rhythm	1.210	1.049–1.432	<0.001
PsAF duration	1.000	0.999–1.001	0.344
Diabetes	1.168	0.806–1.189	0.672
Stroke	1.432	1.089–1.881	0.002
Prior MI	0.972	0.898–1.314	0.947
PVD	1.212	0.910–1.614	0.187
Creatinine	1.002	1.001–1.003	0.019
CHADS2VASc score	1.212	0.987–1.378	0.124
Anticoagulant >30 d	0.923	0.800–1.065	0.271
Anticoagulant type
Apixaban	0.754	0.563–1.078	0.172
Dabigatran	0.853	0.658–1.106	0.228
Rivaroxaban	0.992	0.824–1.195	0.935
Warfarin	1.173	0.994–1.383	0.059
Echocardiographic parameters
LVEDV	1.001	0.999–1.003	0.127
LVEDD	0.997	0.985–1.009	0.075
LVEF	0.992	0.987–0.997	<0.001
E/e′	1.005	0.993–1.017	0.882
LV mass	0.999	0.998–1.001	0.623
LAVI	1.024	1.006–1.027	<0.001
TAPSE	0.904	0.753–0.977	<0.001
RVSP	1.008	0.998–1.015	0.424

AF indicates atrial fibrillation; LAVI, left atrial volume index; LV, left ventricular; LVEDD, left ventricular end‐diastolic diameter; LVEDV, left ventricular end‐diastolic volume; LVEF, left ventricular ejection fraction; MI, myocardial infarction; OR, odds ratio; PsAF, persistent atrial fibrillation; PVD, peripheral vascular disease; RVSP, right ventricular systolic pressure; and TAPSE, tricuspid annular plane systolic excursion.

All continuous predictive variables were categorized according to clinically recognized thresholds to dichotomize between normal and abnormal ranges as previously described. ¹⁸ , ²³ Multivariable analysis of dichotomized covariates that comprised the final model are outlined in Figure 1.

Figure 1. Odds ratios of component CLOTS‐AF covariates.

AF indicates atrial fibrillation; LAVI, left atrial volume index; LVEF, left ventricular ejection fraction; and TAPSE, tricuspid annular plane systolic excursion.

A weight‐adjusted model was derived using beta coefficients for each variable in the model, rounded to the nearest integer (Table S1) to derive the CLOTS‐AF risk score, which includes: creatinine >1.5 mg/dL (2 points), LVEF <50% (2 points), LAVI >34 mL/m² (overload, 1 point), TAPSE <17 mm (2 points), prior stroke (3 points), and AF rhythm (2 points; overall score range, 0–12 points; Figure 2).

Figure 2. CLOTS‐AF risk model.

AF indicates atrial fibrillation; LAVI, left atrial volume index; LV, left ventricular; LVEF, left ventricular ejection fraction; and TAPSE, tricuspid annular plane systolic excursion.

The predictive performance for the unadjusted risk model demonstrated an AUC of 0.820 (95% CI, 0.752–0.887). The weight‐adjusted CLOTS‐AF risk model maintained predictive performance with an AUC of 0.780 (95% CI, 0.707–0.853) and compared favorably with the CHADS₂VASc score for predicting LAAT thrombus (AUC, 0.619 [95% CI, 0.562–0.676]; DeLong Z=−2.427; P=0.015; Figure 3), with a greater degree of accuracy (72% versus 51%, respectively).

Figure 3. AUC of unadjusted model and CLOTS‐AF risk score (blue) compared with CHADS2VASc score (brown).

AUC indicates area under the curve.

On average, each 1‐point increment in the CLOTS‐AF risk score was associated with a nearly 2‐fold increased risk of LAAT (OR, 1.63 [95% CI, 1.47–1.83]; P<0.001). A risk score of ≥6 was considered a statistically significant threshold on Youden's index, with a nearly 6‐fold increase in LAAT risk (OR, 5.67 [95% CI, 3.62–9.00]; P<0.001) compared with those with a score between 0 and 5 (OR, 1.44 [95% CI, 0.55–2.93]).

Characteristics Among Individuals With LAAT According to Heart Failure Status

LAAT was present in 119 of 324 (36.7%) patients with heart failure in the study population. In those with concurrent heart failure, the presence of LAAT was significantly higher among those inadequately warfarinized compared with DOAC use; those in AF; and the presence of more advanced cardiac remodeling (lower LV and RV systolic function, increased atrial and ventricular dimensions, and elevated filling parameters), renal impairment, and a higher CHADS₂VASc score (Table S2).

Longer‐Term Outcome: Thrombus Resolution

In the LAAT cohort, 94 individuals underwent serial TEE imaging (mean TEE number, 2.5±0.9). Of those, anticoagulation had not been commenced before TEE in 13.9%, was of inadequate duration or compliance in 30%, or required TEE in the setting of subtherapeutic international normalized ratio in 56.1%. Thrombus had resolved to enable cardioversion in 59 (63%) at a median of 133 days (interquartile range, 52–289) after index TEE.

Thrombus resolution was associated with a significantly younger mean age (58.8±11.9 years versus 62.5±10.5 years in persisting LAAT; P=0.137), 82% men, and lower body mass index (28.6±6.4 versus 31.9±8.0 kg/m² in persisting LAAT; P=0.031). Prevalence of comorbidities was similar in those with and without thrombus resolution (all P>0.05), as was AF and anticoagulation duration (P=0.125 and P=0.455 respectively, Table S3).

Despite the presence of LAAT at index TEE, anticoagulant adjustment was uncommon (20%) and did not predict thrombus resolution (OR, 0.378; P=0.111). Clinical and echocardiographic parameters and anticoagulant type did not predict thrombus resolution in this population (all P>0.05; Table 5).

Table 5.

Predictors of LAAT Resolution

Characteristics	OR	95% CI	P value
Female sex	0.318	0.091–1.111	0.073
Age >65 y	0.824	0.258–2.626	0.743
Rhythm
AFL	1.490	0.188–2.817	0.321
PsAF	1.300	0.335–5.053	0.704
AF duration	0.998	0.996–1.001	0.159
AF duration >1 y	0.892	0.135–5.913	0.906
Anticoagulant change	0.378	0.115–1.250	0.111
Obesity (BMI >30 kg/m2)	1.889	0.641–5.569	0.249
Hypertension	1.190	0.333–4.254	0.789
Dyslipidemia	1.163	0.298–4.530	0.828
Creatinine	0.997	0.991–1.004	0.425
Prior MI	0.870	0.234–3.240	0.836
Prior stroke	0.579	0.101–3.305	0.539
LVEF <50%	1.479	0.593–3.687	0.401
LAVI >35 mL/m2	0.941	0.322–2.746	0.912
Warfarin post TEE	1.731	0.666–4.500	0.260

AF indicates atrial fibrillation; AFL, atrial flutter; BMI, body mass index; LAAT, left atrial appendage thrombus; LAVI, left atrial volume index; MI, myocardial infarction; LVEF, left ventricular ejection fraction; PsAF, persistent atrial fibrillation; and TEE, transesophageal echocardiography.

Discussion

The present study of 1001 patients who underwent TEE before DCR for AF or AFL demonstrated the following:

1. Discrete thrombus in 14% with dense SEC in a further 7.5% precluding DCR.

2. Significant rates of LAA thrombus despite guideline‐directed anticoagulation.

3. Clinical and noninvasive echocardiographic parameters, including prior stroke/transient ischemic attack, renal impairment, AF rather than AFL, LVEF, TAPSE, and LAVI were independent predictors of LAAT.

4. Based on these parameters, a novel weighted risk score was developed; each 1‐point increment in the CLOTS‐AF risk score was associated with a nearly 2‐fold increased risk of LAAT (OR, 1.63 [95% CI, 1.47–1.83]; P<0.001).

5. The CLOTS‐AF risk score (AUC, 0.780) performed favorably when compared with the CHADS₂VASc score (AUC, 0.619) for LAAT risk prediction (DeLong Z=−2.427; P=0.015) with a greater degree of accuracy (72% versus 51%, respectively).

The incidence of LAAT observed in this study is consistent with that previously reported in smaller studies of between 9.8% and 23% (excluding cases of spontaneous echo contrast without discrete thrombus) in those not established on oral anticoagulation, ⁶ , ⁷ , ²⁴ , ²⁵ and the rate of discrete thrombus (4.3%) as a proportion of those receiving guideline‐directed anticoagulation was comparable with the pooled prevalence of 2.73% reported in a recent meta‐analysis. ⁴ A randomized study comparing early TEE‐guided versus conventional treatment for cardioversion of AF reported LAAT in 13.8% of 549 patients with AF duration >48 hours with an inadequate period of preprocedural anticoagulation (warfarin or unfractionated heparin commenced within 5 days or 24 hours before TEE, respectively). ⁵ Rates of SEC were not reported. The significant rates of LAAT or dense SEC observed in this and prior studies highlights the need to better identify high‐risk individuals where a suitable period of therapeutic anticoagulation should be initiated rather than early TEE‐guided DCR.

Performance of Risk Scores for LAAT

LAA thrombus is considered an important surrogate for ischemic stroke risk given a significant proportion of cardioembolic strokes in the presence of AF are thought to originate from the LAA. ²⁶ The CHADS₂VASc score has largely replaced the CHADS₂ score and was proposed to enhance risk stratification for thromboembolism to guide the need for anticoagulation among the low‐risk AF population. ¹² However, the adaptation to the original model only modestly outperformed the CHADS₂ score among 182 678 patients with AF with a C statistic of 0.67 (compared with 0.66 for CHADS₂ score). ²⁰ Moreover, despite its widespread clinical application, a meta‐analysis of 34 studies reported substantial variability in overall and CHADS₂VASc point‐score–stratified stroke and thromboembolic rates between and within geographic regions. ⁹ , ¹¹ , ²⁷ The performance of alternative risk scores such as ATRIA have been similarly modest. ²⁸

CHADS₂VASc in LAAT Risk Stratification

The CHADS₂VASc score was developed to predict the risk of stroke and thromboembolism rather than LAAT risk. However, the CHADS₂VASc score has been widely applied to the prediction of LAAT in numerous prior studies ¹¹ , ²⁹ given that the predominant source of thromboembolism is the LAA. In the present study, the CHADS₂VASc score was not an independent predictor of LAAT, irrespective of whether the CHADS₂VASc score was evaluated as a continuous covariate or binary at a cutoff ≥2. Indeed, the predictive performance of the CHADS₂VASc score (AUC, 0.619) in our real‐world study is comparable to findings described in the original risk stratification schema and subsequent small‐scale observational LAAT studies. ⁹ , ³⁰ A recent observational study described no association between increasing CHADS₂VASc score and LAAT, with LAAT observed in individuals with low CHADS₂VASc score. ³¹

Furthermore, while age is considered an important consideration in thromboembolic risk and is a component of the CHADS₂VASc score, it appears less clearly associated with LAAT. Although LAAT appears more prevalent with aging, the higher relative rates are likely explained by the greater burden of comorbidities and cardiac remodeling among the older population. Few studies have evaluated the impact of age on LAAT risk; 2 contemporary observational studies (151 patients and 512 patients undergoing TEE before catheter ablation and cardioversion, respectively) report that age is not an independent predictor of LAAT when adjusted for other clinical covariates. Both studies report relatively high rates of LAAT including dense SEC influencing treatment strategy (10% and 11.7%, respectively). ⁹ , ³¹ In the present study, age applied as a continuous or categorical covariate in regression analysis was not an independent predictor of LAAT.

The CHADS₂VASc score was developed to determine stroke risk and guide the need for anticoagulation in patients with AF rather than specifically estimating the likelihood of LAAT. The present study adds to the mounting literature regarding the modest performance of the CHADS₂VASc score in LAAT risk prediction. ⁸ , ¹¹ , ²⁹ While the presence of LAAT is considered a marker of thromboembolic risk, the lack of a clear temporal relationship between stroke and AF episodes ³² suggest LAAT is not the sole responsible factor and that the CHADS₂VASc score may more broadly represent stroke risk. Moreover, age, sex, and hypertension have not consistently been demonstrated to independently predict LAAT risk despite the heightened association with stroke risk in the AF population. ³¹ , ³³ Contemporary studies suggest that female sex should be considered as a risk modifier rather than an independent risk factor in the absence of accompanying thromboembolic risk factors for AF, leading to its removal from the score in some AF guidelines. ³⁴ , ³⁵ This may in part explain the lower predictive performance of the CHADS₂VASc score for LAAT risk prediction in the present and prior LAAT studies.

While the CHADS₂VASc score did not predict LAAT in the present study, certain individual components including prior stroke and systolic heart failure were independent predictors of LAAT; both are recognized risk factors for LAAT and systemic thromboembolism among the AF population and on this basis were incorporated into the CLOTS‐AF risk score.

Cardiac Remodeling and Thromboembolic Risk

LV systolic dysfunction is a recognized risk factor for thrombogenesis in AF. RV systolic dysfunction often accompanies LV systolic dysfunction or may occur in the presence of sleep disordered breathing, a common association of AF. ³⁶ TAPSE is a commonly reported and readily quantifiable surrogate for RV systolic function. The relationship between LAAT and RV dysfunction as assessed by TAPSE may have several explanations: (1) AF‐mediated cardiomyopathy with ventricular dysfunction as a consequence of AF, which may form a vicious cycle with worsening LA mechanical function and increasing thrombogenic risk; (2) preexisting RV systolic dysfunction resulting in altered forward flow, loading conditions, and a low flow state, which may promote LA stasis; and (3) LV systolic dysfunction resulting in elevated LA pressure and volume, resulting in higher pulmonary vascular resistance, RV dilatation, and remodeling with consequent RV dysfunction.

Clinical Implications

Contemporary AF guidelines recommend TEE imaging in those with AF duration >48 hours and in the setting of inadequate anticoagulation (DOAC <3 weeks' duration). ¹⁴ It should be noted that AF duration >48 hours was selected on the basis of small studies published >2 decades ago, ²⁴ , ²⁵ one of which was the Cardioversion Guided by Transesophageal Echocardiography: The ACUTE Pilot Study: A Randomized, Controlled Trial, which reported low rates of systemic thromboembolism among 62 patients randomized to TEE‐guided cardioversion with short‐term anticoagulation, despite procedural postponement in 13% due to the presence of LAAT. ²⁵ Guideline recommendations regarding preprocedural TEE imaging with AF duration beyond 48 hours ostensibly originated from the implicit assumption and clinical experience that atrial thrombus formation in the setting of AF occurs over several days, without robust data to support this hypothesis. ³⁷ Indeed, an observational study of 317 patients with AF not established on systemic anticoagulation reported an LAAT incidence of 14% with AF duration <3 days (versus 27% with AF duration ≥3 days). ³³ This reinforces the importance taking into account clinical and hemodynamic factors that may influence thromboembolic risk profile when considering an expedited TEE‐guided DCR.

The novel CLOTS‐AF risk score may enhance targeted clinical decision making in the following settings:

1. Identify high‐risk individuals with unclear AF/AFL duration in whom TEE‐guided DCR should be deferred until a suitable period of therapeutic anticoagulation has been satisfied. This will reduce unnecessary delays in care and exposure to an invasive procedure with inherent risks to the patient and improve health care resource use.

2. Identify low‐risk individuals appropriate for an expedited TEE‐guided cardioversion.

Thrombus Resolution

Resolution of LAAT was apparent in 63% (DOAC, 20; warfarin, 39) of 94 patients who underwent repeat TEE imaging with a view to cardioversion. Anticoagulant management was changed in just 20%. There are limited data to guide failed anticoagulation management in the presence of LAAT. The present study highlights the importance of an adequate compliant period of anticoagulation before TEE and the significant risk of LAAT in the setting of nonadherence. Prior observational studies report variable resolution of LAAT between 50% and 90%, largely in warfarinized and anticoagulant‐naïve people. ³⁸ Comparison of thrombus resolution rates among warfarinized and DOAC‐treated populations is limited to small retrospective series. ³⁹ The present study supports continued anticoagulation with the same agent in those with LAAT due to an inadequate preprocedure duration of anticoagulants. The study does not address management of LAAT despite adequate dosing, duration, and compliance.

Study Limitations

The study population is a selected cohort of patients with AF/AFL and inadequate anticoagulation before cardioversion and does not report the true prevalence of LAAT among the general AF/AFL population. This was a single‐center retrospective analysis of LAAT incidence. External validation would strengthen the predictive power and clarify the application of this risk model to the general AF/AFL population undergoing rhythm control; however, there are no established large‐scale databases of this nature to facilitate such an analysis.

Conclusions

Clinical and noninvasive echocardiographic parameters predicted LAAT among a large population undergoing TEE before DCR for AF/AFL. The novel CLOTS‐AF score may enhance LAAT risk prediction and help identify patients at higher risk for LAAT in whom a period of therapeutic anticoagulation is more suitable rather than an expedited TEE‐guided cardioversion.

Sources of Funding

None.

Disclosures

Dr Segan is supported by a cofunded National Health and Medical Research Council/National Heart Foundation post‐graduate scholarship. Professor Kistler is supported by an investigator grant from the National Health and Medical Research Council and has received funding from Abbott Medical for consultancy and speaking engagements and fellowship support from Biosense Webster. Dr Kalman has research and fellowship support from Medtronic and Biosense Webster. Dr Sandeep Prabhu has received consultancy and speaker fees from Abbott Medical and Biosense Webster. The remaining authors have no disclosures to report.

Supporting information

Tables S1–S3

Figure S1