Introduction

Left ventricular (LV) thrombus has been associated with up to 22% risk of embolization in the past [1]. Guidelines recommend that patients with ischemic cardiomyopathy (ICM) and LV thrombus should receive oral anticoagulation for 3 months [2], while patients with nonischemic cardiomyopathy (NICM) should be treated with oral anticoagulation for at least 3–6 months, to reduce the risk of stroke or systemic embolism events [3]. Based on the 2022 statement for LV thrombus, anticoagulation should be discontinued if patients had a resolution of LV thrombus with left ventricular ejection fraction (LVEF) improving to > 35% or major bleeding occurring [3]. However, considering patients with persistent thrombus despite anticoagulation, there are no sufficient study data to determine whether anticoagulation should be continued indefinitely. Therefore, identifying patients with a high risk of thrombus unresolved may be helpful to provide evidence on the management of anticoagulation.

Depending on a prospective trial and a retrospective study, we aimed to investigate potential factors associated with thrombus unresolved in the population of patients who received oral anticoagulation for 3 months and then provide a prediction model to determine the risk of thrombus persistence or recurrence in patients with LV thrombus.

Methods

Study design and patient population

This study was derived from two studies, including a prospective study named R-DISSOLVE (ClinicalTrials.gov: NCT04970381) and a retrospective registry study (ClinicalTrials.gov: NCT 05,006,677), carried out at Fuwai Hospital, National Center of Cardiovascular Diseases in China. R-DISSOLVE was an interventional, single-arm, open-label, investigator-initiated study between October 2020 and April 2022. The retrospective study collected data from June 2013 to December 2019 by using electronic medical records. The two study protocols were developed separately and approved by the ethics committee at the participating center. The combined study was reported in accordance with the TRIPOD checklist [4].

In the prospective study, patients with LV thrombus for less than 3 months and with systemic anticoagulation of less than 1 month were enrolled. Patients with inherited or acquired thrombophilia (e.g., antiphospholipid syndrome) were excluded since the risk of thrombus persistence/recurrence in these patients was established on a unique pathophysiological mechanism. The inclusion and exclusion criteria in the retrospective study were similar to those used for the prospective one. To increase the sample size [5,6,7], we pooled the data from the two studies to create a combined study, which was then used to develop a statistical model of thrombus persistence/recurrence prediction. Only patients having imaging records at follow-up visits and having continued systemic oral anticoagulation for at least 3 months were eligible for the primary analysis, as evidenced by objective data such as prescriptions from cardiologists.

Definitions

LV thrombus was defined as an abnormal echo mass in the left ventricular cavity, whose edge was different from the left ventricular endocardium [8]. In the prospective trial, thrombus was quantified using contrast-enhanced echocardiography (CE) at baseline and follow-up visits, whereas in the retrospective study, thrombus confirmed by transthoracic echocardiography (TTE), computer tomography (CT), or cardiac magnetic resonance imaging (CMR) was obtained. When a thrombus was detected, its morphology was categorized as either mural (if its borders are generally continuous with the adjacent endocardium) or protuberant (if its borders are distinct from the adjacent endocardium and protrude into the ventricular cavity) [3].

The primary endpoint of this combined study was the LV thrombus persistence/recurrence rate at 3 months confirmed by image techniques. The thrombus persistence was defined as the presence of a thrombus at 3 months that was comparable to the one at baseline. The thrombus recurrence was defined as the presence of thrombus at 3 months following negative images from baseline to 3 months. Safety outcomes included major bleeding according to the International Society on Thrombosis and Haemostasis [ISTH] [9] criteria and clinically related non-major hemorrhage events [10]. Additionally, stroke or embolic events were collected at 3 months.

Model development

A total of 44 variables were collected in the initial database. The data from each study were randomly split into a training set (70% of the sample) and a validation set (30% of the sample). The final sets were created by merging the training set and validation set from each study. The following five stages summarize the process of developing and validating prediction models. First, create the prediction models. The significant variables from the univariate logistic analysis and variables of interest were combined to generate Model 1. Model 2 was created using the Lasso regression algorithm to identify additional potential variables associated with prognosis. Second, assess model discrimination ability by the mean area under the receiver operating characteristic curve (AUC). Third, assess model calibration to compare the predicted with the actual rates of thrombus persistence/recurrence [11]. Fourth, assess the clinical effectiveness of models by the decision curve analysis (DCA) to quantitatively depict the net benefit of clinical decisions. Finally, draw a nomogram to visualize prediction models, which can relatively calculate the risk of thrombus persistence/recurrence at 3 months in each patient by calculating the total score of each independent factor.

Statistics analysis

Descriptive statistics were computed using the CBCgrps-Package in R [12]. Mean (standard deviation, SD) or median (interquartile range, IQR) for continuous variables and frequency (percentage) for categorical variables were reported. The Pearson chi-squared test or the Fisher exact test was performed for categorical data, and the Student unpaired t-test or the Mann–Whitney U test was applied to compare continuous variables. Odds ratios (OR) and 95% confidence intervals (CI) were estimated using regression models. To address missing data for predictor variables, multiple imputations by chained equations with predictive mean matching (MICE-Package in R) were used to create five sets of imputed data. The car package in R was used to detect collinearity between variables, and a variance inflation factor < 10 was tolerated. In addition, a restricted cubic spline curve was used between the continuous variables and the primary outcome. All analyses were scheduled for completion with R Studio and R, Version 3.5.1 (The R Project for Statistical Computing, Vienna, Austria).

Results

Study population

A total of 487 patients and 213 patients were screened in the retrospective and prospective study respectively (Fig. 1). At a 3 months follow-up, 108 patients and 64 patients separately were included in the combined study. Patients from the two groups were divided into a training set and a validation set with a 7:3 ratio. The final training group and validation group, respectively, comprised 124 patients and 48 patients.

Fig. 1
figure 1

Study flow diagram. A total of 172 patients were included in our analysis, which was split into a training set (n = 124, 70% of the sample) and a validation set (n = 48, 30% of the sample). †Others include 5 patients suspected of antiphospholipid antibody syndrome (n = 2) or thrombophilia (n = 3) at discharge. N, numbers of patients

Full size image

The average age of the entire population was 49.8 years, and 143 out of 172 (83%) patients were male. Ischemic cardiomyopathy (ICM) was the leading underlying cause (46.5%). The median LVEF level was 30% and 126 (73.3%) patients had a mural thrombus. Regarding anticoagulation therapy, 123 (71.5%) patients received non-vitamin K antagonist oral anticoagulants (NOACs) therapy, while the remaining were with warfarin. No patients switched the type of oral anticoagulation within 3 months. Of the 123 patients, 99.2% were given rivaroxaban (49.2% with a reduced dose). In addition, 105 (61%) patients received no antiplatelet, while 44 patients received mono antiplatelet agents (36 on clopidogrel and 8 on aspirin) and 23 patients received dual antiplatelet therapy (22 on aspirin plus clopidogrel and 1 on aspirin plus ticagrelor) in the combination with anticoagulation. Baseline characteristics were presented in Table 1 and no significant differences were observed between the training group and the validation group.

Table 1 Baseline characteristics of patients with LV thrombus
Full size table

Development of two models via univariate and Lasso regression

We collected 44 characteristics in the primary database. A total of 11 variables with a P value < 0.10 were selected from the univariable analysis (Table 2). By including an additional four variables of interest (diagnosis, LVEF, anticoagulation therapy, and D-dimer levels) [13,14,15,16], six predictors were finally incorporated into Model 1- diagnosis, antiplatelet therapy, the thickness of thrombi, thrombus morphology, ventricular aneurysm, and D-dimer levels (Table 3). On the other hand, ten-fold cross-validation of the Lasso coefficient profiles of 44 characteristics led to the selection of Lambda = 0.000011 as the minimum criterion for the Lasso regression (Figure S1). A total of ten variables remained after variable selection using the LASSO penalty (Fig. 2), six of which were components of Model 2 in the multivariable logistic analysis (Table 3). They were listed as follows: diagnosis, antiplatelet therapy, the thickness of thrombi, thrombus morphology, spontaneous echo contrast, and D-dimer levels.

Table 2 Univariate analysis associated with LV thrombus persistence/recurrence at 3 months in the training group
Full size table
Table 3 Two models for LV thrombus persistence/recurrence at 3 monthsa
Full size table
Fig. 2
figure 2

Variable selection in the LASSO regression using ten-fold cross-validation. Dotted vertical lines were drawn at the optimal values by using the minimum criteria and the 1-SE criteria. A Lambda value of 0.000011 was chosen according to ten-fold cross-validation. SE, standard error

Full size image

In Model 1, the AUC was 0.852 (95% CI 0.771–0.933) in the training set and 0.631 (95% CI 0.421–0.842) in the validation set (Fig. 3). The AUC in Model 2 was similar to that in Model 1 in the training set (0.856, 95% CI 0.781–0.931) but lower in the validation set (0.617, 95% CI 0.406–0.827), while no differences were found in the comparison of two models’ AUC in the training set (P = 0.838) and the validation set (P = 0.734) (Figure S 2). To be brief, Model 1 demonstrated a stronger capacity to discriminate. Moreover, both models' DCA curves displayed a comparable range of cutoff probabilities, indicating equal clinical efficacy (Figure S3-S4). Model 1 showed a considerably better calibration of the model with more plots surrounding the ideal curves (Fig. 4, Figure S5). Above all, Model 1 was selected because it had a higher AUC-estimated predictive value and a comparable capacity to illustrate the net benefit of clinical decisions. Table 3 shows the outcome of Model 1's prediction after taking the six variables into account. Additionally, by using cross-validation, the accuracy of Model 1 was 0.838 while the Kappa value was 0.416.

Fig. 3
figure 3

ROC curves of Model 1 for predicting the risk of LV thrombus persistence/recurrence at 3 months. A Training set. B Validation set. The blue curve represents the model discrimination ability of Model 1 in the training set (AUC 0.852, 95% CI 0.771–0.933). The red curve represents the model discrimination ability of Model 1 in the validation set (AUC 0.631, 95% CI 0.421–0.842). The point in the curve represents the optimal threshold along with the corresponding specificity and sensitivity respectively. LV, left ventricular; ROC, receiver operating characteristic; AUC, area under the ROC curve

Full size image
Fig. 4
figure 4

Calibration plots for predicting the risk of LV thrombus persistence/recurrence at 3 months in Model 1. A Training set. B Validation set. X-axis: predicted thrombus persistence/recurrence risk; Y-axis: actual thrombus persistence/recurrence rate. Estimates above the grey solid line represent underestimates; those below the grey solid line represent overestimates. The vertical bars represent 95%CIs. LV, left ventricular

Full size image

Clinical utility

According to Model 1 (variables included diagnosis, antiplatelet therapy, thickness of thrombi, thrombus morphology, ventricular aneurysm, and D-dimer levels), the rate of persistent/recurrent LV thrombus within 3 months increased as thrombus thickness increased (OR 1.11, 95% CI 1.05–1.20, P = 0.002). It was interesting to note that patients with protuberant thrombus had a lower rate of thrombus persistence/recurrence compared to those with mural thrombus (OR 0.17, 95% CI 0.03–0.69, P = 0.025). Patients with higher baseline D-dimer levels had a lower likelihood of developing persistent or recurrent thrombus at 3 months (OR 0.67, 95% CI 0.43–0.91, P = 0.030), and there was a linear relationship between the primary end-point and the thrombus thickness or D-dimer levels (all p for non-linear > 0.05; Figure S6). In addition, the incidence of LV thrombus persistence/recurrence at 3 months differed among patients with different diagnoses or different antiplatelet therapy, and it was uncertain which diagnosis or antiplatelet therapy had more effect on the primary end-point because of their wide CIs.

Furthermore, we developed a nomogram risk prediction model that comprised independent risk factors (R2 0.38, C index 0.85, 95% CI 0.77–0.93) (Fig. 5). The scores of the items displayed in the nomogram should be added up (Table 4). For example, an echocardiogram revealed a 10 mm thick LV mural thrombus in the ventricular aneurysm in a patient with ICM who had a D-dimer level of 2 ng/mL at admission. During the hospitalization, rivaroxaban plus aspirin treatment was then administered to the patient. According to our prediction model, the overall score of the patient was 176, and the likelihood of LV thrombus persistence/recurrence within 3 months was roughly 40%.

Fig. 5
figure 5

Nomogram for the prediction of LV thrombus persistence/recurrence at 3 months in Model 1. To use the nomogram, an individual patient’s value is located on each variable axis, and a line is drawn upward to determine the number of points received for each variable value. The sum of these numbers is located on the Total Points axis, and a line is drawn downward to the last axis to determine the risk of thrombus persistence/recurrence. LV, left ventricular; ICM, ischemic cardiomyopathy; DCM, dilated cardiomyopathy

Full size image
Table 4 A risk score for predicting LV thrombus persistence/recurrence at 3 months
Full size table

Outcome of thrombus resolution, bleeding, and major cardiovascular events

At 3 months, 38 patients (22.1%) had persistent or recurrent LV thrombus, compared to a total of 134 patients (77.9%), who had their thrombus resolved. Figure 6 depicted an illustration of a patient with persistent thrombus by CE despite receiving anticoagulant medication for 3 months. Patients in the thrombus resolution group had a higher prevalence of previous heart failure, lower LVEF and higher levels of N-Terminal pro-Brain Natriuretic Peptide (NT-proBNP), a smaller baseline thrombus, a higher proportion of spontaneous echo contrast, and a lower incidence of regional wall motion abnormality and ventricular aneurysm (Table 2).

Fig. 6
figure 6

Patient example of CE images. The thrombus is persistent at 1.5 months and 3 months, compared to LV thrombus (red arrows) at baseline. LV, left ventricular; CE, contrast-enhanced echocardiography

Full size image

Three patients experienced bleeding during a 3-month follow-up; one of them reported eye hemorrhage, while the other two experienced nose bleeding. During hospitalization, nine patients suffered major cardiovascular events—six patients reported having a stroke and three patients encountered a pulmonary embolism (Table S1).

Discussion

It was the first time in a prospective study and a retrospective study to predict the risk of LV thrombus persistence/recurrence among patients with oral anticoagulation for 3 months. The key findings were as follows. Patients who had thicker thrombus, mural thrombus morphology, ventricular aneurysm, or lower D-dimer levels were more likely to have persistent/recurrent LV thrombus. Therefore, irrespective of insufficient evidence, we advocated that anticoagulant therapy could be prolonged and individualized for patients with these high-risk characteristics.

The rate of thrombus persistence/recurrence in the combined study was considerably low thanks to the strict inclusion criteria in the retrospective study and the intensive monitoring in the prospective trial, which demonstrated that patients had relatively fresh thrombi prior to the enrollment and adhered to therapy well over the study period. Another significant aspect was the prospective trial's use of CE at baseline and follow-up visits, which improved the study's power to precisely identify LV thrombus. The incidence of thrombus unresolved was 31.0% in an updated meta-analysis that included 21 studies with 3057 patients over a median follow-up of 12 months [3]. According to a retrospective cohort study, 34.2% of patients with heart failure found persistent LV thrombus with a median duration of 17 months [17]. Based on these studies, it is unclear whether the rate of persistent thrombus will decrease over time given that some persistent thrombi are more likely to be calcified or organized despite anticoagulation in long-term follow-up.

In decades of research, the optimal anticoagulation treatment for LV thrombus has remained controversial. Prior recommendations suggest that anticoagulant therapy with vitamin K antagonists (VKAs) was appropriate for patients with myocardial infarction and asymptomatic LV thrombi for up to 3 months [2, 18]. The 2017 ESC guidelines for myocardial infarction recommended the duration of anticoagulation (either NOACs or VKAs) might be for 6 months guided by repeated imaging [19], in addition, the scientific statement from the American Heart Association (2022) indicated that NOACs were considered to be a reasonable alternative to VKAs in patients with LV thrombus, according to currently available data [3]. The duration of anticoagulation for persistent thrombus, meanwhile, is yet undetermined. According to consensus opinion [3], patients with persistent or recurrent thrombus should continue anticoagulation until resolution on the basis of their high compliance and frequent imaging assessments. A trial of alternative anticoagulation, on the other hand, should be addressed on a case-by-case basis. For those thrombi that are organized or calcified, discontinuing oral anticoagulation is an option because the risk of embolization is probably minimal. Consequently, long-term anticoagulation management options for patients with persistent thrombus should weigh the concerns of indefinite anticoagulation (e.g., increased pill burden and bleeding) against the potential reduction in stroke risk.

We came to the conclusion that patients with persistent thrombus shared six characteristics in addition to presumably poor adherence. It is well acknowledged that patients with different etiologies had different mechanisms in the development of LV thrombus. Based on Virchow's triad of thrombogenesis, three factors include stasis attributable to reduced ventricular function, endocardial injury, and inflammation/hypercoagulability. The interaction between these mechanisms and thrombus outcome, however, is not always easy to clarify [20, 21]. According to our finding, the rate of thrombus persistence/recurrence at 3 months differed among patients with ICM, DCM, and other cardiovascular disorders while it was unreliable to conclude whether DCM or other diseases patients had a higher likelihood of thrombus persistence than those with ICM. Indeed, researchers reported that specific causes of DCM (eg, amyloidosis, eosinophilic myocarditis) could increase the risk of LV thrombus persistent/recurrent [21, 22]. In terms of oral anticoagulation, patients taking NOACs experienced less likelihood of LV thrombus persistence at 3 months than those with VKAs, though the statistical difference was insignificant. A comparable conclusion was reached by several meta-analyses [23,24,25,26] and two small randomized clinical trials (apixaban or rivaroxaban versus warfarin) [27, 28]. Interestingly, although the CI was somewhat wide, it allowed us to conclude that patients with mono or dual antiplatelet therapy had a different LV thrombus persistence/recurrence rate at 3 months compared to those without. A prior study [13] reported that patients taking anticoagulation in combination with mono antiplatelet therapy had a higher risk of persistent thrombus than those who did not. In the study of Niku et al. [29], the rate of receiving antiplatelet therapy paired with anticoagulation was higher in patients with persistent left atrial thrombus than those with thrombus resolution (65% vs 38%, P = 0.03). Overall, for patients undergoing percutaneous coronary intervention with an indication for antiplatelet therapy and who also have an indication for oral anticoagulation, a general strategy (preferably a NOAC plus clopidogrel) may be considered on the basis of current practice and guideline recommendations [30,31,32].

In addition, patients with lower D-dimer levels in our analysis had a higher likelihood of LV thrombus persistence/recurrence at 3 months; however, the result would be adjusted by the mean D-dimer level in future work. The reason might be concluded that the thrombus of patients with a low D-dimer was neither fresh nor mobile, resulting in a persistent thrombus. The grade of thrombus mobility was found to be strongly associated with the thrombus' outcome. Limited data suggest that a large or mural thrombus has a less likelihood of thrombus resolution than a small or protuberant thrombus [33]. When evaluating the shape and the morphology of LV thrombus, the findings from Salah et al. [15], which were in agreement with our results, showed that patients with persistent thrombus had bigger baseline thrombus areas. From the consensus of the statement [3], it is unreasonable to give anticoagulation for mural thrombi even though the risk of embolization may be less than for protuberant thrombi. To conclude, a shared decision-making approach is appropriate following a risk/benefit discussion between patients and clinicians.

The major limitations were listed. First, since the sample of this study is small even when combining both the prospective and retrospective cohorts, which limits the power and utility of the model, external validations of our model are expected. Second, due to the wide CIs, it is still unclear whether patients with DCM and antiplatelet therapy are associated with a high risk of LV thrombus persistence/recurrence or not, so additional prospective research is needed to confirm the results. Third, the anticipated medication after 3 months remains unknown in the retrospective study owing to the short-term follow-up, while patients in the prospective study were informed of the appropriate anticoagulant therapy for at least 6 months. Large-scale trials are essential to determine whether indefinite anticoagulation is merited in patients with persistent/recurrent LV thrombus.

Conclusions

A prediction model comprising six variables was derived from a combination of prospective and retrospective studies. Patients were more likely to develop persistent or recurrent LV thrombus at 3 months if they had thicker thrombus, mural thrombus, ventricular aneurysm, or low baseline D-dimer levels. Considerations on the duration of anticoagulation should be based on the best clinical judgment and shared decision-making, and prospective randomized trials are necessary to validate the model.