In Journal of medical Internet research ; h5-index 88.0
BACKGROUND : Despite decades of research, anticoagulation therapy recommendations in patients with left ventricular thrombus (LVT) are based on small, frequently outdated studies. There may not be strong enough evidence to support effective treatment.
OBJECTIVE : We postulated that using artificial intelligence clustering approaches, discrete patient groups may be identified that correspond to the effectiveness of anticoagulation in people with LVT.
METHODS : 1,675 patients were retrospectively screened by the International Classification of Diseases 10th revision code in Fuwai Hospital, Beijing, from 2009 to 2021. All patients underwent transthoracic echocardiography. Only patients with a confirmed LVT were included after an imaging review by two independent experts. Principal components analysis and hierarchical clustering were applied to this study. The number of clusters and dimensions was determined objectively, with results validated using the bootstrapped approach. Major adverse cardiovascular and cerebrovascular events (MACCE) were the primary outcome, defined as the composite of cardiovascular death, ischemic stroke, and acute myocardial infarction. Cumulative incidence functions were generated for the competing risks of death during the follow-up period.
RESULTS : 1,068 patients with LVT were included, with median age of 55 years (IQR 44-64) and LVEF 38% (IQR 29-46). 890 (83.3%) patients were men. 4 Clusters were confirmed finally. Jaccard scores and heatmap confirmed the robustness and consistency of clustering. Compared with patients in cluster 2, those in cluster 1 were younger, with more females with worse cardiac (higher NT-proBNP level and lower ejection fraction) and renal (higher proportion of eGFR<60 ml/min/1.73m2) function and higher risk of embolism (higher D-Dimer level and higher proportion of round LVT). Compared with cluster 2, cluster 1 had a higher risk of all-cause death, cardiovascular death, and MACCE (all P<.001). No significant changes in the composition of the 2 clusters have occurred over time (all P for trend > .05). Across the whole cohort of patients in LVT, anticoagulation did not significantly reduce MACCE compared with participants without anticoagulation, with an HR of 1.29 (95% CI 0·97-1.71, P=.08). There was a statistically significant reduction in MACCE with anticoagulation in cluster 1, with an HR of 0.67 (95% CI 0.46-0.98, P=.04). However, in cluster 2, anticoagulation can increase the incidence of MACCE with an HR of 1.59 (95% CI 1.03-2.46, P=.04) and significant interaction (P=.03).
CONCLUSIONS : A clustering method based on artificial intelligence in patients with LVT may tell prognostic responses from anticoagulation. Our study confirms that not all patients with LVT benefit from anticoagulation, but only those at high risk. This included a cluster of patients with suboptimal efficacy and another group where anticoagulation reduced MACCE.
Shi Boqun, Zhang Rui, Song Chenxi, Cui Kongyong, Zhang Dong, Dong Qiuting, Jia Lei, Yin Dong, Wang Hongjian, Dou Ke-Fei, Song Weihua
2022-Nov-26
