Background. The purpose of this study was to investigate the risk factors of left atrial (LA) or left atrial appendage (LAA) thrombi in patients with nonvalvular atrial fibrillation (NVAF) and to establish and validate relevant predictive models. It might improve thromboembolic risk stratification in patients with NVAF. Methods. This study retrospectively included 1210 consecutive patients with NVAF undergoing transesophageal echocardiography (TEE), of whom 139 patients had thrombi in LA or in LAA. Through literature review and the ten events per variable (10EPV) principle, 13 variables were finally identified for inclusion in multivariate analysis. Models were constructed by multivariate logistic stepwise regression and least absolute shrinkage and selection operator (lasso) regression. Results. After logistic regression, five variables (AF type, age, B-type natriuretic peptide, E/e’ ratio, and left atrial diameter) were finally screened out as model 1. After Lasso regression, AF type, age, gender, B-type natriuretic peptide, E/e’ ratio, left atrial diameter, and left ventricular ejection fraction were finally screened as model 2. After comparing the two models, the simpler model 1 was finally selected. The area under the ROC curve (AUC) of the model 1 was 0.865 (95% CI: 0.838–0.892), the Hosmer–Lemeshow test = 0.898, and the AUC = 0.861 after internal validation. The clinical decision curve showed that the new clinical prediction model could achieve a net clinical benefit when the expected threshold was between 0 and 0.6. Conclusion. This study constructed a new clinical prediction model of LA or LAA thrombi, with a higher discriminative degree than the CHADS2 and CHA2DS2-VASc scoring systems (AUC: 0.865 vs. 0.643; AUC: 0.865 vs 0.652).

中文翻译：

---

非瓣膜性心房颤动患者左心房及左心耳血栓临床预测模型的构建

背景。本研究旨在探讨非瓣膜性心房颤动（NVAF）患者发生左心房（LA）或左心耳（LAA）血栓的危险因素，并建立和验证相关预测模型。它可能会改善 NVAF 患者的血栓栓塞风险分层。方法。该研究回顾性纳入了 1210 名接受经食管超声心动图 (TEE) 的连续 NVAF 患者，其中 139 名患者在 LA 或 LAA 中有血栓。通过文献回顾和每个变量十个事件 (10EPV) 原则，最终确定了 13 个变量用于多变量分析。通过多元逻辑逐步回归和最小绝对收缩和选择算子（套索）回归构建模型。结果。经过logistic回归，最终筛选出5个变量（房颤类型、年龄、B型利钠肽、E / e '比值、左房内径）作为模型1。经过Lasso回归后，房颤类型、年龄、性别、B-型利钠肽、E/e'比值、左心房内径、左心室射血分数最终筛选为模型2。经过对两种模型的比较，最终选择了较为简单的模型1。模型 1 的 ROC 曲线下面积 (AUC) 为 0.865（95% CI：0.838–0.892），Hosmer–Lemeshow 检验 = 0.898，内部验证后 AUC = 0.861。临床决策曲线表明，当预期阈值在0~0.6之间时，新的临床预测模型可以实现净临床获益。结论. 本研究构建了一种新的LA或LAA血栓临床预测模型，其鉴别度高于CHADS2和CHA2DS2-VASc评分系统（AUC：0.865 vs. 0.643；AUC：0.865 vs 0.652）。