第一部分 植入型心律转复除颤器治疗的肥厚型心肌病患者的临床特征分析

【目的】本研究旨在探讨植入型心律转复除颤器（ICD）治疗的肥厚型心肌病（HCM）患者的临床特征、预后差异及心源性猝死（SCD）危险因素。

【方法】连续纳入2014年至2024年于我院行ICD植入的344例HCM患者，根据患者预防类型分为一级预防和二级预防，根据是否发生SCD事件或ICD恰当治疗分为SCD组和非SCD组，分别分析其基线资料、影像学特征、药物治疗及随访结局，采用多因素logistic回归分析HCM患者发生SCD危险因素，估计比值比（OR）及95%置信区间（CI）。对多因素结果绘制受试者工作特征（ROC）曲线，通过曲线下面积（AUC）对结果进行评价。

【结果】（1）ICD年植入量以20.09%的速度增长，一级预防占比68.9%，但年轻患者（≤25岁）中二级预防比例仍较高（50%以上）；（2）二级预防组年龄更低（44.99±20.08岁 vs. 53.43±15.07岁）、晕厥史（82.2% vs. 48.1%）及心肌损伤标志物（hs-CRP、hs-cTnI等）水平显著高于一级预防组，且ICD恰当治疗率更高（39.3% vs. 26.6%，P=0.018）；（3）SCD高风险组患者男性比例、饮酒史、ACEI/ARB/ARNI使用率、hs-CRP、左室舒张内径显著升高，而白蛋白、高密度脂蛋白胆固醇水平、左心室射血分数、最大室壁厚度、右室内径以及舒张功能障碍比例显著降低（均P<0.05）；（4）多因素分析表明，年龄（OR 0.96，95%CI 0.95-0.98），男性（OR 2.01，95%CI 1.07-3.80）、饮酒史（OR 1.90， 95%CI 1.05-3.42）、晕厥史（OR 2.93，95%CI 1.74-4.93）、ACEI/ARB/ARNI使用（OR 1.83，95%CI 1.07-3.12）、白蛋白（OR 0.91，95%CI 0.85-0.96）、乳酸脱氢酶（OR 1.01，95%CI 1.00-1.01）、右心室内径（OR 0.89，95%CI 0.83-0.96）、最大室壁厚度（OR 0.95，95%CI 0.91-1.00）为SCD独立影响因素，模型AUC为0.778；（5）中位随访2.58年间ICD不恰当治疗率为3.8%，主要原因是房颤和室上性心动过速。

【结论】ICD显著降低HCM患者SCD风险，但年轻患者二级预防比例仍较高，提示需结合影像学及生物标志物优化风险分层，提升HCM患者一级预防精准性。

第二部分 ACC/AHA和ESC心源性猝死风险指南在植入型心律转复除颤器的肥厚型心肌病队列中的预测价值

【目的】本研究旨在评估2020年美国心脏病学会/美国心脏协会（ACC/AHA）和2014年欧洲心脏病学会（ESC）心源性猝死（SCD）风险分层指南在中国植入型心律转复除颤器（ICD）的肥厚型心肌病（HCM）患者中的预测效能。

【方法】本回顾性队列研究纳入2014年至2022年期间在中国医学科学院阜外医院接受ICD治疗无SCD事件病史的HCM患者。根据两种指南进行SCD风险分层，主要终点为因室颤/持续性室速触发的ICD恰当治疗。通过受试者工作特征（ROC）曲线、敏感度/特异度、需治疗人数（NNT）及生存分析（Log-rank检验）评估模型性能。

【结果】本研究共纳入147例接受ICD植入的HCM患者（中位随访3.93年），随访期间44例（29.9%）患者发生终点事件。根据2020 ACC/AHA指南，112例（76.19%）患者被归类为SCD高风险，而基于2014 ESC指南6%和4%风险截断值的高风险患者分别为67例（45.58%）和41例（27.89%）。ROC分析显示，2020 ACC/AHA模型的敏感度较高（79.55%），但特异度仅为25.24%，曲线下面积（AUC）为0.524（0.452-0.592）；2014 ESC-6%模型敏感度29.55%，特异度72.82%，AUC为0.512（0.431-0.595）；2014 ESC-4%模型在敏感度（56.82%）和特异度（59.22%）间具有更佳平衡，AUC为0.580（0.494-0.665），且其需治疗人数（NNT=7）亦低于2014 ESC-6%模型（NNT=40）和2020 ACC/AHA模型（NNT=18）。在生存分析中，2014年ESC指南4%模型能够有效区分高、低危组患者的临床终点事件（Log-rank P=0.033），而6%模型（P=0.47）和2020ACC/AHA指南（P= 0.37）的分组未显示出统计学差异。

【结论】2020年ACC/AHA指南敏感性高但特异性低，2014年ESC指南（4%截断值）综合性能更优，但两者在中国植入ICD的HCM人群中的预测效能有限。需结合影像学、遗传学等指标优化风险分层模型，提升ICD植入决策的精准性。

第三部分 机器学习模型对植入型心律转复除颤器治疗的肥厚型心肌病患者恰当治疗的预测作用

【目的】 现有的肥厚型心肌病（HCM）心源性猝死（SCD）风险分层模型（如HCM Risk-SCD评分）敏感性和特异性有限。机器学习（ML）通过整合多维临床数据，有望改善风险预测。本研究旨在构建基于机器学习的预测模型，评估其对HCM患者植入型心律转复除颤器（ICD）恰当治疗事件的预测价值。

【方法】回顾性连续纳入2014年1月至2023年12月于我院接受ICD植入的HCM患者，采集基线临床资料、实验室指标、心电图与超声心动图参数。采用多重插补处理缺失值后，基于指南推荐风险因素，联合LASSO回归、递归特征消除（RFE）及Boruta算法筛选关键预测因子。应用六种ML算法（随机森林、XGBoost、支持向量机（线性核和径向基核）、分类回归树、神经网络）进行模型训练，通过8:2分层随机拆分实现训练集与测试集划分。评价指标包括受试者工作特征曲线下面积（AUC）、灵敏度及特异度，并采用SHAP方法解析模型决策机制。

【结果】共纳入268例接受ICD植入的HCM患者（中位随访2.56年），93例（34.7%）发生ICD恰当治疗（事件发生率11.2/100人年）。随机森林模型综合性能最优（准确度0.774，AUC 0.829），但其灵敏度较低（0.500）限制其临床应用；XGBoost模型在灵敏度（87.5%）与特异度（62.2%）间达成最佳平衡，测试集AUC达0.784（95% CI: 0.644–0.912），且显著优于传统HCM Risk-SCD模型（P<0.05）。生存分析证实基于机器学习模型的风险分层可显著区分患者高低风险组（Log-rank P<0.001）。SHAP分析揭示核心预测因子包括左室舒张末期内径（LVDd）、血清白蛋白水平、频发室性早搏及超敏C反应蛋白（hs-CRP）。

【结论】基于ML的风险分层模型通过整合多维数据及非线性交互作用，显著提升了对HCM患者接受恰当ICD治疗的预测能力。其高敏感性减少高危患者漏诊，支持个体化临床决策。未来需外部验证并整合遗传数据以进一步优化预测效能。

第四部分 基于血浆蛋白组学的肥厚型心肌病患者心源性猝死危险分层

【目的】针对肥厚型心肌病（HCM）患者心源性猝死（SCD）传统风险分层工具的局限性，本研究旨在通过血浆蛋白组学筛选与SCD相关的分子标志物，构建整合临床与分子标志物的预测模型，以提升风险评估的精准性。

【方法】采用单中心病例对照设计，纳入62例HCM患者（SCD组26例，非SCD组36例）。通过高通量质谱技术（Orbitrap Astral平台）检测血浆蛋白质组，通过差异表达分析（t检验结合Benjamini-Hochberg校正，FDR<0.05）与单因素Logistic回归筛选候选蛋白，并利用LASSO回归优化特征集。最终构建包含CALU、TXN和SPARCL1的联合预测模型（血浆蛋白与传统风险因素）。模型效能通过五折交叉验证、ROC曲线、校准曲线及决策曲线分析评估，并与传统临床模型对比。

【结果】差异表达分析共鉴定257个差异蛋白（SCD组115个上调，142个下调），其中200个蛋白在差异分析与Logistic回归中显著重叠（FDR<0.05）。功能富集于补体激活、钙稳态调控及细胞外基质重构通路。LASSO回归筛选出12个关键蛋白，最终构建以CALU（钙结合蛋白）、TXN（抗氧化蛋白）和SPARCL1（细胞外基质蛋白）为核心的联合模型。联合模型预测SCD的曲线下面积（AUC）为0.936（95%CI: 0.868–0.982），显著优于传统临床模型（AUC=0.604，P<0.001），敏感度与特异度分别为86.4%和79.3%，且校准误差较传统模型降低59.9%。

【结论】本研究首次揭示CALU、TXN和SPARCL1通过调控钙信号紊乱、氧化应激失衡及心肌纤维化参与SCD病理进程，其联合模型可显著提升HCM患者SCD风险分层的精准性。研究结果为基于蛋白质组学的个体化风险评估提供了新策略，并为靶向干预分子通路的探索奠定理论基础，未来需通过多中心队列验证及机制研究推动临床转化。

综述：肥厚型心肌病心源性猝死的预测因素研究进展

肥厚型心肌病（HCM）的心源性猝死（SCD）预防仍是当前临床实践中的关键挑战，尤其在年轻患者群体中，其复杂性尤为突出。尽管近年来在风险分层方面取得了显著进展，但现有工具仍存在局限性。例如，欧洲HCM Risk-SCD模型的敏感度不足（41%-71%），且在不同种族人群中表现出预测效能的偏差；而美国AHA/ACC指南因过度依赖单一危险标志物，可能导致过度治疗。本研究综述系统性探讨了传统临床指标、影像学标志物、遗传学特征以及人工智能技术在预测HCM患者SCD风险中的应用进展。未来研究需突破现有模型的静态评估局限，通过整合基因组学、影像组学和动态生物标志物数据，构建智能化预测系统，并优化风险分层阈值及ICD植入标准，以实现HCM患者SCD预防的精准化管理。

Part 1. Clinical Characteristics of Hypertrophic Cardiomyopathy Patients Treated with Implantable Cardioverter-Defibrillators (ICD)

Objective: This study aims to explore the clinical characteristics, prognosis differences, and risk factors for sudden cardiac death (SCD) in hypertrophic cardiomyopathy (HCM) patients treated with implantable cardioverter-defibrillators (ICD).

Methods: A total of 344 HCM patients who underwent ICD implantation from 2014 to 2024 were consecutively enrolled. The patients were divided into two groups based on prevention type: primary prevention and secondary prevention. The patients were also categorized into SCD and non-SCD groups according to whether SCD events occurred or ICD therapy was appropriate. Baseline data, imaging characteristics, medication treatment, and follow-up outcomes were analyzed. Multivariate logistic regression was used to analyze the risk factors for SCD in HCM patients, and odds ratios (ORs) with 95% confidence intervals (CIs) were estimated. Receiver operating characteristic (ROC) curves were plotted for multivariate results, and the area under the curve (AUC) was used for evaluation.

Results: (1) The annual ICD implantation rate increased by 20.09%, with primary prevention accounting for 68.9%. However, the proportion of secondary prevention was still relatively high among younger patients (≤25 years) (over 50%). (2) The secondary prevention group was younger (44.99±20.08 years vs. 53.43±15.07 years), had a higher incidence of syncope (82.2% vs. 48.1%), and exhibited significantly higher levels of myocardial injury markers (hs-CRP, hs-cTnI, etc.) compared to the primary prevention group. The rate of appropriate ICD therapy was also higher in the secondary prevention group (39.3% vs. 26.6%, P=0.018). (3) In the SCD group, male proportion, alcohol history, ACEI/ARB/ARNI use, hs-CRP levels, and left ventricular end-diastolic diameter were significantly higher, while albumin levels, high-density lipoprotein cholesterol, left ventricular ejection fraction, maximal wall thickness, right ventricular end-diastolic diameter, and the proportion of diastolic dysfunction were significantly lower (all P<0.05). (4) Multivariate analysis revealed that age (OR 0.96, 95% CI 0.95-0.98), male sex (OR 2.01, 95% CI 1.07-3.80), alcohol history (OR 1.90, 95% CI 1.05-3.42), syncope history (OR 2.93, 95% CI 1.74-4.93), ACEI/ARB/ARNI use (OR 1.83, 95% CI 1.07-3.12), albumin (OR 0.91, 95% CI 0.85-0.96), lactate dehydrogenase (OR 1.01, 95% CI 1.00-1.01), right ventricular end-diastolic diameter (OR 0.89, 95% CI 0.83-0.96), and maximal wall thickness (OR 0.95, 95% CI 0.91-1.00) were independent risk factors for SCD, with an AUC of 0.778 for the model. (5) During a median follow-up of 2.58 years, the inappropriate ICD therapy rate was 3.8%, with atrial fibrillation and supraventricular tachycardia being the main causes.

Conclusions: ICD significantly reduces the risk of SCD in HCM patients. However, the proportion of secondary prevention remains high among younger patients, suggesting the need for improved risk stratification based on imaging and biomarkers to enhance the precision of primary prevention in HCM patients.

Part 2. Predictive Value of ACC/AHA and ESC Sudden Cardiac Death Risk Guidelines in a Hypertrophic Cardiomyopathy Cohort Treated with ICDs

Objective: This study aims to evaluate the predictive performance of the 2020 American College of Cardiology/American Heart Association (ACC/AHA) and 2014 European Society of Cardiology (ESC) guidelines for sudden cardiac death (SCD) risk stratification in hypertrophic cardiomyopathy (HCM) patients treated with implantable cardioverter-defibrillator (ICD) in China.

Methods: This retrospective cohort study included HCM patients who received ICD therapy at the Fuwai Hospital, Chinese Academy of Medical Sciences, between 2014 and 2022, and had no history of SCD events. SCD risk stratification was performed according to two guidelines, with the primary endpoint being appropriate ICD treatment triggered by ventricular fibrillation or sustained ventricular tachycardia. The model's performance was evaluated using receiver operating characteristic (ROC) curves, sensitivity/specificity, number needed to treat (NNT), and survival analysis (Log-rank test).

Results: This study included 147 HCM patients who received ICD implantation (median follow-up of 3.93 years), during which 44 patients (29.9%) experienced the primary endpoint event. According to the 2020 ACC/AHA guidelines, 112 patients (76.19%) were classified as high risk for SCD, while the high-risk patients based on the 6% and 4% risk cutoffs from the 2014 ESC guidelines were 67 (45.58%) and 41 (27.89%), respectively. ROC analysis showed that the 2020 ACC/AHA model had high sensitivity (79.55%), but low specificity (25.24%), with an area under the curve (AUC) of 0.524 (0.452-0.592). The 2014 ESC-6% model had sensitivity of 29.55%, specificity of 72.82%, and an AUC of 0.512 (0.431-0.595). The 2014 ESC-4% model showed a better balance between sensitivity (56.82%) and specificity (59.22%), with an AUC of 0.580 (0.494-0.665), and a lower number needed to treat (NNT = 7) compared to the 2014 ESC-6% model (NNT = 40) and the 2020 ACC/AHA model (NNT = 18). In survival analysis, the 2014 ESC 4% model effectively distinguished between high-risk and low-risk patient groups for clinical endpoint events (Log-rank P = 0.033), whereas the 6% model (P = 0.47) and the 2020 ACC/AHA model (P = 0.37) did not show statistical significance.

Conclusion: The 2020 ACC/AHA guidelines demonstrated high sensitivity but low specificity, while the 2014 ESC guidelines (with the 4% cutoff) exhibited better overall performance. However, both models showed limited predictive efficacy in the Chinese HCM population receiving ICD implantation. To enhance the accuracy of ICD implantation decisions, it is necessary to integrate imaging, genetic, and other biomarkers to optimize the risk stratification model.

Part 3. Predictive Role of Machine Learning Models for Appropriate ICD Therapy in Hypertrophic Cardiomyopathy Patients

Objective: Current risk stratification models for sudden cardiac death (SCD) in hypertrophic cardiomyopathy (HCM), such as the HCM Risk-SCD score, have limited sensitivity and specificity. Machine learning (ML), by integrating multidimensional clinical data, holds promise to improve risk prediction. This study aims to develop a machine learning-based predictive model to evaluate its value in predicting appropriate ICD therapy events in HCM patients.

Methods: This retrospective cohort study consecutively included HCM patients who underwent ICD implantation at our hospital from January 2014 to December 2023. Baseline clinical data, laboratory indicators, electrocardiographic, and echocardiographic parameters were collected. After multiple imputation for missing data, key predictive factors were selected using LASSO regression, recursive feature elimination (RFE), and the Boruta algorithm, based on guideline-recommended risk factors. Six ML algorithms (random forest, XGBoost, support vector machine (linear and radial basis function kernels), classification and regression trees, and neural networks) were used for model training. Training and testing sets were split using an 80:20 stratified random method. Evaluation metrics included the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. The SHAP method was employed to explain the model’s decision-making mechanism.

Results: A total of 268 HCM patients who received ICD implantation were included (median follow-up 2.56 years), with 93 (34.7%) patients experiencing appropriate ICD therapy (event rate 11.2/100 person-years). The random forest model had the best overall performance (accuracy 0.774, AUC 0.829), but its low sensitivity (0.500) limited clinical application. The XGBoost model achieved the best balance between sensitivity (87.5%) and specificity (62.2%), with a test set AUC of 0.784 (95% CI: 0.644–0.912), significantly outperforming the traditional HCM Risk-SCD model (P<0.05). Survival analysis confirmed that ML-based risk stratification significantly differentiated high- and low-risk groups for clinical endpoints (Log-rank P<0.001). SHAP analysis identified key predictive factors, including left ventricular end-diastolic diameter (LVDd), serum albumin level, frequent ventricular premature beats, and high-sensitivity C-reactive protein (hs-CRP).

Conclusion: The ML-based risk stratification model, by integrating multidimensional data and nonlinear interactions, significantly enhances the prediction of appropriate ICD therapy in HCM patients. Its high sensitivity reduces misdiagnosis of high-risk patients, supporting personalized clinical decision-making. Future external validation and integration of genetic data are necessary to further optimize predictive performance.

Part 4. Plasma Proteomics-Based Risk Stratification for Sudden Cardiac Death in Hypertrophic Cardiomyopathy Patients

Objective: Due to the limitations of traditional risk stratification tools for sudden cardiac death (SCD) in hypertrophic cardiomyopathy (HCM) patients, this study aims to identify plasma proteomic biomarkers associated with SCD and construct a predictive model integrating clinical and molecular markers to improve the accuracy of risk assessment.

Methods: A single-center case-control design was employed, including 62 HCM patients (26 in the SCD group and 36 in the non-SCD group). Plasma proteomics was analyzed using high-throughput mass spectrometry (Orbitrap Astral platform). Differential expression analysis (t-test with Benjamini-Hochberg correction, FDR < 0.05) and univariate logistic regression were used to identify candidate proteins. LASSO regression was applied to optimize the feature set. A combined predictive model was developed, including CALU, TXN, and SPARCL1 (plasma proteins and traditional risk factors). Model performance was assessed through five-fold cross-validation, ROC curve analysis, calibration curves, and decision curve analysis, and compared with traditional clinical models.

Results: Differential expression analysis identified 257 differentially expressed proteins (115 upregulated and 142 downregulated in the SCD group). Of these, 200 proteins were significantly overlapping between differential analysis and logistic regression (FDR < 0.05). Functional enrichment pathways included complement activation, calcium homeostasis regulation, and extracellular matrix remodeling. LASSO regression selected 12 key proteins, ultimately constructing a combined model centered on CALU (calcium-binding protein), TXN (antioxidant protein), and SPARCL1 (extracellular matrix protein). The AUC of the combined model for predicting SCD was 0.936 (95% CI: 0.868–0.982), significantly outperforming the traditional clinical model (AUC = 0.604, P < 0.001). The sensitivity and specificity were 86.4% and 79.3%, respectively, and the calibration error was reduced by 59.9% compared to the traditional model.

Conclusion: This study is the first to reveal that CALU, TXN, and SPARCL1 participate in the pathological process of SCD through regulating calcium signaling disruption, oxidative stress imbalance, and myocardial fibrosis. The combined model significantly improves the precision of SCD risk stratification in HCM patients. The results provide a new strategy for personalized risk assessment based on proteomics and lay the theoretical foundation for exploring targeted molecular pathway interventions. Future research should focus on multicenter cohort validation and mechanistic studies to promote clinical translation.

Review: Advances in Predictive Factors for Sudden Cardiac Death in Hypertrophic Cardiomyopathy

Prevention of sudden cardiac death (SCD) in hypertrophic cardiomyopathy (HCM) remains a key challenge in clinical practice, particularly in the young patient population, where its complexity is especially prominent. Despite significant progress in risk stratification in recent years, existing tools still have limitations. For instance, the European HCM Risk-SCD model has inadequate sensitivity (41%-71%) and demonstrates predictive bias across different racial populations, while the American AHA/ACC guidelines, due to their over-reliance on single risk markers, may lead to overtreatment. This review systematically explores the application of traditional clinical indicators, imaging biomarkers, genetic features, and artificial intelligence techniques in predicting SCD risk in HCM patients. Future research needs to break through the static evaluation limitations of current models by integrating genomics, radiomics, and dynamic biomarker data to construct intelligent predictive systems, optimize risk stratification thresholds, and improve ICD implantation criteria, aiming to achieve precision management for SCD prevention in HCM patients.