癌症患者及癌症幸存者癌症治疗相关心功能障碍风险预测模型:系统综述与荟萃分析
该系统综述与荟萃分析纳入56项研究,报告51个开发模型和12个外部验证模型。多数模型在成人中开发,主要针对乳腺癌或血液恶性肿瘤的 pretreatment 风险。HFA-ICOS工具外部验证最频繁但持续低估风险,在抗HER2治疗的乳腺癌患者中合并C统计量为0.60(95%CI 0.52至0.68)。现有模型需更多证据才能广泛临床采用。
要点
- 56项研究纳入,51个开发模型和12个外部验证模型,几乎所有模型存在高偏倚风险。
- HFA-ICOS工具外部验证11次,持续低估风险,观察事件率超过预测风险。
- 抗HER2治疗的乳腺癌患者中,HFA-ICOS合并C统计量为0.60(95%CI 0.52至0.68)。
核心结论
- 现有癌症治疗相关心功能障碍预测模型在广泛临床采用前需要额外证据。
- 关键性能指标报告不佳和外部验证研究有限目前限制其全面评估。
- HFA-ICOS工具表现欠佳,尤其是将轻度癌症治疗相关心功能障碍纳入事件时。
- 未来研究应优先使用跨多种恶性肿瘤的大型聚类数据集验证和更新现有模型,以增强其性能、可推广性和常规实践中的临床实用性评估。
研究设计与样本
- 系统综述与荟萃分析,检索Embase、Medline和Cochrane对照试验中心注册库,从建库至2024年8月23日。
- 纳入开发或外部验证多变量模型预测癌症治疗相关心功能障碍风险的研究,人群为接受全身抗肿瘤药物治疗的≤39岁年轻人或≥40岁老年人癌症患者或癌症幸存者。
- 排除辐射诱导心脏毒性研究;两名评价者独立筛选研究、提取数据并使用预测模型偏倚风险评估工具评估偏倚风险。
- 采用随机效应荟萃分析合并性能指标。
主要发现
- 共纳入56项研究,报告51个开发模型和12个外部验证模型。
- 多数模型在成人中开发(34/51,67%),主要针对乳腺癌(20/34,59%)或血液恶性肿瘤(6/34,18%)以确定治疗前风险(33/34,97%)。
- 在年轻人中,多数开发模型(16/17,94%)聚焦长期风险评估,主要在血液恶性肿瘤幸存者中。
- 13/51(25%)开发模型和6/44(14%)外部验证报告了区分度和校准度指标;几乎所有模型存在高偏倚风险。
- 12/51(24%)开发模型进行了外部验证,年轻人中4/17(24%),成人中8/34(24%)。
关键数值与推荐
- HFA-ICOS工具外部验证最频繁(11次),主要在接收HER2靶向治疗的乳腺癌患者中(5/11,45%)。
- 所有外部验证中,该工具持续低估风险,观察事件率超过预测风险,尤其在轻度癌症治疗相关心功能障碍为最常见结局的研究中。
- 抗HER2药物治疗的乳腺癌患者中,合并C统计量为0.60(95%置信区间0.52至0.68)。
- 该人群中,观察合并事件率:低风险组(预测<2%)为12%,中风险组(2-9%)为15%,高风险组(10-19%)为25%,极高风险组(≥20%)为41%。
临床怎么用
- 现有预测模型在广泛临床采用前需要额外证据,HFA-ICOS工具表现欠佳。
- 未来研究应优先使用大型聚类数据集验证和更新现有模型,以增强其性能、可推广性和临床实用性评估。
局限性
- 几乎所有模型存在高偏倚风险。
- 关键性能指标报告不佳:仅13/51(25%)开发模型和6/44(14%)外部验证报告区分度和校准度指标。
- 外部验证研究有限:仅12/51(24%)开发模型进行了外部验证。
- HFA-ICOS工具表现欠佳,尤其在将轻度癌症治疗相关心功能障碍纳入事件时。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
OBJECTIVES: To systematically review all prediction models developed or validated for cancer therapy related cardiac dysfunction (CTRCD) and to quantitatively analyse their performance. DESIGN: Systematic review and meta-analysis. DATA SOURCES: Embase, Medline, and the Cochrane Central Register of Controlled Trials, from inception to 23 August 2024. ELIGIBILITY CRITERIA FOR SELECTING STUDIES: Studies that developed or externally validated multivariable models to predict CTRCD risk in young people (children, adolescents, and young adults (≤39 years)) or older adults (≥40 years) with cancer or cancer survivors treated with systemic antineoplastic agents. Studies on radiation induced cardiotoxicity were excluded. METHODS: Two reviewers independently screened studies, extracted data, and assessed risk of bias using the Prediction model Risk Of Bias ASsessment Tool. Performance measures were pooled using random effects meta-analyses. RESULTS: 56 studies were included, reporting 51 developed models and 12 externally validated models. Most models were developed in adults (n=34/51, 67%), primarily for breast cancer (n=20/34, 59%) or haematological malignancies (n=6/34, 18%) to determine pretreatment risk (n=33/34, 97%). In young people, most developed models (n=16/17, 94%) focused on long term risk assessment, mostly in survivors of haematological malignancies. Discrimination and calibration metrics were reported for 13/51 (25%) developed models and 6/44 (14%) external validations. Nearly all models were at high risk of bias. 12/51 (24%) developed models underwent external validation, four of 17 (24%) in young people and eight of 34 (24%) in adult populations. The Heart Failure Association-International Cardio-Oncology Society (HFA-ICOS) tool was the most frequently validated (11 times), mainly in patients with breast cancer receiving HER2 (human epidermal growth factor receptor 2) targeted therapies (5/11, 45%). Across all external validations, this tool consistently underestimated risk, with observed event rates exceeding predicted risks, especially in studies where mild CTRCD was the most frequently reported outcome. Among patients with breast cancer treated with anti-HER2 agents, the pooled C statistic was 0.60 (95% confidence interval 0.52 to 0.68). In this population, observed pooled event rates were 12% in the low risk group (<2% predicted), 15% in the medium risk group (2-9%), 25% in the high risk group (10-19%), and 41% in the very high risk group (≥20%). CONCLUSIONS: Existing prediction models for CTRCD need additional evidence before widespread clinical adoption. Poor reporting of key performance metrics and limited external validation studies currently restrict their thorough evaluation. The HFA-ICOS tool shows suboptimal performance, especially when mild forms of CTRCD are included as events. Future research should prioritise validating and updating existing models using large, clustered datasets across various malignancies to enhance the assessment of their performance, generalisability, and clinical utility in routine practice. SYSTEMATIC REVIEW REGISTRATION: PROSPERO CRD42023475469.
引用
- Gomes C, Geels J, Debray TPA, et al. Risk prediction models for cancer therapy related cardiac dysfunction in patients with cancer and cancer survivors: systematic review and meta-analysis. BMJ. 2025. PMID: 40987514. doi:10.1136/bmj-2025-084062
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。