当前位置: X-MOL 学术Eur. Heart J. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Improving Cardiovascular Risk Prediction: A Tale of Sisyphus
European Heart Journal ( IF 39.3 ) Pub Date : 2017-10-10 , DOI: 10.1093/eurheartj/ehx558
Marc L De Buyzere 1 , Thierry C Gillebert 1, 2
Affiliation  

Survival analysis in prognostic studies has usually relied on Kaplan– Meier and/or multivariable Cox proportional hazard models. Earlier studies suggesting the superiority of 24 h ambulatory blood pressure (ABP) over and above office blood pressure (BP), and the incremental prognostic value of night-time ABP are no exception. One of the most important limitations of conventional survival analysis is the censoring of the survival time at the end of the observation period, for subjects not experiencing an event of interest. During the last decade, methodologies integrating the concept of competing risk analysis have gained popularity for remediation of this limitation. For prognosis-oriented research questions, Fine and Gray developed a sub-distribution hazard model. For more aetiology-oriented research questions, multivariable cause-specific proportional hazard (sometimes termed cause-specific Cox regression) models have been recommended. Applying competing risk analysis is precisely what Mortensen et al. have done in their analysis of the large International Database on Ambulatory blood pressure monitoring in relation to Cardiovascular Outcomes (IDACO). In addition, they developed a prediction model of 10 year person-specific absolute risks of cardiovascular (CV) mortality and CV events. It is a retrospective multi-centre analysis on six cohorts of subjects representing the general population of various countries: Copenhagen (Denmark), Ohasama (Japan), Noorderkempen (Belgium), Uppsala (Sweden), Montevideo (Uruguay), and Maracaibo (Venezuela). On a total of 10 425 subjects, 7929 were considered to have adequate BP readings and sufficient data on covariates and outcome to be included. Mean age was 57 years, mean body mass index was 26 kg/m, and mean office systolic BP was 135 mmHg. Prevalence of diabetes and treated hypertension was 9 and 23%, respectively. Of note, mean systolic BP was 147 mmHg in Uppsala and 171 mmHg (!) in Maracaibo, the other means being around 130 mmHg. Minimum median follow-up was 9.1 years. During the follow-up period, a total of 563 participants died from CV events and 758 died from non-CV events. Further, a total of 1173 were diagnosed with a fatal or non-fatal CV event. Without doubt, it is a large database with long and detailed follow-up. For the present analyses, the authors removed six of the original studies included in the IDACO database, presumably because of a too short follow-up period. Earlier, in part of the original IDACO database (Belgium, Denmark, Japan, and Sweden), ABP proved superior to office BP in predicting CV but not total and non-CV mortality. When two correlated variables such as ABP and office BP are entered in a classical Cox regression model, one of them may appear more important than the other. The background idea of the study under scrutiny is that a somewhat higher hazard ratio for an endpoint (as usually occurs for ABP vs. office BP) does not necessarily translate into better long-term prediction for an individual person (personspecific absolute risk models). It is a well-accepted proposition. The long-term prediction of a person’s risk not only depends on the association of BP with CV events but also on the competing risk of nonCV mortality. In the study of Mortensen et al., cause-specific Cox regression models failed to show clinically relevant added value for 24 h ABP in 10 year risk prediction in a model containing office and 24 h ABP next to common risk indicators. Moreover, night-time BP did not improve 10 year prediction obtained from daytime measurements. For an otherwise healthy population, sufficient prognostic accuracy for CV risks can be achieved with office BP. The interpretation of aetiological research in the presence of competing risks should bear in mind that, because of competing events handled as censored observations, the number of patients at risk during follow-up decreases (removal from the remaining risk sets). The

中文翻译:

改善心血管风险预测:西西弗斯的故事

预后研究中的生存分析通常依赖于 Kaplan-Meier 和/或多变量 Cox 比例风险模型。早期研究表明 24 小时动态血压 (ABP) 优于诊室血压 (BP),夜间 ABP 的增量预后价值也不例外。传统生存分析最重要的限制之一是在观察期结束时对未经历感兴趣事件的受试者的生存时间进行审查。在过去的十年中,整合竞争风险分析概念的方法已广受欢迎,以弥补这一限制。对于面向预后的研究问题,Fine 和 Gray 开发了一个子分布风险模型。对于更多以病因学为导向的研究问题,已推荐使用多变量特定原因比例风险(有时称为特定原因 Cox 回归)模型。应用竞争风险分析正是 Mortensen 等人所做的。他们对与心血管结局相关的动态血压监测大型国际数据库 (IDACO) 进行了分析。此外,他们还开发了一个预测模型,预测 10 年个体特定的心血管 (CV) 死亡率和 CV 事件的绝对风险。这是对代表不同国家一般人群的六组受试者进行的回顾性多中心分析:哥本哈根(丹麦)、奥哈萨马(日本)、诺德肯彭(比利时)、乌普萨拉(瑞典)、蒙得维的亚(乌拉圭)和马拉开波(委内瑞拉) )。在总共 10 425 个科目中,7929 被认为具有足够的 BP 读数和足够的协变量和结果数据。平均年龄为 57 岁,平均体重指数为 26 kg/m,平均诊室收缩压为 135 mmHg。糖尿病和治疗高血压的患病率分别为 9% 和 23%。值得注意的是,乌普萨拉的平均收缩压为 147 毫米汞柱,马拉开波为 171 毫米汞柱(!),其他平均值约为 130 毫米汞柱。最短中位随访时间为 9.1 年。在随访期间,共有 563 名参与者死于心血管事件,758 名参与者死于非心血管事件。此外,共有 1173 人被诊断为致命或非致命 CV 事件。毫无疑问,它是一个大型数据库,后续长期详细。对于目前的分析,作者删除了 IDACO 数据库中包含的六项原始研究,大概是因为随访时间太短。早些时候,在部分最初的 IDACO 数据库(比利时、丹麦、日本和瑞典)中,ABP 在预测 CV 而非总死亡率和非 CV 死亡率方面被证明优于办公室 BP。当在经典 Cox 回归模型中输入两个相关变量(例如 ABP 和办公室 BP)时,其中一个可能看起来比另一个更重要。接受审查的研究的背景思想是,终点风险比(通常发生在 ABP 与办公室 BP 之间)并不一定转化为对个人更好的长期预测(个人绝对风险模型)。这是一个被广泛接受的提议。对一个人风险的长期预测不仅取决于 BP 与 CV 事件的关联,还取决于非 CV 死亡率的竞争风险。在 Mortensen 等人的研究中,在包含办公室和 24 小时 ABP 的模型中,特定原因的 Cox 回归模型未能在常见风险指标旁边显示 24 小时 ABP 在 10 年风险预测中的临床相关附加值。此外,夜间血压并没有改善从白天测量中获得的 10 年预测。对于其他方面健康的人群,使用诊室血压可以达到足够的 CV 风险预后准确性。在存在竞争风险的情况下对病原学研究的解释应牢记,由于竞争事件作为审查观察处理,随访期间处于风险中的患者数量减少(从剩余风险集中删除)。这 在一个包含办公室和 24 小时 ABP 的模型中,特定原因的 Cox 回归模型未能显示 24 小时 ABP 在 10 年风险预测中的临床相关附加值,并与常见风险指标相邻。此外,夜间血压并没有改善从白天测量中获得的 10 年预测。对于其他方面健康的人群,使用诊室血压可以达到足够的 CV 风险预后准确性。在存在竞争风险的情况下对病原学研究的解释应牢记,由于竞争事件作为审查观察处理,随访期间处于风险中的患者数量减少(从剩余风险集中删除)。这 在一个包含办公室和 24 小时 ABP 的模型中,特定原因的 Cox 回归模型未能显示 24 小时 ABP 在 10 年风险预测中的临床相关附加值,并与常见风险指标相邻。此外,夜间血压并没有改善从白天测量中获得的 10 年预测。对于其他方面健康的人群,使用诊室血压可以达到足够的 CV 风险预后准确性。在存在竞争风险的情况下对病原学研究的解释应牢记,由于竞争事件作为审查观察处理,随访期间处于风险中的患者数量减少(从剩余风险集中删除)。这 对于其他方面健康的人群,使用诊室血压可以达到足够的 CV 风险预后准确性。在存在竞争风险的情况下对病原学研究的解释应牢记,由于竞争事件作为审查观察处理,随访期间处于风险中的患者数量减少(从剩余风险集中删除)。这 对于其他方面健康的人群,使用诊室血压可以达到足够的 CV 风险预后准确性。在存在竞争风险的情况下对病原学研究的解释应牢记,由于竞争事件作为审查观察处理,随访期间处于风险中的患者数量减少(从剩余风险集中删除)。这
更新日期:2017-10-10
down
wechat
bug