Incident cardiovascular disease risk prediction using extensive oximetry patterns.
Incident cardiovascular disease risk prediction using extensive oximetry patterns.
复制标题
作者：
在多项研究中，阻塞性睡眠呼吸暂停（OSA）与心血管疾病（CVD）风险增加相关[1-3]。然而，其关联的性质需要进一步的研究，包括如何最好地预测哪些患者最终发展为CVD。这是一项具有挑战性的任务，一方面，呼吸暂停低通气指数（AHI;可能会掩盖OSA内在型）[4，5]提供的异质信息，另一方面，多导睡眠图信号的复杂性（定义最佳信号的指导很少）。识别提高CVD风险预测准确性的最佳信号或信号组合将改善风险分层和个性化医疗。理想情况下，在广泛的候选信号范围内进行搜索还将提供对潜在生物学机制的见解，并在适当考虑有助于这些信号的更大生物学背景时有助于设计改进的OSA临床试验措施[6]。在数千名参与者中测试数十个候选信号仍然具有技术挑战性，即使考虑由国家睡眠研究资源（NSRR）等资源提供的清洁和协调的PSG数据集[7]。在本期中，Kate Sutherland及其同事[8]描述了最全面的血氧测定模式分析，这可能有助于预测OSA患者迄今为止发生的CVD。作者检查了睡眠心脏健康研究参与者的一个子集，患有OSA（定义为AHI≥ 5），没有预先存在的CVD。中位随访时间为11.5年。在主要分析中检查了31个候选信号，并在性别分层和NREM和REM特异性分析中进一步检查。这些信号被分为四个域[9]。除了使用低氧负荷和类似测量方法探索的去饱和特征外，Sutherland及其同事还检查了饱和值的时间序列（例如SpO 2值分布的形状）、血氧测定信号的功率谱密度（例如曲线的OSA带部分分布的形状）和非线性分析（例如类似血氧测定模式的规律性）。作者没有在合并样本或男性中确定任何与Bonferroni校正显著性下的CVD事件相关的个体信号，但确实确定了女性中的暗示性关联，包括氧去饱和指数（ODI）为5%，最低SpO 2，功率谱密度的OSA频带平均值，和SpO 2分布标准差（p 0.007-0.014）。提示与缺氧负荷的相关性存在于女性（p= 0.02），但不存在于男性（p= 0.91）。多个候选信号是比AHI更好的CVD事件个体预测因子（p≥ 0.26）。尽管缺乏统计学显著性结果，但这项研究仍然是确定CVD候选预测信号的重要一步，应在未来的研究中继续进行。在一系列部分相关的候选信号中识别与临床结果相关联的最佳信号的广泛的“网格搜索”是超越AHI以识别更多临床相关的PSG度量的有效手段。Sutherland等人已经证明了他们的算法可以扩展到数千个记录，这是实际解决流行病学规模问题的必要门槛。多个提示性关联和先前对具有更广泛AHI值和广泛相似终点的相同数据集的研究[2]表明，研究结果的改善可能存在显著关联。
Obstructive sleep apnea (OSA) has been associated with increased risk of developing cardiovascular disease (CVD) in multiple studies [1–3]. However, the nature of its association requires further investigation, including how to best predict which patients ultimately develop CVD. This is a challenging task, given the heterogeneous information provided by the apnea–hypopnea index (AHI; which may obscure OSA endotypes)[4, 5] on one hand, and the complex nature of polysomnographic signals (with minimal guidance on defining optimal signals) on the other. Identifying an optimal signal or combination of signals that improves the accuracy of CVD risk prediction would improve risk stratification and personalized medicine. Ideally a search across a broad range of candidate signals would also provide insights on potential biological mechanisms and contribute to the design of improved OSA clinical trial measures when appropriately considering the larger biological context that contributes to these signals [6]. Testing dozens of candidate signals among thousands of participants remains technically challenging, even when considering cleaned and harmonized PSG datasets provided by resources such as the National Sleep Research Resource (NSRR)[7]. In this issue, Kate Sutherland and colleagues [8] describe the most comprehensive analysis of oximetry patterns that may help to predict incident CVD among patients with OSA to date. The authors examined a subset of Sleep Heart Health Study participants with OSA (as defined by AHI≥ 5) and without preexisting CVD. The median follow-up was 11.5 years. Thirty-one candidate signals were examined in primary analyses and further examined in sex-stratified and NREM-and REM-specific analyses. These signals were divided into four domains [9]. In addition to desaturation characteristics that have been explored using the hypoxic burden and similar measures, Sutherland and colleagues examined the time series of saturation values (eg the shape of the distribution of SpO2 values), the power spectral density of the oximetry signal (eg the shape of the distribution of the OSA-band portion of the curve), and nonlinear analyses (eg the regularity of similar oximetry patterns). The authors did not identify any individual signals that were associated with incident CVD at Bonferroni-adjusted significance in the combined sample or in men, but did identify suggestive associations in women that include the oxygen desaturation index (ODI) at 5%, the nadir SpO2, the mean of the OSA frequency band of the power spectral density, and the SpO2 distribution standard deviation (p 0.007–0.014). A suggestive association with the hypoxic burden was present in women (p= 0.02) but not in men (p= 0.91). Multiple candidate signals were better individual predictors of incident CVD than the AHI (p≥ 0.26). Despite a lack of statistically significant results, this study is nevertheless an important step in identifying candidate predictive signals for CVD that should be continued in future studies. A broad “grid search” that identifies optimal signals associated with clinical outcomes among a range of partially correlated candidate signals is a productive means of moving beyond the AHI to identify more clinically relevant PSG metrics. Sutherland et al. have demonstrated the scalability of their algorithms to thousands of recordings, which is a necessary threshold for practically addressing epidemiological-scale questions. Multiple suggestive associations and a prior study of the same dataset with a broader range of AHI values and broadly similar endpoints [2] indicate that significant associations are possible with improvements in study …
登录
查看更多内容
影响因子：
5.6
作者：
通讯作者：
--
DOI：
10.1164/rccm.202010-3900oc
发表时间：
2021-06-15
影响因子：
24.7
作者：
Azarbarzin, Ali;Sands, Scott A.;Wellman, Andrew
通讯作者：
Wellman, Andrew
影响因子：
5.6
作者：
Won, Christine H. J.;Reid, Michelle;Redline, Susan
通讯作者：
Redline, Susan
影响因子：
24
作者：
McClelland RL;Jorgensen NW;Budoff M;Blaha MJ;Post WS;Kronmal RA;Bild DE;Shea S;Liu K;Watson KE;Folsom AR;Khera A;Ayers C;Mahabadi AA;Lehmann N;Jöckel KH;Moebus S;Carr JJ;Erbel R;Burke GL
通讯作者：
Burke GL
DOI：
10.1164/rccm.202105-1274oc
发表时间：
2022-01-01
影响因子：
24.7
作者：
Trzepizur, Wojciech;Blanchard, Margaux;Gagnadoux, Frederic
通讯作者：
Gagnadoux, Frederic