Hyfe咳嗽监测系统的验证和准确性：一项多中心临床研究-动脉网

Validation and accuracy of the Hyfe cough monitoring system: a multicenter clinical study

Nature 等信源发布 2025-01-06 15:25



可切换为仅中文







Abstract

摘要

Background: The ability to passively and continuously monitor coughing for prolonged periods of time would significantly improve cough management and research. To date there is no automated clinically validated cough monitor that can be routinely used in clinical care and research. Here we describe the validation of such an automated cough monitor.

背景：长期被动持续监测咳嗽的能力将显着改善咳嗽管理和研究。迄今为止，还没有经过临床验证的自动咳嗽监测仪可以常规用于临床护理和研究。在这里，我们描述了这种自动咳嗽监测仪的验证。

Methods: This multicenter observational study compared the results of the Hyfe CoughMonitor wrist-worn device with manually counted coughs in subjects with a variety of etiologies as they went about their usual daily activities. We collected 24 h of continuous sounds from subjects while they simultaneously wore a CoughMonitor and an audio recorder.

方法：这项多中心观察性研究比较了Hyfe CoughMonitor腕带装置与手动计数咳嗽的结果，这些患者在进行日常活动时有多种病因。我们收集了受试者同时佩戴咳嗽监护仪和录音机的24小时连续声音。

Coughs were labelled by multiple trained annotators who listened to the continuous audio recordings using validated methodology. The time stamps of these human-detected coughs were compared to those of the CoughMonitor to determine the system’s overall performance using event-to-event and hourly rate correlation analyses.

咳嗽由多名训练有素的注释员标记，他们使用经过验证的方法收听连续的录音。将这些人类检测到的咳嗽的时间戳与CoughMonitor的时间戳进行比较，以使用事件到事件和每小时速率相关分析来确定系统的整体性能。

Results: Over the 546 h monitored, 4,454 cough events were recorded; The overall sensitivity was 90.4% (95% CI of 88.3–92.2%). The overall false positive rate was 1.03 false positives per hour (95% CI of 0.84 to 1.24). The overall correlation between manual and CoughMonitor measured hourly coughing was high (Pearson correlation coefficient of 0.99).

结果：在监测的546小时内，记录了4454次咳嗽事件；总体敏感性为90.4%（95%可信区间为88.3-92.2%）。总体假阳性率为每小时1.03个假阳性（95%CI为0.84至1.24）。手动和CoughMonitor测量的每小时咳嗽之间的总体相关性很高（Pearson相关系数为0.99）。

Two case studies of long-term monitoring of patients with chronic cough are presented. Conclusion: The present analysis of cough events demonstrated that the Hyfe CoughMonitor accurately reflects them with a high sensitivity and a low false positive rate. Future studies should focus on its potential role in the management of patients with cough in clinical practice..

本文介绍了对慢性咳嗽患者进行长期监测的两个案例研究。结论：目前对咳嗽事件的分析表明，Hyfe咳嗽监测仪以高灵敏度和低假阳性率准确反映了咳嗽事件。未来的研究应侧重于其在临床实践中对咳嗽患者管理的潜在作用。。

Registration Clinicaltrials.gov, NCT05723159.

注册Clinicaltrials.gov，NCT05723159。

Introduction

简介

Cough is one of the most common reasons for which patients seek medical care, it is associated with a broad range of medical conditions and greatly contributes to healthcare expenditure all over the world

咳嗽是患者寻求医疗保健的最常见原因之一，它与广泛的医疗条件有关，并极大地促进了全世界的医疗保健支出

. For some diseases, such as COPD and COVID-19, cough rates also help to predict adverse outcomes

对于某些疾病，如COPD和COVID-19，咳嗽率也有助于预测不良后果

. However, in an era when symptom quantification drives refinement in diagnosis and precision in therapy

然而，在一个症状量化推动诊断精细化和治疗精确化的时代

, cough is currently not measured as part of clinical practice. This is not because of lack of interest, as efforts to quantify cough date back to the 1950’s

，咳嗽目前尚未作为临床实践的一部分进行测量。这并不是因为缺乏兴趣，因为量化咳嗽的努力可以追溯到20世纪50年代

Improvements in acoustic signal processing and machine learning techniques have fostered renewed attention to contactless fully automated cough monitoring

声信号处理和机器学习技术的改进促使人们重新关注非接触式全自动咳嗽监测

. Objective, prolonged cough monitoring can provide a valuable data stream for the diagnosis, prognosis, assessment of treatment response and even syndromic surveillance of respiratory diseases as well as for the development of novel therapeutics. However, there are currently no validated, unobtrusive, and fully-automated cough monitoring systems that are commonly used to monitor cough as patients go about their normal activities.

目的，长期咳嗽监测可以为呼吸系统疾病的诊断，预后，治疗反应评估甚至症状监测以及新疗法的开发提供有价值的数据流。然而，目前还没有经过验证的，不引人注目的，全自动的咳嗽监测系统，这些系统通常用于在患者进行正常活动时监测咳嗽。

Validation of such systems relies upon having accurate ground truth cough data as gold standard. There is consensus that manual cough counting from audio recordings, albeit laborious, can achieve good interobserver agreement

这些系统的验证依赖于拥有准确的地面实况咳嗽数据作为金标准。人们一致认为，从录音中手动计数咳嗽虽然很费力，但可以获得良好的观察者间一致性

. When manual labeling cough acoustic data, the cough second (a second containing at least one explosive phase of a cough) is a unit of annotation that correlates well with true cough rates and reduces the ambiguity associated with the use of individual explosive phases or cough epochs

.当手动标记咳嗽声学数据时，咳嗽秒（第二个包含咳嗽的至少一个爆发阶段）是一个注释单位，与真实咳嗽率密切相关，并减少了与使用单个爆发阶段或咳嗽时期相关的歧义

The primary objective of the study was to assess the overall performance of the Hyfe CoughMonitor (Hyfe Inc., 2022), when used by individuals with problematic cough, in comparison to the gold standard of manual cough annotation. Additionally, we also compared the CoughMonitor’s performance during the daytime versus nighttime, between individuals and as a function of cough rates..

这项研究的主要目的是评估Hyfe咳嗽监测仪（Hyfe Inc.，2022）的总体性能，与手动咳嗽注释的金标准相比，当被有问题的咳嗽患者使用时。此外，我们还比较了咳嗽监测仪在白天和夜间的表现，个人之间以及咳嗽率的函数。。

Methods

方法

The Hyfe CoughMonitor

Hyfe咳嗽监测仪

The CoughMonitor App runs on an Android Smartwatch (Shenszhen Domino Times Smart 4G Watch, Model DM20). In brief, this app uses the watch’s microphone to continuously capture and encrypt ambient sounds in a manner that cannot be replayed and is deleted after processing, thus ensuring that sound recordings are secure and transient.

CoughMonitor应用程序运行在Android智能手表（Shenszhen Domino Times Smart 4G Watch，型号DM20）上。简而言之，该应用程序使用手表的麦克风以一种无法重放且在处理后被删除的方式连续捕获和加密环境声音，从而确保录音安全且短暂。

An artificial intelligence (AI) based algorithm was designed and implemented in CoughMonitor to detect cough sounds from raw audio recordings. A short description follows next. First, an acoustic event detector looks for onset acoustic events that are similar to the explosive phase of a typical cough sound.

。下面是一个简短的描述。首先，声学事件检测器寻找类似于典型咳嗽声爆发阶段的起始声学事件。

Second, once such events have been detected, they are segmented into 0.5-second chunks, transformed into an image-like representation that illustrates the acoustic energy distribution of the segmented sound over time and frequency, and passed to a convolutional neural network (CNN), a popular machine learning model originating from image processing and computer vision.

其次，一旦检测到这样的事件，它们就会被分割成0.5秒的块，转换成类似图像的表示，该表示说明了分割声音随时间和频率的声能分布，并传递给卷积神经网络（CNN），这是一种源自图像处理和计算机视觉的流行机器学习模型。

Finally, the neural network – trained on .

最后，训练神经网络。

millions

百万

of coughs and cough-like sound segments – decides whether the input audio corresponds to a cough sound or not, and in the former case, a timestamp is generated. All processing happens in-device. While charging, timestamps and cough durations are transmitted via Wi-Fi to the Hyfe cloud, where the time of each cough is converted to cough seconds.

咳嗽和类似咳嗽的声音片段-决定输入音频是否对应咳嗽声音，在前一种情况下，会生成时间戳。所有处理都在设备中进行。充电时，时间戳和咳嗽持续时间通过Wi-Fi传输到Hyfe云，每次咳嗽的时间转换为咳嗽秒。

In this study, all uploading was handled by study personnel at the end of the monitoring period of each participant. The same version of the cough detection software was used throughout this study (version 1.0.0)..

在这项研究中，所有上传都是由研究人员在每个参与者的监测期结束时处理的。在整个研究过程中使用了相同版本的咳嗽检测软件（版本1.0.0）。。

Continuous sounds were recorded using the same Android smartwatch (Shenszhen Domino Times Smart 4G Watch, Model DM20), but running custom software that continuously recorded all ambient sounds.

使用相同的Android smartwatch（Shenszhen Domino Times Smart 4G Watch，型号DM20）记录连续的声音，但运行的是连续记录所有环境声音的定制软件。

Study design

研究设计

This is a multicenter observational study of individuals with problematic cough due to a variety of cough related conditions designed to assess the overall performance of the Hyfe CoughMonitor System in comparison to manually counted cough events.

这是一项多中心观察性研究，针对因各种咳嗽相关疾病而出现咳嗽问题的个体，旨在评估Hyfe CoughMonitor系统与手动计数咳嗽事件相比的整体性能。

Enrollment and eligibility

注册和资格

Individuals of both sexes who had problematic coughs consulting at two clinical sites between March 17 and Nov 7, 2023: (1) Oregon Health & Science University (OHSU) in the US, and (2) University Clinic of Navarra in Spain A third group of participants was enrolled remotely in a decentralized manner in the US through targeted outreach.

2023年3月17日至11月7日期间，在两个临床地点咨询有问题咳嗽的男女个人：（1）美国俄勒冈健康与科学大学（OHSU），以及（2）西班牙纳瓦拉大学诊所第三组参与者通过有针对性的外联在美国以分散的方式远程登记。

Inclusion and exclusion criteria are shown in Table .

纳入和排除标准如表所示。

Table 1 Study inclusion and exclusion criteria.

表1研究纳入和排除标准。

Full size table

全尺寸表

Sample size

样本量

Hourly cough counts are necessarily non-negative integers and follow negative binomial distributions closely

每小时咳嗽计数必然是非负整数，并且密切遵循负二项分布

, neither these counts nor any simple transformations thereof are normally distributed, precluding application of standard formulas for the SEs of the Pearson correlation or linear regression coefficients. A simulation based on data collected in previous studies

，这些计数或其任何简单变换都不是正态分布的，因此无法将标准公式应用于Pearson相关系数或线性回归系数的SE。基于先前研究中收集的数据的模拟

to estimate the sampling distribution of hourly cough rates and showed that a minimum of 18 participants contributing 20 h of monitoring each, yielding 360 paired person-hours for analysis, would result in a standard error under 0.1 in average correlations and slopes (see online data supplement for the study protocol).

为了估计每小时咳嗽率的抽样分布，并表明至少有18名参与者每人贡献20小时的监测，产生360对成对的人时进行分析，将导致平均相关性和斜率的标准误差低于0.1（见在线数据补充研究方案）。

This target was later expanded to 23 participants in order to have over 50% of the sample recruited in the US for regulatory purposes..

这一目标后来扩大到23名参与者，以便在美国招募超过50%的样本用于监管目的。。

Data acquisition

数据采集

After obtaining informed consent, research subjects were instructed to wear two devices: (i) the Hyfe CoughMonitor on one wrist and (ii) a second identical watch running custom software that continuously recorded all ambient sounds on the other wrist.

在获得知情同意后，研究对象被指示佩戴两种设备：（i）一只手腕上的Hyfe CoughMonitor和（ii）第二个相同的手表，运行定制软件，连续记录另一只手腕上的所有环境声音。

At bedtime, subjects were instructed to charge both devices at bedside. The exact start and stop time of monitoring was recorded by study personnel. Subjects were instructed to monitor for 24 h and write down the times they went to bed, awoke and any times they had to leave the watch aside such as while showering.

在就寝时间，指示受试者在床边为两个设备充电。研究人员记录了监测的确切开始和停止时间。受试者被要求监测24小时，并记录下他们上床睡觉的时间，醒来的时间以及他们必须将手表放在一边的任何时间，例如洗澡时。

Participants who recorded for less than 20 h were excluded from analysis..

记录时间少于20小时的参与者被排除在分析之外。。

Cough annotation

咳嗽注释

To obtain the gold standard, the exact start time of each cough was manually annotated on the continuous recordings as previously described

为了获得金标准，如前所述，在连续记录中手动注释每次咳嗽的确切开始时间

. In brief, continuous audio recordings were broken into 60 s segments and presented to two trained annotators using a proprietary audio playback, visualization and data entry program. Each 60 s segment was listened to by the annotators independently and blinded to one another´s results. Every cough, throat clear, sneeze or cough-like sounds was labeled as a segment, from its beginning to its end, noting if the sound was distant or unclear.

简而言之，连续录音被分成60秒的片段，并使用专有的音频播放，可视化和数据输入程序呈现给两位训练有素的注释者。注释者独立听取每个60秒的片段，并且对彼此的结果视而不见。每一个咳嗽、喉咙清晰、打喷嚏或咳嗽样的声音都被标记为一个片段，从开始到结束，注意声音是否遥远或不清楚。

Sounds for which the two labelers disagreed on the presence of a cough, the timing of a cough’s start by greater than 100 milliseconds, or the indication that the cough sound was distant or unclear were adjudicated by a third expert trained annotator. The adjudicator was aware of the discrepant annotations and listened to each 60s audio segment containing discrepancies in full.

两名贴标员对咳嗽的存在，咳嗽开始的时间超过100毫秒或咳嗽声音遥远或不清楚的迹象不一致的声音由第三位受过专家训练的注释员裁定。裁判员意识到不一致的注释，并听取了每个60年代的音频片段，其中包含完整的不一致之处。

The interobserver variability of this system has been quantified as negligible with an inter-labeler Pearson´s correlation of 0.96 .

该系统的观察者间变异性被量化为可忽略不计，标签间Pearson相关性为0.96。

Event classification

事件分类

The timestamps of automatically (CoughMonitor) and manually (human annotated) detected cough events were described and compared statistically. Each timestamp was converted to cough seconds, defined as a second during which at least one individual cough occurs

描述并统计比较了自动（CoughMonitor）和手动（人类注释）检测到的咳嗽事件的时间戳。每个时间戳都转换为咳嗽秒，定义为至少发生一次咳嗽的秒

. As previously shown, cough seconds can be interchangeable units with explosive phases as units of cough

如前所述，咳嗽秒可以互换，爆炸相可以作为咳嗽的单位

and a more consistent annotation metric

以及更一致的注释度量

. We used the following event definitions:

。我们使用了以下事件定义：

True Positive: a cough detected by the CoughMonitor matching human annotators within a 0.5 s margin.

真阳性：CoughMonitor在0.5 s范围内检测到与人类注释器匹配的咳嗽。

False Positive: a “cough” detected by the CoughMonitor but either (a) not labeled by humans or (b) labeled by humans as a different sound i.e., throat clear, sneeze or (c) not matching coughs labeled by humans within 0.5 s.

假阳性：咳嗽监测仪检测到“咳嗽”，但（a）未被人类标记，或（b）被人类标记为不同的声音，即喉咙清晰，打喷嚏，或（c）在0.5秒内与人类标记的咳嗽不匹配。

Statistical analysis

统计分析

Results were analyzed on the basis of cough seconds in two complementary ways: (1) an event-to-event comparison of each cough second yielding the CoughMonitor’s sensitivity, false positive rate and positive predictive value, and (2) a correlation analysis of hourly cough rates comparing human annotators and the monitor presented as a Bland-Altman plot and, in the online data supplement, as linear plots..

结果以咳嗽秒为基础，以两种互补的方式进行分析：（1）每个咳嗽秒的事件对事件比较，产生咳嗽监护仪的敏感性，假阳性率和阳性预测值，以及（2）每小时咳嗽率的相关性分析，比较人类注释器和监护仪以Bland-Altman图的形式呈现，并在在线数据增补中以线性图的形式呈现。。

All statistical analyses were performed in R (R Core Team (2022). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria).

所有统计分析均在R（R Core Team（2022））中进行。R：用于统计计算的语言和环境。R统计计算基金会，奥地利维也纳）。

Cough event based analysis

基于咳嗽事件的分析

To calculate the CoughMonitor`s sensitivity, false positive rate and positive predictive value (PPV) as measures of its success at detecting individual cough seconds, we compared (i) ground truth annotations (converted to cough seconds) with (ii) cough seconds as per the CoughMonitor per hour. When a cough second was detected by the CoughMonitor within 0.5 s of a human labeled cough, the event was considered to be matched and therefore a “true positive”.

为了计算CoughMonitor的灵敏度，假阳性率和阳性预测值（PPV）作为检测单个咳嗽秒数成功的指标，我们比较了（i）地面真相注释（转换为咳嗽秒数）和（ii）咳嗽秒数根据CoughMonitor每小时。当CoughMonitor在人类标记咳嗽的0.5秒内检测到咳嗽秒时，该事件被认为是匹配的，因此是“真正的阳性”。

Following a match, both the CoughMonitor cough second and human labeled cough second are removed from eligibility for further matching (i.e., if a human-labeled cough fell within 0.5 s of the CoughMonitor, the human-labeled cough could only match with one CoughMonitor cough), meaning that only one event is considered a true positive; the remaining, unmatched CoughMonitor event would be considered a false positive despite it being within 0.5 s of a human labeled cough.

匹配后，CoughMonitor咳嗽二级和人类标记的咳嗽二级都被取消进一步匹配的资格（即，如果人类标记的咳嗽在CoughMonitor的0.5秒内，人类标记的咳嗽只能与一次CoughMonitor咳嗽匹配），这意味着只有一次事件被认为是真正的阳性；剩下的，无与伦比的CoughMonitor事件将被认为是假阳性，尽管它在人类标记咳嗽的0.5秒内。

A “false positive” was considered to be a CoughMonitor detection which was not matched to a human labeled cough. By comparison with ground truth annotations, each CoughMonitor timestamp is therefore either a true positive or a false positive. Sensitivity and the false positive rate were then calculated according to the usual formulas,.

“假阳性”被认为是咳嗽监测仪检测，与人类标记的咳嗽不匹配。因此，与地面实况注释相比，每个CoughMonitor时间戳要么是真阳性，要么是假阳性。然后根据通常的公式计算灵敏度和假阳性率，。

CoughMonitor Sensitivity (%) =

咳嗽监护仪灵敏度（%）=

\(\:\frac{Total\:number\:of\:true\:positives}{Total\:number\:of\:cough\:seconds}\)

\（\：\frac{总数\：次数\：of \：真\：正}{总数\：次数\：of \：咳嗽\：秒}）

False positive rate (/hr) =

假阳性率（/小时）=

\(\:\frac{Total\:number\:of\:false\:positives}{Total\:number\:of\:hours\:of\:monitoring}\)

\（\：\frac{总数\：数量\：of \：假\：正}{总数\：数量\：of \：小时\：of \：监控}）

Positive predictive value (%) =

阳性预测值（%）=

\(\:\frac{Total\:number\:of\:true\:positives}{Total\:number\:of\:true\:and\:false\:positives}\)

Cough rate based analysis

基于咳嗽率的分析

To evaluate the CoughMonitor’s overall performance, its hourly cough second counts were compared, person-hour by person-hour, with the gold standard hourly cough second counts obtained from trained human annotators. The agreement between these paired person-hour counts and the ideal model

为了评估CoughMonitor的整体性能，将其每小时咳嗽秒数与从训练有素的人类注释员那里获得的金标准每小时咳嗽秒数进行了比较。

\(\:y\:=\:x\:+\:\epsilon\)

\（\：y\：=\：x\：+\：\ε\）

, where

，其中

\(\:\epsilon\)

\（\：\ε\）

denotes an integer-valued random error term with mean 0 and constant variance, was assessed with both a linear analysis and a Bland-Altman analysis.

表示均值为0且方差为常数的整数值随机误差项，通过线性分析和Bland-Altman分析进行评估。

To account for within-subject correlations, the clustered bootstrap was used to calculate all of these confidence intervals, with each cluster consisting of one individual’s paired person-hour counts.

为了解释受试者内部的相关性，使用聚类自举来计算所有这些置信区间，每个聚类由一个人的成对人时计数组成。

Secondary analyses

二次分析

We then conducted three pre-specified secondary analyses comparing the CoughMonitor’s performance (1) between day and nighttime, using a two-sample t-test (2) between each individual, using descriptive metrics and (3) as a function of cough rate by means of one-way analyses of the variance.

然后，我们进行了三次预先指定的二次分析，比较了咳嗽监护仪在白天和夜间的表现（1），使用每个人之间的双样本t检验（2），使用描述性指标和（3）作为咳嗽率的函数通过单向方差分析。

There was no missing data in this study.

这项研究没有缺失数据。

Ethics approval and registration

道德审批和注册

The study was approved by the OHSU Institutional Review Board (number 24749), the Ethics Committee for Medical Research in Navarra (number PI_2022/101), and by WCG IRB for the distributed trial (number 20232553). All procedures were performed in accordance with relevant named guidelines and regulations.

该研究得到了OHSU机构审查委员会（编号24749），纳瓦拉医学研究伦理委员会（编号PI\U 2022/101）和WCG IRB的分布式试验（编号20232553）的批准。所有程序均按照相关指定的指南和规定进行。

Informed consent was obtained from all participants. The study was prospectively registered in Clinicaltrials.gov (NCT05723159) on February 03, 2023..

获得了所有参与者的知情同意。该研究于2023年2月3日在Clinicaltrials.gov（NCT05723159）上进行了前瞻性注册。。

Results

结果

Enrollment

A total of 28 participants with problematic coughs due to a variety of etiologies were recruited. A battery misconfiguration resulted in the device turning off in the first 3 subjects who were enrolled and this misconfiguration was corrected after the 3rd participant, the detection algorithm was not changed.

共招募了28名因各种病因引起咳嗽问题的参与者。电池配置错误导致注册的前3名受试者的设备关闭，并且在第3名参与者之后纠正了这种错误配置，检测算法没有改变。

No subsequent technical interruptions were observed. Two participants inadvertently prematurely terminated monitoring prior to 20 h and were thus excluded from analysis (Fig. .

没有观察到随后的技术中断。两名参与者在20小时之前无意中提前终止了监测，因此被排除在分析之外（图）。

). There were no adverse events related to the use of CoughMonitor.

)。没有与使用CoughMonitor有关的不良事件。

Fig. 1

图1

Enrollment flow diagram of study.

。

Full size image

全尺寸图像

The final analysis included data from 23 subjects (Table

最终分析包括23名受试者的数据（表

). Almost two thirds were female (15, 65%), the mean age was 52 (range 24 to 72 years), and 12 (52%) were recruited in the US (5 from Oregon Health Science University and 7 from the dispersed clinical trial). The mean monitoring time was 23.8 h (range 22.9 to 25.64) and the mean hourly cough rate annotated by humans was 8.1 (range 1.5–32).

)。近三分之二是女性（15，65%），平均年龄52岁（24至72岁），美国招募了12名（52%）（5名来自俄勒冈健康科学大学，7名来自分散临床试验）。平均监测时间为23.8小时（范围22.9至25.64），人类注释的平均每小时咳嗽率为8.1（范围1.5-32）。

All diagnoses are provided in Table .

表中提供了所有诊断。

(see

（参见

Table S2

表S2

in the supplementary material).

在补充材料中）。

Table 2 Demographic and cough monitoring results of study subjects.

表2研究对象的人口统计学和咳嗽监测结果。

Full size table

全尺寸表

Continuous recordings

连续录制

A total of 546 h of continuous audio/monitoring time containing 4,454 cough seconds was captured from all participants. The mean number of cough seconds by participant was 200 (range 36 to 821). The mean hourly cough rate was 8.15 (range 1.5 to 32).

所有参与者共捕获了546小时的连续音频/监测时间，其中包含4454咳嗽秒。参与者的平均咳嗽秒数为200（范围36至821）。平均每小时咳嗽率为8.15（范围1.5至32）。

It was apparent to annotators listening to the continuous recordings that most monitoring occurred in subjects’ homes and workplaces. However, monitoring also occurred in a wide variety of acoustic environments such as grocery stores, restaurants, in vehicles, and even while listening to loud techno music..

显然，对于收听连续录音的注释者来说，大多数监测发生在受试者的家中和工作场所。然而，监测也发生在各种各样的声学环境中，例如杂货店、餐馆、汽车中，甚至在听响亮的科技音乐时。。

Cough event based performance results

基于咳嗽事件的性能结果

The overall sensitivity was 90.4% (95% CI of 88.3–92.2%). The overall false positive rate was 1.03 false positives per hour (95% CI of 0.84 to 1.24). The overall positive predictive value was 87.5% (95% CI of 81.9–91.6%). Figure

总体敏感性为90.4%（95%可信区间为88.3-92.2%）。总体假阳性率为每小时1.03个假阳性（95%CI为0.84至1.24）。总体阳性预测值为87.5%（95%CI为81.9-91.6%）。图

shows the cumulative number of cough seconds detected by the CoughMonitor and the human annotators. The sex, age and diagnosis of the participants had no impact on performance (Figure

显示CoughMonitor和人类注释器检测到的咳嗽秒数。参与者的性别、年龄和诊断对表现没有影响（图

S1级

, S2 and S3 in the online data supplement).

，在线数据增补中的S2和S3）。

The sensitivity rate for individual subjects ranged from 78.1 to 96.5% with false positivity rates ranging from 0.38 to 2.23 cough seconds per hour (Fig.

个体受试者的敏感性为78.1%至96.5%，假阳性率为每小时0.38至2.23咳嗽秒（图）。

, Figure S4).

，图S4）。

Fig. 2

图2

Cumulative cough event-based performance results. Cumulative cough seconds from all participants over the course of the study with time. The ground truth human counts are shown in black and the Hyfe results are shown in red.

基于累积咳嗽事件的表现结果。随着时间的推移，研究过程中所有参与者的累积咳嗽秒数。地面真相人类计数以黑色显示，Hyfe结果以红色显示。

Full size image

全尺寸图像

Fig. 3

图3

Individual performance results.

个人绩效结果。

Full size image

全尺寸图像

Cough rate based performance results

基于咳嗽率的性能结果

CoughMonitor performance based on hourly cough rate is shown in a standard Bland-Altman plot in Fig.

基于每小时咳嗽率的CoughMonitor性能显示在Fig.的标准Bland-Altman图中。

, providing a visual assessment of the agreement between the hourly cough rates as measured by the CoughMonitor and by the human annotators. We calculated the upper and lower 95% limits of agreement (LOAs), and also computed the 95% confidence interval for the upper and lower COAs using cluster bootstrap resampling techniques.

，提供了由CoughMonitor和人类注释器测量的每小时咳嗽率之间的一致性的视觉评估。我们计算了95%的一致性上限和下限（LOAs），并使用聚类自举重采样技术计算了上下COAs的95%置信区间。

As shown by the dashed horizontal lines, the overall bias (mean difference) is 0.23 (95% CI -0.039 to 0.51), the lower limit of agreement (LOA) is -3.7 (95% CI -5.2 to -3), and the upper limit of agreement is 4.8 (95% CI 4 to 6)..

。。

Fig. 4

图4

Cough rate-based performance results (Bland-Altman plot). Each point is one person-hour; its x-coordinate is the average of its manual and CoughMonitor hourly cough second counts, and its y-coordinate is their difference. The dashed horizontal lines indicate the bias (mean difference) of 0.23, the lower limit of agreement of -3.7, and the upper limit of agreement of 4.8..

基于咳嗽率的性能结果（Bland-Altman图）。每一点为一人小时；。水平虚线表示偏差（平均差）为0.23，一致性下限为-3.7，一致性上限为4.8。。

Full size image

全尺寸图像

Linear cough rate performance results

A linear validation analysis assumes that the paired person-hour counts exhibit a linear relationship. Figure

线性验证分析假设成对的人时数呈现线性关系。图

below confirms this assumption visually, and the overall Pearson correlation coefficient of 0.99 (95% CI 0.962 to 0.996) quantifies the strength of this linear relationship. While having a Pearson correlation coefficient close to + 1 is a necessary criterion for satisfactory performance, this is insufficient alone, as it only implies that the paired counts cluster tightly about some line; the coefficients of the regression line (the dashed black line in Fig. .

下面从视觉上证实了这一假设，总体Pearson相关系数为0.99（95%CI为0.962至0.996）量化了这种线性关系的强度。虽然Pearson相关系数接近+1是令人满意的性能的必要标准，但这仅仅是不够的，因为它只意味着成对的计数紧密地聚集在某条线上；。

) measure how close this line is to the ideal line y = x (the solid black line in Fig.

)测量这条线与理想线y= x（Fig.中的实心黑线）的距离。

). The slope and the intercept of the ordinary least squares (OLS) line of best fit in Fig.

)。。

are 0.94 (95% CI 0.91 to 0.97) and 0.74 (95% CI 0.50 to 0.99), respectively, indicating good agreement and hence satisfactory performance.

。

Because the results reported by the Hyfe CoughMonitor are expressed as hourly coughing rates, we calculated the correlation based on all 477 of the hours monitored. As shown in Fig.

由于Hyfe CoughMonitor报告的结果表示为每小时咳嗽率，因此我们根据监测的所有477小时计算了相关性。如图所示。

, the overall linear correlation was high, with a Pearson correlation coefficient of 0.99, an OLS slope of 0.94, and an OLS intercept of 0.74.

，总体线性相关很高，Pearson相关系数为0.99，OLS斜率为0.94，OLS截距为0.74。

Fig. 5

图5

Cough rate-based performance results (linear plot). Each point is one person-hour, the black dashed line is the OLS line of best fit, and the black solid line is the line of perfect agreement (

基于咳嗽率的性能结果（线性图）。每个点是一个人小时，黑色虚线是最适合的OLS线，黑色实线是完全一致的线(

\(\:y=x\)

\（\：y=x \）

Full size image

全尺寸图像

Discussion

讨论

We evaluated the Hyfe CoughMonitor in over 546 h of continuous audio containing 4,454 coughs from 23 participants with a broad range of cough-causing diagnoses. This resulted in a sensitivity of 90.4% and an overall false positive rate was 1.03 false positives per hour. Thus, the monitor is unique in its ability to continuously monitor cough frequency in a manner that is unobtrusive, fully automated, and privacy preserving as users go about their usual activities of daily living.

我们在超过546小时的连续音频中评估了Hyfe CoughMonitor，其中包含来自23名参与者的4454次咳嗽，这些参与者具有广泛的咳嗽诊断。这导致敏感性为90.4%，总体假阳性率为每小时1.03个假阳性。因此，监护仪的独特之处在于它能够以不引人注目，全自动和隐私保护的方式持续监测咳嗽频率，因为用户可以进行日常生活活动。

These attributes promise to improve patients’ understanding of their cough and its triggers, as well as to improve the diagnosis and management of disease. Furthermore, continuous cough monitoring addresses some of the fundamental statistical inadequacies of short-term cough monitoring that have limited clinical trials of new antitussive drugs as recently raised by the US Food and Drug Administration.

。此外，持续咳嗽监测解决了短期咳嗽监测的一些基本统计缺陷，这些缺陷限制了美国食品和药物管理局最近提出的新型镇咳药物的临床试验。

Using cough seconds as unit of analysis and applying a previously described annotation protocol, has proven accurate to establish the ground truth, showing an intra-labeler disagreement of less than two per hour monitored; (Pearson’s correlation 0.98) and inter-labeler agreement with a Pearson’s correlation of 0.96 .

使用咳嗽秒作为分析单位，并应用先前描述的注释协议，已被证明可以准确地确定基本事实，显示标签内不一致的情况每小时监测不到两次；（皮尔逊相关系数0.98）和标签间一致性，皮尔逊相关系数为0.96。

. In addition to showing high reproducibility, this analysis establishes the error rate of even the most rigorous human annotation with implications for their use in trials and as comparators when validating automated monitors.

除了显示出高重现性外，该分析还确定了即使是最严格的人类注释的错误率，并对其在试验中的使用以及在验证自动监护仪时作为比较器的使用产生了影响。

Sensitivity is an essential metric of event-level performance, but detecting or failing to detect a single cough is not clinically meaningful. Hourly cough rates, on the other hand, are highly informative to patients, providers, and investigators, the Bland-Altman plots reported here show the robustness of hourly rates determined using CoughMonitor..

敏感性是事件水平表现的重要指标，但检测或未能检测到单一咳嗽在临床上没有意义。另一方面，每小时咳嗽率对患者，提供者和研究人员来说是非常有用的，这里报道的Bland-Altman图显示了使用CoughMonitor确定的每小时咳嗽率的稳健性。。

Finally, we illustrate the potential clinical value of unobtrusive, continuous cough monitoring for long periods of time spanning over six months in two different clinical cases.

最后，我们说明了在两个不同的临床病例中，对跨越六个月的长时间进行不引人注目的连续咳嗽监测的潜在临床价值。

There are several limitations to this study. First, the results were obtained with a single make and model of Android watch which could limit its generalizability to other devices. However, in the recent literature resulting from a sharp increase in the general interest in cough recognizing algorithms, comparable results have been obtained using a variety of devices that employ comparable microphones and chip sets.

这项研究有几个局限性。首先，结果是通过单一品牌和型号的Android手表获得的，这可能会限制其对其他设备的普遍性。然而，在最近的文献中，由于人们对咳嗽识别算法的普遍兴趣急剧增加，使用各种使用类似麦克风和芯片组的设备已经获得了类似的结果。

. Second, in this study we have made only limited efforts to distinguish between coughs from the wearer and others in close proximity. Thus, the device should not be used in environments with a high burden of non-user coughs until such an analysis has been conducted. Third, cough recognition will be influenced by the acoustic environment in which the CoughMonitor is worn, as a large volume of peaks may artificially inflate the detected cough second rate, even with a low false positivity rate.

其次，在这项研究中，我们只做了有限的努力来区分佩戴者和附近其他人的咳嗽。因此，在进行此类分析之前，不应在非使用者咳嗽负担很重的环境中使用该设备。第三，咳嗽识别将受到佩戴咳嗽监护仪的声学环境的影响，因为大量的峰值可能会人为地夸大检测到的咳嗽二次率，即使假阳性率很低。

Although subjects were instructed to avoid unusually loud environments, they were actually worn in a wide variety of settings such as while using public transport and listening to loud techno music. Nonetheless, users must be informed that results may be less accurate in settings with extremely loud background noise.

虽然受试者被要求避免异常嘈杂的环境，但实际上他们在各种各样的环境中穿着，例如在使用公共交通工具和听大声的科技音乐时。尽管如此，必须告知用户，在背景噪音非常大的环境中，结果可能不太准确。

Finally, an overall false positivity of one per hour means that the accuracy will be better among subjects with higher cough burden, this aligns well with clinical practice among subjects with chronic cough and for evaluating the efficacy of drugs for this indication but if there is intention to use the device among subjects with relatively low cough rates, appropriately powered studies in different acoustic environments should be conducted beforehand..

最后，每小时一次的总体假阳性意味着咳嗽负担较高的受试者的准确性会更好，这与慢性咳嗽受试者的临床实践以及评估药物对该适应症的疗效非常吻合，但如果打算在咳嗽率相对较低的受试者中使用该设备，则应事先在不同的声学环境中进行适当的动力研究。。

Conclusion

结论

The Hyfe CoughMonitor System has an overall sensitivity for detecting a cough event above 90% and a false positivity rate of about one per hour. These results were observed in men and women with a variety of diagnoses as individuals went about their usual activities of daily living. Given the system’s high accuracy, usability, and scalability, it has the potential to greatly improve clinical care, drug development and regulatory efforts, particularly among subjects with a high cough burden..

Hyfe CoughMonitor系统检测咳嗽事件的总体灵敏度高于90%，假阳性率约为每小时一次。这些结果在患有各种诊断的男性和女性中观察到，因为个体进行了日常生活活动。鉴于该系统的高准确性，可用性和可扩展性，它有可能大大改善临床护理，药物开发和监管工作，特别是在咳嗽负担高的受试者中。。

Data availability

数据可用性

Anonymized individual data and code relevant to this article will be publicly available in GitHub (https://github.com/hyfe-ai/validation/) as well as study protocol, immediately following publication, indefinitely. Further assistance with data access can be obtained from the corresponding author (CCh).

与本文相关的匿名个人数据和代码将在GitHub中公开提供(https://github.com/hyfe-ai/validation/)以及研究方案，在出版后立即无限期。可以从通讯作者（CCh）获得有关数据访问的进一步帮助。

CCh, ISO, MR and KW had full access to all of the data in this study and take complete responsibility for the integrity of the data and the accuracy of the data analysis..

CCh、ISO、MR和KW可以完全访问本研究中的所有数据，并对数据的完整性和数据分析的准确性承担全部责任。。

References

参考文献

Cornford, C. S. Why patients consult when they cough: A comparison of consulting and non-consulting patients.

康福德，C.S。为什么患者咳嗽时会咨询：咨询患者和非咨询患者的比较。

Br. J. Gen. Pract.

英国陆军上将Pract。

, 1751–1754 (1998).

CAS

中科院

PubMed

PubMed Central

公共医学中心

MATH

数学

Google Scholar

谷歌学者

Irwin, R. S. et al. Overview of the management of cough: CHEST Guideline and Expert Panel Report.

Irwin，R.S.等人，《咳嗽管理概述：胸部指南和专家小组报告》。

Chest

胸部

146

, 885–889.

https://doi.org/10.1378/chest.14-1485

(2014).

Article

文章

PubMed

PubMed Central

公共医学中心

MATH

数学

Google Scholar

谷歌学者

Altshuler, E. et al. Digital cough monitoring - A potential predictive acoustic biomarker of clinical outcomes in hospitalized COVID-19 patients.

Altshuler，E。等人。数字咳嗽监测-住院COVID-19患者临床结果的潜在预测声学生物标志物。

J. Biomed. Inf.

J.生物识别。信息。

138

, 104283.

https://doi.org/10.1016/j.jbi.2023.104283

(2023).

Article

文章

MATH

数学

Google Scholar

谷歌学者

Crooks, M. G. et al. Domiciliary cough monitoring for the prediction of COPD exacerbations.

Crooks，M.G.等人。用于预测COPD恶化的家庭咳嗽监测。

Lung

肺

199

, 131–137.

https://doi.org/10.1007/s00408-021-00435-9

(2021).

Article

文章

PubMed

PubMed Central

公共医学中心

MATH

数学

Google Scholar

谷歌学者

Gomes, B. & Ashley, E. A. Artificial intelligence in molecular medicine.

Gomes，B。＆Ashley，E.A。分子医学中的人工智能。

N. Engl. J. Med.

N.Engl。医学杂志。

388

, 2456–2465.

https://doi.org/10.1056/NEJMra2204787

(2023).

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Hall, J. I., Lozano, M., Estrada-Petrocelli, L., Birring, S. & Turner, R. The present and future of cough counting tools.

Hall，J.I.，Lozano，M.，Estrada Petrocelli，L.，Birring，S。＆Turner，R。咳嗽计数工具的现状和未来。

J. Thorac. Dis.

J、索拉克。说吧。

, 5207–5223.

https://doi.org/10.21037/jtd-2020-icc-003

(2020).

Article

文章

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Loudon, R. G. & Spohn, S. K. Cough frequency and infectivity in patients with pulmonary tuberculosis.

Loudon，R.G。＆Spohn，S.K。肺结核患者的咳嗽频率和感染性。

Am. Rev. Respir. Dis.

我有呼吸牧师。数字化信息系统。

, 109–111.

https://doi.org/10.1164/arrd.1969.99.1.109

(1969).

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Gabaldon-Figueira, J. C. et al. Acoustic surveillance of cough for detecting respiratory disease using artificial intelligence.

Gabaldon Figueira，J.C.等人。使用人工智能检测呼吸系统疾病的咳嗽声学监测。

ERJ Open Res.

ERJ公开决议。

https://doi.org/10.1183/23120541.00053-2022

(2022).

Gabaldon-Figueira, J. C. et al. Longitudinal passive cough monitoring and its implications for detecting changes in clinical status.

Gabaldon-Figueira，J.C.等人。纵向被动咳嗽监测及其对检测临床状态变化的意义。

ERJ Open Res.

ERJ公开决议。

https://doi.org/10.1183/23120541.00001-2022

(2022).

Huddart, S. et al. Continuous cough monitoring: A novel digital biomarker for TB diagnosis and treatment response monitoring.

Huddart，S.等人，《持续咳嗽监测：结核病诊断和治疗反应监测的新型数字生物标志物》。

Int. J. Tuberc. Lung Dis.

Int.J.Tuberc。隆迪。

, 221–222.

https://doi.org/10.5588/ijtld.22.0511

(2023).

Article

文章

CAS

中科院

PubMed

PubMed Central

公共医学中心

MATH

数学

Google Scholar

谷歌学者

Kang, Y. R. et al. Long-COVID severe refractory cough: Discussion of a case with 6-week longitudinal cough characterization.

Kang，Y.R.等人，《长型冠状病毒严重难治性咳嗽：讨论一例6周纵向咳嗽特征的病例》。

Asia Pac. Allergy

亚太地区。过敏

, e19.

，e19。

https://doi.org/10.5415/apallergy.2022.12.e19

(2022).

Article

文章

PubMed

PubMed Central

公共医学中心

MATH

数学

Google Scholar

谷歌学者

Zimmer, A. J. et al. Making cough count in tuberculosis care.

。

Commun. Med. (Lond)

公社。医学（伦敦）

, 83.

https://doi.org/10.1038/s43856-022-00149-w

(2022).

Article

文章

PubMed

MATH

数学

Google Scholar

谷歌学者

Morice, A. H. et al. ERS guidelines on the assessment of cough.

Morice，A.H.等人，《ERS咳嗽评估指南》。

Eur. Respir. J.

欧，深呼吸。J

, 1256–1276.

https://doi.org/10.1183/09031936.00101006

(2007).

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Kelsall, A. et al. How to quantify coughing: correlations with quality of life in chronic cough.

Kelsall，A。等人，《如何量化咳嗽：与慢性咳嗽生活质量的相关性》。

Eur. Respir. J.

欧，深呼吸。J

, 175–179.

https://doi.org/10.1183/09031936.00101307

(2008).

Article

文章

CAS

中科院

PubMed

MATH

数学

Google Scholar

谷歌学者

Sanchez-Olivieri, I. et al. Performance evaluation of human cough annotators: Optimal metrics and sex differences.

Sanchez-Olivieri，I。等人。人类咳嗽注释器的性能评估：最佳指标和性别差异。

BMJ Open Respir. Res.

BMJ开放式呼吸。我没有。

https://doi.org/10.1136/bmjresp-2023-001942

(2023).

Rudd, M., Song, W. J. & Small, P. M. The statistics of counting coughs: Easy as 1, 2, 3?

Rudd，M.，Song，W.J。和Small，P.M。计数咳嗽的统计数据：简单到1,2,3？

Lung

肺

200

, 531–537 (2022).

https://doi.org/10.1007/s00408-022-00559-6

Galvosas, M. G. F. et al. C. Performance evaluation of the smartphone-based AI cough monitoring app - Hyfe Cough Tracker against solicited respiratory sounds [version 1; peer review: 1 approved with reservations, 1 not approved].

Galvosas，M.G.F.等人C.基于智能手机的AI咳嗽监测应用程序的性能评估-Hyfe咳嗽跟踪器对抗请求的呼吸音[版本1；同行评审：1批准保留，1未批准]。

F1000Research

F1000研究

, 730 (2022).

https://doi.org/10.12688/f1000research.122597.1

McGarvey, L. P. et al. Efficacy and safety of gefapixant, a P2X(3) receptor antagonist, in refractory chronic cough and unexplained chronic cough (COUGH-1 and COUGH-2): results from two double-blind, randomised, parallel-group, placebo-controlled, phase 3 trials.

McGarvey，L.P.等人，《P2X（3）受体拮抗剂吉非哌新治疗难治性慢性咳嗽和不明原因慢性咳嗽（咳嗽-1和咳嗽-2）的疗效和安全性》：两项双盲、随机、平行组、安慰剂对照的3期临床试验结果。

Lancet

柳叶刀

399

, 909–923.

https://doi.org/10.1016/S0140-6736(21)02348-5

(2022).

Article

文章

PubMed

Google Scholar

谷歌学者

FDA. Pulmonary-Allergy Drugs Advisory Committee (PADAC) meeting on Gefapixant November 17. Full hearing.

FDA。肺过敏药物咨询委员会（PADAC）11月17日在Gefapixant举行会议。。

https://www.fda.gov/advisory-committees/advisory-committee-calendar/november-17-2023-meeting-pulmonary-allergy-drugs-advisory-committee-11172023

. Accessed Feb 2024 (2023).

2024年2月（2023年）访问。

Hegde, S. et al. Cough sounds in screening and diagnostics: A scoping review.

Hegde，S.等人，《筛查和诊断中的咳嗽音：范围界定评论》。

Laryngoscope

喉镜

https://doi.org/10.1002/lary.31042

(2023).

Article

文章

PubMed

MATH

数学

Google Scholar

谷歌学者

Serrurier, A., Neuschaefer-Rube, C. & Rohrig, R. Past and trends in cough sound acquisition, automatic detection and automatic classification: A comparative review.

Serrurier，A.，Neuschaefer-Rube，C。＆Rohrig，R。咳嗽声采集，自动检测和自动分类的过去和趋势：比较综述。

Sensor (Basel)

传感器（巴塞尔）

https://doi.org/10.3390/s22082896

(2022).

Download references

下载参考资料

Acknowledgements

致谢

The authors would like to thank the participants and study team for their efforts to make this research happen.

作者要感谢参与者和研究团队为完成这项研究所做的努力。

Funding

资金

This study was funded by Hyfe Inc. The sponsor participated in study design, collection and analysis; it had no role in the decision to submit this manuscript for publication. ISGlobal acknowledges support from the Spanish Ministry of Science and Innovation through the “Centro de Excelencia Severo Ochoa 2019–2023” Program (CEX2018-000806-S), and support from the Generalitat de Catalunya through the CERCA program..

这项研究由Hyfe Inc.资助。赞助商参与了研究设计，收集和分析；它在决定提交这份手稿出版方面没有任何作用。ISGlobal感谢西班牙科学与创新部通过“2019-2023年塞韦罗·奥乔亚卓越中心”计划（CEX2018-000806-S）提供的支持，以及加泰罗尼亚国家综合委员会通过CERCA计划提供的支持。。

Author information

作者信息

Authors and Affiliations

作者和隶属关系

ISGlobal, Barcelona Institute for Global Health, Barcelona, Spain

ISGlobal，巴塞罗那全球卫生研究所，西班牙巴塞罗那

Carlos Chaccour

卡洛斯·查库

Navarra University Clinic, Pamplona, Spain

西班牙潘普洛纳纳瓦拉大学诊所

Carlos Chaccour, Isabel Sánchez-Olivieri, Juan Berto Botella & Juan P. de-Torres

卡洛斯·查库尔、伊莎贝尔·桑切斯·奥利维里、胡安·贝尔托·博塔和胡安·P·德·托雷斯

CIBERINFEC, Madrid, Spain

西班牙马德里CIBERINFEC

Carlos Chaccour

卡洛斯·查库

Oregon Health Science University, Portland, OR, USA

俄勒冈州波特兰市俄勒冈州健康科学大学

Sarah Siegel, Gina Megson & Kevin L. Winthrop

莎拉·西格尔、吉娜·梅格森和凯文·温思罗普

IDisNa, Pamplona, Spain

IDisNa，潘普洛纳，西班牙

Juan P. de-Torres

胡安·P·德托雷斯

Hyfe, Wilmington, DE, USA

Hyfe，威尔明顿，DE，美国

Lola Jover, Joe Brew, George Kafentzis, Mindaugas Galvosas, Matthew Rudd & Peter Small

Lola Jover，Joe Brew，George Kafentzis，Mindaugas Galvosas，Matthew Rudd&Peter Small

Department of Computer Science, University of Crete, Heraklion, Greece

希腊赫拉克利翁克里特大学计算机科学系

George Kafentzis

乔治·卡芬奇斯

University of the South, Sewanee, TN, USA

美国田纳西州塞瓦尼南方大学

Matthew Rudd

马修·路德

University of Washington, Seattle, WA, USA

华盛顿大学，西雅图，华盛顿州，美国

Peter Small

Authors

作者

Carlos Chaccour

卡洛斯·查库

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Isabel Sánchez-Olivieri

伊莎贝尔·桑切斯-奥利维里

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Sarah Siegel

莎拉·西格尔

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Gina Megson

吉娜·梅格森

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Kevin L. Winthrop

凯文·L·温思罗普

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Juan Berto Botella

胡安·贝尔托·博塔

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Juan P. de-Torres

胡安·P·德托雷斯

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Lola Jover

洛拉·乔弗

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Joe Brew

Joe Brew公司

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

George Kafentzis

乔治·卡芬奇斯

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Mindaugas Galvosas

明道加斯Galvosas

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Matthew Rudd

马修·路德

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Peter Small

彼得·斯莫尔

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

Google Scholar

谷歌学者

Contributions

捐款

Conceptualisation: CCh, JB, PSData curation: CCh, ISO, SS, LJFormal analysis: MR, JBInvestigation: CCh, ISO, SS, GM, KW, JPDT, JB, GK, PS Methodology: CCh, MR, JB, PSSupervision: CCh, SS, PSWriting - original draft: CCh, PSWriting - review & editing: all authors contributed, reviewed and approved the last draft..

概念化：CCh，JB，PSData策展：CCh，ISO，SS，LJ形式分析：MR，JBInvestigation：CCh，ISO，SS，GM，KW，JPDT，JB，GK，PS方法论：CCh，MR，JB，PS监督：CCh，SS，PSWriting-原稿：CCh，PSWriting-审查和编辑：所有作者都参与，审查并批准了最后的草稿。。

Corresponding author

通讯作者

Correspondence to

通信对象

Carlos Chaccour

卡洛斯·查库

Ethics declarations

道德宣言

Competing interests

相互竞争的利益

MR, MG, LJ, MG, JB, GK and PS are employees of Hyfe, Inc and own equity in Hyfe Inc. CCh has received consultancy fees and owns equity in Hyfe Inc. All other authors declare no conflict of interest.

MR，MG，LJ，MG，JB，GK和PS是Hyfe，Inc的员工，拥有Hyfe Inc.的股权。CCh已收到咨询费，并拥有Hyfe Inc.的股权。所有其他作者声明没有利益冲突。

Additional information

其他信息

Publisher’s note

出版商注释

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Springer Nature在已发布的地图和机构隶属关系中的管辖权主张方面保持中立。

Electronic supplementary material

电子补充材料

Below is the link to the electronic supplementary material.

以下是电子补充材料的链接。

Supplementary Material 1

补充材料1

Rights and permissions

权限和权限

Open Access

开放存取

This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material.

本文是根据知识共享署名非商业性NoDerivatives 4.0国际许可证授权的，该许可证允许以任何媒介或格式进行任何非商业性使用，共享，分发和复制，只要您对原始作者和来源给予适当的信任，提供知识共享许可证的链接，并指出您是否修改了许可材料。

You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.

To view a copy of this licence, visit .

要查看此许可证的副本，请访问。

http://creativecommons.org/licenses/by-nc-nd/4.0/

Reprints and permissions

重印和许可

About this article

关于本文

Cite this article

引用本文

Chaccour, C., Sánchez-Olivieri, I., Siegel, S.

查库，C.，桑切斯·奥利维里，I.，西格尔，S。

et al.