Reliability of heart rate and respiration rate measurements with a wireless accelerometer in postbariatric recovery

Recognition of early signs of deterioration in postoperative course could be improved by continuous monitoring of vital parameters. Wearable sensors could enable this by wireless transmission of vital signs. A novel accelerometer-based device, called Healthdot, has been designed to be worn on the skin to measure the two key vital parameters respiration rate (RespR) and heart rate (HeartR). The goal of this study is to assess the reliability of heart rate and respiration rate measured by the Healthdot in comparison to the gold standard, the bedside patient monitor, during the postoperative period in bariatric patients. Data were collected in a consecutive group of 30 patients who agreed to wear the device after their primary bariatric procedure. Directly after surgery, a Healthdot was attached on the patients’ left lower rib. Vital signs measured by the accelerometer based Healthdot were compared to vital signs collected with the gold standard patient monitor for the period that the patient stayed at the post-anesthesia care unit. Over all patients, a total of 22 hours of vital signs obtained by the Healthdot were recorded simultaneously with the bedside patient monitor data. 87.5% of the data met the pre-defined bias of 5 beats per minute for HeartR and 92.3 % of the data met the pre-defined bias of 5 respirations per minute for RespR. The Healthdot can be used to accurately derive heart rate and respiration rate in postbariatric patients. Wireless continuous monitoring of key vital signs has the potential to contribute to earlier recognition of complications in postoperative patients. Future studies should focus on the ability to detect patient deterioration in low-care environments and at home after discharge from the hospital.

Methods
All data collected were analyzed retrospectively after patients completed the study. The American National Standards Institute standard for cardiac monitors, heart rate meters, and alarms defines accuracy as a “readout error of no greater than ±10% of the input rate or ±5 beats per minute (bpm), whichever is greater”. Therefore, in this study the acceptable error between the measurements was set at 5 bpm for HeartR and 5 respirations per minute (rpm) for RespR. Data management and analysis was performed using RStudio.   

The Healthdot starts logging directly after activation. Only the periods when there was logging of patient parameters were evaluated during this study.

For this comparative analysis, only the internally stored data of the Healthdot was evaluated, because it has a higher sampling frequency than the transmitted aggregated data. Because the sample frequencies of the HeartR and RespR generated by the Healthdot are different, the 8-sec HeartR data were resampled by linear interpolation between samples, obtaining a 1-sec interval for the HeartR data as well as the RespR data.

Extracted reference from the patient monitor and Healthdot measurements were represented on the same time frequency (1 value/second) and then time-synchronized. The synchronization procedure included as first step a fixed time shift of the Healthdot measurements by applying the time lag corresponding to the maximum of the cross-correlation function between reference and Healthdot measurements. The second step corresponded to a visual inspection of the offset-corrected Healthdot measurement and the reference to fine tune the selected offset in three different instances of the recording so to identify via these offsets eventual clock drifts. Clock drift was defined as any progressive increase or decrease in the offset over time, which was then corrected by linear interpolation of the time offset along the measurement samples. Only intervals with quality index > 0 (scale 0-100) were retained.

The vital signs of the Healthdot and the reference monitor were compared using the Bland-Altman method for repeated measurements. This method was used to account for within-subject variation by correcting for the variance of differences between the average differences across patients and the number of measurements per patient. The mean difference, or bias, between the wireless sensor and the reference monitor, and the 95% confidence interval (CI) (+/- 1.96 SD), or limits of agreement, were determined for both the HeartR and RespR data. Furthermore, the Pearson’s correlation coefficient was calculated to assess the strength of the association between the measurements of the Healthdot and the measurements of the reference patient monitor.

Because outliers were observed in the data, error bars of the mean differences between the Healthdot and patient monitor, including their confidence interval, were made for each patient for both HeartR and RespR. These error bars were created on the data with a 1-sec interval as well as on the data over a 5-min average. The latter analysis was performed because the Healthdot is currently designed to average data and send that data package to the cloud every 5 minutes, which represents the intended performance in clinical use.

通过对生命体征参数的持续监测，可提升术后病情恶化早期征兆的识别效率。可穿戴传感器可通过无线传输生命体征数据实现这一目标。一款基于加速度计的新型设备Healthdot，设计用于贴附于皮肤表面，以监测两项核心生命体征参数：呼吸频率（Respiration Rate, RespR）与心率（Heart Rate, HeartR）。本研究旨在针对肥胖症手术患者的术后阶段，评估Healthdot所测得的心率与呼吸频率相对于金标准——床边监护仪的可靠性。

研究纳入了30名在接受减重手术后同意佩戴该设备的连续入组患者，并收集相关数据。术后即刻，将Healthdot贴附于患者左侧下肋部。针对患者在麻醉恢复室停留的时段，将基于加速度计的Healthdot所测得的生命体征，与金标准床边监护仪采集的生命体征进行对比。所有患者中，Healthdot采集的生命体征数据共有22小时与床边监护仪的数据实现了同步记录。其中87.5%的数据符合心率预设的偏倚阈值（5次/分钟），92.3%的数据符合呼吸频率预设的偏倚阈值（5次/分钟）。

Healthdot可准确获取减重手术后患者的心率与呼吸频率数据。对核心生命体征进行无线持续监测，有望更早识别术后患者的并发症。未来的研究应聚焦于该设备在低护理环境以及患者出院后的家庭环境中，检测病情恶化的能力。

Methods

所有采集的数据均在患者完成研究后进行回顾性分析。美国国家标准学会（American National Standards Institute, ANSI）针对心脏监护仪、心率计及警报设备制定的标准，将精度定义为"读数误差不超过输入速率的±10%，或±5次/分钟（bpm），取两者中更严格的标准"。因此，本研究将测量可接受误差设定为：心率误差不超过5次/分钟（bpm），呼吸频率误差不超过5次/分钟（rpm）。数据管理与分析采用RStudio完成。

Healthdot在激活后即刻开始记录数据，本研究仅评估了有患者参数记录的时段。本次对比分析仅采用Healthdot内置存储的数据进行评估，因为其采样频率高于传输的聚合数据。由于Healthdot生成的心率与呼吸频率采样频率不同，研究人员通过样本间线性插值对8秒间隔的心率数据进行重采样，使心率数据与呼吸频率数据均统一为1秒间隔。

从床边监护仪与Healthdot提取的参考数据与测量数据均统一为相同的时间频率（1个数据点/秒），随后进行时间同步。时间同步流程分为两步：第一步，通过计算参考数据与Healthdot测量数据之间互相关函数的最大值，确定时滞量，对Healthdot的测量数据进行固定时移校正；第二步，对经过偏移校正后的Healthdot测量数据与参考数据进行目视检查，在三段不同的记录时段中微调所选偏移量，以通过这些偏移量识别潜在的时钟漂移。时钟漂移被定义为偏移量随时间出现的渐进性增加或减少，随后通过对测量样本的时间偏移量进行线性插值来完成校正。仅保留质量指数大于0（量表范围为0~100）的时段数据。

采用针对重复测量的Bland-Altman法，对Healthdot与参考监护仪的生命体征数据进行对比。该方法通过校正患者间平均差异的方差与每位患者的测量次数，以考量受试者内的变异情况。针对心率与呼吸频率数据，分别计算无线传感器与参考监护仪之间的平均差异（即偏倚），以及95%置信区间（CI，±1.96倍标准差），即一致性界限。此外，还计算了Pearson相关系数，以评估Healthdot测量值与参考监护仪测量值之间的关联强度。

由于数据中存在异常值，研究人员为每位患者的心率与呼吸频率数据分别绘制了Healthdot与监护仪平均差异的误差棒，其中包含其置信区间。这些误差棒分别基于1秒间隔的数据与5分钟平均数据绘制。之所以进行后者分析，是因为Healthdot目前的设计为对数据进行平均处理，并每5分钟向云端发送一次数据包，这正是其临床应用中的预期性能。