Turner Richard D, Bothamley Graham H
Respir Med. 2014 Dec;108(12):1808-15. doi: 10.1016/j.rmed.2014.10.003.
Cough causes morbidity and transmits disease yet has been under-researched. The best method for recognising and counting coughs remains unclear. We tested the accuracy of the human ear and measured the influence of visual data on cough counting. We also evaluated PulmoTrack®, a potentially fully-automated cough monitor.
Consistency amongst listeners and the effect of visual data: Three 14–22-min sequences containing 45–79 coughs were played to 15 respiratory physicians on at least two occasions. Only sound was played on the first occasions but on the final occasion a simultaneous display of audio activity was included. Counts of cough sounds across methods and listeners were compared. Evaluation of PulmoTrack®: 20-h recordings were made from 10 patients with cough. Automated counts were compared with assessment by one investigator.
Agreement among listeners was high. The intraclass correlation coefficient (ICC) for cough counts by ear alone was 0.89 (95% CI, 0.65–1.00). With a concurrent visual display of sound amplitude it was 0.94 (0.80–1.00). 4.8% (0.6–9.5) fewer coughs were counted using visual data than by listening alone (mean [SD] total coughs: 190.2 [3.4] vs 200.7 [14.6]; p = 0.04). Cough frequencies reported by PulmoTrack® and the researcher differed substantially (ICC 0.23, −0.51 to 0.34, p = 0.87); PulmoTrack® had a sensitivity of 26% for detecting coughs identified by ear.
Coughs are well recognised by different listeners. The method used to count coughs should be clearly described as visual information has a significant influence. Non-automated cough counting remains the gold standard method of quantifying cough.
咳嗽会引发疾病并传播病菌，但对此研究不足。目前尚不清楚识别和计数咳嗽的最佳方法。我们测试了人耳识别咳嗽的准确性，并测量了视觉数据对咳嗽计数的影响。我们还评估了PulmoTrack®，这是一种可能实现完全自动化的咳嗽监测仪。
听众之间的一致性及视觉数据的影响：向15位呼吸内科医生播放了三个时长为14 - 22分钟、包含45 - 79次咳嗽的音频序列，播放次数至少为两次。第一次只播放声音，但最后一次播放时同时显示音频活动。比较不同方法和听众对咳嗽声音的计数。PulmoTrack®的评估：对10名咳嗽患者进行了20小时的记录。将自动计数结果与一名研究人员的评估结果进行比较。
听众之间的一致性很高。仅通过耳朵计数咳嗽的组内相关系数（ICC）为0.89（95%可信区间，0.65 - 1.00）。同时显示声音振幅时，ICC为0.94（0.80 - 1.00）。使用视觉数据计数的咳嗽次数比仅靠听少4.8%（0.6 - 9.5）（平均[标准差]总咳嗽次数：190.2 [3.4] 对200.7 [14.6]；p = 0.04）。PulmoTrack®报告的咳嗽频率与研究人员的结果有很大差异（ICC 0.23， - 0.51至0.34，p = 0.87）；PulmoTrack®检测耳朵识别出的咳嗽的灵敏度为26%。
不同听众对咳嗽的识别效果良好。由于视觉信息有显著影响，应明确描述计数咳嗽所使用的方法。非自动化咳嗽计数仍是量化咳嗽的金标准方法。
