Dataset related to article: "Quantitative assessment of the quality of home sleep studies: A computer‐assisted approach"

Abstract

Home monitoring is the most practical means of collecting sleep data in large-scale research investigations. Because the portion of recording time with poor-quality data is higher than in attended polysomnography, a quantitative assessment of the quality of each signal should be recommended. Currently, only qualitative or semi-quantitative assessments are carried out, likely because of the lack of computer-based applications to carry out this task efficiently. This paper presents an innovative computer-assisted procedure designed to perform a quantitative quality assessment of standard respiratory signals recorded by Type 2 and Type 3 portable sleep monitors. The proposed system allows to assess the quality (good versus bad) of consecutive 1-min segments of thoraco-abdominal movements, oronasal, nasal airflow and oxygen saturation through an automatic classifier. The performance of the classifier was evaluated in a sample of 30 unattended polysomnography recordings, comparing the computer output with the consensus of two expert scorers. The difference (computer versus scorers) in the percentage of good-quality segments was on average very small, ranging from -3.1% (abdominal movements) to 0.8% (nasal flow), with an average total classification accuracy from 90.2 (oronasal flow) to 94.9 (nasal flow), a Sensitivity from 0.93 (oronasal flow) to 0.98 (nasal flow), and a Specificity from 0.74 (nasal flow) to 0.86 (abdominal movements). In practical applications, the scorer can run a check-and-edit procedure, further improving the classification accuracy. Considering a sample of 270 unattended polysomnography recordings (recording time: 545 ± 44 min), the average time taken for the check-and-edit procedure of each recording was 6.9 ± 2.1 min for all respiratory signals.