Discussion

The use of video recordings has already been assessed as a diagnosis tool for sleep conditions such as parasomnias [22]. Its application to pediatric OSDB has only been reported in two publications [18, 23], of which only one was available by the time we started this work [18]. Indeed, as early as 1996, Sivan et al had demonstrated the interest of a standardized method of sleep scoring based on home video analysis in children suspected of OSAHS. However, since then, the idea didn’t spread wider, probably limited by the availability of efficient home video registration equipment. Nowadays, any family with a smartphone possesses a simple technical solution to record a sleep video at home, putting in a new light the diagnostic value of this tool [23].
Our study was designed based on that of Sivan et al[18], with some modifications we consider as strengths:
-we used PSG videos rather than home recordings, to get rid of the parent-dependent video quality issue
-the videos were synchronized to PSG, which allowed to compare our video risk scores and the PSG results on the same night. It also allowed a complete sleep cycle video analysis based on EEG. Like Thomas et al [23] in their recent study, we chose to consider OAHI rather than AHI, which is more in line with current scoring guidelines.
-the video score obtained was brought back to two different time windows: 30 minutes as in Sivan’s study, and 10 minutes. Ten minutes seemed a more relevant video duration to use in potential future works as it is more compatible with an everyday practice. Indeed, the underlying purpose of this pilot study was to prepare a wider prospective work using smartphone home sleep video recordings. This explains our choice to slightly modify items 3 and 5 compared with Sivan’s original score. Of note, there is only one similar publication based on smartphone recordings in the literature [23], in which the authors used a different scoring system based on 1-minute videos.
We acknowledge some limitations in our study:
-we included fewer patients than Sivan et al, therefore there was only two patients with moderate OSAHS
-with only one investigator, we could not assess a potential interscorer variability. Of note, there was few disagreement between the 3 investigators in Sivan’s study [18].
Despite these limitations, our results are consistent with those of Sivan et al. The threshold value retained in their study was a 30-minute score greater than or equal to 8, which was predictive of an AHI ≥ 1 per hour with a sensitivity of 94% and a specificity of 68% [18]. In our cohort, we were unable to identify a reliable threshold for the diagnosis of mild OSAHS. Indeed, despite a perfect specificity (100%), sensitivity remained low for both RS30 (60% for a threshold of 6,09) and RS 10 (50% for a threshold of 6,40) when considering an OAHI≥ 1.5. Our results were more accurate when considering an OAHI≥ 5. Indeed, a RS30 threshold of 6.09 was predictive of a moderate to severe OSAHS with a sensitivity of 83% and a specificity of 90%. Similarly, a RS10 threshold of 6.50 was predictive of a moderate to severe OSAHS with a lower sensitivity (67%), but a higher specificity (100%).
In comparison with the SHS questionnaire, the RS30 and the RS10 had a similar sensitivity for the diagnosis of moderate to severe OSAHS, but a higher specificity. Indeed, a threshold of SHS ≥ 2.75 had a sensitivity of 88% and a specificity of 63% to predict an AHI ≥ 5 per hour in Spruyt et al study [7]. Of note, we found a lower correlation coefficient (r = 0.43) between the OAHI and the SHS, compared to RS30 (r = 0.71) and RS10 (r = 0.59). Just like in our study, Spruyt et al had difficulty in establishing a predictive threshold for mild OSAHS.
It is important to emphasize that comparison of our results with both those of Sivan et al [18] and Spruyt et al [7] must be taken with caution, as we didn’t follow the same rules for scoring respiratory events. Indeed, their studies were carried out before the edition of the 2012 and 2015 rules for scoring of the AASM [12, 13].
Both SHS and our risk scores focus on the noise (snoring) and movements components of OSAHS. An imprecise observer-dependent evaluation of these two parameters may contribute to lower the accuracy of these scores, raising the question of the interest of an automated analysis. Two recent studies in adults using an automatic analysis system of the patients’ thoracic movements to determine the severity of OSAHS showed interesting results [24, 25].
We chose to study the second sleep cycle of children for several reasons. We avoided the first sleep cycle as it may include less respiratory events due to predominance of slow wave sleep, thus distorting analysis. Besides, there is an increase in the proportion of time spent in REM sleep with each successive sleep cycle [26]. As the number of central and obstructive apneas is higher during REM sleep, the interpretation of the last sleep cycles could lead to an overestimation of the number of respiratory events. Finally, as this pilot study was a prelude to a wider prospective work using smartphone home sleep video recordings, it seemed relevant to analyze a rather early in the nighttime window, when children are asleep but parents still available to capture the video.