Validation of automated sleep stage and apnoea analysis in suspected obstructive sleep apnoea

Full-night polysomnography is necessary for the diagnosis of obstructive sleep apnoea (OSA). However, analysis of the sleep stages and apnoeas is time-consuming. Computer systems for automated analysis have, thus, been developed to alleviate this task. We investigated 27 consecutive patients referred to our sleep laboratory with suspected OSA. The analysis of sleep stages and apnoeas was performed by visual scoring, according to Rechtschaffen and Kales, and by commercially available automated analysis device. The mean difference between visual scoring and automated analysis was -1, 111, -140, -3, 1 and 27 min, for sleep stages awake, I, II, III, IV and rapid eye movement (REM) respectively. For the apnoea index, the automated analysis rated a lower figure (mean difference 7.h-1, 95% confidence interval 2-12.h-1). The diagnosis of OSA was performed with a sensitivity of 85% and a specificity of 93% by automated analysis. Comparison of two independent handscores showed good agreement, with a mean difference of 6, 4, 3, -7, 1 and -1 min, for sleep stages awake, I, II, III, IV and REM, respectively. In conclusion, the automated analysis underestimates stage I sleep and the apnoea index. Visual scoring is advisable for control of the results. Automated analysis should only be used by those who are able to perform a visual analysis.