P146 Assessing the Performance of Automatic Analysis During Routine Polysomnography

INTRODUCTION: This study aims to retrospectively assess the performance of automatic analysis (AA) during routine type 1 & 2 PSG studies. A literature review identified a gap in the current research for assessing the use of automatic analysis in routine clinical practice. Routine scientific practice often involves multiple scorers analysing data sets from a variety of study types with varied signal derivations and signal quality. METHODS: Retrospective study (2018–2020). All PSG data were analysed by experienced scientists (ES). The automatic analysis system used was Somnolyzer 24x7. All relevant parameters were analysed using paired T-tests with significance level of α = 0.05. PSG data sets included: Type 1 studies (n = 1370) split into two groups: Group 1 (n= 1148) included studies with automatic analysis only and Group 2 (n = 222) which included studies with automatic analysis and scientist review (SR) overnight. Type 2 studies (n = 235) included studies only with automatic analysis. RESULTS: Significant differences were found in 13 parameters between AA and ES in type 1 PSG (including total RDI), but with only 4 parameters in type 2 PSG studies. CONCLUSIONS: There were statistically significant differences between automatic analysis and human scoring with routine scientific practice, however there were no differences when involving interscorer variances. These differences may have clinical significance, particularly with medical interpretation and practice.