O035 Automated vs. expert manual analysis of the Multiple Sleep Latency Test

PURPOSE: To compare Compumedics Profusion PSG™ automated sleep analysis of Multiple Sleep Latency Tests (MSLTs) with expert consensus manual analysis. METHODS: Consecutive PSG with MSLTs were analysed using automated software (Compumedics Ltd (Abbottsford, Victoria, Australia) Profusion PSG™ V4.5 Build 531) (‘Auto’) and by two of nine experienced scientists. Discrepancies between scientists were discussed to establish expert consensus (‘Final’). RESULTS: Fifty consecutive patients referred for investigation of Narcolepsy were included. Two were excluded due to poor signal quality (1) and early test termination (1). The remaining 48 (37 M, 10 F, 1) had a median (range) age of 37 (17–63) years, BMI 28.0 (19.9–66.1) kg/m2, and mean sleep latency (MSL) 14.0 (1.5–20.0) minutes. Of five MSLTs with MSL <=8 min, Auto-MSL was also <=8 min. Of 43 MSLTs with MSL >=8 min, Auto-MSL was <=8 min in 12. MSL sensitivity was 100% and specificity 72%. For the one MSLT with >=2 SOREMs, Auto identified 1 SOREM. Nap-wise, Auto-SOREM sensitivity was 17% and specificity 98%; one of six REM-positive naps was detected by auto-analysis and there were seven false positive and five false negative SOREM results. CONCLUSIONS: (1) Automated analysis poorly detected short MSL and SOREM occurrence but was able to rule out all true-negative MSLT results, in this MSLT dataset. (2) This comparison methodology and dataset facilitates robust prospective testing of other current and future algorithms.