0349 HOW ACCURATELY CAN THE COMBINATION OF A LARGE NUMBER OF SLEEP PARAMETERS MEASURED BY POLYSOMNOGRAPHY PREDICT SUBJECTIVE SLEEP QUALITY?

Abstract

Introduction
Different from our previously reported studies examining the bivariate correlation between each of the sleep parameters assessed objectively by polysomnography (PSG) and patient-reported sleep quality (SQ), we investigated how accurately a large number of PSG parameters can jointly predict (explain) SQ.

Methods
PSG recordings from two clinical trials involving 1518 insomnia patients treated with suvorexant or placebo were used post-hoc to build regression/classification models associating SQ and 98 PSG parameters and evaluate the accuracy of SQ prediction as a function of these parameters. PSG recordings from 882 primary insomnia patients from two clinical trials in which patients were treated with gaboxadol or placebo, were used to confirm the findings.

Results
Overall accuracy of the SQ prediction given a large number of PSG parameters is moderate (area under the ROC curve (AROC) =~70%). In contrast, subjective total sleep time and subjective number of awakenings explain SQ with much higher accuracy (AROC > 80%). Ranking of PSG parameters by their contribution to SQ revealed several clusters of correlated parameters contributing equally.

Conclusion
Accuracy of SQ prediction using a large number of PSG parameters was quantified. Obtained results may serve as the baseline for novel PSG parameters developed to improve SQ prediction accuracy.

Support (If Any)
Funding: Merck & Co., Inc.