Measurement error in visually scored electrophysiological data: respiration during sleep.

Electrophysiological sleep studies employ the simultaneous recording of multiple physiological parameters. These data are often scored and analyzed visually, i.e. by trained human scorers. In this study, we investigated the measurement error inherent in visual evaluation of respiratory disturbances during human sleep. Three trained scorers independently evaluated twenty-eight all-night sleep recordings. Results indicated that the scorers achieved high agreement on only selected variables. Scorers' experience did not affect the reliability and independent judgements of recording quality did not account for the discrepancies. These findings, although most relevant for the specific parameters studied here, as well as for these scoring criteria, subjects, and techniques, document the unreliability of certain visually scored sleep data. The present paradigm may be useful in evaluating other types of measurement error.