12(2): 133--139, Raven Press, Ltd., New York. © 1989 Association ... The Oxford-Medilog 9000 (Oxford Medical, Inc., Clearwater, FL) is an ambulatory cassette ...
Sleep 12(2): 133--139, Raven Press, Ltd., New York © 1989 Association of Professional Sleep Societies
Two Methods of Scoring Sleep with the Oxford Medilog 9000: Comparison to Conventional Paper Scoring Timothy J. Hoelscher, W. Vaughn McCall, Judith Powell, Gail R. Marsh, and C. William Erwin Sleep Disorders Center, Duke University Medical Center, Durham, North Carolina, U.S.A.
Summary: This study evaluated two methods of scoring taped polysomnographic data directly on the Medilog 9000 scanner: (a) screen-by-screen scoring, and (b) rapid screen scoring. Sixteen overnight polysomnograms recorded on Medilog 9000 recorders were scored using the above two methods and were also printed on paper for conventional paper scoring. Interscorer agreement was 87.8% for paper scoring, 85.5% for screen-by-screen scoring, and 84.2% for rapid screen scoring. Comparison of screen-by-screen scoring with paper scoring revealed small absolute deviations and correlations of r > 0.90 for all sleep parameters, with the exception of brief «2 min) awakenings (r = 0.69). Rapid screen scoring resulted in slightly lower correlations and greater deviations from paper scoring on several sleep parameters, but appeared acceptable for most clinical purposes and greatly reduced the required scoring time. Although some statistically significant differences between scoring methods were observed, the size of effect was small and of doubtful clinical importance. These findings suggest that polysomnographic data recorded on Medilog 9000 recorders can be reliably and accurately scored on the Medilog scanner, obviating the laborious task of printing the taped data on paper. Key Words: Ambulatory sleep monitoring-Sleep scoring-Medilog 9000 scanner.
The Oxford-Medilog 9000 (Oxford Medical, Inc., Clearwater, FL) is an ambulatory cassette recording system with the capability of monitoring eight channels of electrophysiologic data. Originally developed for ambulatory electroencephalogram (EEG) monitoring 0), the Medilog 9000 has more recently been used for monitoring sleep patterns outside of the sleep laboratory. A comparison of Medilog 9000 recordings of 24 normal sleepers with conventional in-laboratory polysomnography revealed that the Medilog 9000 reliably reflected standard sleep parameters (2,3). In a previous report from our laboratory on 339 ambulatory sleep studies, the Medilog 9000 resulted in technically acceptable recordings in 9~97% of all studies and was well accepted by Accepted for publication August 1988. Address correspondence and reprint requests to Timothy J. Hoelscher, Ph.D., at Sleep Disorders Center, Eastern Virginia Medical School, 600 Gresham Dr., Norfolk, V A 23507, U.S.A.
133
134
T. 1. HOELSCHER ET AL.
most patients (4). Although the Medilog 9000 is at present inadequate for sleep apnea diagnostic studies because it lacks a measure of oxygen de saturation, it appears to have considerable promise for evaluating a variety of sleep/wake disorders in the home environment. A potential shortcoming in using the Medilog 9000 for sleep monitoring concerns problems associated with scoring the polysomnographic data. Although the taped data can be printed on paper for conventional scoring purposes, this process is laborious and inefficient because the printing is done at real time. * To date, no method has been validated for scoring the sleep data directly on the Medilog 9000 scanner. This article describes two methods of scoring polysomnograms on the Medilog scanner, and presents data comparing the two scanner scoring methods with conventional paper scoring. METHODS Subjects Sixteen overnight polysomnograms recorded on Medilog 9000 ambulatory cassette tapes were selected from our files. The tapes were randomly chosen within the restriction that the sample had to include 8 insomniacs and 8 normal sleepers. The sample consisted of 9 females and 7 males aged 14-67 years (mean = 36.1, SD = 18.1). The 8 insomniacs were three patients with a primary diagnosis of periodic leg movements, 3 psychophysiologic insomniacs, and 2 patients with affective disorder. Apparatus The Oxford Medilog 9000 is an eight-channel ambulatory cassette recording system. Monitoring capabilities include EEG, electro-oculogram (EOG), electromyogram (EMG), electrocardiogram (ECG), chest impedance, and nasal-oral thermistor. Additional features include digital time recorded at I-s intervals and an event marker. The battery-powered recorder measures 15.4 x 11.2 x 4.0 cm, weighs 0.9 kg, and can be worn on a belt or shoulder strap. A standard C-120 cassette can record up to 24 h of data. The Medilog 9000 scanner simultaneously displays in a paginated format all eight channels of physiologic data along with a digital time display. The data can be displayed on screen using 16- or 8-s epochs which correspond to paper speeds of 15 and 30 mm/s, respectively. The tape can be advanced by (a) paging forward at 8- or 16-s intervals (manual pagination), (b) slowly scanning forward or backward within the confines of a 64-s memory, or (c) scanning at 20, 40, or 60 times the recorded speed (automated pagination). Auditory EEG analysis can be used to warn the scorer of an upcoming arousal or change in sleep stage as reflected by a change in the EEG frequency pattern. Procedure The sleep montage varied across the 16 subjects arid was determined by the referral question. However, each recording included a minimum of two EEG channels (CrM z, Oz-Cz), bilateral EOG (left eye, M z; right eye, M t ), and one channel of submental chin
* High-speed printing of a Medilog 9000 tape is possible; however, this process requires purchase of an Oxford sleep stager in addition to a polygraph with high-speed writing capabilities (e.g., Siemens Mingograt).
Sleep, Vol. 12, No.2, 1989
MEDILOG 9000 SCORING
135
EMG. Twelve ofthe 16 subjects were studied in their homes and 4 were studied in their hospital rooms. Scoring Paper records were produced by printing the 16 taped polysomnograms on paper by a Grass polygraph at a paper speed of 15 mm/s. Each of the 16 records was scored using each of the following scoring systems: (a) conventional paper scoring, (b) screenby-screen scoring, and (c) rapid screen scoring. The order of scoring of the sleep records was left unspecified. However, to guard against practice and order effects, the scorers were required to alternate between subjects and scoring method. For example, after scoring the record of subject 1 with one scoring method, the scorer then had to score at least one other subject's record before again scoring subject 1 using a different scoring method. Although scorers were not kept blind to the identity of the sleep record, all scoring was done independently in that the scorers were not allowed to compare the results of different scoring methods for a given subject. All scoring was done by the first three authors. T.J.H. is an accredited clinical polysomnographer (ACP), W.V.M. is also an ACP, although he was in the process of taking the ACP examination during the time the sleep records were scored, and J.P. is a clinical psychology intern who received extensive training in the scoring of sleep stages. TJ.H. scored 44% of the records in this study, W.V.M. scored 31%, and J.P. scored 25%. Paper scoring. Conventional paper scoring (20-s epochs) followed the standardized scoring system of Rechtschaffen and Kales (5). The Cr M2 channel was used for ruler measurements of EEG frequency and amplitude for scoring slow-wave sleep. Paper scoring required -3 hipolysomnogram, which averaged 8.4 h in length. Screen-by-screen scoring. This scoring method closely paralleled procedures for conventional paper scoring and involved scoring each individual 16-s epoch displayed by the scanner. A sleep stage was assigned to each specific 16-s epoch. The scorer advanced the tape by paging forward (manual pagination), which advances the tape by 16-s epochs. However, during periods in which the subject stayed in the same sleep stage for an extended period (e.g., during an extended awakening), the scorer reviewed the tape at 20 times the recorded speed (automated pagination) and stopped the tape when the sleep stage changed. The EEG signal amplitude used on the scanner screen was 5 /J-V/mm. Thus, in scoring slow-wave sleep a deflection ~15 mm was required for scoring a delta wave (75 /J-V). This scoring method required -3-4 hipolysomnogram. Rapid screen scoring. The scorer played the tape at 20 times the recorded speed (automated pagination) and noted the time of occurrence of sleep stage changes. Changes in sleep stage < 1 min were ignored. The scorer was free to stop the tape and further review and/or replay any portion of the tape that proved difficult to score. Sleep parameters were calculated by hand based on the recorded time of sleep stage changes. Wake time after sleep onset (W ASO) 0.80 (p < 0.01) across all sleep parameters, with the exception of stage 1 percentage (r = 0.73, p < 0.01) and WASO < 2 min (r = 0.66, p < 0.01). Analyses of variance revealed significant method x group interactions for sleep efficiency percentage (F[I, 14] = 6.38, p < 0.05) and stage 2 sleep latency (F[I, 14] = 4.93, p < 0.05). Further examination indicated that rapid screen scoring resulted in lower sleep efficiency percentage (means = 88.1% and 90.8%) and longer stage 2 sleep latencies (means = 28.0 and 26.3 min) than paper scoring for normal subjects but not for insomniacs (sleep efficiency percentage means = 77.6 and 78.0; stage 2 sleep means = 24.1 and 25.1). Significant main effects for scoring method were observed for stage 1 percentage (F[l, 14] = 6.00, p < 0.05) and REM latency (F[I, 14] = 12.5, p < 0.01), with rapid screen scoring resulting in less stage 1 percentage (means = 7.3% and 9.3%) and shorter REM latencies (means = 74.9 and 78.2 min) than paper scoring. DISCUSSION Acceptable reliability levels were obtained in scoring polysomnographic data directly on the Medilog scanner. Interscorer agreement averaged 85.0% for screen-by-screen scoring and 84.2% for rapid screen scoring, compared to 87.8% for conventional paper TABLE 2. Conventional paper scoring (20-s epochs) versus rapid screen scoring (I-min epochs) on the Medilog 9000 scanner Paper
Absolute deviation
Scanner"
Correlation
Parameter
Mean
SD
Mean
SD
Mean
SD
I'
Total sleep time (min) Sleep efficiency % Stage 1 latency (min) Stage 2 latency (min) REM latency (min) Total WASO c (min) WASO "" 2 min WASO < 2 minc Stage 1 % Stage 2 % Slow-wave % REM %
402.1 84.4 22.6 25.7 78.2 59.7 42.0 17.7 9.3 49.4 21.1 20.2
90.9 12.4 19.2 18.9 30.0 50.1 49.0 7.3 4.9 10.3 11.6 6.8
400.9 82.8 22.6 26.1 74.9 60.4 44.2 16.2 7.3 50.3 21.4 21.0
89.3 11.6 19.0 19.5 30.2 48.5 48.7 6.0 3.9 10.6 10.7 7.2
6.0 2.1 1.4 1.8 4.4 5.8 6.3 4.1 2.9 4.2 4.8 1.9
7.0 1.4 1.0 1.9 2.8 6.0 7.1 3.0 2.4 2.6 4.3 1.5
0.995 0.988 0.996 0.991 0.991 0.995 0.982 0.662 0.753 0.889 0.831 0.947
WASO, wake time after sleep onset. Rapid screen scoring. b All r values significant at p < 0.01. C Includes movement time. a
Sleep, Vol. 12, No.2, 1989
138
T. J. HOELSCHER ET AL.
scoring. These results are consistent with findings from a previous study which reported 82.9% agreement in scoring I-min epochs directly on the Medilog scanner (6). Screen-by-screen scoring was highly concordant with conventional paper scoring as evidenced by correlations of r > 0.90 across all sleep parameters, with the exception of brief «2 min) awakenings (r = 0.69). Although significant differences between screenby-screen scoring and paper scoring were observed for sleep efficiency percentage, WASO < 2 min, stage 2 latency, and REM latency, the size of these differences was small. For example, screen-by-screen scoring resulted in significantly lower sleep efficiencies than paper scoring even though the difference in means was only 0.7% (83.7% vs. 84.4%). Owing to very low variances, however, the analyses revealed statistically significant differences between these scoring methods for sleep efficiency as well as for the other variables listed above. It is unlikely that the observed size of effect for scoring method is large enough for any of these variables to have an appreciable impact on clinical interpretations of sleep data. We do suggest, however, that researchers not combine data derived from different scoring methods since such a practice may result in spurious results for the above variables. Although screen-by-screen scoring obviates the laborious task of printing the taped data on paper at real time, careful scoring requires -3-4 hipolysomnogram and represents no appreciable time savings from scoring a conventional paper record. By contrast, rapid screen scoring is very time-efficient and requires an average of only I hipolysomnogram. Rapid screen scoring resulted in somewhat lower correlations and slightly larger deviations from paper scoring than screen-by-screen scoring but appears acceptable for most clinical purposes. Although rapid screen scoring appears to be quite sensitive to major changes in sleep stage (macro events), it is less sensitive to brief arousals and rapid alternation between sleep stages (micro events; e.g., records in which brief periods of stage 2 alternate with brief periods of slow-wave sleep). Two methodologic limitations of this study merit comment. First, although the scorers were not allowed to compare the results of the different scoring methods for a given subject, they were not kept blind to the identity of the sleep records. The scorers may have retained a general impression of a record, which may have influenced the later scoring of the same record with a different scoring method. Second, this study did not include epoch-by-epoch comparisons ofthe three scoring methods owing, in part, to the variable epoch lengths of the different scoring methods. Furthermore, epoch-by-epoch analysis is difficult with the Medilog 9000 because of problems associated with the synchronization of the scanner and paper formats (7). When a taped record is printed from the Medilog, synchronization of exact time between the scanner and paper is distorted by slight timing errors in the speed of the polygraph paper transport and the tape transport of the playback unit. t Practically, however, direct epoch-by-epoch comparisons are far less important than comparisons of standard sleep parameters derived from the different scoring methods. Clinical and research uses of polysomnographic data are ultimately based on summary statistics (i.e., sleep parameters) generated from the individually scored epochs and not on the individual epochs per se. The observed deviations between scanner and paper scoring not only reflect differt Epoch dislocation reSUlting from timing errors of the playback unit can reportedly be avoided by using an Oxford sleep stager (SS90MKIII) which controls the scanner's capstan speed. However, this procedure requires the additional purchase of a polygraph with high-speed writing capabilities (e.g., Siemens Mingograt) because the only playback speed allowed by the sleep stager is 20 times the recorded speed, and specialized software and hardware to place time codes on paper.
Sleep, Vol. 12, No.2, 1989
MEDILOG 9000 SCORING
139
ences in method, but also reflect discrepancies caused by varying epoch lengths as well as differences between scorers. Despite these multiple sources of error, sleep parameters calculated from the two scanning methods were highly concordant with those derived from conventional paper scoring. These results suggest that polysomnographic data recorded on Medilog 9000 recorders can be reliably and accurately scored directly on the Medilog scanner, providing further support for use of the Medilog in monitoring sleep patterns outside of the sleep laboratory. REFERENCES 1. Ebersole JS, Bridgers SL. Direct comparison of 3- and 8-channel ambulatory cassette EEG with intensive inpatient monitoring. Neurology 1985;35:846-54. 2. Sewitch DE, Kupfer DJ. Polysomnographic telemetry using Telediagnostic and Oxford Medilog 9000 systems. Sleep 1985;8:288-93. 3. Sewitch DE, Kupfer DJ. A comparison of the Telediagnostic and Medilog systems for recording normal sleep in the home environment. Psychophysiology 1985;22:718-26. 4. Hoelscher TJ, Erwin CW, Marsh GR, Webb MD, Radtke RA, Lininger A. Ambulatory sleep monitoring with the Oxford Medilog 9000: Technical acceptability, patient acceptance, and clinical indications. Sleep 1987;10:606-7. 5. Rechtschaffen A, Kales A, eds. A manual of standardized terminology. techniques and scoring system of sleep stages of human subjects. Los Angeles: UCLA Brain Information Service/Brain Research Institute, 1968. 6. Patterson N, Ball S, Cohen SA, Seidel WF, Yost D, Dement WC. Medilog 9000: Recording quality and visual scoring reliability. Sleep Res 1986;15:252. 7. Erwin CW, Hartwell JW. Sleep staging of ambulatory tape-recorded polysomnographic data: What a difference an epoch makes. J Clin Neurophysiol 1987;4:215.
Sleep. Vol. 12, No.2, 1989
