The ongoing work of the digital task force moves closer to providing sleep specialists with an updated version of the R&K Manual.
After several years of discussion, heated arguments, sweat, and intense work, the “ad hoc” committee for standardizing recording, scoring, and summarizing human sleep completed their task and published what is often called “The R&K Manual.” The “R” stands for Allan Rechtschaffen and the “K” stands for Anthony Kales who were the cochairmen of the committee. The manual provided rules for classifying each 10-, 20-, or, most commonly, 30-second epoch as wakefulness or stages 1, 2, 3, 4, or REM (rapid eye movement) sleep.
The year was 1968. There were soldiers in Viet Nam, hippies in San Francisco, popular music was purchased on vinyl recordings called LPs, manuscripts were prepared with typewriters, sleep studies were recorded on paper polygraph machines, and computers filled entire rooms and were attended to by the high priests of technology. Most computer users submitted their deck of EBCDIC punch cards at the altar of the batch input job window, returning a couple of hours later to pick up their deck and printout. More often than not, the “job” did not run to completion and the printout terminates with an infamous “ABEND” (abnormal ending) error because of something mis-specified in the JCL setup cards. The typical user has never even seen the computer mainframe.
It is now 37 years later, there are soldiers in Iraq, dot.com entrepreneurs in San Francisco, we buy music on CDs or directly over the Internet, no one would even consider writing a manuscript on a typewriter (even if they could find one), and sleep studies are recorded using computerized polysomnography systems. The computers can be carried around in your briefcase, and everyone, including 5-year-old children, have hands-on, direct contact with the computer keyboard, mouse, trackball, or game controller. No one uses punch cards anymore, and results (and error messages) are presented nearly instantaneously by the Windows® operating system on the computer’s video screen. Nonetheless, polysomnographic technologists are still classifying each recording epoch as wakefulness or stages 1, 2, 3, 4, or REM sleep based on the rules detailed in the R&K Manual.
Updating R&K and the Digital Task Force
The amazing thing is that the Manual of Standardized Terminology, Techniques, and Scoring System for Sleep Stages of Human Subjects still works remarkably well; however, an update and revision is long overdue. To this end, the American Academy of Sleep Medicine has appointed task forces to review current literature, consider issues, and provide updated sections for the manual. Task forces include groups considering recording and scoring visual EEG phenomena, transient central nervous system arousals, sleep-disordered breathing events, leg movement activity, and changes occurring as a function of age. Finally, there is a task force that is considering issues related to computerized polysomnography.
One way to consider digital polysomnography involves examining the constituent data process. The five basic processes are:
1. Data acquisition (recording)
2. Data display (viewing)
3. Data manipulation (scoring and editing)
4. Data reduction (parameterization for reporting)
5. Data filing (storage)
Let us consider the issues concerning the first of these five processes.
The accuracy of the data preserved is directly proportional to the sampling rate and the number of bits used to represent a sample. That is, EEG sampled 1,000 times per second using a 12-bit analog-to-digital converter will more accurately represent the signal than EEG sampled at 100 Hz or a sample represented as an 8-bit word; however, the cost of a higher sampling rate, greater amplitude resolution, or both is paid for in file size.
The Nyquist law stipulates that for analysis, the minimum sampling rate must be twice the frequency of the highest frequency waveform desired. The key word here is “analysis.” While a sampling rate of 32 Hz may satisfy the Nyquist law for analyzing sleep spindles, it is wholly inadequate for reconstructing a decent graphic representation of a spindle. For the purpose of data display, a minimally adequate, albeit “choppy” image requires a sampling rate exceeding the highest frequency by a factor of four or five. To provide a clear reproduction of the original signal, a sampling rate 10 to 20 times higher than the maximum frequency is required. Thus, the common sampling rate of 100 Hz for sleep EEG would allow minimally adequate visualization of waveforms up to 20-25 Hz. By contrast, 40 Hz EMG would undergo obvious degradation and while we may not be interested in the details of 40 Hz EMG, we certainly need to recognize it visually in our computerized polysomnogram.
A key feature of sampling intervals is that they act as digital filters by introducing significant amplitude attenuation of frequencies with approximate or equivalent cycle lengths. Thus, a recording sampled at 100 Hz may completely attenuate EEG spikes making it not possible to differentiate between parasomnias and nocturnal seizures. Thus, EEG should probably be sampled at 200 to 300 Hz, minimum. Table 1 shows our current working guidelines for sampling rates.
Table 1. Current working guidelines for sampling rates for bioactivities routinely recorded during polysomnography.
Amplitude resolution gets a tad more technical. Computer systems use analog-to-digital converters (ADCs) to sample electrical signals. The ADC represents the sample numerically as a computer “word” with a specified number of bits. If the computer word has 8 bits, the minimum value is 0 and the maximum is 256. If the first bit is usually used to represent positive (+) vs negative (-), thus the numeric range becomes -127 through +128. Furthermore, if the voltage ranges from +2.5 to -2.5 V, then the maximum resolution will be approximately 20 mV. If a 50-mV signal were calibrated to 1-V input, then maximum resolution would be approximately 1 mV (which is quite crude). A 12-bit ADC would improve this by allowing 4,096 bits to represent signal amplitude, thereby improving resolution 16-fold. Therefore, a 12-bit or greater ADC should be adequate for most signals recorded via polysomnography.
This is a topic on which Nic Butkov, RPSGT, of Rogue Valley, Ore, and I have commiserated time and time again. Amplifier quality is an underappreciated issue in digital polysomnography. Many veteran polysomnographers believe the Grass P511 was the high point in EEG amplifier technology. These amplifiers were virtually indestructible, had nearly ideal bandwidth and roll-off characteristics, and created even, smooth output signal that remains unmatched even today. It is now decades after most well-designed analog amplifiers have been replaced with digital counterparts within computerized systems. One thing is certain: digital amplifiers need further evolution. Quality seems to have taken a back seat to quantity in that these newer amplifier systems have 32, or even 64, amplifiers in an instrument that is literally “smaller than a bread box.” Unfortunately, miniaturization coupled with tremendous cost reduction has provided enough incentive for buyers to accept signal quality erosion. Now is the time to ramp up improved amplifier characteristics. Amplifiers with better signal-to-noise ratios, improved noise reduction, and better output linearity should be integrated into computerized polysomnographic systems.
Montage Selection and Re-referencing
Finally, montage selection and re-referencing capabilities round out polysomnography recording issues. Fixed montages, as used on many initial computerized polysomnographic system offerings, should no longer be considered adequate or reasonable. Curiously, the elimination of the selector panel, to some extent, created this issue. The invention of a digital selector panel is long overdue. Such a system would return the flexibility in making and troubleshooting recordings that was enjoyed for years with analog machines. By contrast, a wonderful feature directly resulting from and made possible by digital technology is re-referencing. That is, if one makes all monopolar recordings, referenced to a common lead, it is possible to create new channel combinations by displaying moment-to-moment differential voltages between any channel pair. Thus, if C3, C4, O3, O4 were recorded, C3-C4, C3-O3, C3-O4, C4-O3, C4-O4, and/or O3-O4 could be displayed if desired. Re-referencing is seldom needed in routine polysomnography; however, digital systems should provide this capability.
I hope that this brief description of the issues and ideas concerning digital data acquisition (recording) provides a flavor for the ongoing work of the digital task force. To be sure, work concerning data display (viewing), data manipulation (scoring and editing), data reduction (parameterization for reporting), and data filing (storage) is progressing. Ultimately, guidelines will be based on either an evidence-based medicine or a RAND method approach to assure that the best available information is considered. Many overlapping issues between task forces are being resolved, and as each step is taken, an updated version of the R&K Manual inches ever closer to reality.
Max Hirshkowitz, PhD, DABSM, is associate professor of the Department of Medicine and Department of Psychiatry at the Baylor College of Medicine, Houston.
Hirshkowitz M. Standing on the shoulders of giants: the Standardized Sleep Manual after 30 years. Sleep Medicine Reviews. 2000;4:169-179.
Iber C. Development of a new manual for characterizing sleep. Sleep. 2004;27:190-192.
Rechtschaffen A, Kales A. A Manual of Standardized Terminology, Techniques, and Scoring System for Sleep Stages of Human Subjects. Public Health Service, NIH Publication No. 204. Washington, DC: US Government Printing Office; 1968.