34
Clinical
Performance
9
Clinical Performance
Clinical Performance Summary
A ninety-nine subject (99) clinical study was conducted in which polysomnographic (PSG) data
was acquired and analyzed by polysomnographic technologists, and simultaneously data from the
differential mastoids were acquired and processed by the Zmachine algorithm.
Each PSG record was scored independently by at least two (2) certified polysomnographic
technologists (3 records scored by 2 technologists, 16 records scored by 3 technologists, and 80
records scored by 4 technologists) using the Rechtschaffen & Kales (R&K, 1968) visual sleep
scoring rules. The performance of the Zmachine was evaluated by comparing the Zmachine
algorithm to the consensus of human scorers.
Analysis demonstrates substantial agreement between the Zmachine and the consensus of
polysomnographic technologists. The P1 (
the probability that Zmachine will correctly assign
an epoch when the PSG Consensus assigns the epoch to a particular stage
) and P2 (
when
the Zmachine assigns an epoch to a particular stage, the probability of such assignment is
correct
) values for the detection of Wake, Light Sleep, Deep Sleep and REM for the Zmachine
algorithm (staging version 2.1) are summarized in the following table:
Wake
Light Sleep
Deep Sleep
REM
P1
0.928
0.825
0.783
0.723
P2
0.838
0.855
0.743
0.749
The P1 and P2 values were also computed for each subject individually (99 P1 and 99 P2 values
were obtained for each stage). The mean and interval estimation (95% confidence level) is
summarized the following table:
Wake
Light Sleep
Deep Sleep
REM
P1
0.933
±
0.012
0.822
±
0.016
0.730
±
0.063
0.719
±
0.037
P2
0.814
±
0.027
0.854
±
0.017
0.722
±
0.072
0.751
±
0.033
The overall Kappa agreement between the Zmachine Algorithm and the PSG Consensus for
85,206 epochs of data is 0.722, which is well above the 0.6 threshold for good agreement. The
Kappa score was also computed for each subject individually (99 Kappa scores were obtained).
The interval estimation with 95% confidence level for the Kappa score is 0.708
±
0.020.
In conclusion, the Zmachine Algorithm shows good performance in sleep staging from a single-
channel of EEG data. A copy of the full Technical Report is available upon request.
