«PEKKA TIIHONEN Novel Portable Devices for Recording Sleep Apnea and Evaluating Altered Consciousness Doctoral dissertation To be presented by ...»
The audio stimulator block consists of a commercial DSP demo board (EZ-KIT Lite, Analog Devices Inc., Norwood, MA, USA) and several digital logic, memory and analogue circuit chips. This block produces the stimulation sequences constructed of two different pre-programmed 84 ms sine-wave tones (800 or 560 Hz) (Table 1 and Figure 4 in Study IV). Front panel thumb wheels can be used to select the desired stimulation program and stimulation intensity.
Both blocks have their own internal (embedded) programs to manage all the functions of the data logger (Figure 3 in Study IV), and to produce the desired audio tones. The logger program reads the setup parameters from a PC Card hard disk (or from a compact flash card) at the equipment start-up. For checking the data and signal Kuopio University Publications C. Natural and Environmental Sciences 261: 1 - 79 (2009)
- 52 Methodology quality, a special reviewing program was developed with C++ (Borland International, Inc.).
Prior to clinical evaluation, the device was thoroughly technically validated in our own laboratory. Furthermore, the radio frequency emission levels were inspected in an electro magnetic compatibility (EMC) laboratory (Pohjois-Savo Polytechnic, Kuopio, Finland).
5.3 Data analysis
Commercial PSG software (Somnologica 3.2) was used for the somnographic analysis.
All apnea, hypopnea, oxygen desaturation and snoring events were marked manually after automatic pre-analysis according to the rules described in Table 2.9. In order to detect a hypopnea event, the first rule (airflow amplitude drop t 30% and SpO2 drop t 4%) in Table 2.9 was used. The time markers for falling asleep and waking up were placed based on the patient’s sleep log and on visual inspection of the raw signals. The default parameters for automatic detection of apneas, hypopneas and oxygen desaturation events are listed in Tables 5.3 and 5.4. The default values in both tables for automatic detection of hypopneas indicate that the first rule for detection of hypopnea in (Table 2.9) was applied in Somnologica. Events during a movement period, 20 s after a movement period or in upright position or while awake were not accepted as apneahypopnea or oxygen desaturation events.
Table 5.3: Limits for accepting an event as apnea or a hypopnea event in the automated analysis in Somnologica 3.
Statistical analyses were conducted with the SPSS software (version 14.0, SPSS Inc., Chicago, IL, USA). The significance of differences between polysomnographic parameters recorded simultaneously in the sleep laboratory (an ambulatory device versus the reference instrument) in Studies I and II was investigated with the Wilcoxon signed rank-test (Field 2009). The significances of differences between the polysomnographic parameters recorded with the ambulatory devices (Venla versus Embletta and APV2 versus Embletta) in Studies I and II were investigated with the Mann-Whitney U test. In Study III, the significance of differences between the automatically and manually determined polysomnographic parameters was assessed with the Wilcoxon signed rank-test. The correlation analyses were conducted by means of the Pearson correlation analysis. Data analysis (filtering, sorting, and averaging) of EEG and ERP signals in Study IV was done using Scan 4.3 (NeuroScan Inc., Sterling, VA, USA) software package. Origin 6.0 and 7.5 program packages (Microcal Software Inc., Northampton, MA, USA) were used to plot the signals and other curves, to calculate noise statistics and to conduct correlation and Bland-Altman analyses (Bland
1986) Corel Draw versions 7.0 and 9.0 (Corel Corp., Ottawa, Canada) were used for the creation of the graphical presentations.
6.1 Evaluation of ambulatory devices for screening of sleep apnea Prior to the clinical applications, the ambulatory device designed and constructed in Study I (Venla) was evaluated with extensive technical tests. The radio-frequency emission generated by Venla was found to be below the limits set by the International Special Committee for Radio Interference. The measured power consumption of Venla was 0.33 W, which allows an operation time of 17.2 h with two AA batteries.
The clinical evaluation of the ambulatory devices began with simultaneous recordings with a reference sleep laboratory instrument. Although the recorded signals were qualitatively highly similar (Figure 6.1), more precise inspection revealed that there were certain differences between the signals. For example, the nasal flow signal recorded with Venla or APV2 has more noise than the signal recorded with the reference instrument. Figure 6.2 demonstrates an airflow recording with a nasal pressure cannula and an oronasal thermistor when signs of airflow through the mouth are seen.
The greatest differences between Venla and the other devices are seen in snore sound signals. Importantly, Venla detected more snoring than APV2 or the reference instrument. This is attributable to the technical differences in the recording of snoring and indicates that the low sampling frequency (8 Hz) used in Venla is sufficient.
Furthermore, due to the lower sampling frequency (1 Hz versus 3 Hz) used in Venla and APV2 for HR and SpO2 signals, these curves look coarser than those recorded with the reference instrument. However, despite these differences, the apnea events could be detected similarly with all three devices.
Figure 6.1: Examples of polysomnographic signals recorded from a patient suffering from severe sleep apnea with Embla (a), with Venla (b) and with APV2 (c).
The recording was done in a sleep laboratory simultaneously with all three devices. Despite some differences in signal quality, the apnea events could be detected similarly with the devices. The second signal (calculated nasal flow) in all panels is a square root of the first signal (pressure recorded with nasal cannula). Heart rate (HR) is measured with a pulse oxymeter and represented in beats per minute (BPM).
Kuopio University Publications C. Natural and Environmental Sciences 261: 1 - 79 (2009)
- 56 Results Figure 6.2: Example of airflow recording with a nasal pressure cannula and an oronasal thermistor with the Venla device. In the pressure channel there are four events (A – D) which appear as apneas. However, in the thermistor channel, there are signal variations (marked with up arrows) within the events A and B, suggesting presence of airflow. There is additional evidence for this in the snore channel showing loud noise peaks (down arrows) simultaneously with the signal variations seen in the thermistor channel. Oxygen desaturation signal shows that all respiratory events (A – D) cause desaturation.
No statistically significant differences (Wilcoxon signed-rank test) were found in the AHI and ODI values determined from simultaneous sleep laboratory measurements with Embla, Venla and APV2 (Studies I and II in Table 6.1). The AHI values determined with Venla and APV2 were strongly correlated with those obtained with the reference instrument (Embla) (r2 = 0.994, p 0.0001, AHIVenla = 1.014 u AHIEmbla +
0.402 and r2 = 0.994, p 0.0001, AHIAPV2 = 0.993 u AHIEmbla + 0.505). For the oxygen desaturation index, the correlations were equally strong (r2 = 0.995, p 0.0001, ODIVenla = 0.983 u ODIEmbla + 0.305 and r2 = 0.992, p 0.0001, ODIAPV2 = 0.997 u ODIEmbla + 0.226). It is important that the diagnostic sensitivities of Embletta, Venla and APV2 were rather similar (Table 6.2).
The second part of the clinical evaluation (Study I) was arranged by conducting 275 successful ambulatory home recordings in order to compare the performance of the devices in routine clinical use. No statistically significant differences were found in the AHI, ODI or hypopnea values between Embletta and Venla (Table 6.1). Furthermore, the diagnostic sensitivities of the devices were found to be similar (Table 6.2).
One important finding was that the novel devices showed better technical reliability than the commercial reference device (Embletta). This reduces the failure costs considerably due to a significantly lower number of re-recordings required (Table 6.3).
The most common reasons for technical failures of all devices were loss of either airflow or oxygen saturation signals.
Table 6.1: Comparison of the respiratory parameters (mean r SD) recorded with the evaluated devices in Studies I and II.
No statistically significant differences were detected in the AHI and ODI values determined with the devices.
Table 6.3: Summary of the technical reliability and the estimated yearly failure costs of the evaluated ambulatory devices.
The Venla and APV2 devices showed better technical reliability and lower estimated failure costs compared to the commercial reference device.
6.2 Accuracy of automatic analysis of ambulatory sleep recordings A significant portion of patients classified as having mild obstructive sleep apnea with manual analysis were misdiagnosed as normal with both devices (65.4% with Venla and 11.4% Embletta) when automatic analysis was used (Table 6.4). Furthermore, there was a trend to underestimate the severity of the disease in patients having moderate or severe sleep apnea. However, there were some cases in which the automatic analysis overestimated the severity of the disease (8.6% and 5.0% in mild and moderate classes, respectively) with Embletta. This was not seen with Venla.
Table 6.4: The agreement between automatic and manual analysis in normal, mild, moderate and severe disease with the Embletta device (N = 167) and the Venla device (N = 100, in Study III).
The automatic analysis led to misdiagnosis of a significant portion of patients. This was especially significant with Venla.
With Venla, the AHI values obtained using automatic analysis were significantly lower than those obtained with manual analysis (Table 6.5). Significant differences between automatic and manual analysis were found in the detection and classification of central and mixed apneas. Despite the differences in the absolute values of parameters, high correlations between manually and automatically determined AHI, ODI, and mixed AHI were detected with both devices. However, the Bland-Altman analysis revealed a significant trend in the difference between automatically and manually determined AHI, ODI and mixed apnea indices in the case of Venla.
Table 6.5: Comparison of manually and automatically analysed respiratory parameters in Study III.
With Venla, the AHI values obtained using automatic analysis were significantly lower than those obtained with manual analysis. Especially, the automatic analysis failed in detecting mixed and central apneas.
6.3 A portable device for brain function monitoring with eventrelated potentials In Study IV, a portable battery powered device for clinical auditory ERP measurements was developed and validated. Prior to the in vivo application, extensive technical tests and performance value measurements were conducted. The total power consumption of the device was found to be 2.8 W (460 mA at 6.0 V), of which the data logger consumes about half. The estimated operating time of the device with four D-size (18 Ah) batteries is 27 hours (70% of the battery energy consumed). The radio frequency emission generated by the device was found to be well below the limitations set by the International Special Committee for Radio Interference.
The intensity of the audio stimulus was found to correspond accurately to the thumb wheel setting (mean difference 0.20 ± 0.51 dB, Figure 6.3(a)). Furthermore, the sinewave audio tones (560 Hz and 800 Hz) were found to be free of glitches, hum and broadband audio noise.
The input signal range (32.0 ± 0.1 mV, mean ± sd) and the bit resolution (0.489 ±
0.001 PV/bit) were found to be closely matched between the four EEG channels.
Similarly, the pass bands (0.5 Hz - 85 Hz, -3 dB) of the EEG channels were found to be identical (Figure 6.3(b)). The input signal range, bit resolution and pass band of the ECG channel are 65.5 mV, 0.999 μV/bit and 0.8 Hz – 85 Hz, respectively. The pass bands are shown in Figure 6.3(b), respectively. The broadband RMS noise amplitudes of the input channels of the data logger were 2.10 ± 0.14 μV. However, the filtering (1.0 – 45.0 Hz) diminished the RMS noise amplitudes (0.80 ± 0.14 μV) significantly.
The quality of recorded spontaneous biosignals (EEG and ECG) was high, and the recorded data suited well for the clinical applications (Figure 6.4 (a) and (d)). The ERPs measured using an oddball paradigm with 85% standard (800 Hz) and 15% deviant (560 Kuopio University Publications C. Natural and Environmental Sciences 261: 1 - 79 (2009)
- 60 Results Hz) stimuli produced a clearly distinguishable response to the deviant tones when compared to the standard tones (Figure 6.4 (b)). In addition, the cortical SEP measurements were successful in producing distinct, noise-free responses (Figure 6.4(c)). The SEP measurements showed that the detection of external triggers in the data logger is very precise, as suggested by the high temporal accuracy of the electrical stimulus artefact peak. Virtually no temporal jitter of the artefact peak could be detected between individual stimuli.
Figure 6.3: The intensity of the audio stimulus was found to correspond accurately to the thumb wheel setting (a).
The pass bands (0.5 – 85 Hz, -3 dB) of the EEG channels were identical (b).
Figure 6.4: Examples of different signals recorded with the Emma device.
The quality of recorded spontaneous EEG while awake (eyes closed) was found to be good (a). The ERPs measured using the oddball paradigm produced a clearly distinguishable response to the deviant tones compared to the standard tones (b). The SEP recordings were of good quality producing distinct noise free responses (c). The quality of ECG recordings was found to be good as demonstrated here in V4 derivation of the 12-channel system (d).