Sleep apnea: a review of diagnostic sensors, algorithms, and therapies

Pneumotachography is the gold standard for monitoring ventilation during sleep. A flow meter is attached to a facemask which is placed on the nose and mouth (Webster 2014). The patient's airflow passes through a flow-resistive element with channels that laminarize the flow. With a laminar flow, the energy loss of the air going through the resistive element is due to viscosity. The pressure difference is measured across the resistive element to quantify the energy loss, and the flow F is directly proportional to the pressure difference: $~F=\frac{\pi {{r}^{4}}\left({{P}_{{\rm i}}}-{{P}_{{\rm o}}} \right)}{8\eta L}$. Under laminar flow, the pneumotachometer obeys the Hagen–Poiseuille law, where r is the radius of the channel, P_i and P_o are the pressures of the inlet and outlet ports, η is the viscosity, and L is the length of the channel (Ehrenwerth et al 2013) (figure 1). Miscalculations of the viscosity can cause erroneous measurements of the flow.

Zoom In Zoom Out Reset image size

In addition to the pneumotachometer, sensors commonly used in spirometry and ventilation monitoring include differential pressure sensors and hotwire anemometers. Differential pressure sensors operate by having a fixed orifice in the line of flow. The flow is laminar upstream from the orifice and turbulent downstream from the orifice, causing a pressure difference. Using Bernoulli's principle and a known cross-sectional area of the channel, the flow is calculated from the pressure drop across the orifice.

Although differential pressure sensors are widely used in anesthesia systems, ventilators, and spirometers, they result in a high pressure drop across the orifice at high flows, which is not ideal, and are insensitive to low flows (Ehrenwerth et al 2013). Additionally, differential pressure sensors are pressure dependent, and an unsealed respiratory circuit could cause erroneous measurements.

Hotwire anemometers are also used widely in spirometry and ventilation monitoring. Hot wire anemometers use a very fine wire electrically heated to some temperature above ambient. Air flowing past the wire cools the wire. Hotwire anemometers show great potential for medical airflow monitoring because they have a fast frequency response, high accuracy, a low pressure drop, and the ability to measure low flows and a wide dynamic range (Ardekani and Motlagh 2010). However, hotwire anemometry is susceptible to turbulent flows and debris. If placed proximally to the patient, notice should be taken to laminarize the flow and ensure debris does not affect the sensor or patient. When used in a mask for respiratory monitoring, a tight seal between the mask and the patient should be ensured to prevent leaks and cause erroneous flow measurement.

Use of a facemask with a pneumotachometer, differential pressure sensor, and hotwire anemometer sensor is often considered bulky or uncomfortable for a patient to wear during an entire sleep study. In order to prevent disturbance of sleep and circumvent the need for patient participation for flow measurement, flow sensors have been placed in nasal cannulas or above the mouth and below the nose. When measured from a nose cannula, parameters such as humidity, end-tidal CO₂, temperature, and pressure can assess changes in flow and be used to detect apneas and hypopneas. Although thermistors placed in a nose cannula have been widely used to assess cessation of flow, they have a slow response time and cannot reliably detect the full range of flows that occur during respiratory events (Montserrat et al 1997, Norman et al 1997, Teichtahl et al 2003). Positioning of thermoelements on the face, body positioning, and variation in sensitivity and frequency response between different sensors can cause variability which contributes to a poor relationship between air temperature and airflow (Berg et al 1997, Berry et al 2005).

Alternatively, the pressure measured from a nasal cannula has shown to correlate well with patient breathing. When measuring the pressure of the nasal cannula, the measurement is linearized by computing the square root of the nasal prong pressure (Farre et al 2001). Several studies have compared the Apnea hypopnea indexes (AHI) that resulted from nasal cannula pressure measurements to scores that resulted from thermistors or chest wall movements with reported bias of the AHI ranging from  −9.6 h⁻¹ to  +4.6 h⁻¹ (Bradley et al 1995, Fleury et al 1996, Kiely et al 1996, Mayer et al 1998, Rees et al 1998, Hernández et al 2001). When compared with face mask pneumotachometers, the AHI score of the nasal pressure cannula had a fair correlation to that derived from pneumotachometer. However, when measured over several hours, proportionality coefficients shifted, and therefore the nasal pressure recordings did not quantitatively reflect airflow (Thurnheer et al 2001). Although not quantitatively reflecting patient airflow, the nasal pressure cannula monitoring can be used to detect inspiratory flow limitation or hypopneas because these events are detected based on relative comparison to normal breathing of the patient (Farré et al 2001). When compared with pneumotachography, the golden standard for ventilation measurement in sleep studies, nasal pressure monitoring and subsequent linearization by a square root transform provided an AHI and detected apneas and hypopneas without a significant bias (Thurnheer et al 2001). Comparing nasal pressure monitoring to pneumotachography on an event-by-event basis rather than complete AHI scores, breathing event classification agreed well with an average Cohen's κ statistic of 0.76, where 1.0 represents complete agreement and 0.0 represents agreement no better than chance (Heitman et al 2002).

Polyvinylidene fluoride (PVDF) sensors have been recently introduced for sleep respiratory monitoring. PVDF consists of a thin plastic film, that when polarized, is sensitive to both temperature and pressure changes, and produces a linear voltage output based on those changes. When compared with the traditional methods of a thermistor and nasal pressure cannula in one study, PVDF had a near unity correlation coefficient for the four indices calculated, including apnea–hypopnea index (0.990), obstructive apnea index (0.992), hypopnea index (0.958), and central apnea index (1.0) (Kryger et al 2013). PVDF sensors eliminate the need to have multiple sensors placed at the same location, as they respond to both heat and pressure. However, they have similar limitations to those of thermistors and pressure sensors.

The commonly used respiratory inductance plethysmography (RIP) indirectly measures ventilation by recording changes in thoracic and abdomen cross-section. RIP measures abdomen chest movements through coiled wires wrapped around a patient's chest that carry a low-amplitude sine wave. Changes in chest and abdomen circumference alter the self-inductance of the wires, and therefore the frequency of the sine wave, which can be demodulated and processed to track changes in chest and abdomen size (Watson et al 1988). RIP is widely studied and regarded highly by the AASM due to its great accuracy, sensitivity, and high patient safety (Zhang et al 2012). The wires were initially integrated into two elastic bands, one around the abdomen and one around the chest, but RIP has also been used with wires sewn into shirts, comfortably worn by patients.

With proper calibration, RIP can achieve a tidal volume measurement accuracy of 96% when compared with pneumotachography or spirometry (Gonzalez et al 1984). The posture of the patient greatly affects the measurement of tidal volume; if the RIP is calibrated with the patient in the upright position, sleeping and breathing in the supine position can lead to greater sources of error. However, Gonzalez et al (1984) found that using a two-body posture calibration method can improve the error due to changing body position, with measured tidal volumes falling within 4–10% of those measured by spirometry.

RIP is noted for its added benefit of helping to distinguish between OSA and CSA, as illustrated in figure 2. During OSA, cessation of breathing occurs despite an ongoing effort to breathe, while CSA occurs when the brain does not properly send signals to the muscles controlling respiration. During CSA, the lack of effort by the muscles in the abdomen and chest can be noted with the use of RIP, aiding distinguishing CSA from OSA.

Zoom In Zoom Out Reset image size

RIP accuracy and precision is limited in obese patients. The accuracy in estimating minute ventilation is considered acceptable, with a relative difference of 5.2%. The precision of RIP in obese patients can limit the accuracy of detecting hypopneas. The SD of the difference between RIP and measurements from a pneumotachograph was 10.5% during wakefulness and 33% during sleep (Cantineau et al 1992). However, RIP scoring has been shown to have increased sensitivity and specificity in overweight or obese patients when compared to the recommended and acceptable criteria for sleep scoring by the AASM (Kogan et al 2016). Although RIP accuracy and precision is significantly decreased during sleep in obese patients, RIP is still clinically useful (Cantineau et al 1992).

Impedance plethysmography (IP) measures respiratory cycles by changes in electrical resistance. When volume changes within an induced electrical field, a high frequency (about 100 kHz), low level current (about 1 mA or more) is injected through electrodes placed on the chest wall. As the thoracic cross-sectional area changes during breathing, it changes the electrical resistance, which can be recorded as a change in voltage (Webster 2014). Like RIP, IP is prone to error and inconsistency due to changes in patients' posture during recording signal.

Esophageal manometry, or measuring the esophageal pressure, can be useful in assessing patients with upper airway resistance syndrome. Esophageal manometry can indicate increased respiratory effort during respiratory events. However, esophageal manometry is relatively invasive, not tolerated well by most patients, and therefore is not used for routine testing (McNicholas 2008). Additionally, it has been shown that the use of nasal cannula pressure monitoring can be used to identify non-invasively the same events as esophageal manometry including Respiratory Effort-Related Arousals, apneas, and hypopneas (Rapoport 2000).

Electromyographic (EMG) activity can also be recorded during sleep apnea studies. The sternocleidomastoid (SCM), genioglossal (GG) and abdominal wall (ABD) muscle activity can be observed to represent inspiratory pressure generating, inspiratory airway-maintaining, and expiratory muscles respectively. The phasic activity of the GG muscles is normally reduced in REM sleep compared to wakefulness, but children with OSA were found to have phasic GG EMG activity during sleep. The SCM and GG muscles EMG indexes indicate that they increase activity with worsening hypoxemia and hypercarbia (Jeffries et al 1984, Dempsey et al 2010).

In addition to more traditional thermal and pressure sensors, acoustic monitoring has also been researched as a possible means to detect apneas and hypopneas. An acoustic sensor is used to record the signals generated from the expired air flow. These signals can be acquired in the natural sleep environment without any contact to the subject and can be used to detect breathing events, obstruction and snoring (Vegfors et al 1993, Gordh et al 1995, Roebuck et al 2013). Children with suspected obstructive Sleep Disordered Breathing were found to produce louder high-frequency inspiratory sounds (HFIS) during sleep as well as have narrower airways (Rembold and Suratt 2004). The HFIS were observed using a microphone placed above the observed patient. Although the occurrence of HFIS correlated with the patient's rate of obstructive respiratory events, the HFIS intensity did not correlate well with respiratory effort as measured by an esophageal catheter (Rembold and Suratt 2014). This could be explained by the fact that HFIS intensity increases with high-velocity, high-turbulent flows, and some obstructive events do not have any patient airflow. While acoustic monitoring has not yet been used to accurately assess AHI indexes, acoustic monitoring has been used in the commercially available Masimo RRa^® to determine respiratory rate and apneas through an acoustic sensor that adheres to patients' necks. The RRa^® was found to be 3% more accurate than a capnograph when assessing respiratory rate, and had a higher sensitivity (P  =  0.0461) when detecting cessation of breathing (⩾30 s) (Ramsay et al 2013).

In another effort to diagnose and monitor respiratory events without patient contact, fiber grating vision sensing and laser sensing has been researched. With this technology, the positions or movements of the chest and abdomen are monitored with respect to the patient's center of gravity. Using over 100 sampling locations, respiratory events could be determined and distinction between OSA and CSA could be made (Takemura et al 2005). However, this method requires much calibration between patients, and is highly dependent upon patient movement, which can occur during sleep. With refinement, fiber-grating vision sensing could be used to detect sleep apnea without patient contact.

Researchers have also explored the use of off-the-shelf smart phones and applications to diagnose sleep apnea using a sonar system with frequency modulated continuous waves. The system uses the phone's speakers to emit sound waves and analyzes the reflections. The reflections from the body arrive back to the phone at different times depending on the distance from the phone, and amplitude changes due to changes in breathing are extracted. The system can even analyze breathing from two patients lying in the same bed. When placed within 1 m of the patient, the phone application can estimate respiratory rate with 99.2% accuracy. When compared with a traditional PSG, the application identified CSAs with a 0.9957 correlation coefficient, OSAs with a 0.9860 correlation coefficient, and hypopneas with a 0.9533 correlation coefficient. The system accurately predicted apneic events without patient contact, but does not provide other information provided in a traditional PSG, including patient position, visual information, and electroencephalogram (EEG) signals used to determine sleep stages (REM, non-REM, and awake) (Nandakumar et al 2015).

2.1.2.1. CO₂ concentration. 

2.1.2.2. O₂ concentration.

Electropotentials are measured during sleep studies to detect SAS events and sleep patterns. Signals such as EMG, EEG, and EOG are collected during polysomnography, as illustrated in figure 3.

Zoom In Zoom Out Reset image size

A number of studies have shown that SAS events and patterns can be detected with high accuracy using only single channel ECG data. HRV information can be obtained from the ECG signal, and using the standard deviation of the NN interval of both nighttime and daytime heart beats has shown to detect obstructive sleep apnea with high sensitivity (89.7%) and high specificity (98.1%). However, using ECG as the sole parameter for sleep apnea screening has its limitations, as diseases such as diabetes, sequelae of myocardial infarction and chronic heart failure, which are often associated with OSA, can cause false negatives when screening for OSA (Roche et al 1999, Shiomi et al 1996). Changes in HRV due to tidal volume and breathing effort must be taken into account when processing HRV data (Meziane et al 2015). Even when not used as the sole parameter for sleep apnea detection, ECG data are still a valuable tool used in polysomnography.

Chin and leg EMG is a traditional parameter used during polysomnography. EMG data are most often used during polysomnography to monitor sleep stages, along with EEG and EOG, because muscle tone subsides during NREM sleep and is at its lowest during REM sleep. In addition, chin electrodes are typically placed close to the geniohyoideus (GH) and genioglossus (GG) muscles which are pharyngeal dilator muscles and which play an important part in maintaining upper airway patency. Therefore, chin EMG can be considered a noninvasive method to assess upper-airway muscles' EMG activity shown in figure 3. OSA patients have been found to have significantly higher chin EMG at sleep termination, which correlates with other findings that OSA patients have higher upper airway muscle activities than normal subjects in wakefulness (Mezzanotte et al 1992, Al-Angari 2008).

Lastly, surface EEG allows for detailed analysis of depth of sleep through sleep staging, as well as sleep architecture and efficiency based on phasic transitions through sleep cycles. While not required to obtain a SAS diagnosis, EEG provides a clinically useful measure of the effect of SAS respiratory pathology on the overall quality of sleep. In SAS patients, conventional parameters showed predictable decrements in total sleep time (TST), fragmentation of sleep architecture, slow wave sleep, and REM sleep and increases in stage 1 and nocturnal awakenings (Terzano et al 1996). Moreover, obstructive event-induced work of breathing or hypoxemia has an effect on electrocortical activity. In a study of severe SAS patients, statistically significant differences were observed in the brain activity of apneic patients relative to normative patients globally and in particular in the local temporal region to SAHS events. More specifically, EEG signals measured directly following apnea onset displayed average differences of 268% between initial and maximum signal amplitude measurements, and 202% between initial and final values of amplitude, frequency, phase, and other signal and descriptive statistical parameters (Svanborg and Guilleminault 1996). As such, the AASM recommends three EEG derivations for scoring of sleep, including frontal, central, and occipital sensor locations F4/M1, C4/M1, O2/M1, as well as back-up derivations: F3/M2, C3/M2, and O1/M2 (Iber 2007, Ruehland et al 2011).

Electrooculogram (EOG) is used to capture the distinct sharp eye movements characteristic of REM sleep. Arousal from REM sleep is accompanied by sudden halting of rapid eye movement on EOG and increased muscle tone recorded on EMG.

Body position of patients during a sleep cycle is of great interest to researchers because OSA severity can vary with body position. To monitor body position, accelerometers or video recordings have been used.

Using accelerometry (or actigraphy), patient body movements can be recorded, typically with inexpensive piezoelectric sensors. Its ability to track body motion and snoring helps track information such as sleep duration and the number of times a patient wakes during the night. Although actigraphy has lower specificity than polysomnography in identifying wakefulness in patients, its use might be preferred to polysomnography when patient monitoring spans a long duration, as it is easier for patients to comply with actigraphy monitoring than traditional polysomnography (Roebuck et al 2013).

Video recordings and corresponding image processing technology have been widely used in the noncontact apnea detecting field. Many researchers correlate the PSG signals to the patient's sleeping behavior, such as body position, which is extracted from videos. In some cases, video recordings have helped to confirm diagnoses when used in conjunction with PSG signals. Even though the AHI limits may be normal, video recordings may show supplemental information about respiratory events such as head movements and arousals (Griffiths et al 1991, Sivan et al 1996, Anders and Sostek 1976, Silvestri et al 2009, Kryger et al 2013).

Table 2 shows summary of SAS diagnostic parameters from PSG, types of physiological events encompassing those parameters, sensor modalities for detecting events, and classification methods, and algorithms commonly used to recognize SAS event patterns.

Table 2. Summary of SAS diagnostic parameters, the types of physiological events, sensor modalities, and commonly used classification methods.

Diagnostic parameters	Characteristic physiological events	Relevant signal and sensor modalities	Classification methods	Reliable sensor modalities
Apnea–hypopnea index (AHI), respiratory disturbance index (RDI), other respiratory parameters	Obstructive apneas, central apneas, mixed apneas, hypopneas, respiratory effort-related arousals, oxygen desaturations, bradycardia, sinus tachycardia	Pneumotachometry, nasal cannula pressure sensors, thermistor, thermocouples, polyvinylidene fluoride, hotwire anemometers, respiratory inductance plethysmography (RIP), impedance pneumography (IP), magnetometer, strain gages, end-tidal ${{P}_{{\rm C}{{{\rm O}}_{2}}}}$, pulse oximetry, photoplethysmography, ECG, audio (acoustic sensor), IR, sonar, video, actigraphy, peripheral arterial tonometry	Amplitude and adaptive thresholding, fuzzy networks, artificial neural networks, decision trees, ensemble models, nearest neighbor methods, linear and kernel SVM, deep neural networks	Gold standard PSG studies recommend thermal and pressure based airflow sensors used with 2 effort belts, nasal cannula pressure sensors, End-tidal ${{P}_{{\rm C}{{{\rm O}}_{2}}}}$ in a hospital setting (Collop 2011)
Sleep architecture, sleep efficiency (SE), arousal index (ArI), sleep latency (SL), REM latency (RL), total sleep time (TST), periodic leg movements (PLMS) index, other arousal and sleep parameters	Stage wake, stage REM, stage N1, stage N2, stage N3, arousals, sleep spindles, K-complexes, periodic leg movements, spikes, sharps, vertex wave sharps	EEG (F4-M1, C4-M1, O2-M1, F3-M2, C3-M2,O3-M2), Chin EMG, EOG, ECG, Leg EMG, audio, video, actigraphy	Autoregressive models, hidden Markov models, Gaussian mixture models, linear and quadratic discriminant analysis, random forests, regression trees, deep belief networks, deep neural networks	Gold standard PSG studies recommend primary EEG derivations at F4/M1, C4/M1, O2/M1, submental EMG, and back-up EEG derivations at F3/M2, C3/M2, and O1/M2 (Iber 2007, Ruehland et al 2011)

A common framework through which we can consider the problem of recognizing SAS related patterns in physiological data involves two general concepts—data representations and models. Data representations can be defined as abstract reformulations of the original raw signal data that aim to exhibit desirable analytical properties for classification, such as being more discriminative between disease and normative classes, better conditioned, lower in dimensionality, sparser, among others. Models then operate on these resulting data representations, and estimate or predict the corresponding class label for each representation. In the context of automated approaches to sleep scoring, each data representation typically represents a single epoch of sleep data, while the applied models output which, if any, of the aforementioned apnea related events are found within that epoch.

Figure 4 shows the process of obtaining data representations involves applying techniques from fields including signal processing, pattern recognition, data mining, feature engineering, and machine learning to single and multichannel polysomnography data. The sections below discuss some common data representation techniques that have exhibited promising performance for respiratory and neural event detection in empirical studies.

Zoom In Zoom Out Reset image size

2.2.1.1. Time-frequency based transforms.

2.2.1.2. Multiresolution and wavelet based transforms.

2.2.1.3. Nonlinear and other transforms.

Models and classifiers take individual feature vectors or representations and output either continuous or discrete class labels. For sleep scoring, the data representations are typically vectors that correspond to a single 30 s epoch of polysomnography data. These vectors are used as inputs, while class labels for each epoch include normal breathing, apnea events, hypopnea events, and the relevant stage of sleep. The available classifiers range in complexity from simple models with a small number of static parameters such as adaptive thresholds, to architectures composed of hundreds of millions of tunable parameters as in the case of deep learning. The key aspect of machine learning classifiers is that their core models can be adaptively parameterized through a training optimization procedure that seeks to minimize the prediction error on available data while maximizing the generality to unseen data. In this sense, all models discussed here with data-adaptive parameters whose performance improves with experience could be understood in a machine learning context. This section of the review discusses specific algorithmic approaches for the classification of SAS patterns in single or multichannel PSG data.

2.2.2.1. Amplitude and adaptive thresholding.

2.2.2.2. Linear and kernel methods.

2.2.2.3. Tree based models.

2.2.2.4. Artificial neural networks.

2.2.2.5. Fuzzy logic systems and networks.

2.2.2.6. Deep learning.

2.2.2.7. Low dimensionality and total variability based approaches.

2.2.2.8. Discussion of signal processing and algorithm techniques.

The first developed PAP device, CPAP, delivers constant pressure for the whole night, which requires manual titration in the laboratory before use. CPAP is the basic type of PAP machine, invented by Colin Sullivan in 1981 (Sullivan et al 1981), and has been widely accepted as the gold standard for treatment of OSA (Kribbs et al 1993). CPAP is effective at eliminating apneas and hypopneas (Weaver et al 2012, Becker et al 2003). The major problem for CPAP is that over 40% of the patients with OSA are noncompliant with the treatment (Kribbs et al 1993, Weaver and Grunstein 2008). Several components were added to the CPAP device to improve the adherence such as chin straps, mask re-fitting, and humidification for the tube (Engleman and Wild 2003). While effective at treating OSA, CPAP may not suppress CSA, particularly in patients with heart failure (Arzt et al 2007), causing search for alternative treatment options.

The second developed PAP device, bi-level positive airway pressure (BPAP) delivers a higher IPAP and a lower pressure during expiration (EPAP), which decreases the effort for the patients to breathe. BPAP is the pressure-controlled ventilation that allows spontaneous breathing at any time. A flow sensor detects the timing to change from EPAP to IPAP. IPAP and EPAP usually range from 4 to 30 cmH₂O (Berry et al 2012b). Other parameters, such as inspiration time, pressure rise time, and flow trigger sensitivity are set by the technician (Zdrojkowski and Estes 2000).

The third PAP device, auto-titrating positive airway pressure (APAP) provides adjustable pressure to maintain airway patency, and gives appropriate response to the respiratory events. Several studies have been hypothesized that adjustable lower pressure with APAP could increase acceptance and adherence with chronic positive pressure treatment (Teschler and Berthon-Jones 1998, Berthon-Jones et al 1996, Berry et al 2002, Berry and Sriram 2014). The data in the literature mostly shows that CPAP and APAP have similar sleep quality. Since the treatment effects are similar between APAP and CPAP, the therapy of choice may depend on other factors such as patient preference, specific reasons for non-compliance and cost (Ip et al 2012). APAP and CPAP devices can adjust and compensate the pressure when leaks occur. However, due to excessive increases in pressure in the case of mask/tube leakage, potential risk may arise, and some patients may be sensitive to the pressure changes and thus feel less comfortable with APAP (Hussain et al 2004, Berry and Sriram 2014).

The final PAP device, adaptive servo ventilation (ASV) provides patients steady, minute ventilation based on the measurement of patient breaths. Although initially used in CSA patients (Hussain et al 2004), a 2015 study SERVE-HF involving 1325 patients showed that while ASV was efficacious in treating CSA, it had no significant effect on a broad spectrum of functional measures including quality-of-life measures, 6 min walk distance, or unplanned hospitalization for worsening heart failure. In contrast to smaller studies and meta-analysis, a significant increase in both cardiovascular mortality and all-cause mortality was observed in the ASV group. A first possible explanation for this trial failure is that CSA is a compensatory mechanism for heart failure. The second hypothesis is that PAP device may impair the cardiac function for some patients (Cowie et al 2015). While theoretically feasible, these hypotheses have no experimental basis, and based on the many adverse effects of CSA on CHF pathophysiology, further research is needed to understand why the SERVE-HF study failed.

In addition to the types of the PAP devices, this section discusses current continuous technological advances that aim to improve the patient's comfort, adherence, and clinical benefits which could be categorized as respiration phase detection, ventilation estimation, CSA distinguishing, humidifiers, expiratory pressure relief (EPR), ramp, and automatic start and finish.

Determining the respiration phase is essential for auto PAP devices to apply future EPR and inspiratory flow limitation. ResMed's devices employ fuzzy logic to separate the whole respiration cycle into nine different phases described in table 4, which is based on the respiration flow rate direction and the volume of flow (Berthon-Jones 2014). Respironics's devices divide the respiration cycle into fixed time segments (64 ms). To estimate the total duration of the respiration cycle, Respironics records and analyzes the history inspiration time and expiration time data (Hill 2004).

To quantify a patient's breath quality, different companies employ different algorithms to estimate ventilation. A weighted peak flow (WPF) method which averages the WPF over a current period of time (Matthews et al 2007, 2012) has a good noise rejection to estimate the ventilation. ResMed measures long term average ventilation (for example, the instantaneous ventilation is averaged within a 100 s interval by a low pass filter). The target ventilation is taken as 95% of the long term average ventilation (Berthon-Jones 2003).

Airway patency detection could distinguish CSA from OSA. Two methods are used to detect airway patency: cardiogenic oscillation testing (Morrell et al 1995) and device-generated pressure oscillation testing (Berthon-Jones 2010). The first method tries to detect the cardiogenic flow, which is the airflow induced in the lungs. It is related to the proximity of the lungs and the heart during sleep. Based on analysis of a large number of clinical cases, some researchers concluded that no cardiogenic oscillation was shown in OSA thus it could be a good indicator of CSA (Ayappa et al 1999, Martin and Oates 2014). Cardiogenic oscillations in the airflow have been observed during some central apneas, but there is controversy over whether they correlate with airway patency (Morrell et al 1995). The presence of cardiogenic oscillations on the current CPAP flow signal is a specific indicator of central apnea and may have a role in self-titrating CPAP algorithms. Some results from the trials are promising, but larger studies are needed to determine the accurate correlation between airway patency and cardiogenic oscillations (Ayappa et al 1999). The second method, applies an oscillatory pressure waveform to a patient's airway. This waveform induces an airflow signal. Figure 5 shows that patients with OSA (airway obstructed) have lower mid-inspiratory flow than CSA patients. This indicator, compared with the pre-set threshold, helps the device to distinguish CSA from OSA (Berthon-Jones 2010).

Zoom In Zoom Out Reset image size

Patients without airway obstructed (CSA patients or normal people) have higher mid-inspiratory flow.

Several methods have been developed to increase patient comfort on PAP. Humidifiers increase the humidity of inhaled air. They most often are heated, and consist of a water chamber and a heating plate. To get desirable humidification of air, temperature sensors at the heating plate or humidification at the tube are necessary to control the production of water vapor. Higher humidification of inhaled air helps to reduce nasal irritation and congestion (Massie et al 1999). If bedroom temperature is much cooler than the heating temperature, water may condense in the tube or mask, which is called 'rainout'. Insulation of the PAP hose can prevent condensation. Also, a heated hose could also benefit the user's comfort, which could increase humidification and eliminate 'rainout'.

To aid patients' exhalation against CPAP, the EPR feature was introduced. With EPR, when the patient exhales, the flow generator device detects the beginning of exhalation, and then adjusts the motor speed to drop treatment pressure, thus reducing the breathing effort. Generally, EPR is set from 0 to 3 cmH₂O and should not drop below 4 cmH₂O. Manufacturers have different names for exhalation relief, but they mostly work the same way. ResMed uses EPR for pressure relief. Respironics calls their pressure relief C-Flex, A-Flex, or C-Flex+. Zhu et al (2016) reported the effect of the pressure-relief feature on fixed CPAP and APAP treatment efficacy with a respiratory bench model. They found out pressure-relief features may attenuate CPAP efficacy if not adjusted and calibrated at the beginning of the treatment. On the other hand, the pressure-relief feature may overstate delivered pressures in APAP.

The ramp feature reduces the uncomfortable feeling of sudden air pressure increase. It works by gradually increasing pressure over a defined time range. Commonly this time is around 15–20 min. ResMed's AutoRamp fixes the starting pressure and keeps tracking the state of the patients' sleep. When patients have fallen asleep, the pressure increases gradually until it reaches the preset pressure. It determines the state of sleep by detecting the breath stability or the occurrence of sleep events. Respironics smart ramp increases the speed of ramp if hypopneas or obstructive apneas are detected (Ogden 1997).

Some PAP devices automatically start working when the machine senses airflow from breathing, which indicates that the patient is wearing the mask. If the machine senses very high airflow for a certain period of time (e.g. 1.5 s) this indicates the machine is working incorrectly and will automatically turn off.

Other new technology to mention is SensAwake CPAP modality (reducing pressure on awakenings) which works in conjunction with the Auto CPAP algorithm. SensAwake uses flow to continuously monitor patients' breathing patterns. It senses the change from breathing patterns associated with sleep to those associated with awake states. Once it detects a transition from sleep to awake it promptly reduces that pressure delivered to the patient (Dungan et al 2011).

Table 5 summarizes the benefits and challenges of PAP modalities. In summary, PAP devices are extremely effective when used correctly; however, their major challenge remains the patients' intolerance and nonadherence. Several studies have looked into improving adherence through various forms of additional communication and monitoring systems. Fox et al (2012) found that sending data every night to a web-based database can improve PAP adherence rates. By sending data such as air leaks, applied pressures, and objective adherence, coordinators could reach out to patients to help when necessary. However, other forms of communication were not as successful (Munafo et al 2016). In addition, the use of psychological and behavioral therapies have been shown to have significant effect on CPAP compliance rates. This suggests that possible improvements to increase CPAP compliance rates should consider the psychological aspect of CPAP as well (Somiah et al 2012).

OAs or intraoral devices (ODs) which aim to treat OSA by physically altering the mandible, tongue, or soft palate during sleep to prevent the collapse of upper airway muscles that is characteristic of OSA (Hoffstein 2007). ODs are growing in popularity, as they are found to be more comfortable than CPAP and have higher adherence rates. However, ODs tend to be a secondary form of treatment for OSA to CPAP as CPAP has a higher efficacy; ODs are recommended when a patient becomes noncompliant to CPAP (Kushida et al 2006). Yet, there is no predictive effectiveness ability for ODs since they have been shown to either help, hinder, or not alter sleep apnea in patients.

There are three types of ODs: soft palate lifters (SPLs), tongue retaining devices (TRDs), and mandibular advancement appliances (MAAs or OA). MAAs are by far the most used ODs today to combat OSA. MAAs operates by holding the mandible in an anterior position to physically open up the airway. When the mandible is kept anteriorly to its normal position, the MAA prevents the mandible from receding into the oropharynx and blocking the airway. TRDs hold the tongue forward in the mouth to prevent it from receding into the airway, while SPLs hold the soft palate in place to prevent collapsing in the airway. With the main airway kept open during sleep, OAs prevent apneic episodes (Hoffstein 2007).

3.2.2.1. Mandibular advancement appliances (MAA). 

3.2.2.2. Tongue retaining device (TRD).

3.2.2.3. Soft palate lifter (SPL).

3.2.2.4. Oral pressure therapy (OPT).

MMAs have a role as alternative treatment strategies for OSA. Since CPAP has been proven to be the most effective method of treatment for OSA, it is the first-line treatment strategy. If the patient cannot tolerate CPAP due to side-effects, the next course of action may be to use OAs. OAs are used only for mild cases of OSA, as for more severe cases of OSA CPAP is still the better option. Also, OAs are not effective against CSA. When considering using OAs, it is important to establish what is causing the OSA and determine if oral adjustments can possibly treat the cause. Furthermore, dental factors have to be considered when using OAs. If there are issues with any dental structures from teeth to the temporomandibular joint (TMJ), OAs therapy might not be possible, as common side effects of OAs are teeth discomfort and TMJ pain (Epstein et al 2009).

OA have promise as a main treatment for sleep apnea for patients who can't tolerate CPAP. OAs have been shown to greatly reduce or eliminate snoring in almost all patients according to bed partners and patients. The effectiveness of OAs also proves attractive, with relatively high rates in reduction of AHI. The side-effects for both MAAs and TRDs are mild and can be overcome with continued use of the device. Compliance rates for OAs tend to be higher than CPAP, which supports the use of OAs when patients don't comply with CPAP (Schmidt-Nowara et al 1995). Although OA are not as effective as CPAP in reducing sleep apnea, snoring, and improving daytime function, they have a definite role in the treatment of snoring and mild sleep apnea.

3.2.4.1. MAA versus OPT.

3.2.4.2. MAA versus TRD.

While compliance rates are high for oral devices, the main reason for discontinued use often deals with the perception of the device's effectiveness. Wearers of OAs most frequently use their devices because of complaints from their bed partners about snoring. Since the patients tend to only have a mild form of sleep apnea, they experience few daytime symptoms. This lack of daytime symptoms can lead the patient to not realize they have sleep apnea or that it has been dealt with and they stop usage. With no bed partner to detect snoring, or if they stop noticing it, the patient might stop using their treatment device even if it is comfortable. This can lead to lower compliance rates for sleep apnea patients, even though they are pleased with their devices (Hoffstein 2007).

Compliance rates for OAs have had difficulty in being accurately described due to various reasons for discontinued use. Since OAs are used to treat mild forms of OSA, patients can discontinue use even though they still have sleep apnea because they do not realize it. This leads to variations in studies about how many patients continue compliance with their devices over time as between different studies that use different lengths of follow-up and definition of compliance. For instance, three different studies had a compliance rate of 100% for 3 to 21 months, 75% after 7 months, and 52% after 3 years (Schmidt-Nowara et al 1995). However, a new method to measure compliance uses a small embedded temperature sensor that detects when the device is inside the mouth will allow for more accurate compliance values in future studies (Sutherland et al 2014). CPAP, in contrast, compliance is best measured quantitively by downloading the usage data from the machine itself so there is no need for additional sensors like in OAs (Lal et al 2010, Somiah et al 2012).

Side effects for OD can be common, but are relatively mild and tolerable, especially when compared to CPAP. Possible adverse effects for MAA were found to include jaw discomfort, TMJ pain, dryness of mouth, and slight shifting of the teeth. Similar side effects were also found with TRDs, being dryness of mouth, excess salivation, and soft tissue irritation (Deane et al 2009). For SPL, the side effects were more severe and have led to lower compliance rates. They are: gagging, soft tissue irritation, and choking (Barthlen et al 2000). Table 6 summarizes the comparison of the oral appliances.