Detection of sleep-disordered breathing in patients with spinal cord injury using a smartphone

sensors Article Detection of Sleep-Disordered Breathing in Patients with Spinal Cord Injury Using a Smartphone Yolanda Castillo-Escario 1,2,3,* , Hatice Kumru 4,5,6,* , Ignasi Ferrer-Lluis 1,2,3 , Joan Vidal 4,5,6 and Raimon Jané1,2,3   Citation: Castillo-Escario, Y.; Kumru, H.; Ferrer-Lluis, I.; Vidal, J.; Jané, R. Detection of Sleep-Disordered Breathing in Patients with Spinal Cord Injury Using a Smartphone. Sensors 2021,21, 7182. https:// doi.org/10.3390/s21217182 Academic Editor: James F. Rusling Received: 13 September 2021 Accepted: 27 October 2021 Published: 29 October 2021 Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. Copyright: © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). 1Institute for Bioengineering of Catalonia (IBEC), Barcelona Institute of Science and Technology (BIST), 08028 Barcelona, Spain; iferr[email protected] (I.F.-L.); [email protected] (R.J.) 2Department of Automatic Control (ESAII), Universitat Politècnica de Catalunya-Barcelona Tech (UPC), 08028 Barcelona, Spain 3Centro de Investigación Biomédica en Red de Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), 28029 Madrid, Spain 4Fundación Institut Guttmann, Institut Universitari de Neurorehabilitació, 08916 Badalona, Spain; [email protected] 5Universitat Autònoma de Barcelona, 08193 Bellaterra, Spain 6FundacióInstitut d’Investigacióen Ciències de la Salut Germans Trias i Pujol, 08916 Badalona, Spain *Correspondence: [email protected] (Y.C.-E.); hkumr[email protected] (H.K.) Abstract: Patients with spinal cord injury (SCI) have an increased risk of sleep-disordered breathing (SDB), which can lead to serious comorbidities and impact patients’ recovery and quality of life. However, sleep tests are rarely performed on SCI patients, given their multiple health needs and the cost and complexity of diagnostic equipment. The objective of this study was to use a novel smartphone system as a simple non-invasive tool to monitor SDB in SCI patients. We recorded pulse oximetry, acoustic, and accelerometer data using a smartphone during overnight tests in 19 SCI patients and 19 able-bodied controls. Then, we analyzed these signals with automatic algorithms to detect desaturation, apnea, and hypopnea events and monitor sleep position. The apnea–hypopnea index (AHI) was significantly higher in SCI patients than controls (25 ± 15 vs. 9 ± 7, p< 0.001). We found that 63% of SCI patients had moderate-to-severe SDB (AHI ≥ 15) in contrast to 21% of control subjects. Most SCI patients slept predominantly in supine position, but an increased occurrence of events in supine position was only observed for eight patients. This study highlights the problem of SDB in SCI and provides simple cost-effective sleep monitoring tools to facilitate the detection, understanding, and management of SDB in SCI patients. Keywords: spinal cord injury; sleep-disordered breathing; sleep apnea; sleep position; smartphone; biomedical signal processing; mHealth; monitoring 1. Introduction Sleeping, like breathing, is an action that we undertake throughout our entire life. We spend approximately 30% of our time sleeping [ 1 ], and this is strictly necessary since sleep is the natural state of rest and self-regulation of the organism. However, several diseases can affect sleep quality, producing symptoms of varying severity. These medical conditions, which are called sleep disorders, are highly prevalent in the general population. In recent years, there has been an increasing awareness of the importance of sleep, and sleep epidemiology has become a rapidly growing field [2]. One of the most common sleep disorders is sleep apnea syndrome, also referred to as sleep-disordered breathing (SDB). Sleep apnea is an underdiagnosed medical condition [3,4] that is characterized by repeated episodes of absence (apnea) or reduction (hypopnea) in airflow during sleep, which can be either obstructive or central in origin. These events lead to hypoxia and microarousals, producing sleep fragmentation and leading to symptoms such as excessive fatigue and daytime sleepiness. In addition, evidence Sensors 2021,21, 7182. https://doi.org/10.3390/s21217182 https://www.mdpi.com/journal/sensors Sensors 2021,21, 7182 2 of 19 suggests that SDB increases the risk of cardiovascular and cerebrovascular diseases [ 5 , 6 ]. The prevalence of moderate-to-severe SDB in the general population is 23% in women and 50% in men, being higher in elderly and obese individuals [7]. Spinal cord injury (SCI) is a condition that causes motor and sensory impairment below the level of the injury but also results in many other health complications. Sleep disturbances are common in patients with SCI and were not usually present before the lesion [ 8 – 10 ]. In fact, it has been reported that individuals with SCI have an increased risk of experiencing disrupted sleep [ 9 , 11 ], SDB [ 11 – 14 ], and daytime symptoms, such as sleepiness and lack of energy [ 8 ], even if they can breathe normally when awake [ 15 ]. Different predisposing factors contribute to this high incidence of SDB in SCI patients. For instance, SCI can produce significant neuromuscular weakness in the diaphragm, abdominal, and intercostal muscles, thus affecting respiratory function [ 16 ]. This is especially critical during sleep, when breathing is completely unconscious, and contributes to the appearance of disturbed sleep patterns. The neuromuscular respiratory weakness seen in SCI has an additional impact on SDB, facilitating the obstruction of the upper airway during sleep and thus hypoventilation [ 16 ]. Moreover, the injury can affect both central control of respiration and upper airway collapsibility, thus promoting the appearance of both central and obstructive apneic events [ 15 ]. The association of SCI and SDB is complex and may be influenced by multiple factors including the level and completeness of the injury, the time post-injury, and the associated comorbidities [ 8 , 15 ]. For example, it has been reported that patients with tetraplegia are more likely to suffer from SDB than those with paraplegia [17]. Although individuals with SCI are at a higher risk of sleep disorders, sleep quality is rarely evaluated in these patients [ 15 ], given the many health complications of SCI and the fact that rehabilitation interventions are mostly targeted at the recovery of motor function. Nonetheless, more efforts should be devoted to the detection and management of sleep disorders in SCI patients, since poor quality of sleep affects the patients’ recovery and well-being and may underlie serious complications. Sleep studies could help to early detect SDB in SCI patients and provide the most appropriate treatment when required. Common screening tools used for sleep apnea detection include self-reported sleepiness [ 18 ], questionnaires, such as the Berlin [ 19 ] and the STOP-Bang [ 20 ] questionnaires, or methods to assess airway dimensions such as the modified Mallampati (MMP) scores [ 21 ] or cone-beam computed tomography (CBCT) [ 22 ]. However, questionnaires and self-reported symptoms are subjective measures, and the assessment of airway dimensions only provides limited information for sleep apnea detection and can lead to inconclusive results [ 23 ]. Currently, the gold-standard technique for sleep evaluation is polysomnography (PSG). However, this technique requires a complex and costly setup, which is very uncomfortable for the patient, who needs to spend the night in the hospital attached to many sensors and wires. Moreover, most clinical sleep laboratories are not prepared to address the needs of patients with SCI such as wheelchair access, special beds, and adequately trained staff [ 15 ]. For this reason, simpler cost-effective tools are needed to facilitate the diagnostic procedure and reach more patients. Home respiratory polygraphy may be an alternative to PSG for SDB diagnosis, but stronger evidence is needed and the transport of equipment from the hospital to home and back may create difficulties [24]. In terms of sleep research in SCI individuals, some previous studies have investigated the SDB prevalence in these patients conducting laboratory PSG [ 11 , 17 ], while others relied on portable polygraphy in-hospital [ 14 ], home sleep apnea testing [ 12 , 25 – 28 ], or pulse oximetry recordings [ 13 , 29 ]. These studies demonstrated significant differences in the number of apneic events between SCI individuals and healthy controls [ 8 ], reporting an SDB prevalence in SCI patients that ranged from 15% to 81%. These discrepancies are attributable to different types of patients (e.g., injuries from cervical to lumbar levels), different diagnostic methods, and different criteria for defining SDB [ 15 , 16 ]. More information about these and other studies can be found in some state-of-the-art reviews on sleep research and SDB in SCI [8,15,16,30]. Sensors 2021,21, 7182 3 of 19 Over the last years, many novel portable systems have been proposed to detect sleep apnea through unobtrusive sensors measuring only a subset of physiological signals such as nasal airflow, thoracic movement, oxygen saturation, or acoustic snoring signals [ 31 ]. Recently, our group developed a smartphone-based system for sleep apnea screening, monitoring, and management [ 32 – 35 ] and compared it with a commercial portable device for sleep apnea diagnosis at home. The proposed system uses the built-in sensors of the smartphone and an external pulse oximeter to acquire acoustic, accelerometer, and pulse oximeter data. These data can be analyzed to detect and quantify apnea and hypopnea events and, thus, stratify patients according to their severity [ 32 , 33 ]. Moreover, smartphone accelerometer data can be used to provide a high-resolution sleep position and investigate its association with apneic events [ 34 , 35 ], since supine position can promote the occurrence of apnea and hypopnea events, a phenomenon known as positional sleep apnea [36,37]. In this work, we aimed to use a smartphone system to perform sleep studies in individuals with SCI, so that we could evaluate SDB and its association with sleep position in these patients using a simple non-invasive mobile health (mHealth) tool. This technical approach has been successfully tested in patients with sleep apnea and healthy subjects in previous studies [ 32 – 34 ], but it has never been used in SCI patients. Therefore, the main objectives of this study were to use a novel smartphone system to monitor sleep apnea and sleep position in individuals with SCI and to investigate the characteristics of apnea, hypopnea, and desaturation events in these patients. We hypothesized that SCI patients would have more respiratory problems during sleep than control subjects, which may lead to altered ventilation and oxygenation patterns with an increased number of episodes of flow limitation. The analysis of biomedical signals recorded with a smartphone allowed for us to investigate respiratory events and sleep position in these patients and, thus, to test the potential of this novel mHealth tool to detect respiratory sleep disorders after SCI and its relationship with sleep posture. 2. Materials and Methods 2.1. Participants Nineteen individuals with SCI (16 men, 3 women, mean age 43 ± 16 years) were selected to participate in the study (Table 1). The inclusion criteria were cervical or thoracic SCI patients admitted to hospital for rehabilitation, traumatic or non-traumatic in origin, less than 1 year post-injury, and complete or incomplete injuries classified as A–D according to the American Spinal Injury Association Impairment Scale (AIS) [ 38 ]. The exclusion criteria were lumbar SCI, no signed consent, previously diagnosed SDB or respiratory disorders, requirement for ventilatory support, pacemaker dependency, arrhythmia and other cardiovascular conditions, any other neurological disorder, and other comorbidities that would contraindicate the test. In addition, nineteen able-bodied subjects (15 men, 4 women) were used as a healthy control group. Control subjects were excluded if they had previously been diagnosed with SDB, other sleep disorders, or any neurological or musculoskeletal disease. The mean age of the control group was 37 ± 14 years with no significant differences between groups (p= 0.23). The mean body mass index (BMI) was 23 ±3 kg/m2 in SCI patients and 24 ± 4 kg/m 2 in controls, without significant differences between groups (p= 0.27). Table 1. Clinical characteristics of the SCI individuals. Patient ID Gender Age (Years) BMI (kg/m2) Injury Level AIS Months Post-Injury Etiology SCI 1 M 47 24.8 C4 A 6.6 Traumatic SCI 2 F 27 20.8 C4 A 9.6 Traumatic SCI 3 M 18 17.3 C4 A 11.9 Traumatic SCI 4 M 47 22.0 C4 C 3.0 Traumatic SCI 5 M 55 20.3 C4 C 5.7 Traumatic Sensors 2021,21, 7182 4 of 19 Table 1. Cont. Patient ID Gender Age (Years) BMI (kg/m2) Injury Level AIS Months Post-Injury Etiology SCI 6 M 45 20.7 C4 C 6.7 Traumatic SCI 7 M 60 23.8 C4 D 2.8 Traumatic SCI 8 M 76 22.2 C4 D 8.4 Traumatic SCI 9 M 31 24.2 C5 A 7.7 Traumatic SCI 10 M 19 23.2 C5 C 3.4 Non-traumatic SCI 11 M 46 21.0 C6 A 5.0 Traumatic SCI 12 M 67 23.5 C6 A 6.1 Traumatic SCI 13 M 20 19.6 C6 B 6.2 Traumatic SCI 14 M 38 20.6 C8 B 6.2 Traumatic SCI 15 M 34 21.9 T7 B 1.7 Traumatic SCI 16 F 55 28.8 T9 D 2.4 Non-traumatic SCI 17 M 53 28.9 T10 B 3.7 Traumatic SCI 18 M 43 22.2 T11 A 1.3 Traumatic SCI 19 F 38 22.3 T11 D 4.1 Traumatic Mean ±SD/ Total 16 M (84%) 3 F (16%) 43 ±16 22.5 ±2.8 14 cervical (C4–C8) 5 thoracic (T7–T11) 7 AIS A 4 AIS B 4 AIS C 4 AIS D 5.4 ±2.8 Traumatic: 17 (89%) Non-traumatic: 2 (11%) The protocol was approved by the Ethics Committee of the Institut Guttmann (IG code number 2020343) and was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all the participants prior to enrollment. 2.2. Data Acquisition System and Experimental Setup Overnight recordings were obtained from the 19 SCI patients and 19 control subjects using a smartphone system to measure acoustic, accelerometric, and pulse oximetry signals. These data were then analyzed to detect apneic events and monitor sleep position. This allowed for a simplified approach for sleep apnea investigation as described in [32–34]. The smartphone was a Samsung Galaxy S5 SM-G900F with Android 6.0.1. This model was chosen because it is a mid-range phone with a high-quality microphone [ 39 ]. Sleep recordings with the smartphone were performed during one full night at the hospital in the SCI patients and at home in the control group. The smartphone was placed and fixed with an elastic band on the subjects’ thorax, over the sternum (Figure 1), in the position suggested by Nakano et al. [ 40 ]. In this configuration, the accelerometer’s x-axis was in the medial–lateral direction pointing to the left side of the body, the y-axis in the inferior–superior direction pointing towards the head of the patient, and the z-axis in the anteroposterior direction pointing front to back (Figure 1). During the acquisition, the smartphone was in flight mode with the WiFi and Bluetooth options disabled and the screen switched off. Control subjects were instructed on how to wear the smartphone system with the elastic band and how to start and stop the acquisition. In the case of SCI patients, due to the fact of their motor disability, the setup was prepared by trained clinical staff. To reduce possible interferences and sound artifacts, subjects slept alone in the bed during the recordings, and they were instructed to try to minimize noise sources such as sounds from machines or electronic devices. Moreover, the smartphone placement ensured that the microphone was close to the nose and mouth. In that position, the signal-to-noise ratio (SNR) of the smartphone recordings was comparable to those of commercial tracheal microphones [ 39 ]. Participants were able to choose the sleeping position they liked freely (no specific instructions were given in this regard). Since most of the tetraplegic SCI Sensors 2021,21, 7182 5 of 19 individuals were unable to turn, their position was changed at least every 3 h during the night by the nursing staff as per clinical protocol. Figure 1. Smartphone placement attached to the subject’s thorax with an elastic band. The orientation of the smartphone accelerometer’s axes and the sleep and stand angles are also indicated. The proposed mHealth system recorded three signals simultaneously: audio, using the smartphone built-in microphone; tri-axial accelerometry, with the smartphone embedded accelerometer (MPU-6500 sensor); oxygen saturation (SpO 2 ), using an EMO-80 wireless fingertip pulse oximeter (EMAY Ltd., Hong Kong, China). The sampling frequency was 48 kHz for audio signals, 200 Hz for accelerometer data, and 1 Hz for SpO 2 . The Android app “Automate” was used to automatically launch the acquisition apps when the phone booted up (Easy Voice Recorder for audio signals and Sensors Logger for accelerometry). Data were automatically stored in the internal memory of the smartphone in .wav and .txt formats, respectively. The pulse oximeter acquired data overnight and was then connected to the smartphone via Bluetooth through the EMAY Pulse Oximeter app to export the SpO 2 data in .csv files. Total sleeping time had to be at least 4 h, otherwise the examination was repeated. 2.3. Signal Processing and Analysis The analysis of acoustic signals was used to obtain ventilation patterns and detect apneas and hypopneas, while SpO 2 data allowed for the investigation of oxygenation patterns. Data from the smartphone accelerometer were used to calculate the sleeping position and investigate its relationship with the appearance of apnea and hypopnea events. Signal processing and analysis was performed offline using custom algorithms developed by our group in MATLAB®r2018a (Mathworks Inc., Natick, MA, USA). 2.3.1. SpO2Analysis Pulse oximetry recordings allowed us to track the changes in oxygen saturation during the night and, especially, to identify drops in SpO 2 (i.e., desaturations) caused by apneas and hypopneas. SpO 2 values lower than 40% or higher than 100% were considered artifacts and were padded with the previous correct value. The recordings were automatically analyzed to extract a series of features including the awake SpO 2 (calculated as the median SpO 2 value in the first 30 s of the recordings), the median and minimum SpO 2 , and the cumulative time spent with SpO 2 below 90% (CT90) and below 94% (CT94), both expressed as a percentage of the total sleeping time. In addition, the oxygen desaturation index (ODI) was calculated as the number of oxygen desaturations of at least 3% per hour of sleep. 2.3.2. Apnea and Hypopnea Detection Audio signals were downsampled to 5 kHz, applying a lowpass filter with a cut-off frequency of 2.5 kHz to prevent aliasing. Since there was a lot of wide-band background noise, especially at lower frequencies, spectral subtraction was applied to the signals [ 41 ]. An estimated noise model was automatically selected by calculating the root mean squared (RMS) value of each 0.5 s window (99% overlap) in the first 10 min of the recordings, and Sensors 2021,21, 7182 6 of 19 then joining the 10 windows (non-overlapping with each other) with the lowest RMS to obtain a segment of 5 s to estimate the noise spectrum. After this filtering step, signals were normalized to the maximum absolute value. The first 10 min were discarded for the subsequent analysis. On the other hand, movement artifacts and position changes were detected from accelerometer data [ 33 ] and excluded from the analysis, since they also produced sound artifacts. An entropy-based analysis of acoustic signals was used to detect silence events (SEv) corresponding to apneas and hypopneas as in previous studies [ 32 ]. The automatic detection of SEv was based on the calculation of the fixed sample entropy (fSampEn). fSampEn is a measure of time-series complexity, or regularity, that can be used as a robust envelope estimator for noisy physiological signals [ 42 , 43 ]. Being Nthe number of data points in the time series, mthe embedding dimension, and ra tolerance parameter; the fSampEn(m,r,N) is defined as the negative natural logarithm of the conditional probability that, in a data set of length N, two sequences that are similar to msamples within a tolerance rremain similar for m + 1 samples [42,43]. The SpO 2 signal was used to guide the SEv detector, since, to reduce the computational cost and false alarm rate, we only analyzed the audio segments starting 60 s prior to the beginning of each desaturation event and finishing at the end of the desaturation event. Overlapping segments were concatenated up to a maximum length of 10 min. In each of those segments, the envelope of the audio signal was computed by calculating the fSampEn in 0.75 s windows with 50% overlap (m= 2, and rwas equal to the standard deviation of the audio segment). After that, an adaptive threshold was applied; all points below that threshold were found, and regions between 6 and 100 s were selected as SEv [ 32 ], corresponding to either apneas or hypopneas. Once SEv were detected, they were classified into apneas or hypopneas using an algorithm previously published by our group, which showed an accuracy of 82% for apnea/hypopnea classification [ 32 ]. The algorithm is based on time–frequency representations of the audio segments to detect low-intensity respiratory sounds and distinguish them from artifacts. If low-intensity respiratory sounds were found, that event was classified as a hypopnea, otherwise it was classified as an apnea. A step-by-step explanation and all the details of the algorithms described in this section for SEv detection and for apnea/hypopnea classification can be found in [32]. The apnea–hypopnea index (AHI) was calculated as the total number of SEv (apneas and hypopneas) per hour of sleep. According to the American Academy of Sleep Medicine (AASM) guidelines [ 44 ], subjects can be classified into 4 different categories: normal (AHI < 5), mild sleep apnea (5 ≤ AHI < 15), moderate sleep apnea (15 ≤ AHI < 30), and severe sleep apnea (AHI ≥ 30). After classifying the events, we also calculated the apnea index (AI) and hypopnea index (HI) as the number of apneas or hypopneas per hour of sleep, respectively. Moreover, we calculated the percentage of time spent in apnea and hypopnea events, i.e., the sum of the duration of all the SEv divided by the total time. 2.3.3. Sleep Position Monitoring From accelerometer data, the sleep and stand angles were derived based on the projection of gravity on the axes of the accelerometer using the algorithms presented in [ 34 , 35 ]. This method was validated in previous studies, showing a 96% agreement with video-validated position from PSG [34]. To remove high-frequency noise, a median filter with a window of 60 s was applied around each accelerometer data sample. Then, the sleep angle was calculated as the angle in the X–Z plane between the accelerometry vector and the (1,0) vector, while the stand angle was calculated as the angle in the Y–Z plane between the accelerometry vector and the (1,0) vector. The sleep angle provides information about the sleep position (lateral rotation) while sleeping. As defined, 0 ◦ is a perfect left position, 90 ◦ a perfect supine position, ± 180 ◦ a perfect right position, and − 90 ◦ a perfect prone position [ 34 , 35 ]. For visualization purposes, the sleep angle was discretized into the 4 classical sleep positions Sensors 2021,21, 7182 7 of 19 using the thresholds that showed the best agreement with PSG according to previous studies: supine (60–120 ◦ ), lateral left ( − 40–60 ◦ ), lateral right (120–180 ◦ and from − 180 to − 140 ◦ ), and prone (from − 140 ◦ to − 40 ◦ ) [ 34 ]. The stand angle indicates whether the subject is standing or lying in bed and was used to discard non-lying positions. As defined, ± 180 ◦ corresponds to a perfect stand position, 0 ◦ to a headstand position, and 90 ◦ and −90◦to a complete lying position. In addition, we studied and represented the sleep position with angular resolution to investigate its association with the occurrence of apnea and hypopnea events. It is known that some patients with sleep apnea have a higher frequency of events in supine position, a phenomenon that is known as positional sleep apnea. To investigate whether the SCI patients in our sample suffered from positional sleep apnea or, conversely, there was no relationship between the occurrence of events and sleep position, we calculated the percentage of time spent at each sleep angle (i.e., for each sleep angle, θ , in increments of 1 ◦ , we computed the percentage of time spent in a 15 ◦ window centered at that angle: θ± 7.5 ◦ ) and the percentage of events occurring at that angle (i.e., for each sleep angle, θ , we computed the percentage of apneas and hypopneas occurring in sleep angles of θ± 7.5 ◦ ) [ 34 ]. Then, we compared the percentage of time and the percentage of events occurring at each sleep angle and subtracted the two curves to determine whether more events than expected occurred at each position. 2.3.4. Oral vs. Nasal Breathing We calculated, for each subject, the percentage of time that was spent breathing through the mouth during the night using an algorithm developed by our group to distinguish between nasal and oral breathing from the spectral characteristics of acoustic breathing signals [32]. Audio signals were segmented into 10 s non-overlapping sliding windows, and each window was classified into nasal or oral breathing. To do so, the fast Fourier transform (FFT) was calculated, and a linear envelope extracted using windows of 15 Hz. While most of the power of nasal breathing is concentrated in the low-frequency band, the spectrum of oral breathing presents a prominent peak between 950 and 2 kHz [ 32 ]. Therefore, if the height of the maximum peak in the 950–2000 Hz band was, at least, 60% of the maximum value of the envelope (in the 500–2000 Hz band, to avoid the variable effect of basal noise concentrated at lower frequencies), then the window was labeled as oral breathing and otherwise considered nasal breathing [ 32 ]. Once all windows were labeled, we calculated the percentage of windows classified as oral breathing. 2.4. Statistical Analysis The features described in Section 2.3. were extracted for all subjects. Data were also averaged for the SCI and control groups. Means, standard deviations (SDs), and ranges are reported for each group. Mann–Whitney U tests were applied to compare the two groups (SCI vs. control), since normality assumption was not met according to Kolmogorov–Smirnov tests. In the SCI group, the Spearman correlation coefficient was used to investigate the relationship between the extracted features (especially AHI and SpO 2 parameters) and age, BMI, injury level, AIS, and time post-injury. An alpha level of 0.05 was used to determine significance for all statistical tests. 3. Results The extracted features are displayed in Table 2for all SCI patients, while a summary of the values for each group (SCI vs. control) and the p-values of the statistical comparisons are presented in Table 3. Below we describe in more detail the results in terms of oxygen saturation, apneas, and hypopneas detected from acoustic signals, sleep position measured from accelerometer data, and prevalence of oral breathing. Sensors 2021,21, 7182 8 of 19 Table 2. Outcome measures for all SCI patients. Patients with mild sleep apnea (5 ≤ AHI < 15) are highlighted in yellow, patients with moderate sleep apnea (15 ≤AHI < 30) in orange, and patients with severe sleep apnea (AHI ≥30) in red. Patient ID Awake SpO2 Median SpO2 Minimum SpO2 CT94 (%) CT90 (%) ODI (h−1) AHI (h−1) AI (h−1) HI (h−1) Time in Events (%) Oral Breathing (%) SCI 1 98 95 78 24.33 3.73 40.44 42.56 11.55 31.01 28.58 11.42 SCI 2 98 98 76 0.61 0.17 10.03 12.09 5.62 6.47 8.65 67.91 SCI 3 98 95 78 23.30 3.18 14.82 13.85 9.84 4.01 6.76 31.03 SCI 4 95 95 54 32.15 15.21 59.59 60.05 30.40 29.64 31.24 46.57 SCI 5 95 94 73 44.11 5.55 17.66 15.59 3.29 12.30 7.21 12.11 SCI 6 99 92 74 70.38 24.41 20.83 24.84 8.93 15.91 19.68 8.81 SCI 7 95 95 83 26.73 1.98 43.00 45.85 21.32 24.54 37.29 22.73 SCI 8 92 90 81 98.41 39.04 25.38 29.43 9.29 20.14 19.91 61.64 SCI 9 98 96 85 7.05 0.06 14.06 14.54 3.60 10.93 8.10 44.77 SCI 10 96 93 61 69.88 0.82 14.00 8.50 1.87 6.62 3.46 63.72 SCI 11 98 95 81 4.95 0.45 11.49 11.10 9.64 1.45 5.20 52.63 SCI 12 98 94 74 44.35 9.87 30.74 35.19 12.20 22.99 24.58 55.87 SCI 13 99 95 90 13.13 0.00 9.02 8.77 1.98 6.79 6.37 36.29 SCI 14 97 95 88 6.62 3.23 7.59 9.06 4.53 4.53 7.21 11.24 SCI 15 94 95 88 7.20 0.15 16.09 16.33 3.87 12.46 9.20 34.79 SCI 16 94 90 80 89.95 37.78 26.60 30.42 10.56 19.85 21.57 63.42 SCI 17 96 95 80 22.68 1.11 39.98 42.17 26.10 16.07 28.64 43.35 SCI 18 99 95 89 14.75 0.05 25.01 24.89 15.13 9.76 20.32 30.75 SCI 19 95 94 79 44.91 1.10 22.44 28.60 15.08 13.51 19.48 12.96 Table 3. Group means, standard deviations, and ranges for each variable for the SCI and control groups. Group Statistic Awake SpO2 Median SpO2 Minimum SpO2 CT94 (%) CT90 (%) ODI (h−1) AHI (h−1) AI (h−1) HI (h−1) Time in Events (%) Oral Breathing (%) SCI Mean ±SD 97 ±2 94 ±2 79 ±9 34 ±29 8 ±12 24 ±14 25 ±15 11 ±8 14 ±9 16 ±10 37 ±20 Range 92–99 90–98 54–90 0.6–98 0–39 8–60 8–60 2–30 1–31 3.5–37 9–68 Control Mean ±SD 96 ±2 94 ±2 86 ±825 ± 36 4±11 9 ±7 9 ±7 5 ±3 5 ±5 5 ±4 11 ±16 Range 93–99 90–98 60–95 0–95 0–39 0.7–27 0.3–24 0.3–9 0–19 0.1–13 1.3–66 p-Value 0.26 0.77 0.003 0.04 0.003 <0.001 <0.001 0.006 <0.001 <0.001 <0.001 3.1. SpO2Measures Table 2shows the SpO 2 measures for all SCI individuals, while Table 3compares the mean values of the SCI and control groups. Mean ODI was 24 ± 14 for the SCI patients (range: 8–60) and 9 ± 7 for control subjects (range: 0.7–27), with significant differences between groups (p< 0.001). There were also significant differences in the minimum SpO 2 , CT90, and CT94 (Table 3). However, the awake SpO 2 and median SpO 2 did not significantly differ between groups (Table 3). Therefore, although SCI patients had more desaturations of at least 3% than control subjects, the basal SpO 2 levels before and during sleep were not considerably altered. 3.2. Apneas and Hypopneas Apnea and hypopnea events were automatically identified from the analysis of acoustic signals and SpO 2 for all the subjects. An example of a segment of SpO 2 and audio from the overnight recordings of a SCI patient (SCI 4) with severe SDB (AHI = 60 h −1 ) is shown in Figure 2, with the marks corresponding to automatically detected desaturation, apnea, and hypopnea events. Sensors 2021,21, 7182 9 of 19 Figure 2. Example of a segment of audio and SpO 2 signals from a SCI patient with severe OSA (SCI 4, AHI = 60 h−1), showing apnea (red), hypopnea (cyan), and desaturation (black) events. Table 2shows the AHI, HI, AI, and percentage time spent in apnea and hypopnea events for each of the SCI patients. All SCI patients (100%) met the diagnostic criteria for sleep apnea (AHI ≥ 5), with 63% of the patients having moderate-to-severe sleep apnea (AHI ≥ 15). Specifically, none of the 19 patients had AHI < 5, seven patients (37%) had mild SDB (5 ≤ AHI < 15), six patients (31.5%) had moderate SDB (15 ≤ AHI < 30), and six patients (31.5%) had severe SDB (AHI ≥ 30) (Figure 3a, Table 2). In contrast, the incidence of SDB (AHI ≥ 5) in control subjects was 74%, with 21% of the patients having moderateto-severe SDB (AHI ≥ 15). Specifically, 5 of the 19 controls (26%) had AHI < 5, 10 (53%) had mild SDB (5 ≤ AHI < 15), four (21%) moderate SDB (15 ≤ AHI < 30), and none had severe SDB (AHI > 30) (Figure 3a). Fisher’s exact test confirmed that the occurrence of SDB was significantly higher in the SCI patients than in the control sample (p= 0.02). Figure 3. Number of subjects in each category of sleep apnea severity for the SCI and control groups ( a ), mean and SD of AHI, AI, and HI in each group ( b ), and mean and SD of the percentage of time spent during events and the percentage of oral breathing in each group ( c ). Statistically significant differences are indicated with asterisks: ** p< 0.01, and *** p< 0.001. Mean AHI was 25 ± 15 h −1 for the SCI group (range: 8–60) and 9 ± 7 for the control group (range: 0.3–24), with significant differences between groups (p< 0.001) (Table 3, Figure 3b). Both AI and HI were significantly higher in the SCI group (Table 3, Figure 3b ). The percentage of the total time of the night spent in apnea and hypopnea events was also significantly higher in SCI patients than control subjects (16 ± 10 vs. 5 ± 4, p< 0.001) (Table 3, Figure 3c). Sensors 2021,21, 7182 16 of 19 this study and possible future extensions. First, despite the already high SDB severity found in SCI patients, the AHI might be slightly underestimated for several reasons: (1) hypopneas with snoring could be missed since our algorithm was specifically designed to detect silence events [32], and (2) as the apnea/hypopnea detection is guided by SpO2(i.e., only regions preceding desaturations are analyzed), we could miss apneas not followed by desaturations or hypopneas associated with arousals. To address these issues, in future work we could combine audio information with other channels from the smartphone or external sensors but also implement machine learning and deep learning approaches and compare them with the proposed rule-based algorithms. On the other hand, future extensions could include increasing the sample size to better assess the effects of the injury characteristics (injury level, completeness, and time post-injury) and other factors (including medication use or rehabilitation treatments) into SDB severity in SCI patients. Longitudinal studies could also be useful to investigate AHI variability between different nights, follow up the patient’s condition, and even assess the effects of rehabilitation. These studies would be extremely costly with current procedures. However, the proposed smartphone system is a simple non-invasive tool which would reduce the costs and complexity of medical sleep monitoring. Therefore, it could facilitate access to sleep studies for SCI patients to improve the detection and management of SDB in these patients with subsequent benefits for their overall health. 5. Patents The algorithms presented in this manuscript are under a process to recognize industrial property. Author Contributions: Conceptualization: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; methodology: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; software: Y.C.-E. and I.F.-L.; validation: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; formal analysis: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; investigation: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; resources: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; data curation: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; writing—original draft preparation: Y.C.-E.; writing—review and editing: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; visualization: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; supervision: H.K., J.V. and R.J.; project administration: Y.C.-E., H.K., I.F.-L., J.V. and R.J.; funding acquisition: Y.C.-E., H.K., I.F.-L., J.V. and R.J. All authors have read and agreed to the published version of the manuscript. Funding: This work was supported in part by the “La Caixa” Foundation (ID 100010434) under fellowship codes LCF/BQ/DE18/11670019 and LCF/BQ/DI17/11620029; in part by the European Union’s Horizon 2020 Research and Innovation Program under the Marie Sklodowska–Curie Grant number 713673; in part by the CERCA Program/Generalitat de Catalunya; in part by the Secretaria d’Universitats i Recerca de la Generalitat de Catalunya under grant GRC 2017 SGR 01770; in part by the Spanish Ministry of Science, Innovation and Universities and the European Regional Development Fund under grant RTI2018-098472-B-I00; in part by H2020-ERA-NET Neuron under Grant AC16/00034; in part by La Maratóde TV3 2017 under Grant 201713.31; in part by Premi Beca “Mike Lane” 2019—Castellers de la Vila de Gràcia. Institutional Review Board Statement: This study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Ethics Committee of Institut Guttmann (protocol code IG 2020343 from 5 October 2020). Informed Consent Statement: Informed consent was obtained from all subjects involved in the study . Data Availability Statement: The data presented in this study are available on request from the corresponding authors. Acknowledgments: This work was developed in the framework of the joint project “Biomedical Signal Interpretation to Study Motor Impairment, Neurological Disorders and Novel Personalised Neurorehabilitation Therapies” between the Fundación Institut Guttmann and the Institute for Bioengineering of Catalonia. Conflicts of Interest: The authors declare no conflict of interest. Sensors 2021,21, 7182 17 of 19 Abbreviations AASM American Academy of Sleep Medicine AI Apnea Index AIS American Spinal Injury Association Impairment Scale AHI Apnea–Hypopnea Index BMI Body Mass Index CBCT Cone-Beam Computed Tomography CT90 Cumulative Time with SpO2below 90% CT94 Cumulative Time with SpO2below 94% FFT Fast Fourier Transform fSampEn Fixed Sample Entropy HI Hypopnea Index mHealth Mobile Health MMP Modified Mallampati ODI Oxygen Desaturation Index PSG Polysomnography RMS Root Mean Square SCI Spinal Cord Injury SD Standard Deviation SDB Sleep-Disordered Breathing SEv Silence Event SNR Signal-to-Noise Ratio SpO2Oxygen Saturation References 1. Aminoff, M.J.; Boller, F.; Swaab, D.F. We spend about one-third of our life either sleeping or attempting to do so. Handb. Clin. Neurol. 2010,98, 7. [CrossRef] 2. Ferrie, J.E.; Kumari, M.; Salo, P.; Singh-Manoux, A.; Kivimaki, M. Sleep epidemiology—A rapidly growing field. Int. J. Epidemiol. 2011,40, 1431–1437. [CrossRef] [PubMed] 3. Kapur, V.; Strohl, K.P.; Redline, S.; Iber, C.; O’Connor, G.; Nieto, J. Underdiagnosis of Sleep Apnea Syndrome in U.S. Communities. Sleep Breath. 2002,6, 49–54. [CrossRef] [PubMed] 4. Ribeiro, J.P.; Araújo, A.; Vieira, C.; Vasconcelos, F.; Pinto, P.M.; Seixas, B.; Cêrca, B.; Borges, I. Undiagnosed Risk of Obstructive Sleep Apnea in Obese Individuals in a Primary Health Care Context. Acta Med. Port. 2020,33, 161–165. [CrossRef] 5. Yoshihisa, A.; Takeishi, Y. Sleep Disordered Breathing and Cardiovascular Diseases. J. Atheroscler. Thromb. 2019 ,26, 315–327. [CrossRef] 6. Durgan, D.J.; Bryan, R.M., Jr. Cerebrovascular Consequences of Obstructive Sleep Apnea. J. Am. Hear. Assoc. 2012 ,1, e000091. [CrossRef] 7. Heinzer, R.; Vat, S.; Marques-Vidal, P.; Marti-Soler, H.; Andries, D.; Tobback, N.; Mooser, V.; Preisig, M.; Malhotra, A.; Waeber, G.; et al. Prevalence of sleep-disordered breathing in the general population: The HypnoLaus study. Lancet Respir. Med. 2015 ,3, 310–318. [CrossRef] 8. Hultén, V.D.T.; Biering-Sørensen, F.; Jørgensen, N.R.; Jennum, P.J. A review of sleep research in patients with spinal cord injury. J. Spinal Cord Med. 2020,43, 775–796. [CrossRef] 9. Giannoccaro, M.P.; Moghadam, K.K.; Pizza, F.; Boriani, S.; Maraldi, N.M.; Avoni, P.; Morreale, A.; Liguori, R.; Plazzi, G. Sleep disorders in patients with spinal cord injury. Sleep Med. Rev. 2013,17, 399–409. [CrossRef] 10. Biering-Sørensen, F.; Biering-Sørensen, M. Sleep disturbances in the spinal cord injured: An epidemiological questionnaire investigation, including a normal population. Spinal Cord 2001,39, 505–513. [CrossRef] 11. Short, D.J.; Stradling, J.R.; Williams, S.J. Prevalence of sleep apnoea in patients over 40 years of age with spinal cord lesions. J. Neurol. Neurosurg. Psychiatry 1992,55, 1032–1036. [CrossRef] 12. McEvoy, R.D.; Mykytyn, I.; Sajkov, D.; Flavell, H.; Marshall, R.; Antic, R.; Thornton, A.T. Sleep apnoea in patients with quadriplegia. Thorax 1995,50, 613–619. [CrossRef] 13. Cahan, C.; Gothe, B.; Decker, M.J.; Arnold, J.L.; Strohl, K.P. Arterial oxygen saturation over time and sleep studies in quadriplegic patients. Paraplegia 1993,31, 172–179. [CrossRef] 14. Stockhammer, E.; Tobón, A.; Michel, F.; Eser, P.; Scheuler, W.; Bauer, W.; Baumberger, M.; Müller, W.; Kakebeeke, T.; Knecht, H.; et al. Characteristics of sleep apnea syndrome in tetraplegic patients. Spinal Cord 2002,40, 286–294. [CrossRef] 15. Sankari, A.; Vaughan, S.; Bascom, A.; Martin, J.L.; Badr, M.S. Sleep-Disordered Breathing and Spinal Cord Injury: A State-of-theArt Review. Chest 2019,155, 438–445. [CrossRef] 16. Chiodo, A.E.; Sitrin, R.G.; Bauman, K.A. Sleep disordered breathing in spinal cord injury: A systematic review. J. Spinal Cord Med. 2016,39, 374–382. [CrossRef] Sensors 2021,21, 7182 18 of 19 17. Sankari, A.; Bascom, A.; Oomman, S.; Badr, M.S. Sleep Disordered Breathing in Chronic Spinal Cord Injury. J. Clin. Sleep Med. 2014,10, 65–72. [CrossRef] 18. Philip, P.; Bailly, S.; Benmerad, M.; Micoulaud-Franchi, J.A.; Grillet, Y.; Sapène, M.; Jullian-Desayes, I.; Joyeux-Faure, M.; Tamisier, R.; Pépin, J.L. Self-reported sleepiness and not the apnoea hypopnoea index is the best predictor of sleepiness-related accidents in obstructive sleep apnoea. Sci. Rep. 2020,10, 16267. [CrossRef] 19. Netzer, N.C.; Stoohs, R.A.; Netzer, C.M.; Clark, K.; Strohl, K.P. Using the Berlin Questionnaire To Identify Patients at Risk for the Sleep Apnea Syndrome. Ann. Intern. Med. 1999,131, 485–491. [CrossRef] 20. Chung, F.; Yegneswaran, B.; Liao, P.; Chung, S.A.; Vairavanathan, S.; Islam, S.; Khajehdehi, A.; Shapiro, C.M. STOP Questionnaire. Anesthesiology 2008,108, 812–821. [CrossRef] 21. Dalewski, B.; Kami´nska, A.; Syrico, A.; Kałdunska, A.; Pałka, Ł.; Sobolewska, E. The Usefulness of Modified Mallampati Score and CT Upper Airway Volume Measurements in Diagnosing OSA among Patients with Breathing-Related Sleep Disorders. Appl. Sci. 2021,11, 3764. [CrossRef] 22. McCrillis, J.M.; Haskell, J.; Haskell, B.S.; Brammer, M.; Chenin, D.; Scarfe, W.C.; Farman, A.G. Obstructive Sleep Apnea and the Use of Cone Beam Computed Tomography in Airway Imaging: A Review. Semin. Orthod. 2009,15, 63–69. [CrossRef] 23. Lundstrøm, L.; Vester-Andersen, M.; Møller, A.M.; Charuluxananan, S.; L’Hermite, J.; Wetterslev, J. Poor prognostic value of the modified Mallampati score: A meta-analysis involving 177 088 patients. Br. J. Anaesth. 2011,107, 659–667. [CrossRef] 24. Masa, J.F.; Corral, J.; Pereira, R.; Duran-Cantolla, J.; Cabello, M.; Hernández-Blasco, L.; Monasterio, C.; Alonso, A.; Chiner, E.; Rubio, M.; et al. Effectiveness of home respiratory polygraphy for the diagnosis of sleep apnoea and hypopnoea syndrome. Thorax 2011,66, 567–573. [CrossRef] 25. Burns, S.P.; Little, J.W.; Hussey, J.; Lyman, P.; Lakshminarayanan, S. Sleep apnea syndrome in chronic spinal cord injury: Associated factors and treatment. Arch. Phys. Med. Rehabil. 2000,81, 1334–1339. [CrossRef] 26. Berlowitz, D.; Brown, D.J.; Campbell, D.A.; Pierce, R.J. A Longitudinal Evaluation of Sleep and Breathing in the First Year After Cervical Spinal Cord Injury. Arch. Phys. Med. Rehabil. 2005,86, 1193–1199. [CrossRef] 27. Leduc, B.E.; Dagher, J.H.; Mayer, P.; Bellemare, F.; Lepage, Y. Estimated Prevalence of Obstructive Sleep Apnea–Hypopnea Syndrome After Cervical Cord Injury. Arch. Phys. Med. Rehabil. 2007,88, 333–337. [CrossRef] 28. Bauman, K.A.; Kurili, A.; Schotland, H.M.; Rodriguez, G.M.; Chiodo, A.E.; Sitrin, R.G. Simplified Approach to Diagnosing Sleep-Disordered Breathing and Nocturnal Hypercapnia in Individuals With Spinal Cord Injury. Arch. Phys. Med. Rehabil. 2016, 97, 363–371. [CrossRef] 29. Klefbeck, B.; Sternhag, M.; Weinberg, J.; Levi, R.; Hultling, C.; Borg, J. Obstructive sleep apneas in relation to severity of cervical spinal cord injury. Spinal Cord 1998,36, 621–628. [CrossRef] 30. Graco, M.; McDonald, L.; Green, S.E.; Jackson, M.L.; Berlowitz, D.J. Prevalence of sleep-disordered breathing in people with tetraplegia—a systematic review and meta-analysis. Spinal Cord 2021,59, 474–484. [CrossRef] [PubMed] 31. Ahmed, M.; Patel, N.P.; Rosen, I. Portable Monitors in the Diagnosis of Obstructive Sleep Apnea. Chest 2007 ,132, 1672–1677. [CrossRef] [PubMed] 32. Castillo-Escario, Y.; Ferrer-Lluis, I.; Montserrat, J.M.; Jane, R. Entropy Analysis of Acoustic Signals Recorded With a Smartphone for Detecting Apneas and Hypopneas: A Comparison With a Commercial System for Home Sleep Apnea Diagnosis. IEEE Access 2019,7, 128224–128241. [CrossRef] 33. Ferrer-Lluis, I.; Castillo-Escario, Y.; Montserrat, J.M.; Jane, R. Analysis of Smartphone Triaxial Accelerometry for Monitoring Sleep-Disordered Breathing and Sleep Position at Home. IEEE Access 2020,8, 71231–71244. [CrossRef] 34. Ferrer-Lluis, I.; Castillo-Escario, Y.; Montserrat, J.; Jané, R. Enhanced Monitoring of Sleep Position in Sleep Apnea Patients: Smartphone Triaxial Accelerometry Compared with Video-Validated Position from Polysomnography. Sensors 2021 ,21, 3689. [CrossRef] 35. Ferrer-Lluis, I.; Castillo-Escario, Y.; Montserrat, J.; Jané, R. SleepPos App: An Automated Smartphone Application for Angle Based High Resolution Sleep Position Monitoring and Treatment. Sensors 2021,21, 4531. [CrossRef] 36. Yalciner, G.; Babademez, M.A.; Gul, F. Association of sleep time in supine position with apnea-hypopnea index as evidenced by successive polysomnography. Sleep Breath. 2016,21, 289–294. [CrossRef] 37. Oksenberg, A.; Arons, E.; Radwan, H.; Silverberg, D.S. Positional vs Nonpositional Obstructive Sleep Apnea Patients: Anthropomorphic, nocturnal polysomnographic, and multiple sleep latency test data. Chest 1997,112, 629–639. [CrossRef] 38. Roberts, T.T.; Leonard, G.R.; Cepela, D.J. Classifications In Brief: American Spinal Injury Association (ASIA) Impairment Scale. Clin. Orthop. Relat. Res. 2017,475, 1499–1504. [CrossRef] 39. Castillo-Escario, Y.; Camara, M.A.; Blanco-Almazán, D.; Jane, R. Characterization of microphones for snoring and breathing events analysis in mHealth. In Proceedings of the 2017 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Jeju, Korea, 11–15 July 2017; IEEE, 2017; pp. 1547–1550. 40. Nakano, H.; Hirayama, K.; Sadamitsu, Y.; Toshimitsu, A.; Fujita, H.; Shin, S.; Tanigawa, T. Monitoring Sound To Quantify Snoring and Sleep Apnea Severity Using a Smartphone: Proof of Concept. J. Clin. Sleep Med. 2014,10, 73–78. [CrossRef] 41. Boll, S.F. Suppression of Acoustic Noise in Speech Using Spectral Subtraction. IEEE Trans. Acoust. 1979,27, 113–120. [CrossRef] 42. Richman, J.S.; Moorman, J.R. Physiological time-series analysis using approximate entropy and sample entropy. Am. J. Physiol. Heart Circ. Physiol. 2000,278, H2039–H2049. [CrossRef] Sensors 2021,21, 7182 19 of 19 43. Lake, D.E.; Richman, J.S.; Griffin, M.P.; Moorman, J.R. Sample entropy analysis of neonatal heart rate variability. Am. J. Physiol. Regul. Integr. Comp. Physiol. 2002,283, R789–R797. [CrossRef] 44. Berry, R.B.; Brooks, R.; Gamaldo, C.E.; Harding, S.M.; Marcus, C.L.; Vaughn, B.V. The AASM manual for the scoring of sleep and associated events. Rules Terminol. Tech. Specif. 2012,176. 45. Berry, R.B.; Budhiraja, R.; Gottlieb, D.J.; Gozal, D.; Iber, C.; Kapur, V.K.; Marcus, C.L.; Mehra, R.; Parthasarathy, S.; Quan, S.F.; et al. Rules for Scoring Respiratory Events in Sleep: Update of the 2007 AASM Manual for the Scoring of Sleep and Associated Events. J. Clin. Sleep Med. 2012,8, 597–619. [CrossRef] 46. Tran, K.; Hukins, C.; Geraghty, T.; Eckert, B.; Fraser, L. Sleep-disordered breathing in spinal cord-injured patients: A short-term longitudinal study. Respirology 2010,15, 272–276. [CrossRef] 47. Flavell, H.; Marshall, R.; Thornton, A.T.; Clements, P.L.; Antic, R.; McEvoy, R.D. Hypoxia episodes during sleep in high tetraplegia. Arch. Phys. Med. Rehabil. 1992,73, 623–627. 48. Verma, M.; Seto-Poon, M.; Wheatley, J.; Amis, T.C.; Kirkness, J.P. Influence of breathing route on upper airway lining liquid surface tension in humans. J. Physiol. 2006,574, 859–866. [CrossRef] 49. Wijesuriya, N.S.; Eckert, D.J.; Jordan, A.S.; Schembri, R.; Lewis, C.; Meaklim, H.; Booker, L.; Brown, D.; Graco, M.; Berlowitz, D.J. A randomised controlled trial of nasal decongestant to treat obstructive sleep apnoea in people with cervical spinal cord injury. Spinal Cord 2019,57, 579–585. [CrossRef] 50. Zwillich, C.W.; Pickett, C.; Hanson, F.N.; Weil, J.V. Disturbed sleep and prolonged apnea during nasal obstruction in normal men. Am. Rev. Respir. Dis. 1981,124, 158–160. [CrossRef] 51. Lee, S.H.; Choi, J.H.; Shin, C.; Lee, H.M.; Kwon, S.Y.; Lee, S.H. How Does Open-Mouth Breathing Influence Upper Airway Anatomy? Laryngoscope 2007,117, 1102–1106. [CrossRef] 52. Jefferson, Y. Mouth breathing: Adverse effects on facial growth, health, academics, and behavior. Gen. Dent. 2010,58, 18–25. 53. Gainche, L.; Berlowitz, D.J.; Leguen, M.; Ruehland, W.R.; O’Donoghue, F.J.; Trinder, J.; Graco, M.; Schembri, R.; Eckert, D.J.; Rochford, P.D.; et al. Nasal Resistance Is Elevated in People with Tetraplegia and Is Reduced by Topical Sympathomimetic Administration. J. Clin. Sleep Med. 2016,12, 1487–1492. [CrossRef] 54. Braun, S.R.; Giovannoni, R.; Levin, A.B.; Harvey, R.F. Oxygen saturation during sleep in patients with spinal cord injury. Am. J. Phys. Med. 1982,61, 302–309. [CrossRef] 55. Berlowitz, D.J.; Spong, J.; Gordon, I.; Howard, M.E.; Brown, D.J. Relationships Between Objective Sleep Indices and Symptoms in a Community Sample of People With Tetraplegia. Arch. Phys. Med. Rehabil. 2012,93, 1246–1252. [CrossRef] 56. Jensen, M.P.; Hirsh, A.T.; Molton, I.R.; Bamer, A.M. Sleep problems in individuals with spinal cord injury: Frequency and age effects. Rehabil. Psychol. 2009,54, 323–331. [CrossRef] 57. Kang, Y.; Ding, H.; Zhou, H.; Wei, Z.; Liu, L.; Pan, D.; Feng, S. Epidemiology of worldwide spinal cord injury: A literature review. J. Neurorestoratology 2017,6, 1–9. [CrossRef] 58. Zimmermann, G.; Bolter, L.; Sluka, R.; Höller, Y.; Bathke, A.C.; Thomschewski, A.; Leis, S.; Lattanzi, S.; Brigo, F.; Trinka, E. Sample sizes and statistical methods in interventional studies on individuals with spinal cord injury: A systematic review. J. Evid.-Based Med. 2019,12, 200–208. [CrossRef] 59. Wijesuriya, N.S.; Lewis, C.; Butler, J.E.; Lee, B.; Jordan, A.S.; Berlowitz, D.J.; Eckert, D.J. High nasal resistance is stable over time but poorly perceived in people with tetraplegia and obstructive sleep apnoea. Respir. Physiol. Neurobiol. 2017 ,235, 27–33. [CrossRef]