Novel computational tools for enhanced diagnostics of obstructive sleep apnea

(1)

uef.fi

PUBLICATIONS OF

THE UNIVERSITY OF EASTERN FINLAND Dissertations in Forestry and Natural Sciences

Dissertations in Forestry and Natural Sciences

DISSERTATIONS | SAMI NIKKONEN | NOVEL COMPUTATIONAL TOOLS FOR ENHANCED... | No 395

PUBLICATIONS OF

THE UNIVERSITY OF EASTERN FINLAND

Obstructive sleep apnea is a highly prevalent nocturnal breathing disorder that is associated

with severe health consequences and has a detrimental effect on general well-being.

However, as current diagnostic methods are highly labor-intensive and require manual

analysis, obstructive sleep apnea is often undiagnosed. This Ph.D. thesis introduces automatic computational tools that could be

used to improve the efficiency of sleep apnea diagnostics.

SAMI NIKKONEN

Novel computational tools

for enhanced diagnostics of

(2)

(3)

NOVEL COMPUTATIONAL TOOLS FOR ENHANCED DIAGNOSTICS OF OBSTRUCTIVE

SLEEP APNEA

(4)

(5)

Sami Nikkonen

NOVEL COMPUTATIONAL TOOLS FOR ENHANCED DIAGNOSTICS OF OBSTRUCTIVE

SLEEP APNEA

Publications of the University of Eastern Finland Dissertations in Forestry and Natural Sciences

No 395

University of Eastern Finland Kuopio

2020

Academic dissertation

To be presented by permission of the Faculty of Science and Forestry for public examination in the auditorium SN201 in the Snellmania Building at the University

of Eastern Finland, Kuopio, on November, 20, 2020, at 12 o’clock

(6)

(7)

Grano Oy Kuopio, 2020

Editors: Pertti Pasanen, Raine Kortet, Jukka Tuomela, Matti Tedre

Distribution: University of Eastern Finland / Sales of publications www.uef.fi/kirjasto

ISBN: 978-952-61-3603-5 ISBN: 978-952-61-3601-1 (PDF)

ISSNL: 1798-5668 ISSN: 1798-5676 ISSN: 1798-5668 (PDF)

(8)

Author’s address: Sami Nikkonen

University of Eastern Finland Department of Applied Physics P.O. Box 1627

70211 KUOPIO, FINLAND

&

Kuopio University Hospital Diagnostic Imaging Center P.O. Box 100

70029 KUOPIO, FINLAND email: sami.nikkonen@uef.fi Supervisors: Professor Juha Töyräs

University of Eastern Finland Department of Applied Physics KUOPIO, FINLAND

email: juha.toyras@uef.fi

&

The University of Queensland

School of Information Technology and Electrical Engineering

BRISBANE, QUEENSLAND, AUSTRALIA email: j.toyras@uq.edu.au

Adjunct Professor Sami Myllymaa University of Eastern Finland Department of Applied Physics KUOPIO, FINLAND

email: sami.myllymaa@uef.fi Adjunct Professor Isaac Afara University of Eastern Finland Department of Applied Physics KUOPIO, FINLAND

email: isaac.afara@uef.fi

Adjunct Professor Timo Leppänen University of Eastern Finland Department of Applied Physics KUOPIO, FINLAND

email: timo.leppanen@uef.fi

(9)

Reviewers: Professor Pablo Laguna University of Zaragoza

Department of Electrical Engineering ZARAGOZA, SPAIN

email: laguna@unizar.es

Associate Professor Jean-Philippe Chaput University of Ottawa

Department of Pediatrics OTTAWA, CANADA email: jpchaput@cheo.on.ca

Opponent: Professor Esther Rodriquez-Villegas Imperial College London

Department of Electrical and Electronic Engineering LONDON, UNITED KINGDOM

email: e.rodriguez@imperial.ac.uk

(10)

(11)

Nikkonen, Sami

Novel computational tools for enhanced diagnostics of obstructive sleep apnea University of Eastern Finland, Kuopio, 2020

Publications of the University of Eastern Finland Dissertations in Forestry and Natural Sciences 2020; 395

ABSTRACT

Obstructive sleep apnea (OSA) is a common nocturnal breathing disorder where the upper airways collapse repetitively during sleep causing cessations in breathing. Full obstructions block all airflow and are called apneas while partial obstructions only partially limit the airflow and are called hypopneas. The diagnosis of OSA is based on daytime symptoms and the apnea-hypopnea index (AHI) which is defined as the number of respiratory events per hour of sleep. The AHI is determined with overnight polysomnography (PSG), where several physiological signals such as breathing, brain electrical activity and blood oxygenation, are recorded. The respiratory events are scored manually by reviewing the recorded signals using visual scoring rules. However, the manual scoring of each event is very time consuming and therefore it is an expensive process. Currently, the AHI calculated from the scoring is used as a full-night average and thus the variation in OSA severity during the night is ignored. The event frequency can vary significantly during the night and this variation can affect the diagnosis. In addition, there is extensive night- to-night variation in AHI. Therefore, it is generally accepted that a single night of recording is not sufficient for accurate diagnosis. However, due to the high cost of manual scoring, it is not feasible to record and score multiple consecutive nights and therefore only a single monitoring night is currently used in OSA diagnostics.

In addition to the AHI, the daytime sleepiness is used in the OSA diagnosis. The gold standard test for daytime sleepiness is the multiple sleep latency test (MSLT), where the patient’s sleep latency is measured in a sleep laboratory multiple times during a single day. However, due to the cost and complexity of MSLT, it is not commonly conducted for OSA patients and instead the daytime sleepiness is determined using subjective sleep questionnaires. However, the results of these questionnaires can vary greatly between the patients due to different personal preferences and tolerances to sleepiness. In addition, the results of these subjective sleep questionnaires correlate poorly with the objective MSLT results.

Because of these limitations in OSA diagnostics, the development of more advanced and automated methods could improve the efficiency and availability of OSA diagnostics. Machine learning methods, such as artificial neural networks (ANN) can iteratively learn features from data and use these learned features to perform intelligent tasks without the need to be explicitly programmed. They can be

(12)

applied to tasks where the inputs and desired outputs can be defined but the connection between them is unknown, or dependent on multiple complex factors.

Therefore, machine learning solutions could also be used to solve some of the issues in sleep apnea diagnostics.

The aim of this thesis was to enhance the diagnostics of OSA using novel computational methods. This was done by applying machine learning for automatic respiratory event scoring, estimation of OSA severity and prediction of objective daytime sleepiness. In addition, the intra-night variation in AHI and its effect on the diagnostics of OSA, was studied. An ANN trained to estimate the AHI of an OSA patient using only a peripheral blood oxygen saturation signal achieved excellent accuracy, with 91% of the patients being classified into the correct OSA class using the ANN estimated AHI. The median absolute error of the ANN estimated AHI was 0.78 events/hour. Excellent results were also achieved with another ANN trained to automatically score the individual apnea and hypopnea events from PSG signals. The epoch-by-epoch agreement between manual and the ANN scoring was 89%. The AHI obtained from the ANN scoring was also highly correlated with the manually determined AHI with an intra-class correlation coefficient of 0.96. In addition, when an ANN was trained to automatically estimate daytime sleepiness based on PSG signals, a moderate accuracy was reached. The ANN classified the patients as sleepy or non-sleepy, with 77% accuracy. When investigating the intra-night variation in OSA severity, the frequency and duration of the obstructive events varied significantly hour-by-hour and showed overall increasing trends towards morning.

Using only the AHI for the two hours when the obstructive event frequency was highest led to significantly different rearrangement of patients between the OSA severity classes. By using this two-hour-AHI for severity classification, more consistent relationship was found between the OSA severity and mortality compared to the standard full-night AHI.

In conclusion, by applying machine learning solutions in OSA diagnostics, some of the current limitations could be mitigated. The automatic analysis tools and methods presented in this thesis achieved high accuracy when compared to manual analysis. As the presented methods are simple and fast, they could enable more affordable and more widely available screening tools for OSA. In addition, the automatic analysis methods presented in this thesis could be used together with portable recording devices to efficiently monitor and analyze multiple consecutive nights and thus the errors attributable to the intra- and inter-night variation in OSA severity could be minimized. Therefore, an even better estimation of the true OSA severity could be reached.

(13)

National Library of Medicine Classification: W 26.55.A7, WF 143

Medical Subject Headings: Sleep Apnea, Obstructive/diagnosis; Sleepiness; Machine Learning; Neural Networks, Computer; Polysomnography; Oximetry;

Oxygen/blood

Yleinen suomalainen ontologia: unitutkimus; unihäiriöt; uniapnea-oireyhtymä;

diagnoosi; tekoäly; koneoppiminen; neuroverkot

(14)

ACKNOWLEDGEMENTS

This thesis was carried out in the Department of Applied Physics of the University of Eastern Finland and in the Diagnostic Imaging Center of the Kuopio University Hospital during 2017-2020. I want to thank these institutions for providing the tools, resources, and facilities required for the work on this thesis. I also want to thank The Research Foundation of the Pulmonary Diseases, Orion Research Foundation, Instrumentarium Science Foundation, Finnish Cultural Foundation, Academy of Finland, and the Research Committee of the Kuopio University Hospital for financially supporting the work on this thesis.

I would like to thank my supervisors, Professor Juha Töyräs, Adjunct Professor Sami Myllymaa, Adjunct Professor Timo Leppänen and Adjunct Professor Isaac Afara. This thesis would not have been possible without them. They have guided me through the scientific analyses of the original publications and given me extremely valuable feedback on the manuscripts. I also appreciate all the helpful comments they have given me for this thesis. In addition, I want to thank all of my co-authors, Esa Mervaala, Philip Terrill, Henri Korkalainen, Akseli Leino, Samu Kainulainen, Laura Kalevo, Arie Oksenberg and Brett Duce, who have participated in the original publications. I also want to acknowledge all of my colleagues who I have had the pleasure of working with. I have truly enjoyed the scientific and sometimes also non- scientific discussions we have had.

I want to sincerely thank the reviewers of this thesis, Professor Pablo Laguna and Associate Professor Jean-Philippe Chaput. A further thank you is directed towards Professor Esther Rodriguez-Villegas for agreeing to act as my opponent. I also want to thank the anonymous reviewers who agreed to peer review the original publications and provided valuable comments improving the manuscripts. This work is important and often underappreciated. In addition, I want to thank Ewen MacDonald for thoroughly proofreading the thesis.

Finally, I would like to thank my family and all of my friends who I have had the pleasure of spending time with. They have supported me and helped me to also enjoy the non-work related aspects of life. I want to especially thank my long-time girlfriend Jaana Marttinen. She has always supported me and has been part of my life not only during my doctoral studies, but throughout almost my whole adult life.

Kuopio, August 28th, 2020 Sami Nikkonen

(15)

LIST OF ABBREVIATIONS

AASM American Academy of Sleep Medicine

AF Airflow

AHI Apnea-hypopnea index

AHI-standard Standard full-night AHI

AHI-2h Highest AHI during the night in a two-hour window

AI Apnea index

ANN Artificial neural network

APAP Auto-titrating positive airway pressure

BMI Body mass index

CI Confidence interval

CNN Convolutional neural network

Conv Convolution layer

CPAP Continuous positive airway pressure

ECG Electrocardiography

EDS Excessive daytime sleepiness

EEG Electroencephalography

EMG Electromyography

EOG Electro-oculography

ESS Epworth sleepiness scale

FF Fully connected feedforward layer

HI Hypopnea index

HSAT Home sleep apnea test

ICC Intra-class correlation coefficient

ICD International Statistical Classification of Diseases

LSTM Long short term memory

MAD Mandibular advancement device

MAE Mean absolute error

MLP Multi-layer perceptron

MSE Mean squared error

MSL Mean sleep latency

MSLT Multiple sleep latency test

MaxPool Max pooling layer

ODI Oxygen desaturation index

OSA Obstructive sleep apnea

PSD Power spectral density

PSG Polysomnography

ReLU Rectified linear unit

REM Rapid eye movement sleep

RIP Respiratory inductance plethysmography

RNN Recurrent neural network

(16)

ROC Receiver operating characteristic

SAGIC Sleep Apnea Genetics International Consortium

SpO² Peripheral blood oxygen saturation

TIA Transient ischemic attack

UPPP Uvulopalatopharyngoplasty

XOR Exclusive or

(17)

(18)

LIST OF ORIGINAL PUBLICATIONS

This thesis is based on data presented in the following articles, referred to by the Roman numerals I-IV.

I Nikkonen S, Töyräs J, Mervaala E, Myllymaa S, Terrill P, Leppänen T. (2020).

Intra-night variation in apnea-hypopnea index affects diagnostics and prognostics of obstructive sleep apnea. Sleep and Breathing, 24: 379-386.

II Nikkonen S, Afara IO, Leppänen T, Töyräs J. (2019). Artificial neural network analysis of the oxygen saturation signal enables accurate diagnostics of sleep apnea. Scientific Reports, 9:1-9.

III Nikkonen S, Korkalainen H, Kainulainen S, Myllymaa S, Leino A, Kalevo L, Oksenberg A, Leppänen T, Töyräs J. (2020). Estimating daytime sleepiness with previous night electroencephalography, electrooculography, and electromyography spectrograms in patients with suspected sleep apnea using a convolutional neural network. Sleep, In press.

IV Nikkonen S, Korkalainen H, Leino A, Myllymaa S, Duce B, Leppänen T, Töyräs J. (2020). Automatic respiratory event scoring in obstructive sleep apnea using a long short-term memory neural network. Submitted for publication.

(19)

AUTHOR’S CONTRIBUTION

I The author was responsible for the data analyses, interpreted the results with the co-authors and was the main writer of the manuscript.

II The author participated in the study design with the supervisors, was responsible for the data analyses, interpreted the results with the co-authors and was the main writer of the manuscript.

III The author participated in the study design with the supervisors, was responsible for the data analyses, interpreted the results with the co-authors and was the main writer of the manuscript.

IV The author designed the study with the supervisors, was responsible for the data analyses, interpreted the results with the co-authors and was the main writer of the manuscript.

In all manuscripts, the collaboration with the co-authors has been significant.

(20)

1 INTRODUCTION

Obstructive sleep apnea (OSA) is a common nocturnal breathing disorder characterized by repetitive breathing cessations during sleep [1]. The breathing cessations are caused by obstructions in the upper airways [1]. Partial obstructions are called hypopneas and full obstructions are called apneas. The repeated obstructions cause sleep fragmentation and intermittent hypoxia which can lead to daytime sleepiness, cognitive impairment and depression [2–4]. In addition, OSA is associated with several severe health consequences such as increased risk of stroke and heart failure [1,5]. Furthermore, OSA can increase the risk of workplace or traffic accidents [6–8]. OSA could affect up to half of the adult population, large percentage of which is undiagnosed [9,10]. These factors make OSA a major global health problem with huge economic and social costs [9,11].

The gold standard test in OSA diagnostics is an in-laboratory polysomnography (PSG) where at least the patients’ breathing, brain electrical activity, heart rate and blood oxygenation are recorded overnight [12–14]. The PSG recordings are reviewed manually and apnea and hypopnea events are scored visually by sleep technicians.

However, this manual scoring is very time-consuming and therefore an expensive process. For these reasons, the PSG is only performed over a single night even though it has been shown that a single recording night is not enough for accurate OSA diagnosis due to the significant night-to-night variation in the OSA severity [15–18].

In addition, the severity of OSA is only considered as a full night average and therefore the variation in OSA severity during the night is not taken into account.

Daytime symptoms and daytime sleepiness are also considered when diagnosing OSA [19,20]. However, the gold standard objective test for daytime sleepiness, the multiple sleep latency test (MSLT), is not routinely performed for OSA patients due to its high cost. Instead, daytime sleepiness is determined using subjective sleepiness questionnaires [21,22]. Unfortunately, the sleepiness questionnaires can be unreliable and correlate poorly with the results of MSLT [23–25].

Due to the high and constantly increasing prevalence of OSA, and the limitations in OSA diagnostics, simpler and more automated diagnostic tools are needed [9,26,27]. Therefore, the aim of this thesis was to solve some of these issues in OSA diagnostics using computational machine learning methods. The field of machine learning investigates algorithms that can be trained to learn features from a dataset and consecutively use these learned features to perform intelligent tasks. In this thesis, the focus is on supervised learning applications. In supervised learning, the inputs and desired outputs are defined and the objective is to learn the features that connect the inputs to the outputs [28–30]. Artificial neural networks (ANN) are commonly used tools to solve supervised learning tasks. The idea behind an ANN is to mimic the real nervous system with neurons and connect the neurons together with artificial synapses to create a neural network [31]. The neural network is then

(24)

able to automatically learn features of the training data by adjusting these connections according to the inputs presented to it. Supervised ANN models can therefore be used to aid and greatly simplify model building as the models can be constructed without the need to program or even know all of the connections between the inputs and outputs. Therefore, machine learning methods, such ANNs can also be applied to overcome some of the shortcomings of OSA diagnostics.

In this thesis, the ability of machine learning approaches to solve some of the shortcomings in OSA diagnostics, was investigated. Machine learning was applied to automatic respiratory event scoring, estimation of OSA severity, and objective estimation of daytime sleepiness using polysomnographic signals. In addition to the machine learning approaches, the intra-night variation in the OSA severity and its effect on diagnostics and mortality risk were investigated.

The hypotheses were that neural networks can be used to reliably estimate OSA severity, to accurately detect individual respiratory events and to predict the OSA related daytime sleepiness based on PSG-signals. In addition, it was hypothesized, that the severity of OSA increases towards the morning and by considering this information, the mortality risk estimation related to OSA can be improved.

(25)

2 OBSTRUCTIVE SLEEP APNEA

2.1 DEFINITION

Obstructive sleep apnea (OSA) is a nocturnal breathing disorder where the patient suffers from repeated breathing cessations during sleep [1,12,32]. While the patient sleeps, the upper airway muscles can relax and collapse leading to an obstruction in the upper airways [1,12]. This obstruction blocks airflow causing a cessation in breathing. The breathing cessations caused by these obstruction are also called obstructive events or obstructive respiratory events. If the obstruction prevents all airflow and causes a complete cessation in breathing, it is called an apnea [1]. The obstruction can also be only partial where the airflow is not completely blocked, but only limited, in which case the event is called a hypopnea [1]. An illustration of an upper airway collapse causing an apnea event is presented in Figure 2.1.

The breathing cessations typically end when the patient arouses from sleep and muscle tension is returned allowing the breathing to continue [1,12]. These repeated arousals from sleep cause fragmentation of sleep [1]. In addition to the arousals, the apnea or hypopnea events often cause a drop in blood oxygen saturation [1]. The drops in oxygen saturation are called desaturation events. The repeated arousals and desaturations disturb sleep and cause physiological stress which often leads to excessive daytime sleepiness [33].

The severity of OSA is mainly estimated with the apnea hypopnea index (AHI) [1,19,34]. The AHI is calculated by dividing the number of apnea and hypopnea events with total sleep time, i.e. it simply tells the average number of respiratory events per hour of sleep. According to the American Academy of Sleep Medicine (AASM) guidelines, an apnea is defined as an at least 10 second long event where the breathing airflow drops at least 90% from the baseline [34]. A hypopnea is defined as an at least 10 second long event where the breathing airflow drops at least 30% from the baseline and the event is associated with an arousal or at least a 3% blood oxygen desaturation [34].

According to the Finnish Current Care Guidelines and older epidemiological studies, the prevalence of OSA is at least 2% in women and 4% in men [19,35].

However, more recent studies have shown the true prevalence is much higher and OSA can affect as much as half of the adult population [9,10]. Most individuals suffering from OSA are not aware of it and therefore cannot seek medical care [8]. In addition to the direct healthcare costs of OSA diagnosis and treatment, OSA also causes indirect costs due to the higher healthcare utilization of OSA patients [36]. The yearly direct and indirect healthcare costs of OSA are estimated to be thousands of euros per patient making it a huge social and economic burden due to the high and constantly increasing prevalence of OSA [9,11,36].

(26)

Figure 2.1: Airway during normal breathing and obstructed airway during an apnea event.

2.2 RISK FACTORS

High body mass index (BMI) is the most important risk factor for OSA [1,37–39]. At least 70% of the patients suffering from OSA are obese [39]. Multiple studies have shown that obesity is clearly linked to OSA and that more obese patients suffer from more severe OSA [40–43]. Obesity can directly obstruct the airways since excess fat can be deposited in the areas surrounding the upper airways making them more restricted [37]. Obesity has also been shown to decrease lung volume which can increase the risk for the airway to collapse [37,44]. It has also been suggested that neck circumference could be an even better and more direct estimator for OSA risk than BMI since it could be more directly linked to the upper airways restrictions [45,46].

(27)

In addition to being strongly linked to high BMI, the risk for OSA gradually increases with age and therefore OSA is much more prevalent in the elderly population [32,41,43,47]. Children can also suffer from OSA, but it is rare and very often linked to craniofacial anomalies or adenotonsillar hypertrophy [48,49].

Men have an increased risk to suffer from OSA [32,41,50]. OSA prevalence has often been reported to be around twice as high in men as compared to women [9,10,35]. It is not completely understood why the risk for men is higher but it has been suggested that the increased pharyngeal airway length and increased soft palate area could predispose men to the airway collapse [38,51,52]. However, in women, menopause increases the risk for OSA [50,52]. It has been reported that the prevalence of OSA can be up to four times higher in post-menopausal women compared to pre-menopausal women [53,54]. It has also been reported that weight gain after menopause and different body fat distribution are behind the increased OSA risk for post-menopausal women [55,56]. In addition, it has been suggested that estrogen and progesterone have a protective effect against OSA and the decrease in these hormones during menopause increases the risk for OSA [55,56]. Furthermore, post-menopausal women are naturally older than pre-menopausal women which also directly increases the OSA risk.

Smoking and increased alcohol consumption are linked to OSA [1,57]. This association can partly be explained by the fact that these unhealthy habits are often also accompanied by obesity. However, smoking has been linked to increased risk for OSA even when adjusted with BMI, age and sex [57]. In addition, smoking has a direct detrimental effect on sleep by causing airway inflammation and sleep instability which can lead to fragmentation of sleep [38,58]. In addition, cigarettes contain nicotine. The stimulating effects of nicotine might temporarily even lessen the number of obstructive events immediately after smoking, but the rebound effect and the following nicotine withdrawal may increase the severity of OSA later during the night [38,58]. Additionally, the prolonged effects of nicotine may lead to increased arousal threshold making the individual respiratory events much more severe [59]. Alcohol consumption before sleeping significantly increases the number of respiratory events during the night [60]. Alcohol has a paralyzing effect on the upper airway muscles which can cause the airways to narrow and make them more prone to collapsing and lead to an increased number of respiratory events [61]. In addition, alcohol can increase the arousal threshold causing individual events to last longer and cause much more severe episodes of intermittent hypoxemia [62].

OSA is more prevalent in patients who have suffered a stroke or a transient ischemic attack (TIA) [63–65]. Although this link is not fully understood, a stroke or TIA is a clear risk factor for developing OSA [63–65]. In addition, the sleeping position has a significant effect on the severity of OSA [66,67]. Obstructive events are more common in the supine sleeping position compared to other sleeping positions [66]. Therefore, patients who prefer the supine sleeping position are at a higher risk for OSA. Positional OSA, where the obstructive events mostly occur when sleeping

(28)

in the supine position, is very common as around half of all OSA patients are suffering from it [67–69]. In supine predominant OSA, the supine AHI is at least twice as high as the non-supine AHI and in supine isolated OSA, the obstructive events occur almost exclusively only in the supine position [70].

2.3 SYMPTOMS

The intermittent cessations in breathing during sleep are naturally the main symptom of OSA. However, the patient might not notice these breathing cessations especially in the early stages of the disease, as they only occur while sleeping.

Therefore, a large portion of people suffering from OSA do not know it themselves [8]. Even when the symptoms are noticed, rather than the patients realizing the symptoms themselves, it is often the spouse or bed partner who notices loud snoring or gasping for air during the night [12]. For these reasons, OSA is a heavily underdiagnosed disease and it has been estimated that as many as 93% of women and 82% of men suffering from moderate or severe OSA remain undiagnosed [71].

Therefore, increased awareness about OSA could help people recognize the symptoms and seek medical care.

Snoring can often be one of the earliest signs of possible OSA as it is a marker of an airflow limitation in the upper airways. Since the air cannot move freely in the airways, it becomes turbulent and the surrounding tissues, such as the soft palate, vibrate and cause the characteristic snoring sounds [72]. Snoring is clearly linked to OSA and it has been reported that over 90% of snorers might be suffering from OSA [73]. However, snoring alone still does not guarantee that the patient is also suffering from OSA.

Excessive daytime sleepiness (EDS) is a very common symptom of OSA [4,33].

The sleepiness is caused by the repeated arousals from sleep and physiological stress from the repeated hypoxemias. However, EDS is not clearly correlated with OSA severity and many OSA patients, even with severe OSA, do not suffer from EDS, or at least they do not self-report it [8]. One explanation for this could be, that, the sleepiness experienced by the patients is very subjective and thus they may report it completely differently based on their own personality and sleepiness tolerance level.

This subjectivity makes it difficult to objectively compare self-reported sleepiness levels between different people. Further complicating things, EDS is often self- reported as very vague and differing symptoms such as lack of motivation and energy [33]. Since the disease progresses slowly over time, and sleepiness increases gradually, the patients might also become accustomed to it and not realize that they are suffering from EDS [21].

The EDS caused by OSA greatly reduces the quality of life and can often lead to depression [74]. However, daytime sleepiness is also common in individuals not suffering from OSA and might even be more directly linked to depression than OSA

(29)

[75]. Nevertheless, it has been reported that 20-40% of OSA patients suffer from depression [74,76]. In addition to depression, OSA can cause cognitive impairment and increases the risk for developing dementia [3,4,77]. Furthermore, it has been reported that patients with severe OSA can suffer from reduced cognitive performance comparable to patients with multi-infarctual dementia [78]. Although EDS and the sleep fragmentation caused by the repeated arousals from sleep are speculated to be the main causes, the degree of hypoxemia has also been reported to be directly correlated with cognitive impairment [2]. The daytime sleepiness and physiological stress caused by OSA also increases the risk of traffic or workplace accidents [6–8].

OSA greatly increases the risk of cardiovascular mortality [79–81]. In addition, it has been estimated that OSA patients have a two to four times elevated risk of complex arrhythmias even after adjusting for potential confounders [5].

Furthermore, patients with OSA have a higher risk of nocturnal sudden death due to cardiac causes compared to non-OSA patients [82].

Hypertension is also very common in patients with OSA as around 40-60% of OSA patients are suffering from it [32]. It has been speculated that the association of OSA and cardiovascular mortality might be due to the hypertension and the shared risk factors for OSA and cardiovascular diseases [79,81]. However, the risk for cardiovascular mortality has been shown to be considerably higher for OSA patients even when adjusted for hypertension, age, BMI and other shared risk factors [80].

OSA patients have also been reported to have up to six times higher all-cause mortality risk even when adjusted with age, sex, BMI, smoking, cholesterol, diabetes, angina, and mean arterial pressure [81,83]. These findings suggest that OSA directly elevates the mortality risk as the cardiovascular mortality associated with OSA cannot be explained by the shared risk factors alone [81]. OSA has also been linked to an increased risk of stroke, although the causal relationship between them is not fully known [63,81].

OSA patients suffer from an elevated risk for cancer [81,84]. Shared risk factors can explain some of the cancer risk. However, the cancer risk of OSA patients has been reported to be around three times higher even when adjusted for age, sex, BMI and smoking, indicating that shared risk factors explain only part of the elevated cancer risk [81]. It has been speculated that the oxidative stress caused by the repeated hypoxemia episodes is connected to this elevated cancer risk [84].

The adverse effects of OSA can be difficult to evaluate completely since OSA rarely presents alone. Cardiovascular diseases, hypertension and mental health issues are common in OSA patients and the mortality caused by OSA can be difficult to separate from these diseases. Thus, OSA is rarely attributed as a direct cause of death although it could be a partial factor responsible for a heart attack, traffic accident or suicide. Similarly, the milder symptoms of OSA are also present in older and obese individuals not suffering from OSA. These factors could partially hide the full effects of OSA in the general population.

(30)

2.4 DIAGNOSIS

The diagnosis of OSA is based on the number of apnea and hypopnea events during the night and daytime symptoms [19,20,85]. Apnea index (AI) is calculated by dividing the number of apnea events during the night with total sleep time.

Hypopnea index (HI) is calculated by dividing the number of hypopnea events during the night by total sleep time. Apnea-hypopnea index (AHI) is the sum of AI and HI and it simply tells the average number of apneas and hypopneas per hour of sleep.

According to the AASM diagnostic criteria, an AHI of at least 5 in addition to daytime sleepiness not explained by other factors, is required for OSA diagnosis [20,85]. Choking sensations, gasping for air, or recurrent awakenings during the night in addition to AHI >5 are also sufficient for OSA diagnosis even if daytime symptoms are not present [20]. Alternatively, an AHI of over 15 is enough for diagnosis even without any other nocturnal or daytime symptoms [85].

Depending on the daytime symptoms and the AHI, the severity of OSA is further classified to mild, moderate or severe OSA classes [19,20]. The severity classification of OSA according to the Finnish Current Care guidelines and AASM diagnostic criteria is presented in Table 2.1. The severity classification is based on the most severe component [19,20]. Thus, for example, a patient with an AHI of 35 is classified as suffering from severe OSA even if the daytime symptoms would not meet the criteria for severe OSA.

Table 2.1: The severity classification of OSA [19,20].

OSA severity Daytime symptoms AHI

Mild

Sleepiness only in quiet situations when sitting still, not necessarily daily and causing only mild impairment when working or in social situations

Examples: Sitting in a car as a passenger, reading

5-15

Moderate

Daily sleepiness when in low activity situations requiring some attention causing some impairment when working or in social situations

Examples: Attending a meeting, going to a movie

16-30

Severe

Daily uncontrollable sleepiness in high activity situations requiring attention causing severe impairment while working or in social situations Examples: Eating, walking, driving a car

>30

(31)

AHI is determined by monitoring the patient’s sleep overnight. The gold standard overnight monitoring setup is an in-laboratory PSG [12]. In PSG, the patient’s breathing, heart rate, blood oxygenation, respiratory effort, body position, brain electrical activity, muscle tone, and eye movements are recorded [13]. Additionally, other signals can also be recorded depending on the specific setup [13,14]. The signals commonly recorded in modern polysomnography, the common sensors used to record these signals and the reasons for the signal inclusion are presented in Table 2.2

Table 2.2: Signals recorded in modern polysomnography, common sensor types used to record these signals and the use of the signals [13,14,26,34,86].

Recorded signal Common sensor types Used for Oro-nasal airflow Thermistor,

Nasal pressure sensor Detecting airway obstructions and scoring respiratory events

Blood oxygen

saturation Pulse oximeter Determining oxygen saturation and scoring desaturation events

Respiratory effort

RIP-belt,

Piezoelectric sensor Mercury-in-rubber strain gauge,

Determining thorax and abdomen movement and respiratory effort, differentiation of central, mixed, and obstructive apneas

Body position Accelerometer Determining sleeping position, determining possible positional sleep apnea

Audio recording Microphone Determining breathing and snoring sounds

Brain electrical

activity EEG Determining sleep stages and

arousal events

Eye movements EOG Determining eye movements for

sleep staging Muscle tone,

leg movements Chin and leg EMG Determining muscle tone for sleep staging and determining leg movements during sleep

Heart rate ECG,

Pulse oximeter

Detecting arrhythmias, not used in scoring

Video recording Video camera Determining issues in recording, checking reasons for unexpected or unclear signals

EEG=electroencephalography, ECG=electrocardiography, EMG=electromyography RIP=respiratory inductance plethysmography, EOG=electrooculography

(32)

Apnea and hypopnea events are scored from the PSG recordings by manually reviewing the recordings and marking the respiratory events [26,34]. According to the AASM guidelines for the scoring of respiratory events, an apnea event is scored when the amplitude of the airflow signal drops at least 90% from the reference level for at least 10 seconds [34]. A hypopnea event is scored when the amplitude of the airflow signal drops at least 30% from the reference level for at least 10 seconds and the event is associated with at least a 3% drop in blood oxygen saturation or an arousal from sleep [34]. Oxygen desaturation event can be scored when the blood oxygen saturation drops at least 3% from the reference level [34]. However these desaturation events are not directly used in diagnostics and instead only for hypopnea scoring. Examples of apnea and hypopnea events scored from a PSG recording are presented in Figure 2.2.

Figure 2.2: Examples of an apnea event, a hypopnea event, and oxygen desaturations caused by the respiratory events scored from a PSG recording. The apnea event is annotated with red, the hypopnea event is annotated with blue and the drops in the peripheral blood oxygen saturation (SpO²) are annotated with green.

The reference level is the average breathing airflow signal amplitude before the respiratory event [34]. The reference level and the drops in airflow are only visually

(33)

estimated from the recorded signals [34]. Thus, some borderline events might be scored differently by different scorers depending on personal interpretation.

Traditionally, the airflow has been measured using a thermistor. However, more modern devices also include a nasal pressure sensor. The nasal pressure sensor is generally more accurate when measuring nasal airflow but might not detect shallow mouth breathing. Therefore, a nasal pressure sensor is more suited for hypopnea detection while a thermistor is better suited for apnea detection as it can more accurately detect the complete cessation of the airflow. Therefore, in order to ensure high accuracy, it is recommended that both, a thermistor and a nasal pressure sensor, are used in PSG setups [86].

Pulse oximetry is used to determine the blood oxygen saturation. In addition, a pulse oximeter can also be used to measure pulse rate. Although electrocardiography (ECG) provides a more accurate measurement of heart rate, it may not be always necessary to record since the heart rate is not used in scoring of respiratory events.

Respiratory-related movements of thorax and abdomen are recorded to determine the respiratory effort during apnea and hypopnea events. If respiratory effort is present during an apnea event, the apnea is scored as obstructive, but if no effort is detected, the apnea is scored as central. Central apnea means that the breathing cessation is caused by improper breathing control by the central nervous system instead of an obstruction in the airways [87]. Apnea events that begin as central and the respiratory effort starts at some point during the apnea, are called mixed apneas.

If over half of the apnea events are central, the patient is usually diagnosed with central sleep apnea rather than obstructive sleep apnea [87]. Body position sensor (e.g. an accelerometer) is used to determine in which position the respiratory events occur and if the patient has positional sleep apnea.

In addition to the respiratory event scoring, sleep staging is performed to determine the total sleep time of the patient. Sleep stages are determined based on electroencephalography (EEG), electrooculography (EOG) and electromyography (EMG) signals [34]. The sleep stages are scored according to the AASM rules for the scoring of sleep [34]. Sleep is scored to five stages, wake, rapid eye movement sleep (REM), non rapid eye movement sleep stage one (N1), non rapid eye movement sleep stage two (N2) and non rapid eye movement sleep stage three (N3) [34]. The stages of sleep and how they are commonly characterized are presented in Table 2.3. The sleep stages are scored visually according to the patterns and waveforms of the EEG, EOG and EMG signals characterizing each sleep stage [34]. The sleep stages are scored in 30 second epochs, where one sleep stage is determined for each epoch [34].

If a single epoch contains more than one stage of sleep, the sleep stage for that epoch is scored according to the stage of which the majority of the epoch consists [34]. The AASM scoring rules are mostly based on the traditional Rechtschaffen and Kales [88]

sleep scoring rules introduced in 1968. However, there are enough differences between the rules that sleep staging is not directly comparable between the AASM rules and Rechtschaffen and Kales rules [89].

(34)

Table 2.3: Sleep stages and how they are characterized according to the American Academy of Sleep Medicine specifications [34].

Sleep

Stage Characterized by Explanation

Wake

Alpha rhythm An EEG pattern consisting of sinusoidal 8–13 Hz activity in the occipital region

Eye blinking Vertical eye movements at a frequency of 0.5–2 Hz Rapid and slow

eye movements

Irregular and sharp eye movements in addition to regular, sinusoidal eye movements

N1

LAMF activity Low-amplitude, mixed-frequency (mostly 4-7 Hz) EEG activity

Slow eye

movements Regular, sinusoidal eye movements with a deflection duration >0.5s

V waves Sharp waves with short duration (<0.5 s), in the central region

N2

K complexes Well-defined, sharp negative peak immediately followed by a sharp positive peak standing out from the background EEG, in the frontal region

Sleep spindles A train of sinusoidal waves (11–16 Hz) with a duration >0.5 s, usually in the central region

N3 Slow wave

activity Slow waves (0.5 Hz–2 Hz) with an amplitude >75 µV, in frontal regions

REM

Rapid eye

movements Irregular, sharp eye movements with a deflection duration <0.5 s

Low EMG tone EMG activity at the lowest level

Sawtooth waves Trains of triangular, serrated, 2–6 Hz waves in the central region

Transient muscle

activity Short irregular bursts of EMG activity usually with duration <0.25 s

(35)

The EEG electrodes for sleep staging are positioned using the international 10-20 system [90]. The international 10-20 system is a globally standardized method for electrode placement along which the electrodes are placed at 10% or 20% increments of the total front-back or left-right distance of the head [90]. The electrode placements according to the international 10-20 system are illustrated in Figure 2.3. AASM recommends that at least the EEG derivations F4-M1, C4-M1 and O2-M1 should be used to cover the frontal, central and occipital regions for reliable and consistent sleep staging [34]. However, using the AASM defined EEG derivations might not be possible in some recording setups due to the limited number of recorded channels and might not even be necessary as it has been shown that using other derivations, even with a lower number of channels, will only yield slightly different results in sleep scoring [91,92].

Figure 2.3: International 10-20 system for EEG electrode placement [90].

Although in-laboratory PSG is the recommended and most accurate method to diagnose OSA, it has certain shortcomings. The complex recording setup with multiple sensors and electrodes make it an expensive study [12]. In addition, a sleep laboratory staffed with professional sleep technicians is required to conduct the PSG [12]. In-laboratory PSG also suffers from a first night effect, where the unfamiliar environment and PSG equipment cause discomfort and stress [16,18]. This affects the sleep quality during the PSG recording thus may not be representative of a normal

(36)

night [16,18]. The first night effect is very difficult to reduce even if a more familiar environment such as a hotel is used [18]. The first night effect could be reduced by recording multiple consecutive nights but it is unfeasible with the current costs of in- laboratory PSG [18].

For these reasons, simpler home sleep apnea tests (HSAT) are also used instead of full in-laboratory PSG in OSA diagnostics [86]. As the name implies, HSAT devices are ambulatory devices designed to be used at home and therefore eliminate the sleep laboratory requirement. This makes sleep studies conducted with HSAT devices cheaper than in-laboratory PSGs [93]. However, HSATs also have their own limitations [93,94]. The devices generally cannot be used fully independently and thus still require at least some training and setup by a professional. In addition, mistakes made by the patients and incorrect use of the devices in unattended conditions leads to a greater need for retesting mitigating some of the cost benefits of the HSATs [93–95]. Ambulatory HSAT devices also suffer from a higher rate of data loss which can lead to inconclusive results [96].

Although ambulatory HSATs can include the same signals as full in-laboratory PSGs, most HSAT devices are not as comprehensive and sacrifice some sensors for simplicity and cost. The Task Force of the Standards of Practice Committee of the American Sleep Disorder Association has defined four monitor types for sleep recording [13,97]. The device types and the requirements for each category are presented in Table 2.4.

Type I devices have the strictest requirements while type IV devices have the most lenient requirements. A sleep study performed with a type II device is called a level II study. The requirements presented in Table 2.4 are the minimum requirements and all types of devices can be more comprehensive and can record more signals than the minimum requirement specifies. For example, type I and type II devices generally record more than the one required EEG channel and although either only thermistor or nasal pressure sensors are required for airflow recording in level I-III studies, both are usually recorded. In addition to these rules, it is required that all portable sleep monitoring devices must be able to record a minimum of six hours and that raw data that must be later accessible [13,97].

The type I device is a standard attended in-lab PSG and is the recommended option for OSA diagnostics [13,86]. Type II device is an ambulatory full PSG device and is also a recommended option for OSA diagnostics [13,86]. Type III device is an unattended polygraphy device currently accepted to be used for OSA diagnostics [86]. Type IV device is required to record only a single channel and is not acceptable to be used for the diagnosis of OSA [97,98].

Originally, when the device types were defined in 1994, type III devices were not deemed suitable to be used for OSA diagnostics [13,97]. However, in 2007, AASM revised the recommendations and accepted the use of type III portable monitors for OSA diagnostics [86]. However, the use of type III portable monitors is not allowed, if other sleep disorders are suspected [86].

(37)

Table 2.4: Sleep monitor types as defined by the Task Force of the Standards of Practice Committee of the American Sleep Disorder Association [13,97,98]

Type I Type II Type III Type IV

Description Full, attended, in-laboratory PSG

Full, unattended PSG

Unattended

polygraphy Unattended recording

Use

Various sleep disorder diagnostics including sleep apnea

Sleep apnea diagnostics

Monitoring only, not sufficient for sleep apnea diagnostics Minimum

number of recorded signals

8* 7 4 1

Required signals

EEG, EOG, chin EMG, ECG, airflow, respiratory effort,

body position, oxygen saturation

EEG, EOG, chin EMG, heart rate, airflow, respiratory effort oxygen saturation

respiratory effort, airflow, heart rate, oxygen saturation

respiratory effort or airflow or oxygen saturation

* 7 if body position determined manually

Since type III devices were accepted for OSA diagnosis, their use has increased due to their sufficient accuracy, cost-effectiveness and the simplicity of the devices [99]. A level III study is simpler to conduct compared to full PSG since no electrode attachments are required. It is also faster to score since sleep staging is not performed due to the absence of EEG recording. However, omitting the EEG recording leads to an underestimation of AHI since hypopneas associated only with an arousal cannot be detected from the recording [100]. In addition, without EEG, the sleep stages

(38)

cannot be scored and therefore the sleep efficiency and total sleep time of the patient will remain unknown. Total recording time is used instead of the total sleep time when calculating the AHI which leads to further underestimation of AHI [100].

This underestimation of the AHI might mean that some patients with mild or hypopnea dominant OSA might not be diagnosed correctly. It has been proposed that the current AHI thresholds for the severity of OSA are not optimal for type III polygraphy from a mortality risk perspective and that the thresholds should be lowered [101]. It has also been suggested that different AHI thresholds should be used for level III studies and level I studies [102]. Furthermore, as the current AHI thresholds are arbitrary and not optimized even for level I studies, a complete revision of the thresholds may be in order [102].

In 2007, AASM published detailed and unified rules and instructions for the scoring of sleep and respiratory events [103]. However, in this publication, the hypopnea scoring rule stated that the airflow must drop at least 30% from the reference level for at least 10 seconds causing at least a 4% drop in blood oxygen saturation or an arousal from sleep [103]. In 2012, AASM announced an update that changed the desaturation threshold in hypopnea scoring from the 4% drop to the current 3% drop [104]. Using the 3% hypopnea rule instead of the 4% rule increases the AHI significantly [105,106]. However, none of the OSA severity thresholds have been altered with this change further indicating the need to revise the thresholds.

Although AHI is used very widely around the world and is the primary diagnostic parameter in the estimation of OSA severity, it has several major limitations. The AHI does not take into account the respiratory event types or their durations. A ten-second hypopnea event can have a very different physiological effect than an apnea event lasting over a minute. However, they are counted as equal when calculating the AHI.

Furthermore, the AHI is calculated as a full night average and thus it does not take into account the variation in the respiratory event frequency during the night.

The patient might experience a vastly different number of respiratory events at different periods during the night. Therefore, by averaging the AHI over the whole duration of the night, patients that suffer from respiratory events only for a small part of the night, might not be diagnosed as having OSA.

In addition to the intra-night variation in the AHI, multiple studies have reported a significant night-to-night variation in the AHI [15–17]. It has been reported that for 50% of the patients, the OSA severity classification changed from the first night to the subsequent nights when multiple consecutive nights were monitored [15].

Therefore, it is generally acknowledged that a single night of recording might not be sufficient for OSA diagnostics [16,17]. Nevertheless, due to the cost of recording and scoring multiple nights, only a single monitoring night is currently used in clinical sleep medicine [18]. Therefore, it is evident that simpler, cheaper, and more automated screening devices and diagnostic methods are needed in OSA diagnostics [9,26,27].

(39)

Since the respiratory events and sleep stages are scored manually there is variation between scorers due to different interpretation of the signals. There are conflicting reports on the inter-scorer agreement between sleep laboratories [107–

109]. Previously, it has been reported that the variation is extremely large and that the same patient could be diagnosed as healthy in one laboratory, and with severe OSA in another laboratory [109]. However, more recently, better inter-scorer agreement has been reported [107,108]. In recent years, the AASM inter-scorer variability program has comprehensively evaluated the agreement between scorers for sleep staging and respiratory event scoring in large populations [110,111]. In sleep staging, the agreement has been generally found to be good, although the agreement for N1 and N2 has been relatively low which is to be expected since these stages are often difficult to separate [111]. However, the inconsistency in N1 and N2 sleep stages has only a limited impact on the diagnostics of OSA, since only the total sleep time is used when calculating the AHI. Agreement in respiratory event scoring, which is much more critical for OSA diagnostics than sleep staging, has been found to be excellent [110]. It could be that since AASM has published detailed scoring manuals, the variation in scoring has reduced when the scoring habits have unified and sleep laboratories have had time to adjust to the rules. Still, inter-scorer variation is always present in OSA diagnostics as long as scoring is based on visual analysis of signals.

In addition to the AHI, symptoms of daytime sleepiness are used when diagnosing OSA [19,20]. The gold standard test for EDS is the multiple sleep latency test (MSLT) [112]. MSLT measures the tendency to fall asleep in quiet and peaceful situations during the day [113]. In MSLT, the patient’s daytime sleep latency is measured in a sleep laboratory multiple times during a single day using an EEG recording setup similar to attended PSG [112]. In addition, standard in-laboratory PSG is always performed the previous night before the MSLT [112]. The patient’s sleep latency is usually measured 4-5 times. A single sleep latency measurement is called a nap attempt, which lasts 20 minutes [112]. The patient is not supposed to try to stay awake during the nap attempt and instead should fall asleep if it happens naturally [112]. The patient’s sleep stages are monitored during the nap attempt and the sleep latency is scored from the first stage of sleep [112]. If the patient does not fall asleep within the 20 minutes, that nap attempt is terminated and the sleep latency for that nap attempt is set to be 20 minutes [112]. If the patient falls asleep, the sleep is recorded for 15 minutes to detect if REM sleep occurs [112]. The sleep latency is at minimum 0 minutes which would mean that the patient falls asleep immediately and at maximum 20 minutes which means that the patient did not fall asleep during that nap attempt. There is a two-hour period between the nap attempts [112]. Medications and caffeine should not be used before or during the test [112]. The average of the sleep latencies for each nap attempt is the mean sleep latency (MSL). The patient can be classified into four sleepiness categories based on their MSL: Severe (MSL < 5 min), moderate (5 min MSL < 10 min), mild (10 min MSL < 15 min) and normal (MSL

(40)

15 min) [114,115]. Alternatively, a single MSL threshold of 8 or 10 minutes is often used to differentiate between normal patients and patients suffering from EDS [85,116].

However, due to the cost and complexity of the MSLT, it is not commonly conducted for patients with suspected OSA. MSLT was originally developed for the diagnosis of narcolepsy and is currently mostly used when diagnosing sleep disorders other than OSA [113,117]. Instead, in OSA diagnostics, the daytime sleepiness is usually determined using questionnaires [113]. The most common sleep questionnaire is the Epworth Sleepiness Scale (ESS) [21,22]. ESS describes eight common situations and the patient estimates their chance of dozing or falling asleep in these situations with a score of 0-3 [22]. A higher score means that the patient is more likely to fall asleep [22]. The ESS score is calculated by summing the scores from each situation making the range of ESS score 0-24. The patients are classified to sleepiness categories according to the ESS score results [22]. A score of 0-10 is considered to be within the normal limits and patients with a score >10 are considered to be suffering from daytime sleepiness [118]. ESS is a popular test due to its simplicity and is used very widely around the world and has been translated into numerous languages [21].

However, the ESS has certain shortcomings. Being a questionnaire, the ESS is limited by the patient’s ability to comprehend the questions and to answer them honestly [119]. In addition, the ESS is a subjective test and the results can vary greatly between patients depending on their personal feelings and tolerance to sleepiness [119]. It has been reported multiple times that the ESS correlates rather poorly with MSLT and that it is not sufficient for the evaluation of objective sleepiness [23–25].

Since the MSLT is not performed as a part of routine OSA diagnostics, objective sleepiness estimation for all OSA patients cannot currently be obtained. It is important to note however, that even though the subjective estimation of sleepiness can differ from the MSLT based estimation, it does not mean that the subjective estimation is completely wrong as the effects of daytime sleepiness are largely psychological and if the patient feels excessive sleepiness during daytime, it is a cause for concern even if this finding could not be confirmed with MSLT. Therefore, even though the MSLT is a more accurate test for EDS, the benefits and costs of MSLT should be carefully considered against the limitations of ESS in OSA diagnostics [23,119].

2.5 TREATMENT

The most common treatment option for OSA is treatment with a continuous positive airway pressure (CPAP) device [120,121]. The CPAP device provides constant positive pressure to the patient’s upper airways by blowing air via a facial mask. This

(41)

constant positive pressure from the CPAP device prevents the airways from collapsing, and therefore prevents the respiratory events from occurring.

Treatment with CPAP is considered extremely effective and generally completely eliminates all, or nearly all, apnea and hypopnea events when the device is in use [120,121]. If CPAP is used for at least the recommended six hours per night, it reduces sleepiness and improves the daytime symptoms of OSA [120]. CPAP has been shown to normalize sleep architecture, reduce daytime sleepiness, improve cognitive function, restore memory loss, elevate the general mental state and reduce the risk for accidents [120,122,123]. It has also been reported that cardiovascular outcomes are improved with CPAP use and the risk for lethal cardiovascular events is lower for patients using CPAP [120,124,125]. However, conflicting findings have also been reported which indicate that the use of CPAP does not significantly reduce the risk for cardiovascular events [126]. CPAP is generally only provided for patients with moderate or severe OSA as it is thought that patients with mild OSA can be managed without CPAP [127].

In spite of all these benefits, adherence to CPAP is relatively low. It has been reported that only around 20-60% of the patients provided with CPAP are adherent to the treatment when adherence is defined as greater than four hours of nightly CPAP use [122]. Adherence to the treatment does not seem to be correlated with OSA severity as even patients with severe OSA often do not use the device [128]. Poor adherence is the greatest limitation of CPAP. The most common reason for low adherence is the discomfort of the device as many patients find it very difficult to sleep wearing the CPAP mask and thus decide not to use it [122]. The constant pressurized airflow also can cause dryness in the mouth and throat which also causes discomfort. In addition, the CPAP device produces a weak noise which can disturb the sleep of some patients. However, if the device is patiently used for a few days, the patients usually become completely accustomed to it [122]. Therefore, it would be imperative that the patient does not give up too quickly as once the patients get used to the CPAP device and notice the improvement of symptoms, they are much more likely to be adherent to the treatment [122].

Other more sophisticated air pressure treatment devices are also available, such as the auto-titrating positive airway pressure (APAP) device [129,130]. In APAP devices, the pressure provided by the device is not constant and instead varies based on the patient’s needs. The device increases the pressure as necessary during an obstructive event and gradually decreases the pressure when no events are detected therefore maintaining only a very low pressure during normal breathing [129]. In addition, APAP is able to automatically adjust to the changing conditions of the patient, such as weight gain, and therefore does not need to be manually adjusted over time [130]. However, there is little evidence indicating that APAP holds any advantage over CPAP in general OSA treatment efficiency although these devices may be more comfortable to use [130]. Additionally, there does not seem to be a clear preference for the patients between CPAP and APAP devices when the patients are

(42)

fully used to them [131]. Therefore, as APAP devices are more expensive than the simpler CPAP devices, they remain more rarely used.

Lifestyle changes can also be used as a treatment for OSA [39]. The lifestyle changes can include weight loss, a healthier diet, quitting smoking and reduction in alcohol consumption. Even a simple intervention and counseling session can be an effective treatment for some mild OSA patients but in other cases, a more strict and structured weight reduction program may be needed [132]. Even a minor weight loss can lower the severity of OSA significantly, although some patients do not benefit as much as others [39,132]. Weight loss can also be achieved with bariatric surgery [39].

Alcohol consumption before sleeping increases AHI by paralyzing the upper airway muscles and allowing them to collapse more easily [60,61]. In addition, alcohol increases the arousal threshold making the individual respiratory events more severe and causing greater hypoxemias [62]. Therefore, limiting alcohol use may be beneficial for OSA patients. Although, it has been shown that alcohol consumption when using CPAP may not be as detrimental and may not increase AHI [133].

In addition to lowering the AHI and therefore reducing the severity of OSA, lifestyle changes and weight loss also improve the patient’s general health. Weight loss and reduction in alcohol consumption have a clear positive effect on cardiovascular diseases, hypertension and type 2 diabetes, which are all common comorbidities of OSA [39]. Therefore, even though the weight loss alone may not completely eliminate the disease especially for patients suffering from severe OSA, weight loss is always worth the effort and should always be considered as a part of the treatment [39].

Positional therapy can help patients that suffer from positional OSA where most of the obstructive events occur in the supine position [67,70,134]. Since the supine position predisposes the patient to more obstructive events, avoiding the position will improve the severity of OSA. The supine position can be avoided for example by using an alarm system that detects the sleeping position of the patient and if a supine position is detected, forces them to change position [67]. Alternatively, a simpler positional therapy method is to strap a tennis ball to the patients back which prevents comfortable sleeping in the supine position [67]. Positional therapy can be an effective treatment for many individuals since around half of all OSA patients are suffering from positional OSA [67–69].

Positional therapy also has some shortcomings however. It suffers from poor adherence as patients generally find the tennis ball method uncomfortable and therefore often do not wish to use it [134]. In addition, some patients become used to the alarm systems or the tennis ball and can eventually sleep in the supine position regardless of these preventative methods. Furthermore, positional therapy usually does not completely prevent obstructive events as the events can also happen in other sleeping positions. In addition, patients that have non-positional OSA, naturally gain

(43)

little benefit. Positional therapy is therefore generally considered less efficient than CPAP [135].

OSA can also be treated with special oral appliances or with surgery [136–139].

Mandibular advancement devices (MAD) alter the upper airway geometry and increase their size by moving the tongue and fat tissues in the throat thereby reducing the chance of airway collapse [138,139]. These devices are not as effective as CPAP and can still cause discomfort to the patient. The surgical operations also focus on reducing the likelihood of the airway collapse by altering the upper airway structure directly [136,137]. One common form of surgery is uvulopalatopharyngoplasty (UPPP), where the airspace in the oropharynx is enlarged [136,137]. Tonsillectomy can also be effective in the treatment of OSA and is usually performed in conjunction with UPPP [136]. Success rates of 35-95% have been reported with UPPP, resulting in an over 50% reduction in the AHI [137].

As surgical treatment is naturally invasive, it is generally not used as the first treatment option although the risk for serious complications is relatively low [140].

CPAP does not have the complication risks associated with surgery but has a monetary cost although it is much lower than the cost of surgery [121]. Lifestyle changes and positional therapy have virtually no risk and very low costs and can also be used in conjunction with any other treatment option. A summary of the treatment options for OSA and their advantages and disadvantages is presented in Table 2.5.

Table 2.5: Summary of the most common treatment options for OSA and their advantages and disadvantages.

Treatment option Advantages Disadvantages

Lifestyle changes [39,132]

Virtually free,

Improves quality of life regardless of OSA

Does not completely eliminate OSA,

Requires willpower and motivation

CPAP

[120–122] Eliminates virtually all

respiratory events, Cost,

Poor adherence Positional therapy

[67,134] Cheap

Can cause discomfort, Only effective when treating positional OSA

Oral appliances [138,139]

My be used if patient is intolerant to CPAP

Can cause discomfort, Poor effectiveness Surgery

[136,137,140] Treats OSA without the need to use any devices

Cost,

Complication risk,

Unsuitable for some patients

Novel computational tools for enhanced diagnostics of obstructive sleep apnea

uef.fi

Dissertations in Forestry and Natural Sciences

SAMI NIKKONEN

SAMI NIKKONEN

Novel computational tools

for enhanced diagnostics of

NOVEL COMPUTATIONAL TOOLS FOR ENHANCED DIAGNOSTICS OF OBSTRUCTIVE

SLEEP APNEA

Sami Nikkonen

NOVEL COMPUTATIONAL TOOLS FOR ENHANCED DIAGNOSTICS OF OBSTRUCTIVE

SLEEP APNEA

ABSTRACT

ACKNOWLEDGEMENTS

LIST OF ABBREVIATIONS

LIST OF ORIGINAL PUBLICATIONS

AUTHOR’S CONTRIBUTION

CONTENTS

1 INTRODUCTION

2 OBSTRUCTIVE SLEEP APNEA